Next Article in Journal
Assessments of Roof-Harvested Rainwater in Disctrict Dir Lower, Khyber Pakhtunkhwa Pakistan
Previous Article in Journal
Prediction for the Influence of Guide Vane Opening on the Radial Clearance Sediment Erosion of Runner in a Francis Turbine
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:

A New Approach for Assessing Heat Balance State along a Water Transfer Channel during Winter Periods

School of Civil Engineering, Hefei University of Technology, Hefei 230009, China
School of Engineering, University of Northern British Columbia, Prince George, BC V2N 4Z9, Canada
Power China Beijing Engineering Corporation Limited, Beijing 100024, China
China South-to-North Water Transfer Group Middle Line Co., Ltd., Beijing 100038, China
Yangtze River Scientific Research Institute, Wuhan 430010, China
Authors to whom correspondence should be addressed.
Water 2022, 14(20), 3269;
Submission received: 5 September 2022 / Revised: 6 October 2022 / Accepted: 11 October 2022 / Published: 17 October 2022


Ice problems in channels for water transfer in cold regions seriously affect the capacity and efficiency of water conveyance. Sometimes, ice problems such as ice jams in water transfer channels create risk during winter periods. Recently, water temperature and environmental factors at various cross-sections along the main channel of the middle route of the South-to-North Water Transfer Project in China have been measured. Based on these temperature data, the heat balance state of this water transfer channel has been investigated. A principal component analysis (PCA) method has been used to analyze the complex factors influencing the observed variations of the water temperature, by reducing eigenvector dimension and then extracting the principal component as the input feature. Based on the support vector machine (SVM) theory, a new approach for judging the heat loss or heat gain of flowing water in a channel during winter periods has been developed. The Gaussian radial basis is used as the kernel function in this new approach. Then, parameters have been optimized by means of various methods. Through the supervised machine learning process toward the observed water temperature data, it is found that the air–water temperature difference and thermal conditions are the key factors affecting the heat loss or heat absorption of water body. Results using the proposed method agree well with those of measurements. The changes of water temperature are well predicted using the proposed method together with the state of water heat balance.

