Next Article in Journal
Label GM-PHD Filter Based on Threshold Separation Clustering
Previous Article in Journal
Learning to Combine Local and Global Image Information for Contactless Palmprint Recognition
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

A Novel Method for Source Tracking of Chemical Gas Leakage: Outlier Mutation Optimization Algorithm

1
Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China
2
School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
*
Author to whom correspondence should be addressed.
Sensors 2022, 22(1), 71; https://doi.org/10.3390/s22010071
Submission received: 15 November 2021 / Revised: 10 December 2021 / Accepted: 15 December 2021 / Published: 23 December 2021
(This article belongs to the Section Physical Sensors)

Abstract

:
Chemical industrial parks, which act as critical infrastructures in many cities, need to be responsive to chemical gas leakage accidents. Once a chemical gas leakage accident occurs, risks of poisoning, fire, and explosion will follow. In order to meet the primary emergency response demands in chemical gas leakage accidents, source tracking technology of chemical gas leakage has been proposed and evolved. This paper proposes a novel method, Outlier Mutation Optimization (OMO) algorithm, aimed to quickly and accurately track the source of chemical gas leakage. The OMO algorithm introduces a random walk exploration mode and, based on Swarm Intelligence (SI), increases the probability of individual mutation. Compared with other optimization algorithms, the OMO algorithm has the advantages of a wider exploration range and more convergence modes. In the algorithm test session, a series of chemical gas leakage accident application examples with random parameters are first assumed based on the Gaussian plume model; next, the qualitative experiments and analysis of the OMO algorithm are conducted, based on the application example. The test results show that the OMO algorithm with default parameters has superior comprehensive performance, including the extremely high average calculation accuracy: the optimal value, which represents the error between the final objective function value obtained by the optimization algorithm and the ideal value, reaches 2.464e-15 when the number of sensors is 16; 2.356e-13 when the number of sensors is 9; and 5.694e-23 when the number of sensors is 4. There is a satisfactory calculation time: 12.743 s/50 times when the number of sensors is 16; 10.304 s/50 times when the number of sensors is 9; and 8.644 s/50 times when the number of sensors is 4. The analysis of the OMO algorithm’s characteristic parameters proves the flexibility and robustness of this method. In addition, compared with other algorithms, the OMO algorithm can obtain an excellent leakage source tracing result in the application examples of 16, 9 and 4 sensors, and the accuracy exceeds the direct search algorithm, evolutionary algorithm, and other swarm intelligence algorithms.

1. Introduction

The world’s rapid economic development, urbanization, and industrialization make people increasingly dependent on the chemical industry, machinery industry, and electronics industry, etc. However, the storage of raw materials, the production of spare parts, the assembly of machines, and the testing of technology often need to be carried out in specific places. Among the above industrial processes, the chemical industrial park plays an important role in the storage of raw materials and chemical reaction tests [1,2,3,4]. In chemical industrial parks, there are a large number of chemicals, many of which are flammable, explosive, and hazardous gases [5]. Poisonous gas leakage accidents are challenges for various industries like the petrochemical and gas industries, which will result in financial losses and possibly human injuries [6]. In severe cases, large-scale poisoning disasters or internal domino chain explosion effects may even occur [7].
Toxic gas leakage accidents can be very dangerous to health and the environment. For instance, in December 2018, over 40 workers of a factory in Gujarat’s Valsad district were hospitalized following leakage of chlorine gas; in March 2017, about 450 girl students were hospitalized in southeast Delhi’s Tughlaqabad area following leakage of ‘chloromethyl pyridine’ chemical [8]. Inhaling an excessive amount of toxic gas will cause irreversible harm or even death. Humans have different tolerances for different toxic gases. Table 1 shows the Acute Exposure Guidelines Levels (AEGLs) standard of three kinds of typical toxic gases when supposed the exposure duration is 1 h [9,10,11,12].
The unit of AEGL standard is parts per million (ppm), NR means not recommended due to insufficient data. For the general population, AEGL-1’s airborne concentration will cause notable discomfort but transient and reversible, AEGL-2′s airborne concentration will cause irreversible health effects and even disability, and AEGL-3’s airborne concentration will cause threatening health effects or death [9]. It can be seen from Table 1 that different toxic gases have different concentration threshold standards; even small-scale leak accidents may have irreversible hazards.
To minimize harm to humans and avoid further accidents, it is significant to quickly and accurately get information about the leak point in the chemical leakage accident. Tracking the location of the chemical gas leak source and calculating the average leak rate of the chemical gas leak source are the most important tasks in obtaining the leak point information, which is directly related to the safety and accuracy of the fixed-point elimination work of the emergency safety department [6,13,14]. Using a gas dispersion model to match the actual gas sensor concentration distribution is the mainstream for quickly and accurately getting information about the leak point.
In recent years, many studies have focused on the problem of source tracking of chemical leakage. Singh et al. (2015) used a least-squares approach to inverse modeling, aiming to retrieve multiple point releases in the atmosphere [15,16]. Wang et al. (2018) used a hybrid genetic algorithm to estimate the source of hazardous material releases [17]. Mario et al. (2018) used a Computational Fluid Dynamics (CFD) model to simulate a large amount of gas leak accident data and proposed deep neural network models as a solution to the problem of leak-source tracking [18]. Omar et al. (2020) used conditional bivariate probability function and series signature to realize the spatial identification and temporal prediction of air pollution sources [19]. Huang et al. (2021) compared similarities between source profiles and ambient air profiles, aiming to improve the tracing efficiency [14]. Yet-Pole et al. (2021) combined the benefits of both the CFD model and Gaussian dispersion model, aiming to provide a model basis for the leak of hazardous materials, etc. [4]. The above-mentioned studies mainly include three aspects of such research: the development of the gas dispersion model, the optimization and improvement of the gas dispersion model, and the combination of the gas dispersion model and an external algorithm. Therefore, the gas dispersion model is the basic requirement for the traceability of chemical gas leakage.
There are five mainstream gas dispersion models: Gaussian plume model, Gaussian puff model, Britter and McQuaid (BM) model, Sutton model, and 3-D Finite Element (FEM3) model. The Gaussian plume model and Gaussian puff model are suitable for a continuous leak accident and instantaneous leak accident, respectively. The advantages of Gaussian models are their applicability to various gases, unlimited size of application scenarios, and appropriate model complexity. In addition, Gaussian models’ relatively high consistency between the calculation results and the experimental values make them widely used [4,20,21,22,23,24]. The BM model is based on a series of test data of instantaneous leak and continuous leak of heavy gas, combined by drawing calculation charts; the empirical model has a narrow application range and poor extension [25]. The Sutton model uses the statistical theory of turbulence diffusion to simulate the leak and diffusion process of gas, and there will be large errors [26]. The FEM3 model uses the finite element method to solve the gas leak process-related equations, such as the ideal gas state equation, Navier–Stokes equation, etc. The model has developed rapidly in recent years; the biggest advantage is that it can accurately solve the gas diffusion problem under complex terrain conditions. The shortcomings are detailed meshing, complex solution process, and long calculation time [4,5,27,28,29].
The combination of a gas dispersion model and optimization algorithm has the advantage of various combinations, simple principles, and various applicable scenarios, and is thus widely applied in meeting the demands of source tracking. Table 2 shows the classification and characteristics of some commonly used methods for source tracking of gas leakage.
It can be concluded from Table 2 that considering the economic level, prior technology level and calculation time, the mainstream solution to the problem of gas leakage traceability still relies on the combination of gas dispersion models and optimization algorithms. However, the accuracy of these methods has not reached a high level, which may affect the efficiency of emergency response. For instance, the accuracy of the least-squares method under the one-dimensional gas leakage model is only about 95% [15,16], the accuracy of the DNN under 25 hidden layers is only 75.43% [18], and the approximate errors of PSO, ACO and FA algorithms under the three-dimensional gas leakage models are (Δx, Δy, ΔQ) = (43.97%, 49.43%, 19.75%) [30].
In order to trace the source of the gas leak in a satisfactory time and with as high accuracy as possible in chemical gas leakage accidents under flat terrain conditions, this article has made the following innovations:
  • Traditional gas leakage accident application examples usually take the position and number of the storage tank or environmental variables as known conditions. In order to illustrate that the OMO algorithm has the best comprehensive performance in the gas leakage traceability problem under flat terrain conditions, randomness is introduced to the various variables that will affect the calculation results of the Gaussian plume model, which can make the application examples created in this article to be strongly representative.
  • The introduction of sensor information defect conditions: the number of sensors in the three types of application examples is 16, 9, 4, respectively, and the intensity of the information defect increases in turn, which is a challenge to the exploration integrity of the optimization algorithm, but it is also an important part to prove the advantages of OMO algorithm.
  • The OMO algorithm is different from other swarm intelligence algorithms. The exploration phase of the OMO algorithm is dominated by outliers, and the exploration method introduces the Levi flight, which aims to avoid the local optimum with a high degree of exploration. In the exploitation phase of the OMO algorithm, two-way selection improves the convergence speed and accuracy. The complementarity of the encirclement and mutation strategies has become a major feature of the algorithm.
  • The research on the parameter law of the OMO algorithm and the introduction of precision control parameters enable the OMO algorithm to adjust parameters reasonably in practical applications and quickly adapt to various application scenarios.
The following content is divided into four sections: methodology, experimental design, results and discussion, conclusions.