1. Introduction

Long-distance water transfer project is an important approach to effectively solve the water shortage problem in large- and medium-sized cities [1]. The total length of the main canal of the Middle Route of the South-to-North Water Transfer Project is 1267 km, as shown in Figure 1. Water is transported from the Danjiangkou Reservoir in the Hanjiang River to the North China Plain [2]. The purpose of this water transfer project is to transfer water to Henan Province, Hebei Province, and the cities of Beijing and Tianjin. The average annual amount of transferred water is 9.5 billion m3 [3]. Since the start of the operation of this water transfer project in 2014, given the strong coupling of cascaded channels, it has been difficult to keep water levels at all gates along the route being stable. The cross-section of the main canal has a trapezoidal shape, with a bottom width range from 7.0 to 26.5 m. Along the water transfer route, there are no lakes or reservoirs for water detention, indicating the importance of safety operation of this project [4]. The main canal of this water transfer line is spanned from the place located at the north latitude of 33° to the north latitude of 40°. During winter periods, the temperature in northern China such as in Beijing is generally low. As a consequence, the formation of ice cover in the water transfer cannel creates a big ice problem [5].
During an ice-covered flow condition, the characteristics of river hydraulics have been dramatically changed compared to that under an open flow condition. An ice cover adds an extra boundary to the flow, which leads to considerable changes in the velocity profile, flow rate, bed shear stress distribution, and water temperature [6]. Water temperature is greatly affected by thermal factors and environmental variables [7]. Under the influence of air temperature, flow and boundary conditions during a winter period [8], water temperature is affected by the hydraulic response to the channel or the complex process of water transport during an ice-covered period. In the meantime, water temperature is an important physical factor affecting the change of river ice regime [9]. Therefore, water temperature is often used as one of important indicators for forecasting river ice regime [10]. To avoid the development and formation of an ice jam during winter periods, the water temperature during the water transfer process should be carefully controlled [11].
Boyd et al. [12] calculated and analyzed the peak temperature and duration of each supercooling event by observing the water temperature of three controlled rivers, Kananaskis, North Saskatchewan and Peace Rivers in Alberta, Canada. They pointed out that the water supercooling process is the key condition for the formation of suspended ice and anchor ice in rivers in cold regions. Osterkamp et al. [13] measured the temperature of water in the Goldstream Creek and Chatanika River in Alaska during the formation of frazil ice particles that forms in turbulent water. Frazil ice particles in flowing water are essential for the development of ice pans which may lead to the development of ice jams and cause disaster [14]. The correlation between the cooling rate of a river and the heat loss of the water surface has been calculated for the periods before and during the formation of frazil ice. It was found that the core of frazil ice in the river was cold organic matter, soil particles or a combination of these substances, which could enter the river through the mass exchange process of the air–water interface. The degree of supercooling of the observed river was given. McFarlane et al. [15] compared five different methods for calculating sensible heat flux and latent heat flux based on air temperature, relative humidity, air pressure, wind speed and wind direction, short-wave and long-wave radiation for the Dauphin River in Manitoba. By calculating the heat balance of the river, the supercooling phenomenon of the river was predicted. They claimed that the solar radiation is an important factor in the heat budget of water bodies. Wang et al. [16] pointed out that both temperature field and velocity field of flowing water during an ice-covered period are essential for studying the formation and evolution of frazil ice. By applying the Navier–Stokes equations and energy equation for flow under an ice cover, the velocity field and temperature field of flowing water have been analyzed. It is found that the temperature field of water body is mainly affected by heat conduction on the surface of ice cover and convective heat transfer on the riverbed. Yang [17] developed both linear and non-linear models for calculating heat exchange between water and atmosphere based on meteorological data including solar radiation, long-wave radiation, evaporation and convective heat exchange. The one-dimensional model for simulating temperature of flowing water in an open channel is developed. The model is verified and applied along some typical channel reaches of the Middle Route of the South-to-North Water Transfer Project (MR-SNWTP).
Field observations along the main canal of the MR-SNWTP indicate that, with the decrease in air temperature and water temperature in winter, frazil ice appears and ice pans form. An ice cover will be gradually developed on water surface of the MR-SNWTP. After the MR-SNWTP is covered by ice, both the growth of ice cover thickness and the decrease in water temperature slowed down. During the stable ice-covered period, the water temperature was low with some minor fluctuations. With the increase in air temperature in spring, river ice starts to melt, and the water temperature begins to rise as well [18]. Since the change of water temperature is also an indicator of the complex nonlinear evolution process of an ice cover during winter periods, the machine learning method with advantages in dealing with nonlinear problems can be considered to predict the trend of water temperature [19]. Sun et al. [20] used the machine learning method to predict the river breakup time and peak flow of the Athabasca River in Fort McMurray, Canada. Their results show that the inclusion of certain inputs in the optimal models can reveal some hints for the potential mechanism on river ice from the data-driven perspective. Seidou et al. [21] applied the artificial neural network and the lake ice thermodynamic model to the study ice cover thickness of several lakes and reservoirs in Canada. It is found that, in case of scarce or less data available for use, result using the artificial neural network has higher prediction accuracy. Guo et al. [22] used the improved back propagation (BP) neural network method to predict the river-breakups in the Heilongjiang and improved the prediction accuracy by integrating multiple factors affecting water temperature and development of ice conditions. The commonly used data-driven prediction models include multiple regression, projection pursuit, and cloud model. These prediction models have achieved a series of results in different engineering cases [23]. The traditional multiple regression model is difficult to accurately describe the complex relationship between independent variables and dependent variables. When the process is highly random, the prediction accuracy is often low. The machine learning models such as projection pursuit and cloud model possess the nonlinear mapping and adaptive ability, and their computational efficiency and prediction accuracy are improved compared to the traditional multiple regression models. However, because the goal of train samples for these models is to minimize the fitting error, it is easy to create problems such as overfitting or local optimality. The support vector machine (SVM) is an algorithm based on the statistical learning theory of small samples, which can be used to solve nonlinear and high dimension problems [24]. Wang et al. [25] predicted the water level during an ice-jammed period by means of the BP neural network and the support vector machine. Their results show that the prediction accuracy using the support vector machine method is higher than that using the BP neural network in case that the sample is small. The principal component analysis (PCA) method can be used to extract independent and effective information from features, which is a statistical analysis method to reduce the input dimension [26]. Ren et al. [27] used the PCA method to eliminate redundant information and effectively analyzed the main factors affecting water temperature in the upper reach of the Yellow River. The identification of the heat balance state of water body in the channel during a winter period has the characteristics of nonlinear process and high dimension, which is in line with the characteristics of the machine learning method. However, no research work has been reported regarding the heat balance of a water body during an ice-covered period.
In summary, the seasonal change of water temperature in rivers affects the formation of river ice. Results of daily field observations at channel cross-sections in winters may be affected by ice conditions in the MR-SNWTP. The heat balance of water body in rivers during a winter period is more complicated. The sluice gate for flow control located in Beijuma of the MR-SNWTP is the junction point of the water transfer channel from an open channel to the culvert, and its location is shown in Figure 1. This channel cross-section of the MR-SNWTP was chosen in this study since relatively more field observation data are available and the temperature in this region is relatively low. Additionally, this cross-section is vulnerable to the ice problem during winter periods. In this study, the machine learning method is used to study the heat balance and water temperature at this cross-section of the MR-SNWTP.

2. Methodology

The change of temperature follows the principle of energy conservation [28], and the main parameters used in the water temperature model are shown in Table 1.
The general equation for describing temperature of water in an open channel is as follows:
t ( ρ w c p A w T w ) + x ( ρ w c p Q w T w ) x ( E x ρ w c p A w T w x ) = B w φ w a
Equation (1) is a mathematical model for describing water temperature field [29]. Symbols in Equation are explained in Table 1. From left to right of Equation (1), it is the time term, the convection term, the diffusion term, and the source term, respectively.
For a channel with a regular cross-section similar to ghat of the MR-SNWTP, the temperature gradient along the flow direction is small, and the diffusion term can often be ignored. Combined with the continuity equation of the flow, by assuming the flow is incompressible with the fixed specific heat capacity, Equation (1) can be simplified. In the presence of an ice cover on water surface (either partially or fully covered), both the boundary conditions and flow conditions for the model have been changed, resulting in a significant impact on the water temperature [30]. The exchange of heat occurs between the atmosphere, ice cover and water, which is related to the coverage of ice cover on water surface. Therefore, the equation for describing temperature of water can be rewritten as follows:
t ( ρ w c p A w T w ) + x ( ρ w c p Q w T w ) x ( E x ρ w c p A w T w x ) = B w [ N i φ w i + ( 1 N i ) φ w a ]
where Ni represents the extent that the cross-section is covered by ice. When an ice cover is present in a channel, the heat transfer of the water body is closely related to the heat loss of the ice body. The above equations characterize the physical mechanism of the heat transfer of the water body during a winter period [31].
The heat exchange between atmosphere and water is affected by weather conditions, solar radiation and wind speed. In late fall or early winter, with the decrease in air temperature, cooling intensity will be increased. Therefore, heat loss from water body results in the decrease in water temperature until the water surface is covered by an ice cover which is normally formed by frazil ice particle and ice pans in rivers. After the entire water surface is covered by ice, the water temperature continues to fluctuate and drop, and the water temperature gradient at the front of the ice sheet is large. During the stable ice-covered period, with the decrease in air temperature, the thickness of the ice cover gradually increases, and the heat exchange between water and air decreases. During this stable ice-covered period, the water temperature can reach a dynamic equilibrium condition with some minor fluctuations. In the spring, with the increase in the air temperature and solar radiation, an ice-covered river undergoes the thermodynamic breakup process (ice cover melting process). The water body gradually absorbs heat and results in the increase in temperature of water body. Since the MR-SNWTP is 1267 km long and spans a latitude difference of 7°, there will be some changes in the temperature of water body along the MR-SNWTP. The change of air temperature and that of water temperature has an obvious nonlinear relationship. The continuous change of meteorological elements in winter and spring affects the relevant parameters in Equation (2).
Equations (1) and (2) show the relationship between the change of water temperature and other factors. Since the parameters in these equations vary with various factors including the environmental variables, the analytical solution of these equations will be very complicated. The hydrodynamic process also affects the heat balance of the water body. When the flow velocity is high enough, shore ice along channel banks is normally difficult to be developed. Additionally, the relatively larger flow velocity during a freeze-up period is likely to cause the incoming ice pans/floes to be entrained and submerged under the sheet ice cover to form an ice jam. During the break-up period, the surface of the entire channel (together with pools) is grouped into the open channel section, ice-cover melting section and the fully ice-covered section. Along the melting section of the channel, ice cover becomes soft and loose with a lot of pores. Under the influence of hydrodynamic force, this loose ice cover is gradually eroded, and the wavy-shaped surface appears on the bottom of an ice cover, which eventually causes the ice cover to be melted, broken, and transported downstream. The hydrodynamics of the flowing water and the ice dynamics affect the heat transfer of water and changes of water temperature.
By taking the thermal and dynamic factors that affect the change of temperature of water body as the input features, an SVM-based discriminant model for water heat balance is established to assess the heat loss and heat gain of water body. In this study, data were obtained through on-site observations. Among them, the observed data for water flow conditions at some cross-sections near the control gate have been acquired from the management department of the MR-SNWTP. The temperature sensor and flow velocity meter have been used to collect data for describing water flow conditions, which were measured 4 times every day. The meteorological data were acquired from the meteorological station which was setup by considering the locations for observations of the water flow. The number of daily measurements of meteorological data is consistent with that of for water flow. The daily average approach is adopted to form a dataset for subsequent analysis. These environmental variables have different effects on the latent heat flux and the sensible heat flux, which are correlated to the heat balance. The correlation analysis of each characteristic data has been carried out, and the correlation matrix of the measured dataset is established. The calculated correlation coefficient is presented in Figure 2. The corresponding thermal factors and dynamic factors are abbreviated to better express the figure. These factors include: the temperature difference between air and water (Air and Water TEMP diff), the previous water temperature (Pre-water TEMP), the cumulated seven-day average daily air temperature (7d Air TEMP CUM), the time effect (TM EFF), the maximum air temperature (Max Air TEMP), the minimum air temperature (Min Air TEMP), the difference between wind speed and water velocity (Wind S and VEL diff), the flow Froude number of water flow (Fr), the solar radiation (SR), the air pressure (Air PRESS), the cloudiness (Cloudiness), the weather conditions (WEA COND), the cloud height (Cloud H), and the precipitation (Precipitation).
In Figure 2, the correlation coefficient of 1 indicates that the two features are completely linear positive correlated; the correlation coefficient of −1 indicates that the two features are completely linear negative correlation; and the correlation coefficient of 0 indicates that the two features are independent. It can be seen from Figure 2 that there is a certain degree of correlation among the influencing features, that is, there is information overlap between the input features. The commonly used methods for data dimension reduction include single-factor analysis, grey system analysis, set pair analysis, analytic hierarchy process and principal component analysis. Results of single-factor analysis have polarization problem. The dimension reduction methods based on grey system theory, set pair analysis theory and analytic hierarchy process are controversial in the selection of index weight, and the application of the model automatic calculation is limited. The core idea of the PCA method is to reduce the high-dimensional associated features to a few unrelated features, and to reflect the original information as much as possible. The PCA method can be used to process the standardized sample data, extract the main information of input variables, and improve the accuracy of classification based on comprehensive consideration of various influencing factors [32]. To ensure that the data have the same importance as the input scale, the daily average dataset of the selected measured parameters and measurement records are pre-processed by the standardized method, and the covariance matrix S is calculated according to Equation (3):
S = 1 n 1 X X T
where n is the number of sample data, and X is the normalized sample matrix. By solving the characteristic equation, m non-negative features λk (k = 1, 2, ⋯, m) of the covariance matrix S are obtained and arranged in the order of λ1 > λ2 > ⋯ >λm > 0. The corresponding orthogonal unit eigenvector µk is solved, and the formula of principal component calculation is as follows:
Z k = μ k T X
In Equation (4), Zk represents the kth principal component (km), and the contribution rate vi of the ith principal component is calculated using Equation (5):
v i = λ i / k = 1 m λ k
The upper limit of machine learning effect is controlled by the available data. The standardized input data are divided into training and verification sets. To reduce the difference introduced by different sample division, the model is trained by the cross-validation method. For the part of training set, it is divided equally and one of them is selected as the test set, and the rest is used as the training set to complete the cross-validation process.
The SVM algorithm is based on the principle of minimization of structural risk. Regarding the classification of samples, the sample dataset is set as Zi = [Zi1; Zi2], yi is the corresponding label of Zi. The heat gain of water body is labeled as “+1”, and the heat loss of water body is labeled as “−1”. The hyperplane formula for the sample classification is expressed in Equation (6), and the classification decision function is described as Equation (7) [33]:
w T η ( Z ) + b = 0
f ( Z ) = sign ( w T η ( Z ) + b )
where w = (w1; w2; ⋯;wm) is the weight vector corresponding to Z, b is the displacement term which determines the distance between the hyperplane and the origin. The sign ( ) represents the sign function. The Lagrangian function for solving this problem can be described as Equation (8):
min L ( w , b , η , α ) = 1 2 w 2 i = 1 p α i [ y i ( w T η ( Z ) + b ) 1 ]
where α = (α1; α2; ⋯; αp) is a Lagrangian operator, αi > 0. By solving Equation (8) and substituting it into Equations (7) and (9) is obtained:
{ f ( Z ) = i = 1 p α i y i η ( Z i ) T η ( Z h ) + b s . t . { α i 0 ; i = 1 , 2 , , p i = 1 p α i y i = 0
The standardized sample data are inseparable in the original space. To solve this problem, the standardized sample data are mapped and transformed in the original space, and the value of the transformed space inner product function can be transformed into the value directly calculated using the kernel function. The calculation formula is as follows:
k ( Z i , Z h ) = η ( Z i ) T η ( Z h )
where η (Zi) and η (Zh) are the mapping transformation functions of the original space; and k (Zi, Zh) is the kernel function. When meteorological data are relatively complete, based on the observed thermal-hydrodynamic data about winter ice-water regime of the MR-SNWTP, the heat budget process of water body and the mutual feedback relationship of the related parameters are identified.