2. Methodology

2.1. Gaussian Plume Model

Last century, the Gaussian model was proposed and applied to the diffusion modeling of actual point sources, including the Gaussian plume model and Gaussian puff model [48]. The Gaussian puff model is suitable for the instantaneous gas leak diffusion from point sources, while the Gaussian plume model is suitable for the continuous gas leak diffusion from point sources [49]. In this paper, the location and intensity of the leak source will be inversely calculated based on the Gaussian plume model.
After the leaking gas is fully mixed with air, the diffusion of the leaking gas is mainly affected by air turbulence, forming passive diffusion. When the turbulent field is stable, the spatial concentration of the leaking gas presents an approximately Gaussian distribution of diffusion. The Gaussian plume model is widely used in the physical movement of gas particles [50,51], and it has been proven to have excellent robustness in the case of small-scale gas point source continuous leak accidents [52]. In contrast, some authors have pointed out that although the Gaussian plume model can better reflect the physical characteristics of gas leak diffusion particles, it cannot handle unpredictable and complex changes, such as a gas leak in extreme climates [53].
Therefore, the Gaussian plume model has the following assumptions [54]:
  • The concentration of leaking gas conforms to Gaussian distribution in the y-axis section and the z-axis section;
  • The leakage intensity of the leaking gas is continuous and uniform;
  • The ambient wind speed remains constant during the gas leak process, and the default direction is along the positive x-axis;
  • The model ignores the effects of any chemical reactions, including sedimentation and decomposition;
  • The leaked gas follows the ideal gas equation of state and the law of conservation of mass.
When the leak point source is at the origin of the coordinate system, in an infinite space, the expression of the concentration of the leaking gas at a certain position (x, y, z) downstream of the leak source is as follows [55]:
C ( x , y , z ) = Q 2 π u σ y σ z e 1 2 y 2 σ y 2 + z 2 σ z 2
where C(x, y, z) is the concentration of the leaking gas at the certain position (x, y, z), unit kg/m3; Q is the leakage intensity of leak point source (leak rate), unit kg/s; u is the average wind speed of the environment, unit m/s; σy and σz respectively represent the standard deviation of the diffusion of the leaked gas along with the y and z directions, unit m.
Assuming that the coordinate of the gas leak point source is (x0, y0, h0), and considering the total reflection of the ground surface, the concentration distribution calculation formula of the Gaussian plume model is finally obtained by using the mirror image method:
C ( x , y , z ) = Q 2 π u σ y σ z e y y 0 2 2 σ y 2 e z h 0 2 2 σ z 2 + e z + h 0 2 2 σ z 2
where all coordinate units in the formula are m.
According to the wind speed and solar radiation intensity, the atmospheric stability can be divided into six different grades A–F, as shown in Table 3 and Table 4 [56,57]. σy and σz are calculated by the empirical formula of the Pasquill–Gifford model [58], which has a certain correlation with the atmospheric stability level. Their calculation expressions are shown in Table 5 (when the leak source position is at the origin of the coordinates).

2.2. Outlier Mutation Optimization (OMO) Algorithm

The OMO algorithm is inspired by animal group hunting behavior and evolutionary algorithms, constructed and improved based on human computing habits. The OMO algorithm aims to overcome the shortcomings of inconsistency between the exploration phase and the exploitation phase of most SI algorithms, and reduce the probability of falling into local optimal solutions, finally improving the accuracy of the solution.
The OMO algorithm is mainly divided into two phases: exploration phase and exploitation phase. The phases are alternated according to the escape energy E of the prey:
E = 2 E 0 1 t T
where E is the escape energy of the prey, E0 ~ U(−1, 1), t is the current number of iterations, T is the total number of iterations set.
Figure 1 shows the relationship between the escape energy and the number of iterations in the OMO algorithm. According to the relationship between |E| and 1, the OMO algorithm divides the algorithm into the exploration phase and the exploitation phase. To appropriately increase the step length of the exploration phase and prevent excessive jumps in the exploitation phase, a parameter Epara is introduced:
E p a r a = 2 E 2
In order to adapt to the multi-dimensional exploration interval with different proportions, the step parameters α and β of Levy flight and Brownian motion are introduced respectively:
α = U B L B k α S H
β = ln U B L B k β S R
where kα and kβ are constants, value 100 in general, SH is the speed control constant of predator’s Levy flight, which can be adjusted according to the actual situation, default value 1, SR is the speed control constant of prey’s standard Brownian motion, which can be adjusted according to the actual situation, default value 1, LB and UB represent the lower limit and upper limit of the exploration area, respectively.
Standard Brownian motion and Levy flight belong to the Markov process. Figure 2 shows a schematic diagram of the trajectory of 500 steps’ Levy flights and 500 steps’ Brownian motions in a two-dimensional space when the step parameter is 1. Levy flight has the advantage of the wide exploration area, while Brownian motion has the advantages of small space and high density. Combining Levy flight and Brownian motion will enable the OMO algorithm to have a higher degree of exploration and more accurate convergence results.
If the continuous stochastic process {B(t), t ≥ 0} defined in the time domain t satisfies the following three properties, it can be called one-dimensional standard Brownian motion (or Wiener process) [59,60]:
  • B ( 0 ) = 0 ;
  • s ( 0 , t ) , B ( t ) B ( s ) ~ N ( 0 , t s ) ;
  • s 1 0 , t 1 , s 2 0 , t 2 , s 1 s 2 , t 1 t 2 , B ( t 1 ) B ( s 1 ) independent of B ( t 2 ) B ( s 2 ) .
BM = B(t) − B(s) is considered as the single-step length of standard Brownian motion. OMO algorithm takes the default time interval of 1 s, so the single step length of standard Brownian motion BM ~ N(0, 1).
When |E| ≥ 1, predators enter the exploration phase and randomly adopt a strategy to explore:
X m ( t ) = 1 N i = 1 N X i ( t )
X i ( t + 1 ) = X r a b b i t ( t ) X m ( t ) r 1 D r 2 D U B L B + L B , q [ 2 , 3 ) X i ( t ) + s i g n X i ( t ) X m ( t ) a b s E p a r a α L F ( D ) , q [ 1 , 2 ) X i ( t ) + s i g n X i ( t ) X r a b b i t ( t ) a b s E p a r a α L F ( D ) , q ( 0 , 1 )
L F = μ ν 1 β
μ ~ N ( 0 , σ μ 2 )
ν ~ N ( 0 , σ ν 2 )
σ μ = Γ 1 + β sin π β 2 Γ 1 + β 2 β 2 β 1 2 1 β
where Xm(t) represents the average coordinates of predators in the t-generation population, N is the total number of individuals in the predators’ population, Xi(t) represents the coordinates of the ith predator in the t-generation population, Xrabbit(t) represents the position of the target prey (best fitness) in the t-generation population, LB and UB respectively represent the lower limit and upper limit of the exploration area, the symbol ‘ ’ is the Hadamard product, D is the coordinate dimension, r1(D) and r2(D) are 1 × D-dimensional vectors that obey (0, 1) uniform distribution, sign(x) is a sign function, q is a constant that obeys (0, 3) uniform distribution, LF(D) is a 1 × D-dimensional step vector of the Mantegna algorithm simulating the Levy flight, Equations (9)–(12) are the expressions of the one-dimensional Mantegna algorithm, generally, σν takes the value 1 and β takes the value 1.5.
When q ∈ [2, 3), compared with most SI optimization algorithms, the dimension of random parameters is increased, and the dispersion degree of random distribution is improved. When q ∈ [1, 2) and q ∈ (0, 1), the opposite direction is explored based on the average position of the population and the position of the prey, and Epara appropriately increases the unit step length of the predator individual for Levy flight during the exploration phase, for the purpose of jumping out of the local optimal solution and expanding the scope of exploration in the exploration phase.
When |E| < 1, suppose that r is the chance of a predator in mutating (r ≤ 0.5) or not mutating (r > 0.5) before pounce, and then predators enter the exploitation phase:
Y = X r a b b i t ( t ) + E p a r a X r a b b i t ( t ) + E p a r a β B M ( D ) X i ( t ) , r > 0.5 X r 1 ( t ) + M u × X r 2 ( t ) X r 3 ( t ) , r 0.5
Z = Y + E p a r a α L F ( D ) , r > 0.5 X i ( t ) H + Y ~ H , r 0.5
H 1 , i = 1 , r a n d 0 , 1 > C r 0 , r a n d 0 , 1 C r , i = 1 , 2 , , D
F min = min F [ X i ( t ) ] , F ( Y ) , F ( Z )
X i ( t + 1 ) = X i ( t ) , Y , Z , F min = F [ X i ( t ) ] F min = F ( Y ) F min = F ( Z )
where BM(D) is a 1 × D-dimensional step vector of the standard Brownian motion, r is a constant that obeys (0, 1) uniform distribution, r1, r2, r3 are random integers between [1, N] and are not equal to each other, Mu is the coefficient of variation, values between [0.4, 1], H is a 1 × D-dimensional logical vector aimed to increase the diversity of genes in the population, Cr ∈ [0, 1], which means the crossover probability of the individual’s genes, and the F function is the objective optimization function.
The exploitation phase in the OMO algorithm is divided into two strategies: besiege strategy and mutant strategy. In besiege strategy, the method of combining the small-scale Brownian motion of prey and the shorter step length Levy flight of predators is adopted, which can not only jump out of the local optimal solution but also conduct sufficient small-scale exploration around the prey for high convergence accuracy of the algorithm. In the mutant strategy, predators take mutation and crossover operations; this strategy absorbs the advantages of the integration of exploration and convergence in evolutionary algorithms, guarantees the diversity of the population in the early exploration, and avoids the shortcomings of decreased exploration ability after the exploitation phase is transferred to the exploration phase.
In order to better show the logic of the OMO algorithm, Figure 3 shows a brief flow chart of the OMO algorithm.
The OMO algorithm has superior performance in most standard test functions. Table 6 shows the details of the six representative standard test functions [61,62]. The population number N and maximum iterations of the OMO optimizer are set to 30 and 300, respectively. The speed control constant of the predator’s Levy flight SH is set to default value 1, and the speed control constant of prey’s standard Brownian motion SR is also set to default value 1, other parameters are consistent with those given in Table 6.
Figure 4 shows the test results of the OMO algorithm when the dimension is 2. The equations and parameter range of the test functions are consistent with those shown in Table 6. The population number N and the number of iterations of the OMO algorithm are set to 30 and 300, respectively.
The test results of the OMO algorithm are shown in Figure 4. The search history diagram reflects the characteristics of sufficient peripheral exploration degree, high randomness of exploration, and strong convergence ability in the OMO algorithm. The convergence curve reflects the advantage of the OMO algorithm that can effectively jump out of the local optimal solution.
The experiment of chemical gas leakage’ source tracking will be detailed in Section 3 and result details of the OMO algorithm will be explained in Section 4.