3. Analysis of Heat Budget of Water Body

The MR-SNWTP delivers water from the low latitude to high latitude. In winter, it is necessary to identify the influence of both thermal and dynamic factors of flowing water on the heat balance of flowing water in the main canal of the MR-SNWTP. The key to the study is how to consider the influence of temperature, solar radiation, cloud cover, precipitation, wind speed and other factors on the heat balance process of flowing water.
The supercooling process caused by heat loss of water body is a critical process for ice formation in channels. Hence, it is proposed to use environmental variables to predict the correct direction of water temperature change. Then, by considering the difference between calculation accuracy and complexity, the parameter adjustment algorithm was selected to assess the heat balance state of water body. The cumulative temperature during a certain time period is often used for analyzing river ice hydrology [34]. The difference between air temperature and water temperature, the cumulative air temperature, and wind speed and velocity gradient were selected for analyzing the impact these factors on the heat balance of water body, as shown in Figure 3.
It can be seen from Figure 3 that, under condition of the heat balance of water body, both thermal factors and dynamic factors contribute to the heat transfer of water body, and it is difficult to use a linear segmentation to directly separate the heat transfer caused by thermal factors from that by dynamic factors. This is the characteristics of the heat budget of water body in river during a winter period. It has also become difficult to accurately calculate the heat exchange of flowing water in winter by using both theoretical equations and numerical simulation. Due to the mixing process of flowing water and latent heat from the phase transition from liquid water to solid ice, the heat balance state of water body cannot be simply expressed as a good linear function of either dynamic or thermal condition. In the case of being linearly inseparable, the samples in the low-dimensional space are mapped to a high-dimensional feature space by using a nonlinear mapping function to make it to be linear. Then, it is possible to use the linear algorithm to analyze the nonlinearity of samples in the high-dimensional feature space, and the optimal classification hyperplane can be found in this feature space. The so-called optimal classification surface not only requires to correctly separate the categories, but also needs to maximize the classification interval. The purpose for correctly separation of categories is to ensure that the empirical risk is minimized. Actually, the maximum classification interval is to minimize the confidence range in the generalization bound.
Considering the complexity and coupling of the water heat budget process, the Gaussian radial basis function (RBF) is used as the kernel function of the support vector machine:
k ( Z i , Z h ) = exp ( γ Z i Z h 2 )
In Equation (11), Zi and Zh represent different input characteristics, and γ is the Gaussian kernel bandwidth parameter. By introducing the multiplier α i * , and the penalty parameter C, a slack variable ε needs to be added to the threshold to reduce the error of the dual problem. The slack variable represents that the error term whose deviation is less than ε is not penalized, we can get:
min L ( w , b , η , α , ε i , α * ) = 1 2 w 2 + C i = 1 p ε i i = 1 p α i [ y i ( w T k ( Z i , Z h ) + b ) 1 ε i ] i = 1 p α i * ε i
The parameter C represents the degree of punishment for prediction errors. The larger the C value, the more the model does not allow prediction errors. If the C value is too large, the model will make fewer errors in the training data process and easily cause overfitting. On the contrary, if the C value is too small, the model will easily ignore the prediction error, which will lead to a poor performance of the model. The value of γ will influence the range of the Gaussian function corresponding to each supporting vector. The weight of high-order features decays very fast, so that, the larger the γ value, the less the support vectors; the smaller the γ value, the more the support vectors. The number of support vectors will affect the speed of the training and prediction process. The parameters C and γ jointly determine the prediction results of the model. Appropriate selection of their optimal combination can lead to the good generalization ability and fitting effect of the model.
The selection of values of parameters in the kernel function is the key to the calculation accuracy. Thus, the optimization of parameters is needed to determine the values of C and γ. The commonly used methods for parameter adjustment include the grid search (GS) method, the particle swarm optimization (PSO) method, and the genetic algorithm (GA) method. The grid search method is to use the search data to form a grid space, and then obtain the value for each grid point using the exhaustive method. The concept about the PSO method is that each particle contains two attributes of speed and position, each particle contains a fitness value determined by the objective function and knows the best position and the current position found by itself until now. In the whole group, the best target position found by all individuals is also known by each particle. How each individual particle proceeds to the next position is determined by the parallel search of the population. The GA method is a random search method by means of the evolutionary process of the biological world. Its main characteristics is to directly operate the data to be optimized. It can automatically obtain and guide the optimized search space, adaptively adjust the search direction, and has strong robustness. As mentioned above, the heat exchange of water body during an ice-covered period is a complex process and affected by many factors, which can be expressed as:
d T w d t = g ( T w h , T a s , T t , δ v , F r , P , R a , R e l , R p , N , H c )
where Twh represents the historical conditions of water temperature; Tas represents air temperature conditions including daily maximum, minimum, average temperature and cumulative temperature during a period; Tt represents the time-effect factor. Since the variation of water temperature approximately shows a periodic function and it fluctuates around the average annual water temperature, both sin (2πTday/Tyear) and cos (2πTday/Tyear) are selected to reflect the time-effect Tt; where Tday is the date count of the year (note: 1 January is 1, etc.); and Tyear is the number of annual dates; δv represents the component gradient of wind speed and flow velocity in the flow direction, which has an important influence on the heat transfer process; Fr is the flow Froude number; P is the average daily pressure; Ra is the solar radiation; Rel is the daily average relative humidity; Rp is the average daily precipitation; N is cloud amount; Hc is the cloud height. These parameters have either direct or indirect impact on the change of water temperature. Results of field observation showed that the changes of water temperature are compatible with the simulations of the collected data. If g > 0, it indicates that the water body absorbs heat and the water temperature rises; if g < 0, water body loses heat and water temperature drops. After standardizing parameters in Equation (10) based on data measured at the typical survey stations, the PCA was used to extract the first principal component and the second principal component as the input of the SVM. The cumulative contribution rate of each principal component output by using the PCA is shown in Table 2.