3. Experimental Design

For the chemical gas leakage accident under flat terrain, this paper proposes the combination of the OMO algorithm and Gaussian plume model to trace the source of chemical gas leakage. The experiment is mainly to verify the accuracy and efficiency of the OMO algorithm in solving this kind of problem, the influence of different OMO algorithm parameters on the results, and the advantages compared with other optimization algorithms.

3.1. Chemical Gas Leakage Accident Application Example

This paper proposes a novel algorithm to quickly and accurately realize the source tracking of chemical gas leakage. It is, therefore, necessary to first establish a chemical gas leakage accident application example in order to verify the application effect of the OMO algorithm.
A single application example cannot prove the universality of the OMO algorithm in practical application scenarios. However, more application examples and respective tests also cannot fully cover all the possibilities. Therefore, the category characteristics of the chemical gas leakage accident scene with flat terrain and stable atmospheric conditions need to be fixed, and the variables need to increase randomness. Experiments based on this condition can strongly prove that the OMO algorithm can be efficiently applied to the chemical gas leakage accident scene with flat terrain and stable atmospheric conditions.
In the chemical gas leakage accident scene with flat terrain and stable atmospheric conditions, the main variables include scene range, the location, and intensity of the gas leakage source, average ambient wind speed, atmospheric stability, and the location distribution of gas sensors. Because the entire gas leakage scene can be scaled proportionally, so all application examples belonging to the scene mentioned above can be equivalent to the application examples where the scope of the scene is fixed at 100 × 100 and other variables are random. The atmospheric stability rank has a clear classification and calculation formula, which has little influence on the optimization algorithm and indirectly affects the value of the gas sensors, so this paper does not increase the randomness of the atmospheric stability. Considering the reliability of experimental results and the algorithm evaluation system, multiple groups of independent small-scale tests will be conducted in each experiment. All the obtained optimal values, location values of the gas leakage source and intensity values of the gas leakage source will be averaged respectively, and the results after averaging need to have definite comparison indexes. Therefore, each experiment will keep the intensity and location of the gas leakage source unchanged and add the randomness of the gas sensors distribution and the randomness of the average ambient wind speed to the application example. In addition, although different gas leakage source locations and intensities will form different objective functions, the formed objective functions belong to the same type and have little impact on the comprehensive performance evaluation of the optimization algorithm. Therefore, the test results in this paper only show and analyze one of the representative cases.
Suppose that in a flat terrain chemical industrial park with a range of 100 m × 100 m, a chlorine storage tank at an unknown location has a leakage accident. The average leakage rate (the leakage intensity) is Q = 80 kg/s, the location of the leakage source is (x0, y0, h0) = (8, 15, 1) m, atmospheric stability level rank D, the average environmental wind speed is u = (10 × rw, 0, 0) m/s, and the plane distribution of the gas concentration sensors are as shown in Figure 5, with their height assumed to be 0 m and a measurement unit of kg/m3. Supplementary notes: rw obeys the uniform random distribution between (0, 1), the 16, 9, 4 sensors are randomly distributed in 4 × 4, 3 × 3, 2 × 2 squares formed by hollow circles, respectively. The accuracy of an ordinary commercial chlorine sensor (e.g., ME3-CL2) is 0.1 ppm, which is approximately equal to 3.155e-8 kg/m3. For the convenience of calculation, the default sensor accuracy is set to 1.0e-8 kg/m3 in this article.

3.2. Chemical Gas Leakage’s Source Tracking Model

In order to simplify the chemical gas leakage’s source tracking model, this paper ideally considered that the chlorine storage tanks in the chemical industrial park are made following the Chinese industrial standard of 1 m height. Therefore, when a leakage accident occurs, the parameters that need to be back-calculated are the top view plane position (x0, y0) of the leakage source and the average leakage rate Q of the leakage source.
The construction of the chemical gas leakage’s source tracking model is based on the Gaussian plume model. Suppose the total number of deployed sensors is NS, the chlorine concentration observation value obtained by deployed sensors is Cob(i), and the theoretical value calculated by the diffusion model is Ccal(i), where i is the serial number of the sensors. Then the objective function can be expressed in the form of Equation (18):
min x 0 , y 0 , Q f x 0 , y 0 , Q = i = 1 N S C o b i C c a l i 2
Suppose the position of the ith sensor is (xi, yi, zi), then the theoretical value calculated by the Gaussian plume diffusion model can be expressed in the form of Equation (19):
C c a l i = Q 2 π u σ y i σ z i e y i y 0 2 2 σ y i 2 e z i h 0 2 2 σ z i 2 + e z i + h 0 2 2 σ z i 2
According to Table 5, the Pasquill–Gifford diffusion coefficient equation of atmospheric stability level D at the position of the ith sensor is obtained:
σ y = 0.16 x i x 0 1 + 0.0004 x i x 0 1 / 2
σ z = 0.14 x i x 0 1 + 0.0003 x i x 0 1 / 2
Based on the chemical gas leakage accident application example and chemical gas leakage’s source tracking model, the total number NS of deployed sensors is 16, 9, 4, respectively, the lower limit of the optimization interval is LB = (0, 0, 0) and the upper limit is UB = (100, 100, 200), and the concentration observation values of monitoring points are simulated according to the Gaussian plume model.
In addition, the total number of sensors is set to 16, 9, 4 because the economic factors of sensor purchasing and placing in actual chemical industrial parks are taken into account, and monitoring points’ data with insufficient information are more likely to lead to an increase in the number of locally optimal solutions, thus reflecting the advantages of each optimization algorithm.

3.3. Other Details and Pseudocode of OMO Algorithm

All algorithms were implemented under MathWorks MATLAB R2020a (MathWorks.Inc, Natick, MA, USA) on a computer with a Windows 10-64bit professional edition (Microsoft, Redmond, WA, USA), 16GB RAM, and Intel® Core™ i5-9400F. The population number N and maximum iterations of all optimizers are set to 30 and 300, respectively. The geometric mean has the advantage of good stability, especially when analyzing the value of the objective function; all results are recorded and calculated based on the geometric mean of optimizers over 50 independent runs. Each independent operation will reset the random variables in the application example to achieve a high degree of randomness in the application example, which aims to comprehensively evaluate the multi-scenario application capabilities of each algorithm. The OMO algorithm’s parameters use default value and will be tested in Section 4, and the parameters of other algorithms are the same as the default or recommended value, which are reported in Table 7.
The pseudocode of the proposed OMO algorithm is reported in Algorithm 1.
Algorithm 1. Pseudocode of OMO algorithm.
Inputs: The population size N, maximum number of iterations T, Speed Control Constant SH, SR
Outputs: Data required for drawing plotData, the location of prey preyLocation and its fitness value preyEnergy
Initialize the random population Xi(i = 1, 2, …, N), plotData, preyLocation, preyEnergy and step parameters α, β
for (each iteration t) do
  for (each predator Xi) do
    Update the escape energy E and Epara using Equations (3) and (4)
    if (|E| ≥ 1) then                    Exploration phase
      Update the location vector using Equation (8)
    if (|E| < 1) then                    Exploitation phase
      if (r > 0.5) then                    Besiege strategy
        Update the location vector using Equations (13), (14), (16), (17)
        if (r ≤ 0.5) then                  Mutant strategy
        Update the location vector using Equations (13)–(17)
  Checkup population boundary and correct individual
  Calculate the fitness values of predator population
  Set preyLocation and preyEnergy as the location of prey and its fitness based on the minimum value between last prey and current population
  Update plotData and record them
Return plotData, preyLocation, preyEnergy

4. Results and Discussion

4.1. Results

Please note that the calculation time of all algorithms in this section is obtained by summing after 50 independent runs. For experiments that are not affected by the number of sensors, this paper will only show the test results under the 16 sensor scenarios.

4.1.1. Qualitative Results of OMO Algorithm

Table 8, Table 9 and Table 10 reveal the qualitative results of the OMO algorithm applied to the application example constructed above. It can be seen that when the number of sensors is 16 and 9, the OMO algorithm shows extremely high accuracy in the problem of chemical gas leakage traceability, and when the number of sensors is 4, that is, when the data is severely insufficient, the OMO algorithm will have some deviation, but there is still a certain degree of accuracy.
The qualitative results figures of the OMO algorithm applied to the chemical gas leakage accident application example constructed above are demonstrated in Figure 6. These results include five well-known metrics: convergence behavior, the average fitness of the population, search history, the trajectory of the first predator, and the trajectory of the prey. In addition, the escape energy of the prey is also monitored during iterations. The convergence curve reveals how the optimal value of the objective function varies during optimization. The average fitness of the population monitors how the average fitness of whole predators varies during the process of optimization. The search history distribution diagram shows the history of those positions visited by artificial predators during the calculation. The trajectory of the first predator diagram shows how the first variable of the first predator individual varies during the calculation. The map of the trajectory of the prey monitors how the variable of the prey (best solution’s location) varies during the iterations.
From the search history of predators, it can be concluded that predators attempt to initially boost the diversification and explore the easily overlooked areas of solution space and then exploit the vicinity of the best locations. The wide exploration area can better reduce the search omissions.
As in the trajectory in Figure 6, we see that the foremost predator’s amplitude of these variations covers more than 50% of the solution space while the prey’s amplitude of these variations covers about 25% of the solution space. As time passes, the explores amplitudes of the foremost predator and the prey gradually decrease. This point guarantees the transition of the OMO algorithm from the exploratory phase to the exploitative phase. Eventually, the motion patterns of the first predator and the prey become very stable which shows that the OMO algorithm is exploiting the promising regions during the concluding steps.
From the fitness and average fitness curve, it can be observed that the OMO algorithm jumped out of the local optimal solution multiple and reached the global optimal solution about the 50th iteration, but the convergence of the average fitness is reached at about 150th iteration, which means that the OMO algorithm applies the greedy algorithm which is not satisfied with the optimal solution in the previous iteration and tries to jump out of the local optimal solution. Thus, when facing the problem of location optimization with a known optimal value of the objective function, a precision control constant can be introduced to realize the function of stopping iteration after the required precision is reached, thereby greatly shortening the calculation time. The test results of the chemical gas leakage accident application example are shown in Table 11 above.
Although the OMO method is mainly used to solve the traceability problem of chemical gas leakage, for the entire chemical gas leakage accident emergency and response system, the prediction of gas concentration and the marking of dangerous thresholds are also an important part of it. Figure 7 shows a preliminary schematic diagram of gas concentration prediction and the mark of risk thresholds.