According to the cumulative contribution rate of each principal component in Table 2, results showed that the result for the first two principal components can explain more than 50% of the information for original data; the information for the first five principal components can explain more than 80% of the information for original data; the information for the first seven principal components can explain more than 90% of the information for original data. With respect to the prediction of water temperature and heat balance state, based on the data measured at the typical survey section during six winters from 2015 to 2021, the measured data from 2015 to 2020 are used as the training samples to establish the model and optimize the model parameters, but the measured data from 2020 to 2021 are used as the verification samples to test the capacity for generalization of the model. The flow chart for calculation for the assessment of heat balance state is shown in Figure 4.
Since the annual accumulated negative temperature of the training sample varies in a certain range, it can be used to describe different winter characteristics. During the winter period of 2019~2020, the Huinanzhuang pumping station (on the Beijuma River) was under maintenance. The measured data for water transport in the Beiyishui River during an ice-covered period are used to act as the observation data of the typical survey section, so that the training set also has a certain coverage on the hydraulic characteristics. PCA-1 is selected as the first axis and PCA-2 as the second axis. By applying the GS, PSO and GA methods, results of prediction of classification are generated (Figure 5), and the corresponding parameter values are shown in Table 3. When the direction of water temperature change (heat gain vs. heat loss) is correctly identified, it is classified as a successful prediction.
In Figure 5, log2C and log2γare the logarithms of parameters C and γ, respectively. According to the calculation process of Equations (3) and (4), the composition of the principal components includes all the influencing factors, and the contribution of each factor is different. Due to the different features of various algorithms, the combination of parameters C and γ obtained by the optimization also fluctuates within a certain range. It can be seen from Figure 5 that the generated nonlinear partition hyperplane can better distinguish the heat budget of water body. In the prediction model, 63 out of the 89 classified data were predicted correctly, and the accuracy reached at 70.79%. From the number of iterations, the GS method has fewer iterations with a higher optimization efficiency. In view of the cross-validation rate, the GA algorithm performs well in the model training, and the number of the support vectors is less. By observing the execution time of the algorithm, with respect to the operation management of water transfer process, the calculation time consumed using these three algorithms is within the acceptable range, and the time need using the GA algorithm is slightly higher. Furthermore, by selecting the first five principal components in order to adequately reflect the influence of the original correlation factors, the SVM is used to predict the heat balance state of water body. The prediction results are summarized in Table 4.
With the increases in the number of principal components extracted, the contribution of each component in the model also increases accordingly. It can be seen from Table 4 that the comprehensive impact factor formed by more principal components can reflect more information about the original data, and the prediction accuracy of the model can be further improved. It has been found that the prediction results by means of two bionic algorithms, namely the PSO and GA methods, are better than those using the GS method. Results showed that the number of support vectors generated by the PSO algorithm is less. From Table 3 and Table 4, once can see that there are differences in computational performance using different approaches. When the number of input data are increased, the time of iterations also increases. This study is concerned with the exploration of the heat balance state of the MR-SNWTP. In practical application, different parameters for describing the performance need to be trade-off according to the actual situation of the project. Among these three methods, the times of iterations using the GA algorithm is relatively small, and the cross-validation rate of this method is convenient and outstanding. This means that the GA algorithm method is clearly suitable for the parameter optimization process of the SVM model in this study. The changes of water temperature have been well predicted by means of the proposed method together with the state of water heat balance.
According to the classification results of the model, the prediction results showed that water body began to absorb heat from 13 January 2021. At this time, the air temperature began to maintain above 0 °C with the highest air temperature of about 5 °C when the weather became sunny with sufficient solar radiation and relative low humidity. On 17 January 2021, the main channel of the Beijuma River survey section eventually broke up. After the breakup of the channel along this survey section, due to the release of latent heat from the upstream inflow and ice, the water body experienced a repeatedly heat gain -heat loss process. By 21 January 2021, the main channel was eventually opened, and the water body continued to absorb heat, causing the water temperature to rise. Additionally, then, the remaining shore ice vanished. As a machine learning method, the SVM model can mine the potential knowledge information of the data and complete the classification and prediction of the data according to the features of datasets. In this study, the feasibility of the proposed approach has been verified by using results of the actual water transfer project. Result of this study can provide reference for engineers in water resources engineering. To apply it, by replacing the input data and the value of the kernel parameters, the trend of variation of water temperature can be identified, and the state of water heat budget can be classified. The proposed method in this study can be promising for controlling ice problems in channels for water transfer in cold regions and can provide the operation guidelines for a river system under an ice-covered flow condition in winter.