4.1.2. Relationship between OMO Algorithm and Population Number N

In the case of a fixed number of 300 iterations, the results of the relationship between the OMO algorithm and population number N are shown in Table 12 and Figure 8a.
Taking the Briggs logarithm of the optimal value as the criterion for judging the accuracy of the OMO algorithm, the conclusions can be drawn from Table 12 and Figure 8a:
  • The time-requirement of the OMO algorithm is approximately proportional to the population number N;
  • The accuracy of the OMO algorithm gradually increases as the population number N increases;
  • When the population number 3 < N < 5, the accuracy of the OMO algorithm is relatively low; when the population number 3 < N < 25, the accuracy of the OMO algorithm is significantly improved with the increase of N; when the population number N > 25, the increased speed of OMO algorithm’s accuracy with the increase of N gradually decreases and stabilizes.
Figure 8 combines the results of Section 4.1.2 and Section 4.1.3. It can be concluded from Figure 8 that the population size N and the number of iterations have similar effects on the OMO algorithm in terms of time-requirements and accuracy.

4.1.3. Relationship between OMO Algorithm and Iterations

In the case of a fixed number of N = 30, the results of the relationship between the OMO algorithm and iterations are shown in Table 13 and Figure 8b.
Taking the Briggs logarithm of the optimal value as the criterion for judging the accuracy of the OMO algorithm, the conclusions can be drawn from Table 13 and Figure 8b:
  • The time-requirement of the OMO algorithm is approximately proportional to the iterations;
  • The accuracy of the OMO algorithm gradually increases as the iterations increase;
  • When the iterations <100, the accuracy of the OMO algorithm is significantly improved with the increase of iterations; when the 100 < iterations < 500, the accuracy of the OMO algorithm is slightly improved with the increase of iterations; when the iterations >500, the increased speed of the OMO algorithm’s accuracy with the increase of iterations gradually decreases and stabilizes.

4.1.4. Relationship between OMO Algorithm and Speed Control Constant

Given the high number of iterations and the high population number, the results calculated by the OMO algorithm will show high accuracy, which is not inefficient in studying the relationship between the OMO algorithm and speed control constant. Therefore, in the case of fixed numbers of r = 0, 100 iterations, and 20 population numbers, 1,000,000 sets of data are recorded based on the free combination of SH = 1:1:1000 and SR = 1:1:1000. The results of the relationship between the OMO algorithm and speed control constant are shown in Table 14, Figure 9 and Figure 10.
These conclusions can be drawn from Table 14, Figure 9 and Figure 10:
  • Changing the values of SH and SR will have a certain impact on the accuracy of the OMO algorithm, but random values of SH and SR will have a high probability to get a more satisfactory result.
  • High-precision OMO algorithm results are often accompanied by suitable exploration steps (influenced by SH), and low-precision OMO algorithm results are often accompanied by too-large exploration steps. The reason why some SH are too large but can also achieve accuracy of the algorithm is mainly due to the random allocation of super boundary data, but in the chemical gas leakage accident model established by this paper, the best value of SH is 1–60.
  • High-precision OMO algorithm results are often accompanied by suitable exploitation steps (influenced by SR), and low-precision OMO algorithm results are often accompanied by too large exploitation steps. Logarithmic processing has been performed in the step length parameter of the standard Brownian motion to ensure that the exploitation space is small enough in the final stage and the space has been explored completely; therefore, a small-range change of SR will have little effect on the accuracy of the OMO algorithm. In the chemical gas leakage accident model established by this paper, the best value of SR is 1–100.
Therefore, the default values of the speed control constant for the chemical gas leakage accident model (16 sensors) established by this paper can be decided: SH = 4 and SR = 6. The optimal speed control constant values of different application examples are not necessarily the same and need to be verified by experiments, but most of the speed control constant settings can get a satisfactory result with a high probability.

4.1.5. Robustness Analysis under the Influence of White Gaussian Noise (WGN)

Considering the actual factors of the chemical gas leakage accident application example, we can add White Gaussian Noise (WGN) with different Signal-to-Noise Ratios (SNRs) to the observed value of the gas concentration to simulate the error between the calculated value of the Gaussian plume model and the actual observed value, and analyze the robustness of the OMO algorithm. In the case of fixed numbers of N = 30, iterations = 300, SH = 4, and SR = 6, the results of the relationship between the OMO algorithm’s accuracy and SNR are shown in Table 15.
From the data in Table 15, it can be found that as the noise increases, the error of the calculation result increases relatively. When SNR > 15, the deviation of the leakage source data calculated by the OMO algorithm is within the allowable range of error, but when SNR < 1, the results do not meet the ideal expectations.
In summary, the OMO algorithm has great robustness and can be congruently applied to chemical gas leakage accident scenarios whose SNR > 15.

4.1.6. Comparative Analysis among Optimization Algorithms

In the case of fixed numbers of N = 30, iterations = 300, SH = 1, and SR = 1 in the OMO algorithm, the comparison results of various optimization algorithms are shown in Table 16, Table 17 and Table 18. In addition, all optimization algorithms were implemented based on the application example established in Section 3.1. The population number N of other optimizers is set to 30, except for the PS and SA algorithms, which do not have the parameter of population number N, and the number of iterations of other optimizers is set to 300. The basic parameters of other algorithms are the same as the default or recommended value, which are reported in Section 3.3, Table 7.
The PSO algorithm has the fastest calculation speed: 2.223 s/50 times. The OMO algorithm has the lowest geometric mean error: Δx0 = 1.971%, Δy0 = 0.131%, ΔQ = 1.193%, and the stable highest accuracy, which can be reflected by the geometric mean of the optimal value: 2.464e-15.
The PSO algorithm has the fastest calculation speed: 1.638 s/50 times. The lowest x0 error, the lowest y0 error, and the lowest Q error are: Δx0 = 3.593% calculated by OMO, Δy0 = 0.010% calculated by DE, ΔQ = 0.077% calculated by GA. The OMO algorithm has the stable highest accuracy: 2.356e-13.
The PSO algorithm has the fastest calculation speed: 1.341 s/50 times. The lowest x0 error, the lowest y0 error, and the lowest Q error are: Δx0 = 23.427% calculated by GA, Δy0 = 1.121% calculated by SA, ΔQ = 0.709% calculated by GWO. The OMO algorithm has the stable highest accuracy: 5.694e-23.
Analyzing and comparing the results in Table 11 and Table 16, it can be concluded that the calculation speed of PS, SA, PSO, and GA algorithms are significantly better than other algorithms, but the calculation accuracy of PS, SA, and GA algorithms is significantly lower than most algorithms, while the calculation accuracy of PSO algorithms is relatively higher. The DE algorithm has great advantages in calculation time and calculation accuracy, swarm intelligence algorithms such as HHO, GWO, SSA, WOA performance is mediocre in the calculation speed aspects, and the SSA algorithm surpasses other algorithms in accuracy, second only to the OMO and DE algorithms. The OMO algorithm has the highest calculation accuracy among all algorithms, at the expense of calculation time. However, the OMO algorithm, which introduces precision control parameters, achieves a balance between precision and calculation, when the optimal value of the function reaches 1e-10 (only second to the DE algorithm) and the calculation time is only about 6.697s (only second to PSO, GA algorithms, and PS, SA algorithms which belong to the direct search algorithm category, which is known for its computational speed).
Analyzing and comparing the results in Table 16, Table 17 and Table 18, it can be concluded that regardless of the severity of sensor data insufficiency in the chemical gas leakage scene, the OMO algorithm always maintains great advantages in accuracy compared with other algorithms. The geometric average of the optimal value mainly reflects the average accuracy of the algorithm; that is, the stability and accuracy of the OMO algorithm are higher than other optimization algorithms. When the number of sensors is reduced from 16 to 9, the accuracy of some optimization algorithms is severely reduced, mainly because the lack of information leads to an increase in the local optimal value. When the number of sensors is 9, optimization algorithms that perform better include OMO, PSO, DE, HHO and SSA. When the number of sensors is only 4, the severe data insufficiency leads to dense local optimal values, and each optimization algorithm fails to obtain relatively accurate results. However, from the perspective of the relative error and the average optimal value, the OMO algorithm is more suitable than others for solving the source tracking of chemical gas leakage problems.
In summary, the OMO algorithm has shown extremely high accuracy when facing the application example of chemical gas leakage source tracking. The OMO algorithm introduces precision control parameters that can provide high-precision results while ensuring the calculation speed, thereby meeting the emergency needs in the event of chemical gas leakage accidents.

4.2. Discussion

As per results in previous sections, we can recognize that the OMO algorithm shows significantly superior results for the application example of source tracking of chemical gas leakage.
The following experiments, studies, or analyses on the OMO algorithm have been described in this Section: qualitative results of the OMO algorithm in attempting to solve the application example proposed in Section 3, relationship between OMO algorithm and population number N, relationship between OMO algorithm and iterations, relationship between OMO algorithm and speed control constant, robustness analysis under the influence of WGN and comparative analysis among optimization algorithms. The following conclusions or default values can be obtained from these experiments, studies, and analysis: the OMO algorithm has excellent exploration and exploitation capabilities; the larger the number of populations, the higher the accuracy of the OMO algorithm; the larger the iterations, the higher the accuracy of the OMO algorithm; the speed control constant has a certain degree of influence on the accuracy of the OMO algorithm, default SH = 4 and SR = 6; the OMO algorithm has relatively good robustness and can calculate reasonable results when SNR > 15 in solving the application example; the accuracy of the OMO algorithm is far superior to other algorithms, and the OMO algorithm that introduces precision control parameters is better than other algorithms in terms of overall performance.
In future work, the basic emergency safety system can be established. Real-time prediction of chemical gas dispersion will be added into this system and the Gaussian plume model will be replaced by a higher precision model with the guarantee of viable calculation time-requirements. First, we will conduct more experiments based on the CFD model and establish a database as reference data for real chemical leakage accidents. Second, the chemical gas leakage scenario experiments will be conducted to verify the established database. Third, we will use deep learning or other feasible methods aimed to establish a universal gas diffusion empirical model. Finally, we will test the accuracy of the model and apply it to the prediction of gas diffusion and replace the Gaussian plume model with the OMO algorithm.

5. Conclusions

This paper introduces a novel method for the purpose of chemical gas leakage accident source tracking. The OMO algorithm is mainly divided into two phases. In the exploration phase, Levi’s flight and its step size vector are introduced, and some individual outlier search is adopted to improve the search range at the initial phase of the exploration. In the exploitation phase, Brownian motion and its step size vector are introduced, and the idea of individual mutation is also introduced, which enriches the path of convergence, and improves the convergence speed and accuracy. As regards the computational efficiency of the OMO algorithm, the test results show that the OMO algorithm with default parameters has superior comprehensive performance, including the extremely high average calculation accuracy—the optimal value reaches 2.464e-15 when the number of sensors is 16; 2.356e-13 when the number of sensors is 9; and 5.694e-23 when the number of sensors is 4. There is a satisfactory calculation time: 12.743 s/50 times when the number of sensors is 16; 10.304 s/50 times when the number of sensors is 9; and 8.644 s/50 times when the number of sensors is 4.
In recent years, the safety of chemical industrial parks has received great attention, especially in China. In order to actively respond to chemical gas leakage accidents, a complete set of safety emergency systems needs to be rigorously developed, including the source tracking of chemical gas leakage accidents and the real-time prediction of chemical gas dispersion. In addition to these basic emergency needs, there are challenges of source tracking of chemical gas leakage accidents with complex terrain and climatic conditions, the real-time predicting of fire and explosion caused by leaking gas, etc.
The OMO algorithm combined with a Gaussian plume model is proved in this paper to be able to quickly and accurately realize the inverse calculation of gas leakage sources under flat terrain and stable atmospheric conditions. This method can provide technical support for emergency measures in normal chemical industrial parks, and can also provide some help for the future study of chemical gas leakage accidents under complex situations.
The OMO algorithm can not only be applied to chemical gas leakage accidents in chemical industrial parks but also the three-bar truss design problem, tension/compression spring design problem, gas pipeline leakage accidents, pressure vessel design problem, etc. Conclusions can be drawn from the results of Section 4: OMO’s flexible algorithm parameters can improve its effect on different application examples to a certain extent, including speed control constant, precision control coefficient, the maximum number of iterations, and population number.

Author Contributions

Conceptualization, Z.X. (Zhengyi Xu); Data curation, Z.X. (Zhiyu Xia); Formal analysis, D.L. and J.W.; Investigation, Z.X. (Zhiyu Xia) and D.L.; Methodology, Z.X. (Zhiyu Xia), Z.X. (Zhengyi Xu) and J.W.; Project administration, Z.X. (Zhengyi Xu) and D.L.; Resources, J.W.; Software, Z.X. (Zhiyu Xia) and D.L.; Supervision, Z.X. (Zhengyi Xu) and J.W.; Validation, D.Li; Writing—original draft, Z.X. (Zhiyu Xia); Writing—review & editing, Z.X. (Zhengyi Xu) and J.W. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Science and Technology Innovation Plan of Shanghai Science and Technology Commission (19DZ1202200), Youth Innovation Promotion Association CAS (2021289), and Science and Technology Innovation Plan of Shanghai Science and Technology Commission (20DZ1201002).

Acknowledgments

The authors would like to express their great appreciation to their teachers for their helpful suggestions on this manuscript and to the support provided by the Safety and Emergency Lab of Shanghai Advanced Research Institute (SARI) for this study. This work was sponsored by Science and Technology Innovation Plan of Shanghai Science and Technology Commission (19DZ1202200), Youth Innovation Promotion Association CAS (2021289), and Science and Technology Innovation Plan of Shanghai Science and Technology Commission (20DZ1201002).