4. Conclusions

In this study, starting with the thermodynamic characteristics of water body during winter periods, the heat loss and heat gain process of water body along a typical section of the MR-SNWTP has been studied. The following conclusions have been drawn from this study:
By analyzing the factors that influence water heat balance during winter periods, the correlation coefficient matrix of thermal and hydrodynamic characteristic data is studied. By using the PCA method to extract principal components as model input, the correlation between multiple variables was described by a few variables. By inputting data using the machine learning model, it can effectively reduce data dimension and eliminate redundancy, and thus improve the computational efficiency. With the increases in the number of principal components extracted, the prediction accuracy of the model increases accordingly.
Regarding the selection of the SVM parameters, the GS algorithm, the PSO algorithm and the GA algorithm are selected to optimize the parameters, which are applied to identify heat loss or heat gain of water body in channels during winter periods. After the optimization of parameters, the recognition rate of model prediction algorithm is better. By comparing the differences between the algorithms, the GA algorithm is more suitable for the SVM method used for the assessment of water heat balance state.
Aiming at the problem of heat budget of water body in the study channel in winter, considering water temperature change by means of environmental variables, a new approach is proposed to assess the heat balance state by analyzing the observed data in the field. In view of the nonlinear change of water temperature, the RBF kernel function with better performance in classification is selected. Regarding the sample learning, the SVM can quickly and accurately conduct classification. Thus, the SVM can used to effectively solve the identification problem of water heat exchange.
The changes of water temperature are crucial for the study of ice formation and evolution. The study of the heat balance state associated with flow water body and ice cover involves hydraulics, thermodynamics, meteorology, engineering mechanics and other disciplines. When the prediction is about other water-ice problems, it is necessary to analyze the physical laws and causality of the research content, and use machine learning to assist these prior knowledge to generate an appropriate solution. Based on the theory of this study, the simulation and prediction of water temperature in winter can be carried out, or the modified multi-classification study regarding the ice-water mechanics can be conducted by considering different ice conditions, so as to provide the operation guidelines for a river system under an ice-covered flow condition in winter. This will also be our future research topic. The method proposed in this study can be promising for controlling ice problems in channels for water transfer in cold regions.

Author Contributions

Conceptualization, T.C., J.W. and J.S.; methodology, T.C.; validation, H.Z., Z.H. and M.H.; formal analysis, T.C. and Z.L.; investigation, J.W. and J.S.; writing—original draft preparation, T.C.; writing—review and editing, J.S.; funding acquisition, J.W. All authors have read and agreed to the published version of the manuscript.


This research was funded by the National Natural Science Foundation of China, grant number 51879065; the National Key Research and Development Program of China, grant number 2018YFC1508401.


We thank the National Natural Science Foundation of China, grant number 51879065; the National Key Research and Development Program of China, grant number 2018YFC1508401. The authors are grateful for the financial support.

Conflicts of Interest

The authors declare no conflict of interest.