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Long, S.; Zhao, L.; Shi, T.; Li, J.; Yang, J.; Liu, H.; Mao, G.; Qiao, Z.; Yang, Y. Pollution control and cost analysis of wastewater treatment at industrial parks in Taihu and Haihe water basins, China. J. Clean. Prod. 2018, 172, 2435–2442. [Google Scholar] [CrossRef]
  2. Park, H.-S.; Rene, E.R.; Choi, S.-M.; Chiu, A.S.F. Strategies for sustainable development of industrial park in Ulsan, South Korea—From spontaneous evolution to systematic expansion of industrial symbiosis. J. Environ. Manag. 2008, 87, 1–13. [Google Scholar] [CrossRef] [PubMed]
  3. Wang, Y.; Zhang, R.; Worden, S.; Cao, H.; Li, C. Public participation in environmental governance initiatives of chemical industrial parks. J. Clean. Prod. 2021, 305, 127092. [Google Scholar] [CrossRef]
  4. Yet-Pole, I.; Fu, J.M. Risk analysis of a cross-regional toxic chemical disaster by using the integrated mesoscale and microscale consequence analysis model. J. Loss Prev. Process. Ind. 2021, 71, 104424. [Google Scholar]
  5. Jiang, Y.; Xu, Z.; Wei, J.; Teng, G. Fused CFD-interpolation model for real-time prediction of hazardous gas dispersion in emergency rescue. J. Loss Prev. Process. Ind. 2020, 63, 103988. [Google Scholar] [CrossRef]
  6. Nooralishahi, P.; Lopez, F.; Maldague, X. A Drone-Enabled Approach for Gas Leak Detection Using Optical Flow Analysis. Appl. Sci. 2021, 11, 1412. [Google Scholar] [CrossRef]
  7. Chen, C.; Reniers, G.; Khakzad, N. Integrating safety and security resources to protect chemical industrial parks from man-made domino effects: A dynamic graph approach. Reliab. Eng. Syst. Saf. 2019, 191, 106470. [Google Scholar] [CrossRef]
  8. Gas Leak Accidents that Made the Headlines in the Past. Available online: https://www.deccanherald.com/national/gas-leak-accidents-that-made-the-headlines-in-the-past-834754.html (accessed on 25 June 2021).
  9. Acute Exposure Guideline Levels (AEGLs). Available online: https://response.restoration.noaa.gov/oil-and-chemical-spills/chemical-spills/resources/acute-exposure-guideline-levels-aegls.html (accessed on 10 July 2021).
  10. Ammonia Results-AEGL Program. Available online: https://www.epa.gov/aegl/ammonia-results-aegl-programs (accessed on 10 July 2021).
  11. Chlorine Results-AEGL Program. Available online: https://www.epa.gov/aegl/chlorine-results-aegl-program (accessed on 10 July 2021).
  12. Phosgene Results-AEGL Program. Available online: https://www.epa.gov/aegl/phosgene-results-aegl-program (accessed on 10 July 2021).
  13. Hou, Q.; Zhu, W. An EKF-Based Method and Experimental Study for Small Leakage Detection and Location in Natural Gas Pipelines. Appl. Sci. 2019, 9, 3193. [Google Scholar] [CrossRef]
  14. Huang, Y.; Xiu, G.; Lu, Y.; Gao, S.; Li, L.; Chen, L.; Huang, Q.; Yang, Y.; Che, X.; Chen, X.; et al. Application of an emission profile-based method to trace the sources of volatile organic compounds in a chemical industrial park. Sci. Total Environ. 2021, 768, 144694. [Google Scholar] [CrossRef]
  15. Singh, S.K.; Rani, R. A least-squares inversion technique for identification of a point release: Application to Fusion Field Trials 2007. Atmos. Environ. 2014, 92, 104–117. [Google Scholar] [CrossRef]
  16. Singh, S.K.; Rani, R. Assimilation of concentration measurements for retrieving multiple point releases in atmosphere: A least-squares approach to inverse modelling. Atmos. Environ. 2015, 119, 402–414. [Google Scholar] [CrossRef]
  17. Wang, Y.; Huang, H.; Huang, L.; Zhang, X. Source term estimation of hazardous material releases using hybrid genetic algorithm with composite cost functions. Eng. Appl. Artif. Intell. 2018, 75, 102–113. [Google Scholar] [CrossRef]
  18. Kim, H.; Yoon, E.S.; Shin, D. Deep Neural Networks for Source Tracking of Chemical Leaks and Improved Chemical Process Safety. In 13th International Symposium on Process Systems Engineering (PSE 2018); Mario, R.E., Marianthi, G.I., Gavin, P.T., Eds.; Elsevier Ltd.: San Diego, CA, USA, 2018; Volume 44, pp. 2359–2364. [Google Scholar]
  19. Althuwaynee, O.F.; Pokharel, B.; Aydda, A.; Balogun, A.L.; Kim, S.W.; Park, H.J. Spatial identification and temporal prediction of air pollution sources using conditional bivariate probability function and time series signature. J. Expo. Sci. Environ. Epidemiol. 2020, 31, 709–726. [Google Scholar] [CrossRef]
  20. Yang, Z.; Yao, Q.; Buser, M.D.; Alfieri, J.G.; Li, H.; Torrents, A.; McConnell, L.L.; Downey, P.M.; Hapeman, C.J. Modification and validation of the Gaussian plume model (GPM) to predict ammonia and particulate matter dispersion. Atmos. Pollut. Res. 2020, 11, 1063–1072. [Google Scholar] [CrossRef]
  21. Lee, J.; Lee, S.; Son, H.; Yi, W. Development of PUFF-Gaussian dispersion model for the prediction of atmospheric distribution of particle concentration. Sci. Rep. 2021, 11, 1–13. [Google Scholar] [CrossRef]
  22. Ye, W.; Zhou, B.; Tu, Z.; Xiao, X.; Yan, J.; Wu, T.; Wu, F.; Zheng, C.; Tittel, F.K. Leakage source location based on Gaussian plume diffusion model using a near-infrared sensor. Infrared Phys. Technol. 2020, 109, 103411. [Google Scholar] [CrossRef]
  23. Ban, M.; Yu, J.; Shahidehpour, M.; Guo, D.; Yao, Y. Considering the Differentiating Health Impacts of Fuel Emissions in Optimal Generation Scheduling. IEEE Trans. Sustain. Energy 2020, 11, 15–26. [Google Scholar] [CrossRef]
  24. Pirhalla, M.; Heist, D.; Perry, S.; Tang, W.; Brouwer, L. Simulations of dispersion through an irregular urban building array. Atmos. Environ. 2021, 258, 118500. [Google Scholar] [CrossRef]
  25. Hanna, S. Britter and McQuaid (B&M) 1988 workbook nomograms for dense gas modeling applied to the Jack Rabbit II chlorine release trials. Atmos. Environ. 2020, 232, 117539. [Google Scholar]
  26. Long, C.; Qi, H.; Zhang, Y. A dynamic model of gas diffusion. Dyn. Contin. Discret. Impulsive Syst. Ser. B Appl. Algorithms 2006, 13, 706–709. [Google Scholar]
  27. Zhang, X.; Li, J.; Zhu, J.; Qiu, L. Computational fluid dynamics study on liquefied natural gas dispersion with phase change of water. Int. J. Heat Mass Transf. 2015, 91, 347–354. [Google Scholar] [CrossRef]
  28. Bellegoni, M.; Ovidi, F.; Landucci, G.; Tognotti, L.; Galletti, C. CFD analysis of the influence of a perimeter wall on the natural gas dispersion from an LNG pool. Process Saf. Environ. Prot. 2021, 148, 751–764. [Google Scholar] [CrossRef]
  29. Huh, C.; Choi, S.; Lee, J.M. Concentration model for gas releases in buildings and the mitigation effect. J. Loss Prev. Process Ind. 2020, 65, 104135. [Google Scholar] [CrossRef]
  30. Ma, D.; Tan, W.; Wang, Q.; Zhang, Z.; Gao, J.; Zeng, Q.; Wang, X.; Xia, F.; Shi, X. Application and improvement of swarm intelligence optimization algorithm in gas emission source identification in atmosphere. J. Loss Prev. Process Ind. 2018, 56, 262–271. [Google Scholar] [CrossRef]
  31. Huang, Y.; Wang, Q.; Shi, L.; Yang, Q. Underwater gas pipeline leakage source localization by distributed fiber-optic sensing based on particle swarm optimization tuning of the support vector machine. Appl. Opt. 2016, 55, 242–247. [Google Scholar] [CrossRef]
  32. Heidari, A.A.; Mirjalili, S.; Faris, H.; Aljarah, I.; Mafarja, M.; Chen, H. Harris hawks optimization: Algorithm and applications. Future Gener. Comput. Syst. Int. J. Escience 2019, 97, 849–872. [Google Scholar] [CrossRef]
  33. Wang, J.; Zhang, R.; Li, J.; Xin, Z. Locating unknown number of multi-point hazardous gas leaks using Principal Component Analysis and a Modified Genetic Algorithm. Atmos. Environ. 2020, 230, 117515. [Google Scholar] [CrossRef]
  34. Steffelbauer, D.B.; Günther, M.; Fuchs-Hanusch, D. Leakage Localization with Differential Evolution: A Closer Look on Distance Metrics. In Proceedings of the 18th International Conference on Water Distribution System Analysis (WDSA), Cartagena, Colombia, 24–28 July 2016; Elsevier Science BV: Amsterdam, The Netherlands, 2017; Volume 186, pp. 444–451. [Google Scholar]
  35. Zheng, X.; Chen, Z. Inverse calculation approaches for source determination in hazardous chemical releases. J. Loss Prev. Process Ind. 2011, 24, 293–301. [Google Scholar] [CrossRef]
  36. Hooke, R.; Jeeves, T.A. Direct Search Solution of Numerical and Statistical Problems. J. ACM 1961, 8, 212–229. [Google Scholar] [CrossRef]
  37. Huang, Y.-C.; Lin, C.-C.; Yeh, H.-D. An Optimization Approach to Leak Detection in Pipe Networks Using Simulated Annealing. Water Resour. Manag. 2015, 29, 4185–4201. [Google Scholar] [CrossRef]
  38. Sousa, J.; Ribeiro, L.; Muranho, J.; Marques, A.S. Locating leaks in water distribution networks with simulated annealing and graph theory. In Computing and Control for the Water Industry; Ulanicki, B., Kapelan, Z., Boxall, J., Eds.; Elsevier Ltd.: San Diego, CA, USA, 2015; Volume 119, pp. 63–71. [Google Scholar]
  39. Jiao, S.B.; Fan, L.; Wu, P.Y.; Qiao, L.; Wang, Y.; Xie, G. Assessment of leakage degree of underground heating primary pipe network based on chaotic simulated annealing neural network. In Proceedings of the Chinese Automation Congress, Jinan, China, 20–22 October 2017; IEEE: Piscataway, NJ, USA, 2018. [Google Scholar]
  40. Jaward, M.H.; Bull, D.; Canagarajah, N. Sequential Monte Carlo methods for contour tracking of contaminant clouds. Signal Process. 2010, 90, 249–260. [Google Scholar] [CrossRef]
  41. Quy, T.B.; Kim, J.M. Real-Time Leak Detection for a Gas Pipeline Using a k-NN Classifier and Hybrid AE Features. Sensors 2021, 21, 367. [Google Scholar] [CrossRef]
  42. Husnawati; Fitriana, G.F.; Nurmaini, S. The development of hybrid methods in simple swarm robots for gas leak localization. In Proceedings of the 2017 International Conference on Signals and Systems, Bali, Indonesia, 16–18 May 2017; IEEE: Piscataway, NJ, USA, 2017. [Google Scholar]
  43. Ma, T.; Liu, S.; Xiao, H. Multirobot searching method of natural gas leakage sources on offshore platform using ant colony optimization. Int. J. Adv. Robot. Syst. 2020, 17, 1729881420959012. [Google Scholar] [CrossRef]
  44. Qiu, S.; Chen, B.; Wang, R.; Zhu, Z.; Wang, Y.; Qiu, X. Atmospheric dispersion prediction and source estimation of hazardous gas using artificial neural network, particle swarm optimization and expectation maximization. Atmos. Environ. 2018, 178, 158–163. [Google Scholar] [CrossRef]
  45. Zhang, H.; Liang, Y.; Zhang, W.; Xu, N.; Guo, Z.; Wu, G. Improved PSO-Based Method for Leak Detection and Localization in Liquid Pipelines. IEEE Trans. Ind. Inform. 2018, 14, 3143–3154. [Google Scholar] [CrossRef]
  46. Wang, R.; Chen, B.; Qiu, S.; Ma, L.; Zhu, Z.; Wang, Y.; Qiu, X. Hazardous Source Estimation Using an Artificial Neural Network, Particle Swarm Optimization and a Simulated Annealing Algorithm. Atmosphere 2018, 9, 119. [Google Scholar] [CrossRef]
  47. Lalle, Y.; Abdelhafidh, M.; Fourati, L.C.; Rezgui, J. A hybrid optimization algorithm based on K-means plus plus and Multi-objective Chaotic Ant Swarm Optimization for WSN in pipeline monitoring. In Proceedings of the 15th IEEE International Wireless Communications and Mobile Computing Conference, Tangier, Morocco, 24–28 June 2019; IEEE: Piscataway, NJ, USA, 2019. [Google Scholar]
  48. Lateb, M.; Meroney, R.N.; Yataghene, M.; Fellouah, H.; Saleh, F.; Boufadel, M.C. On the use of numerical modelling for near-field pollutant dispersion in urban environments—A review. Environ. Pollut. 2016, 208, 271–283. [Google Scholar] [CrossRef]
  49. Jung, Y.R.; Park, W.G.; Park, O.H. Pollution dispersion anaysis using the puff model with numerical flow field data. Mech. Res. Commun. 2003, 30, 277–286. [Google Scholar] [CrossRef]
  50. Xu, H.-X.; Li, G.; Yang, S.-L.; Xu, X. Modeling and simulation of haze process based on Gaussian model. In Proceedings of the 11th International Computer Conference on Wavelet Active Media Technology and Information Processing, Chengdu, China, 19–21 December 2014; IEEE: Piscataway, NJ, USA, 2015. [Google Scholar]
  51. Stockie, J.M. The Mathematics of Atmospheric Dispersion Modeling. Siam Rev. 2011, 53, 349–372. [Google Scholar] [CrossRef]
  52. Brusca, S.; Famoso, F.; Lanzafame, R.; Mauro, S.; Garrano, A.M.C.; Monforte, P. Theoretical and experimental study of Gaussian Plume model in small scale systems. In Energy Procedia, Proceedings of the 71st Conference of the Italian Thermal Machines Engineering Association, Politecnico Torino, Turin, Italy, 14–16 September 2016; Masoero, M.C., Arsie, I., Eds.; Elsevier Ltd.: San Diego, CA, USA, 2016. [Google Scholar]
  53. Yang, Y.; Zheng, Z.; Bian, K.; Song, L.; Han, Z. Real-Time Profiling of Fine-Grained Air Quality Index Distribution Using UAV Sensing. IEEE Internet Things J. 2018, 5, 186–198. [Google Scholar] [CrossRef]
  54. Jain, R.K.; Cui, Z.C.; Domen, J.K. Environmental Impact of Mining and Mineral Processing: Management, Monitoring, and Auditing Strategies; Elsevier Ltd.: San Diego, CA, USA, 2016; pp. 87–90. [Google Scholar]
  55. Ma, D.; Zhang, Z. Contaminant dispersion prediction and source estimation with integrated Gaussian-machine learning network model for point source emission in atmosphere. J. Hazard. Mater. 2016, 311, 237–245. [Google Scholar] [CrossRef]
  56. Cao, B.; Cui, W.; Chen, C.; Chen, Y. Development and uncertainty analysis of radionuclide atmospheric dispersion modeling codes based on Gaussian plume model. Energy 2020, 194, 116925. [Google Scholar] [CrossRef]
  57. Kahl, J.D.W.; Chapman, H.L. Atmospheric stability characterization using the Pasquill method: A critical evaluation. Atmos. Environ. 2018, 187, 196–209. [Google Scholar] [CrossRef]
  58. Kaasik, M.; Geertsema, G.; Scheele, R. Validation of Gaussian plume model Aeropol against Cabauw field experiment. Int. J. Environ. Pollut. 2017, 62, 369–384. [Google Scholar] [CrossRef]
  59. Wang, H.; Song, W.; Zio, E.; Kudreyko, A.; Zhang, Y. Remaining useful life prediction for Lithium-ion batteries using fractional Brownian motion and Fruit-fly Optimization Algorithm. Measurement 2020, 161, 107904. [Google Scholar] [CrossRef]
  60. Song, W.; Cattani, C.; Chi, C.H. Multifractional Brownian motion and quantum-behaved particle swarm optimization for short term power load forecasting: An integrated approach. Energy 2020, 194, 116847. [Google Scholar] [CrossRef]
  61. Cavalca, D.L.; Fernandes, R.A.S. Gradient-based mechanism for PSO algorithm: A comparative study on numerical benchmarks. In Proceedings of the IEEE Congress on Evolutionary Computation (IEEE CEC) as part of the IEEE World Congress on Computational Intelligence (IEEE WCCI), Rio de Janeiro, Brazil, 8–13 July 2018; IEEE: Piscataway, NJ, USA, 2018. [Google Scholar]
  62. Bakwad, K.M.; Pattnaik, S.S.; Sohi, B.S.; Devi, S.; Panigrahi, B.K.; Gollapudi, S.V.R.S. Multimodal Function Optimization Using Synchronous Bacterial Foraging Optimization Technique. IETE J. Res. 2010, 56, 80–87. [Google Scholar] [CrossRef]
  63. Mirjalili, S.; Mirjalili, S.M.; Lewis, A. Grey Wolf Optimizer. Adv. Eng. Softw. 2014, 69, 46–61. [Google Scholar] [CrossRef]
  64. Mirjalili, S.; Gandomi, A.H.; Mirjalili, S.Z.; Saremi, S.; Faris, H.; Mirjalili, S.M. Salp Swarm Algorithm: A bio-inspired optimizer for engineering design problems. Adv. Eng. Softw. 2017, 114, 163–191. [Google Scholar] [CrossRef]
  65. Mirjalili, S.; Lewis, A. The Whale Optimization Algorithm. Adv. Eng. Softw. 2016, 95, 51–67. [Google Scholar] [CrossRef]
Figure 1. The curve of escape energy E varies with iterations.
Figure 1. The curve of escape energy E varies with iterations.
Sensors 22 00071 g001
Figure 2. Schematic diagram of two kinds of random walk trajectory, with number of steps = 500. (a) Levy flight with the characteristic of wide exploration range; (b) Standard Brownian motion with the characteristic of high exploration density.
Figure 2. Schematic diagram of two kinds of random walk trajectory, with number of steps = 500. (a) Levy flight with the characteristic of wide exploration range; (b) Standard Brownian motion with the characteristic of high exploration density.
Sensors 22 00071 g002
Figure 3. Flow chart of the OMO algorithm.
Figure 3. Flow chart of the OMO algorithm.
Sensors 22 00071 g003
Figure 4. The standard test results of the OMO algorithm when the dimension is 2. (a1a6) Parameter space schematic diagram of the ith test function; (b1b6) Search history diagram of the OMO algorithm in the ith test function with percentage range mark; (c1c6) The optimal value of the function varies with the number of iterations graph of the OMO algorithm in the ith test function.
Figure 4. The standard test results of the OMO algorithm when the dimension is 2. (a1a6) Parameter space schematic diagram of the ith test function; (b1b6) Search history diagram of the OMO algorithm in the ith test function with percentage range mark; (c1c6) The optimal value of the function varies with the number of iterations graph of the OMO algorithm in the ith test function.
Sensors 22 00071 g004aSensors 22 00071 g004b
Figure 5. Schematic diagram of three application examples. (a) Distribution of 16 monitoring points and leak source; (b) distribution of 9 monitoring points and leak source; (c) distribution of 4 monitoring points and leak source.
Figure 5. Schematic diagram of three application examples. (a) Distribution of 16 monitoring points and leak source; (b) distribution of 9 monitoring points and leak source; (c) distribution of 4 monitoring points and leak source.
Sensors 22 00071 g005
Figure 6. Qualitative results figures of OMO algorithm under 16 sensors. (a) Convergence curve; (b) average fitness of all predators; (c) escape energy of the prey; (d) search history of predators; (e) trajectory of the first predator; (f) trajectory of the prey.
Figure 6. Qualitative results figures of OMO algorithm under 16 sensors. (a) Convergence curve; (b) average fitness of all predators; (c) escape energy of the prey; (d) search history of predators; (e) trajectory of the first predator; (f) trajectory of the prey.
Sensors 22 00071 g006
Figure 7. Supplement to chemical gas leakage accident system. (a) Prediction of gas concentration distribution and contour division under the AEGLs standard; (b) 3D diagram of prediction of gas concentration distribution.
Figure 7. Supplement to chemical gas leakage accident system. (a) Prediction of gas concentration distribution and contour division under the AEGLs standard; (b) 3D diagram of prediction of gas concentration distribution.
Sensors 22 00071 g007
Figure 8. Relationship between OMO algorithm and population number N or iterations. (a) Smooth curve graph of the population number N—the optimal value of the objective function and the population number N—the consuming time; (b) smooth curve graph of the iterations—the optimal value of the objective function and the iterations—the consuming time.
Figure 8. Relationship between OMO algorithm and population number N or iterations. (a) Smooth curve graph of the population number N—the optimal value of the objective function and the population number N—the consuming time; (b) smooth curve graph of the iterations—the optimal value of the objective function and the iterations—the consuming time.
Sensors 22 00071 g008
Figure 9. Experiment data set of the optimal value of the objective function calculated by OMO algorithm distribution statistics histogram.
Figure 9. Experiment data set of the optimal value of the objective function calculated by OMO algorithm distribution statistics histogram.
Sensors 22 00071 g009
Figure 10. Speed control constant—OMO algorithm’s accuracy distribution histogram. (a) SH distribution histogram of the first 1000 sets of high-precision data; (b) SH distribution histogram of the first 1000 sets of low-precision data; (c) SR distribution histogram of the first 1000 sets of high-precision data; (d) SR distribution histogram of the first 1000 sets of low-precision data.
Figure 10. Speed control constant—OMO algorithm’s accuracy distribution histogram. (a) SH distribution histogram of the first 1000 sets of high-precision data; (b) SH distribution histogram of the first 1000 sets of low-precision data; (c) SR distribution histogram of the first 1000 sets of high-precision data; (d) SR distribution histogram of the first 1000 sets of low-precision data.
Sensors 22 00071 g010
Table 1. Ammonia, chlorine, and phosgene, 1 h exposure duration. Results—AEGL Program.
Table 1. Ammonia, chlorine, and phosgene, 1 h exposure duration. Results—AEGL Program.
AmmoniaChlorinePhosgene
AEGL-1300.5NR
AEGL-21602.00.3
AEGL-31100200.75
Table 2. Methods for source tracking of chemical gas leakage.