  1. Zhu, J.; Zhang, Z.; Lei, X.; Yue, X.; Xiang, X.; Wang, H.; Ye, M. Optimal Regulation of the Cascade Gates Group Water Diversion Project in a Flow Adjustment Period. Water 2021, 13, 2825. [Google Scholar] [CrossRef]
  2. Tang, C.; Yi, Y.; Yang, Z.; Cheng, X. Water pollution risk simulation and prediction in the main canal of the South-to-North Water Transfer Project. J. Hydrol. 2014, 519, 2111–2120. [Google Scholar] [CrossRef]
  3. Li, H.; Ji, L.; Li, F.; Li, H.; Sun, Q.; Li, Z.; Yan, H.; Guan, W.; Wang, L.; Ma, Y. Operational Safety Risk Assessment for the Water Channels of the South-to-North Water Diversion Project Based on TODIM-FMEA. Complexity 2020, 2020, 6691764. [Google Scholar] [CrossRef]
  4. Zhou, L.; Zhang, Z.; Zhang, W.; An, K.; Lei, X.; He, M. Real-Time Water Level Prediction in Open Channel Water Transfer Projects Based on Time Series Similarity. Water 2022, 14, 2070. [Google Scholar] [CrossRef]
  5. Sui, J.; Karney, B.W.; Sun, Z.; Wang, D. Field Investigation of Frazil Jam Evolution: A Case Study. J. Hydraul. Eng. 2002, 128, 781–787. [Google Scholar] [CrossRef]
  6. Sui, J.; Thring, R.; Karney, B.W.; Wang, J. Effects of river ice on stage-discharge relationships—A case study of the Yellow River. Int. J. Sediment Res. 2007, 22, 263. [Google Scholar]
  7. Straneo, F.; Heimbach, P. North Atlantic warming and the retreat of Greenland’s outlet glaciers. Nature 2013, 504, 36–43. [Google Scholar] [CrossRef]
  8. Cheng, T.; Wang, J.; Sui, J. Calculation of critical flow depth using method of algebraic inequality. J. Hydrol. Hydromech. 2018, 66, 316–322. [Google Scholar] [CrossRef] [Green Version]
  9. Wang, J.; Wu, Y.; Sui, J.; Karney, B. Formation and movement of ice accumulation waves under ice cover—An experimental study. J. Hydrol. Hydromech. 2019, 67, 171–178. [Google Scholar] [CrossRef] [Green Version]
  10. Ohshima, K.I.; Riser, S.C.; Wakatsuchi, M. Mixed layer evolution in the Sea of Okhotsk observed with profiling floats and its relation to sea ice formation. Geophys. Res. Lett. 2005, 32, 83–100. [Google Scholar] [CrossRef]
  11. Hu, H.; Wang, J.; Cheng, T.; Hou, Z.; Sui, J. Channel Bed Deformation and Ice Jam Evolution around Bridge Piers. Water 2022, 14, 1766. [Google Scholar] [CrossRef]
  12. Boyd, S.; Ghobrial, T.; Loewen, M.; Jasek, M.; Evans, J. A study of supercooling in rivers. Cold Reg. Sci. Technol. 2022, 194, 103455. [Google Scholar] [CrossRef]
  13. Osterkamp, T.E.; Gilfilian, R.E.; Gosink, J.P.; Benson, C.S. Water temperature measurements in turbulent streams during periods of frazil-ice formation. Ann. Glaciol. 1983, 4, 209–215. [Google Scholar] [CrossRef]
  14. Zhang, Y.; Li, Z.; Xiu, Y.; Li, C.; Zhang, B.; Deng, Y. Microstructural Characteristics of Frazil Particles and the Physical Properties of Frazil Ice in the Yellow River, China. Crystals 2021, 11, 617. [Google Scholar] [CrossRef]
  15. McFarlane, V.; Clark, S.P. A detailed energy budget analysis of river supercooling and the importance of accurately quanti-fying net radiation to predict ice formation. Hydrol. Process. 2021, 35, e14056. [Google Scholar] [CrossRef]
  16. Wang, J.; Zhang, L. Study of the velocity field and temperature field of laminar flow under ice cover. J. Hefei Univ. Technol. (Nat. Sci.) 2003, 6, 1192–1194. (In Chinese) [Google Scholar]
  17. Yang, K. Water temperature models of open channels in glacial period. J. Hydraul. Eng. 2022, 53, 20–30. (In Chinese) [Google Scholar]
  18. Cheng, T.-J.; Wang, J.; Chen, P.-P.; Sui, J. Simulation of ice accumulation around bridge piers during river breakup periods using a discrete element model. J. Hydrodyn. 2022, 34, 94–105. [Google Scholar] [CrossRef]
  19. Ibrahim, K.S.M.H.; Huang, Y.F.; Ahmed, A.N.; Koo, C.H.; El-Shafie, A. A review of the hybrid artificial intelligence and optimization modelling of hydrological streamflow forecasting. Alex. Eng. J. 2022, 61, 279–303. [Google Scholar] [CrossRef]
  20. Sun, W.; Lv, Y.; Li, G.; Chen, Y. Modeling River Ice Breakup Dates by k-Nearest Neighbor Ensemble. Water 2020, 12, 220. [Google Scholar] [CrossRef] [Green Version]
  21. Seidou, O.; Ouarda, T.B.M.J.; Bilodeau, L.; Hessami, M.; St-Hilaire, A.; Bruneau, P. Modeling ice growth on Canadian lakes using artificial neural networks. Water Resour. Res. 2006, 42, W11407. [Google Scholar] [CrossRef] [Green Version]
  22. Guo, X.; Wang, T.; Fu, H.; Guo, Y.; Li, J. Ice-Jam Forecasting during River Breakup Based on Neural Network Theory. J. Cold Reg. Eng. 2018, 32, 04018010. [Google Scholar] [CrossRef]
  23. Sang, L.; Wang, J.C.; Sui, J.; Dziedzic, M. A New Approach for Dam Safety Assessment Using the Extended Cloud Model. Water Resour. Manag. 2022, 1–14. [Google Scholar] [CrossRef]
  24. Ma, W.; Liu, H. Classification method based on the deep structure and least squares support vector machine. Electron. Lett. 2020, 56, 538–541. [Google Scholar] [CrossRef]
  25. Wang, J.; Sui, J.; Guo, L.; Karney, B.W.; Jüpner, R. Forecast of water level and ice jam thickness using the back propagation neural network and support vector machine methods. Int. J. Environ. Sci. Technol. 2010, 7, 215–224. [Google Scholar] [CrossRef] [Green Version]
  26. Galelli, S.; Humphrey, G.B.; Maier, H.R.; Castelletti, A.; Dandy, G.C.; Gibbs, M.S. An evaluation framework for input variable selection algorithms for environ-mental data-driven models. Environ. Model. Softw. 2014, 62, 33–51. [Google Scholar] [CrossRef] [Green Version]
  27. Ren, L.; Song, C.; Wu, W.; Guo, M.; Zhou, X. Reservoir effects on the variations of the water temperature in the upper Yellow River, China, using principal component analysis. J. Environ. Manag. 2020, 262, 110339. [Google Scholar] [CrossRef] [PubMed]
  28. Mir, A.; Almanjahie, I.M.; Dar, J.G. Energy balance approach to study the role of perspiration in heat distribution of human skin. Comput. Math. Methods Med. 2020, 2020, 3154908. [Google Scholar] [CrossRef] [PubMed]
  29. Turkyilmazoglu, M. Heat transfer from warm water to a moving foot in a footbath. Appl. Therm. Eng. 2016, 98, 280–287. [Google Scholar] [CrossRef]
  30. Turkyilmazoglu, M. Heat transport in shear-driven flow with axial conduction. J. Taiwan Inst. Chem. Eng. 2021, 123, 96–103. [Google Scholar] [CrossRef]
  31. Yang, S.; Su, H.; Dou, X.; Chen, M.; Huang, Y. Heat Balance Calculation and Energy Efficiency Analysis for Building Clusters Based on Psychrometric Chart. Sensors 2021, 21, 7606. [Google Scholar] [CrossRef] [PubMed]
  32. Carreau, J.; Guinot, V. A PCA spatial pattern based artificial neural network downscaling model for urban flood hazard as-sessment. Adv. Water Resour. 2021, 147, 103821. [Google Scholar] [CrossRef]
  33. Lv, Y.J.; Wang, J.W.; Wang, J.; Xiong, C.; Zou, L.; Li, L.; Li, D.W. Steel corrosion prediction based on support vector machines. Chaos Solitons Fractals 2020, 136, 109807. [Google Scholar] [CrossRef]
  34. Graf, R.; Tomczyk, A.M. The impact of cumulative negative air temperature degree-days on the appearance of ice cover on a river in relation to atmospheric circulation. Atmosphere 2018, 9, 204. [Google Scholar] [CrossRef]
Figure 1. Overview of study area.
Figure 1. Overview of study area.
Water 14 03269 g001
Figure 2. Diagram for correlation analysis of influencing factors of water temperature.
Figure 2. Diagram for correlation analysis of influencing factors of water temperature.
Water 14 03269 g002
Figure 3. Standardized influencing factors and water heat budget analysis.
Figure 3. Standardized influencing factors and water heat budget analysis.
Water 14 03269 g003
Figure 4. The calculation flow chart.
Figure 4. The calculation flow chart.
Water 14 03269 g004
Figure 5. (a) the PCA-GS-SVM classification; (b) parameter optimization using the GS method; (c) the PCA-PSO-SVM classification; (d) the fitness curve of the PSO algorithm; (e) the PCA-GA-SVM classification; (f) the fitness curve of the GA algorithm.
Figure 5. (a) the PCA-GS-SVM classification; (b) parameter optimization using the GS method; (c) the PCA-PSO-SVM classification; (d) the fitness curve of the PSO algorithm; (e) the PCA-GA-SVM classification; (f) the fitness curve of the GA algorithm.
Water 14 03269 g005
Table 1. The main parameters used in the water temperature model.
Table 1. The main parameters used in the water temperature model.
Parameter InterpretationParameterInterpretation
ρwmass density of watercpspecific heat capacity of water
Twwater temperatureQwflow discharge of water
Awflow cross-sectional area of the channelExdiffusion coefficient
Bwwidth of water surface channel cross-sectionφwawater—air heat flux
Nithe ratio of the length of the water surface of the cross-section covered by ice cover to the width of the water surface of the same cross-sectionφwiwater—ice cover heat flux
Table 2. Cumulative contribution rate of the principal components.
Table 2. Cumulative contribution rate of the principal components.
cumulative contribution rate33%52%68%77%83%89%93%
Table 3. The PCA-1~PCA-2 parameter values using different methods.
Table 3. The PCA-1~PCA-2 parameter values using different methods.
Iteration times276286291
Execution time (ms)83,13983,08586,598
Cross-validation rate71.83%72.02%71.46%
Number of boundary support vectors363380349
Number of support vectors385398374
Number of classification predictions898989
Correct classification number636363
Classification accuracy70.79%70.79%70.79%
Table 4. PCA-1~PCA-5 parameter values using different methods.
Table 4. PCA-1~PCA-5 parameter values using different methods.
Iteration times38932146950
Execution time (ms)87,28896,07287,401
Cross-validation rate72.02%71.08%72.02%
Number of boundary support vectors260244257
Number of support vectors293286291
Number of classification predictions898989
Correct classification number646565
Classification accuracy71.91%73.03%73.03%
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Cheng, T.; Wang, J.; Sui, J.; Zhao, H.; Hao, Z.; Huang, M.; Li, Z. A New Approach for Assessing Heat Balance State along a Water Transfer Channel during Winter Periods. Water 2022, 14, 3269.

AMA Style

Cheng T, Wang J, Sui J, Zhao H, Hao Z, Huang M, Li Z. A New Approach for Assessing Heat Balance State along a Water Transfer Channel during Winter Periods. Water. 2022; 14(20):3269.

Chicago/Turabian Style

Cheng, Tiejie, Jun Wang, Jueyi Sui, Haijing Zhao, Zejia Hao, Minghai Huang, and Zhicong Li. 2022. "A New Approach for Assessing Heat Balance State along a Water Transfer Channel during Winter Periods" Water 14, no. 20: 3269.

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