Table 2. Methods for source tracking of chemical gas leakage.
Algorithm
Classification
StrengthsWeaknessesInstances
Swarm Intelligence (SI) algorithms based on gas dispersion modelsHigh degree of freedom in the exploration phase, fast calculation speed, and accurate result in the exploitation phaseInsufficient random exploration may make the result fall into local optimal solutionParticle Swarm Optimization (PSO)
Algorithm [30,31]
Ant Colony Optimization (ACO)
Algorithm [30]
Firefly Algorithm [30]
Harris Hawks Optimization (HHO)
Algorithm [32]
Evolutionary algorithms based on gas dispersion modelsNo need to distinguish the exploration and exploitation phases; controllable search ability (coefficient of variation)High evolutionary generations will reduce population diversityGenetic Algorithm (GA) [17,33]
Differential Evolution (DE)
Algorithm [34]
Direct optimization algorithms based on gas dispersion modelsLow algorithm complexity and fast calculation speedEasy to fall into local optimal solutionPattern Search (PS) Algorithm [35,36]
Simulated Annealing (SA)
Algorithm [37,38,39]
Least Squares Algorithm [15,16]
Methods based on big data or probabilistic analysisNo need to build a scene model, not restricted by geographical conditionsNeed for much prior knowledge and observation data, slow calculation speedDeep Neural Networks [18]
Big Data with Probability Function [19]
Sequential Monte Carlo Methods [40]
K-Nearest Neighbor Classifier [41]
Other methodsAdvanced technology, high accuracyHigh technical and economic requirementsDrone-Enabled Participation [6]
Random Walk Robot Participation [42,43]
Hybrid Optimization Algorithm [44,45,46,47]
Table 3. Solar radiation intensity classification.
Table 3. Solar radiation intensity classification.
Cloud ConditionSolar Radiation Angle α
α > 60°35°< α ≤ 60°15° < α ≤ 35°
Cloud cover 4/8, or thin clouds at high altitudeStrongMediumWeak
Cloud cover 5/8–7/8, cloud height 2134–4877 mMediumWeakWeak
Cloud cover 5/8–7/8, cloud height lower than 2134 mWeakWeakWeak
Table 4. Atmospheric stability classification.
Table 4. Atmospheric stability classification.
Wind SpeedUnder SunshineWithout Sunshine
StrongMediumWeakCloud Cover ≥4/8Cloud Cover ≤3/8
0–2AA–BBFF
2–3A–BBCEF
3–4BB–CCDE
4–6CC–DDDD
>6DDDDD
Table 5. Pasquill–Gifford diffusion coefficient equation.
Table 5. Pasquill–Gifford diffusion coefficient equation.
Stability Level σ y m σ z m
A–B 0.32 x ( 1 + 0.0004 x ) 1 / 2 0.24 x ( 1 + 0.0001 x ) 1 / 2
C 0.22 x ( 1 + 0.0004 x ) 1 / 2 0.20 x
D 0.16 x ( 1 + 0.0004 x ) 1 / 2 0.14 x ( 1 + 0.0003 x ) 1 / 2
E–F 0.11 x ( 1 + 0.0004 x ) 1 / 2 0.08 x ( 1 + 0.0015 x ) 1 / 2
Table 6. Details of the six representative standard test functions.
Table 6. Details of the six representative standard test functions.
Function NameFunction EquationRangeOptimal Solution and Value
Sphere f ( x ) = i = 1 n x i 2 x i 100 0 , x i = 0
Rosenbrock f ( x ) = i = 1 n 1 100 x i + 1 x i 2 2 + 1 x i 2 x i 50 0 , x i = 1
Rastrigin f ( x ) = i = 1 n x i 2 10 cos 2 π x i + 10 x i 50 0 , x i = 0
Griewank f ( x ) = 1 4000 i = 1 n x i 2 i = 1 n cos x i i + 1 x i 30 0 , x i = 0
Schaffer f ( x ) = 0.5 + sin 2 i = 1 n x i 2 0.5 1 + 1 1000 i = 1 n x i 2 2 2 x i 100 0 , x i = 0
Ackley f ( x ) = 20 + e 20 e 0.2 1 n i = 1 n x i 2 e 1 n i = 1 n cos 2 π x i x i 30 0 , x i = 0
Table 7. The parameter settings.
Table 7. The parameter settings.
Optimization AlgorithmParameterValue
DE/GAScaling factor F0.5
Crossover probability Cr0.5
HHOσν1
β1.5
PSInitial point(0, 0, 0)
SAInitial point(0, 0, 0)
PSOInertia factor0.3
c 1 1
c 2 1
Grey Wolf Optimization (GWO) [63]Convergence constant a[2, 0]
Slap Swarm Algorithm (SSA) [64]Convergence constant a[2, 0]
Whale Optimization Algorithm (WOA) [65]Convergence constant c1[2, 0]
Table 8. Qualitative results table of OMO algorithm under 16 sensors.
Table 8. Qualitative results table of OMO algorithm under 16 sensors.
Target ParameterExpected ValueCalculated ValueRelative Error (%)
Time (s)12.743
x 0 (m)88.1581.971
y 0 (m)1515.0200.131
Q (kg/s)8079.0451.193
Optimal value02.464e-15
Table 9. Qualitative results table of OMO algorithm under 9 sensors.
Table 9. Qualitative results table of OMO algorithm under 9 sensors.
Target ParameterExpected ValueCalculated ValueRelative Error (%)
Time (s)10.304
x 0 (m)88.2873.593
y 0 (m)1515.4442.960
Q (kg/s)8076.9093.864
Optimal value02.356e-13
Table 10. Qualitative results table of OMO algorithm under 4 sensors.
Table 10. Qualitative results table of OMO algorithm under 4 sensors.
Target ParameterExpected ValueCalculated ValueRelative Error (%)
Time (s)8.644
x 0 (m)810.72534.068
y 0 (m)1514.1905.401
Q (kg/s)8084.6125.765
Optimal value05.694e-23
Table 11. Qualitative results table of OMO algorithm with a precision of 1e-10 under 16 sensors.
Table 11. Qualitative results table of OMO algorithm with a precision of 1e-10 under 16 sensors.
Target ParameterExpected ValueCalculated ValueRelative Error (%)
Time (s)6.697
x 0 (m)88.2853.568
y 0 (m)1515.0690.459
Q (kg/s)8079.8960.131
Optimal value02.029e-10
Table 12. The relationship between the population number N and the optimal value of the objective function calculated by the OMO algorithm.
Table 12. The relationship between the population number N and the optimal value of the objective function calculated by the OMO algorithm.
Population Number NOptimal ValueTime (s)
38.450e-61.678
52.130e-72.490
89.622e-103.712
103.253e-104.561
151.187e-116.591
207.940e-138.680
257.228e-1510.723
302.464e-1512.743
502.339e-1621.753
1004.073e-1742.479
Table 13. The relationship between the iterations and the optimal value of the objective function calculated by the OMO algorithm.
Table 13. The relationship between the iterations and the optimal value of the objective function calculated by the OMO algorithm.
IterationsOptimal ValueTime (s)
501.384e-102.249
1001.457e-134.405
2003.067e-148.623
3002.464e-1512.743
4007.305e-1616.975
5004.157e-1621.311
10005.238e-1643.213
Table 14. The partial results of the relationship between the speed control constant and the optimal value of the objective function calculated by the OMO algorithm.
Table 14. The partial results of the relationship between the speed control constant and the optimal value of the objective function calculated by the OMO algorithm.
SHSROptimal Value
461.447e-11
2331.690e-11
2432.405e-11
1522.683e-11
133.602e-11
13713.679e-11
4224404.382e-4
7094694.384e-4
4958444.396e-4
869426.804e-4
865767.124e-4
9733297.846e-4
Table 15. The OMO algorithm’s results under different SNR.
Table 15. The OMO algorithm’s results under different SNR.
SNR (dB) x 0   ( m ) y 0   ( m ) Q (kg/s)Optimal Value
014.91612.95661.4065.565e-2
0.315.17314.88654.4006.340e-2
0.611.52013.13082.0523.102e-2
112.12513.92757.6692.614e-2
511.80814.29478.3031.128e-2
109.04714.61873.6113.658e-3
159.56315.17971.5381.071e-3
207.38314.72777.7013.932e-4
308.40515.27180.0328.574e-5
508.17514.97779.1004.209e-7
1007.90614.84881.3723.680e-11
Table 16. Comparison of results of different optimization algorithms under 16 sensors.
Table 16. Comparison of results of different optimization algorithms under 16 sensors.
Target ParametersOMOPSSAPSOGADEHHOGWOSSAWOA
Time (s)12.7434.9103.0282.2236.04910.12912.37313.19613.51413.160
x 0 (m)8.1586.2075.5657.4106.3797.4358.5285.2766.8498.779
y 0 (m)15.02014.50110.90514.17513.89015.02814.53014.74015.02613.188
Q (kg/s)79.04562.40643.92983.69475.82481.03876.95383.62084.95676.095
x 0 relative error (%)1.97122.41830.4337.37920.2627.0646.60634.04414.3919.741
y 0 relative error (%)0.1313.32627.2975.4977.3990.1863.1321.7320.17112.080
Q relative error (%)1.19321.99345.0894.6175.2201.2983.8094.5256.1954.882
Optimal value2.464e-152.309e-69.492e-35.625e-83.302e-43.697e-138.531e-72.712e-76.317e-92.434e-5
Table 17. Comparison of results of different optimization algorithms under 9 sensors.
Table 17. Comparison of results of different optimization algorithms under 9 sensors.
Target ParametersOMOPSSAPSOGADEHHOGWOSSAWOA
Time (s)10.3044.4292.9901.6383.8678.40610.26311.17011.36411.083
x 0 (m)8.2874.11110.6679.5495.6688.8549.6441.8557.3124.460
y 0 (m)15.44412.99012.90813.35213.66115.00115.53310.15013.19012.100
Q (kg/s)76.90995.76252.08190.09879.93885.55473.72292.25083.85383.342
x 0 relative error (%)3.59348.61833.33519.36329.15110.67520.54476.8148.59844.252
y 0 relative error (%)2.96013.40013.94710.9868.9260.0103.55532.33212.06519.336
Q relative error (%)3.86419.70334.89912.6220.0776.9437.84715.3134.8164.178
Optimal value2.356e-139.689e-76.450e-45.182e-86.226e-61.459e-112.012e-74.987e-71.960e-83.161e-4
Table 18. Comparison of results of different optimization algorithms under 4 sensors.
Table 18. Comparison of results of different optimization algorithms under 4 sensors.
Target ParametersOMOPSSAPSOGADEHHOGWOSSAWOA
Time (s)8.6443.2062.9541.3413.0467.2198.8339.3019.5369.390
x 0 (m)10.7252.67712.3692.1016.1264.44610.6693.62812.0614.878
y 0 (m)14.1907.64115.1689.72010.5979.39215.4719.1278.07611.870
Q (kg/s)84.612127.40443.551106.466100.159101.97858.64580.567104.30572.035
x 0 relative error (%)34.06866.53654.60773.73223.42744.42833.36854.65350.76839.026
y 0 relative error (%)5.40149.0601.12135.20029.35437.3863.13739.15146.16020.864
Q relative error (%)5.76559.25545.56233.08225.19927.47226.6940.70930.3829.956
Optimal value5.694e-231.401e-111.384e-65.373e-103.371e-101.075e-119.049e-195.915e-127.136e-142.514e-9
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Xia, Z.; Xu, Z.; Li, D.; Wei, J. A Novel Method for Source Tracking of Chemical Gas Leakage: Outlier Mutation Optimization Algorithm. Sensors 2022, 22, 71. https://doi.org/10.3390/s22010071

AMA Style

Xia Z, Xu Z, Li D, Wei J. A Novel Method for Source Tracking of Chemical Gas Leakage: Outlier Mutation Optimization Algorithm. Sensors. 2022; 22(1):71. https://doi.org/10.3390/s22010071

Chicago/Turabian Style

Xia, Zhiyu, Zhengyi Xu, Dan Li, and Jianming Wei. 2022. "A Novel Method for Source Tracking of Chemical Gas Leakage: Outlier Mutation Optimization Algorithm" Sensors 22, no. 1: 71. https://doi.org/10.3390/s22010071

APA Style

Xia, Z., Xu, Z., Li, D., & Wei, J. (2022). A Novel Method for Source Tracking of Chemical Gas Leakage: Outlier Mutation Optimization Algorithm. Sensors, 22(1), 71. https://doi.org/10.3390/s22010071

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop