Design of Sensor Networks for Chemical Plants Based on Meta-Heuristics
Abstract
:1. Introduction
2. Sensor Network Design and Upgrade Problem
3. Tabu Search Approach
3.1 An Overview
3.2 Special features of Tabu Search-based strategies for Sensor Network Design
3.2.1 Initial Solution
3.2.2 Neighbourhood
3.2.3 Tabu Lists
3.2.4 Evaluation Function
3.2.5 Aspiration and Termination Criterion
3.2.6 Strategic Oscillations within the framework of Tabu Search
3.2.7 Path relinking within the framework of Tabu Search
- a)
- A set P is made up of each solution that, at some stage of the TS phase, improves the best current solution and becomes the best one
- b)
- The first half of R is loaded with the best solution vectors p from set P
- c)
- For each solution vector p that belongs to P but is not included in R, that is p ∈ {P/R}, the Hamming distance, d, between p and R is computed
- d)
- The second half of R is loaded with p ∈ {P/R} that maximizes the Hamming distance. The IS and GS are defined as the worst and best solutions in R, respectively. Regarding the rules to identify the neighbour structure and the guiding attributes, two types of moves are proposed:
- a)
- Let us consider qi and qg represent the current IS and GS, respectively. Starting from qi, the idea is to generate neighbours that tend to qg at last. To generate the first neighbour, n1, it is set equal to qi. Then the elements of both vectors are compared from left to right until a difference is found. For this position, for example the k-th position, the element of qi is replaced by the corresponding element of qg. The comparison finishes and the first neighbour is obtained. The same procedure is repeated between n1 and qg to obtain n2 but in this case the comparison starts at the (k+1) position. The rest of the neighbours are obtained in the same way until the last neighbour has only one element that differs with respect to qg.
- b)
- The rationale behind this move is to build the neighbourhood by changing the position of a measurement. Consequently the total number of measurements remains unchanged. At first, the arithmetic difference between vectors qg and qi is calculated for each element. Positive (+1), negative (-1) and zero differences are obtained. The positions of positive and negative differences are registered in position vectors pp and np. Then all combinations between each element of pp and all the elements of np form a set of interchanges between the corresponding elements of qi. Each interchange generates a neighbour.
3.3 Tabu Search based procedures
3.3.1 Classic Tabu Search
Generate an initial solution q0 Set q*=q=q0 and F(q*) = F(q0) for i =1 to # Max Iter do Generate neighbourhood N(q) Select q’ ∈ N(q) with the lowest F value If q’ ∈ N(q) satisfies the aspiration criterion F(q’) < F(q*) Set q*=q’ and F(q*) = F(q’) else Select a new solution q’ ∈ N(q) that minimizes F(q’) and is non-tabu endif Set the reverse move for pt iterations and update h Set q=q’ endfor return q*
3.3.2 Tabu Search with Strategic Oscillations
if remove=1 Generate neighbourhood N1(q) and evaluate Get best neighbour q’ If F(q’) > L0 remove = 0 else Update tabu list and frequency table q=q’ endif else Generate neighbourhood N2(q) and evaluate Get best neighbour q’ If sum(q) > L1 and q is feasible remove=1 end q=q’ end if F(q)<F(q*) q*=q end end return q*
3.3.3 Tabu Search with Path Relinking
- Generate neighbourhood N1(q) (path 1)
- Generate neighbourhood N2(q) (path 2)
- Select a solution ∈ N1(q) ∪ N2(q) that minimize F() or satisfies the aspiration criterion
- Set q =
4. Scatter Search Approach
4.1 An Overview
4.2 Special features of the Scatter Search-based strategy for Sensor Network Design
4.2.1 Diversification Generation
4.2.2 Improvement of Solutions
4.2.3 Reference Set Generation and Upgrade
4.2.4 Subset Generation and Solution Combinatio
4.3 Scatter Search Based Procedure
Set U = Ø U= Diversification Generation method(Usize) U= Improvement Method 1(U) RefSet= Reference Set Generation and Upgrade method(U) Refset=Sort(Refset) NewSolution = TRUE while (NewSolution) NewSolution = FALSE Generate subsets of RefSet, using the Subset Generation method, with at least one new solution. while (there is at least one subset without evaluation) Select the subset and label it as evaluated. Apply the Solution Combination method to the solutions in the subset. Apply Improvement Method 2 to each solution obtained by combination. Let q be the improved solution. If F(q) is better than F(qworst) and q is not included in RefSet qworst = q Refset=Sort(Refset) NewSolution = TRUE else q= Improvement Method 1(q) If F(q )< qbest solution in RefSet qbest = q NewSolution = TRUE endif endif endwhile endwhile
5. Population Based Incremental Learning Approach
5.1 An Overview
5.2 Special features of PBIL-based strategies for Sensor Network Design
5.2.1 Initial Solution and Evaluation Function
5.2.2 Marginal Distribution Estimation
5.2.3 Selection and Local Search
5.2.4 Distribution Probability Upgrade
5.2. Mutation
5.3 PBIL Based Procedure
Initiate NPBIL probability vectors pk (k =1,…,NPBIL) while (stopping criteria = .FALSE.) for k = 1,…, NPBIL do Generate N individuals by simulation according to pk Evaluate the fitness function F for each member of the population Apply the Improvement Method 2 to each member of the population Select the best solution Upgrade pk using the best solution and the learning rate LR Mutate pk using a probability of mutation PMUTA and a quantity of mutation MS end for Set OffSpring= Ø for k = 1,…, NPBIL step 2 do Select 2 individuals (parents) among all vectors p if random<Pinteraction Use uniform crossover to calculate two children Add children to OffSpring else Add parents to OffSpring endif endfor for k=1,..., NPBIL do Set pk=OffSpring(k) endfor endwhile
6. Analysis of Results
6. 1. Problem 1
Case | Constraints |
---|---|
1 | E≥1 for stream 1 |
2 | E≥1 for stream 2 |
3 | E≥1 for streams 17 23 |
4 | E≥1 for streams 7 16 18 20 |
5 | E≥1 for streams 5 12 14 35 37 44 62 70 77 =1584.2, =1359.6, =200.7, =1580.6, =122.72, =1284.4 |
6 | = 0.9, = 0.9 =0.7, =0.5 |
7 | = 0.8, = 0.8, =0.8 =0.7, =0.15 =0.4 |
6.2. Problem 2
6.3. Results
Parameter Value | SO-TS | PR-TS | SS | PBIL |
pt | --- | --- | ||
ph | 23 | 60 | --- | --- |
Lo | --- | --- | --- | |
L1 | 0.8*n | --- | --- | --- |
|R| | --- | 10 | --- | --- |
frecpr | --- | 15 | --- | --- |
# Max Iter | 300 | 200 | --- | 100 |
|Refset| | --- | --- | 12 | --- |
NPBIL | --- | --- | --- | 8 |
N | --- | --- | --- | 12 |
Pinteraction | --- | --- | --- | 0.7 |
PMUTA | --- | --- | --- | 0.02 |
MS | --- | --- | --- | 0.05 |
LR | --- | --- | --- | 0.1 |
Case | SO-TS | PR-TS | SS (for 20 runs) | PBIL (for 20 runs) | ||||
Min | Mean | CV | Min | Mean | CV | |||
1 | 533.6 | 533.6 | 533.6 | 587.9 | 11.28 | 533.6 | 533.6 | 0.00 |
2 | 894.9 | 894.9 | 894.9 | 1006.3 | 17.90 | 894.9 | 894.9 | 0.00 |
3 | 752.3 | 752.3 | 752.3 | 767.4 | 2.33 | 752.3 | 754.0 | 0.98 |
4 | 1178.0 | 1178.0 | 1178.0 | 1178.0 | 0.00 | 1178.0 | 1178.0 | 0.00 |
5 | 50,845 | 50,846 | 50,847 | 54,183 | 6.32 | 50,845 | 50,846 | 0.08 |
6 | 62,322 | 62,322 | 62,322 | 63,535 | 4.54 | 62,322 | 62,322 | 0.00 |
7 | 80,548 | 80,548 | 80,548 | 80,821 | 0.22 | 80,548 | 80,567 | 0.11 |
Case | SO-TS | PR-TS | SS (AVG) | PBIL (AVG) |
1 | 717 | 1,130 | 3,795 | 16,540 |
2 | 781 | 1,063 | 4,771 | 16,674 |
3 | 39 | 56 | 2,631 | 15,622 |
4 | 110 | 413 | 1,928 | 7,326 |
5 | 8,298 | 10,535 | 129,369 | 253,364 |
6 | 42 | 56 | 1,442 | 6,882 |
7 | 260 | 600 | 2,112 | 9,621 |
SO-TS | PR-TS | SS | PBIL | |
Min | 50,846 | 54,974 | 50,847 | 51,124 |
Max | 290,572 | 290,296 | 290,295 | 55,312 |
Mean | 197,612 | 265,620 | 66,711 | 51,989 |
Deviation | 116,792 | 73,913 | 52,803 | 1,375 |
CV | 59.10% | 27.83% | 79.15% | 2.64% |
7. Conclusions
Acknowledgements
References and Notes
- Holland, J. Adaptation in Natural and Artificial Systems; University of Michigan Press: Ann Arbor, MI, USA, 1975. [Google Scholar]
- Goldberg, D. Genetic Algorithms in Search, Optimization, and Machine Learning; Addison-Wesley: New York, NY, USA, 1989. [Google Scholar]
- Sen, S.; Narasimhan, S.; Deb, K. Sensor network design of linear processes using genetics algorithms. Comput. Chem. Eng. 1998, 22, 385–390. [Google Scholar] [CrossRef]
- Carnero, M.; Hernández, J.; Sánchez, M.; Bandoni, A. An Evolutionary Approach for the Design of Non-Redundant Sensor Networks. Ind. Eng. Chem. Res. 2001, 40, 5578–5584. [Google Scholar] [CrossRef]
- Viswanath, A.; Narasimhan, S. Multi-objective Sensor Network Design Using Genetic Algorithms. In Proc. 4th IFAC Workshop on On-Line Fault Detection and Supervision in the Chemical Process Industries, Seoul, Korea; 2001; pp. 265–270. [Google Scholar]
- Chao-An, L.; Chuei-Tin, C.; Chin-Leng, K.; Chen-Liang, C. Optimal Sensor Placement and Maintenance for Mass-Flow Networks. Ind. Eng. Chem. Res. 2003, 42, 4366–4375. [Google Scholar]
- Gerkens, C.; Heyen, G. Use of Parallel Computers in Rational Design of Redundant Sensor Networks. Comput. Chem. Eng. 2005, 29, 1379–1387. [Google Scholar] [CrossRef]
- Benqlilou, C.; Graells, M.; Musulin, E.; Puigjaner, L. Design and Retrofit of Reliable Sensor Networks. Ind. Eng. Chem. Res. 2004, 43, 8026–8036. [Google Scholar] [CrossRef]
- Carnero, M.; Hernández, J.; Sánchez, M.; Bandoni, J. On the Solution of the Instrumentation Selection Problem. Ind. Eng. Chem. Res 2005, 44(2), 358–367. [Google Scholar] [CrossRef]
- Gerkens, C.; Heyen, G. Sensor placement for fault detection and localization. Comp. Aided Chem. Eng. 2008, 25, 355–360. [Google Scholar]
- Glover, F.; Laguna, M. Tabu Search; Kluwer Academic Publishers: Norwell, MA, USA, 1997. [Google Scholar]
- Lin, B.; Miller, D.C. Solving Heat Exchanger Network Synthesis Problems with Tabu Search. Comp. Chem. Eng. 2004, 28, 1451–1464. [Google Scholar] [CrossRef]
- Lin, B.; Miller, D.C. Tabu Search Algorithm for Chemical Process Optimization. Comp. Chem. Eng. 2004, 28, 2287–2306. [Google Scholar] [CrossRef]
- Cavin, L.; Fischer, U.; Glover, F.; Hungerbuhler, K. Multiobjective Process Design in Multi-purpose Batch Plants using a Tabu Search Optimization Algorithm. Comp. Chem. Eng. 2004, 28, 459–478. [Google Scholar] [CrossRef]
- Carnero, M.; Hernández, J.; Sánchez, M. Optimal Sensor Network Design and Upgrade using Tabu Search. Comp. Aided Chem. Eng. 2005, 20, 1447–1452. [Google Scholar]
- Carnero, M.; Hernández, J.; Sánchez, M. A Tabu search procedure for sensor structure optimization. Trans. Built Environ. 2005, 80, 35–44. [Google Scholar]
- Laguna, M.; Hossell, K.P.; Martí, R. Scatter Search: Methodology and Implementation in C; Kluwer Academic Publishers: Norwell, MA, USA, 2002. [Google Scholar]
- Martí, R.; Laguna, M.; Glover, F. Principles of Scatter Search. Eur. J. Operational Res. 2006, 169, 359–372. [Google Scholar] [CrossRef]
- Larranaga, P.; Lozano, J. A. Estimation of Distribution Algorithms. A New tool for Evolutionary Computation; Kluwer Academic Publishers: Boston/Dordrecht/London, 2001. [Google Scholar]
- Shakya, S. Probabilistic model building Genetic Algorithm (PMBGA): A survey Technical Report; Computational Intelligence Group, The Robert Gordon University: Aberdeen, Scotland, UK, 2003. [Google Scholar]
- Baluja, S. Population-based incremental learning: A method for integrating genetic search based function optimization and competitive learning; (Technical Report CMU-CS-94-163); Carnegie Mellon University: Pittsburgh, PA, USA, 1994. [Google Scholar]
- Martí, R. Scatter Search – Wellsprings and Challenges. Eur. J. Operational Res. 2006, 169, 351–358. [Google Scholar] [CrossRef]
- Wan, S.; Qiu, D. Vehicle Rourting Optimization Problem with Time Constraint using Advanced PBIL Algortihm. In Proc. IEEE Int. Conf. on Service Operations and Logistics, and Informatics 2008; pp. 1394–1398.
- Pang, H.; Hu, K.; Hong, Z. Adaptive PBIL Algorithm and Its Application to Solve Scheduling Problems. In Proceedings of the 2006 IEEE Conference on Computer Aided Control Systems Design 2006; pp. 784–789.
- Ghamlouche, I.; Crainic, T.G.; Gendreau, M. Path relinking, cycle-based neighbourhoods and capacitated multicommodity network design. Ann. Operat. Res. 2004, 131, 109–133. [Google Scholar] [CrossRef]
- Narasimhan, S; Jordache, C. Data Reconciliation and Gross Error Detection: An Intelligent Use of Process Data; Gulf Publishing Company: Houston, TX, USA, 2000. [Google Scholar]
- Glover, F. A Template for Scatter Search and Path Relinking. In Artificial Evolution, Lecture Notes in Computer Science; Hao, J.K., Lutton, E., Ronald, E., Schoenauer, M., Snyers, D., Eds.; Springer: Berlin, Germany, 1998; Vol. 1363, pp. 13–54. [Google Scholar]
- Chaves, J.; Domínguez, D.; Vega, M.; Gómez, J.; Sánchez, J. Parallelizing PBIL for Solving a Real-World Frequency Assignment Problem in GSM networks. In Proc. 16th Euromicro Conference on Parallel, Distributed an Network-Based Processing (IEEE Computer Society); 2008; pp. 391–398. [Google Scholar]
Notation
Ak | Availability of the k-th variable estimate |
c | Acquisition cost vector |
d | Hamming distance |
Ek | Degree of estimability of variable k |
f | Objective function value |
F | Evaluation function |
nearest integer of
towards minus infinity | |
frecpr | PR frequency |
g | Vector of constraints |
GS | Guide solution for PR |
h | Frequency based Tabu list |
I0 | Initial set of instruments |
IS | Initial solution for PR |
L0 | Lower bound for TS |
L1 | Upper bound for TS |
LCC | life cycle cost of the sensor structure |
LR | Learning rate |
MS | Mutation amount |
N | Number of process variables |
N | Neighbourhood of possible solutions |
N | Number of individuals for each instance of PBIL |
NPBIL | Number of instances of PBIL executed in parallel |
# Max Iter | Maximum Number of Iterations for TS |
pt | Tabu Tenure Period |
Pinteraction | Crossover probability for PBIL |
PMUTA | Mutation probability for PBIL |
q | Solution vector |
Q | Penalty function |
Q | Random n-dimensional vector |
|R| | Cardinality of the Reference Set for PR |
|Refset| | Cardinality of the Reference Set for SS |
Ru | Number of unsatisfied constraints |
R | Process model equations |
S | Set of key variables |
Sσ | Set of key variables subject to precision constraints |
Sσ | Set of key variables subject to estimability constraints |
SA | Set of key subject to availability constraints |
t | Recency based Tabu list |
u | Vector of unmeasured variables |
x | Vector of measured variables |
z | Vector of process variables |
Standard deviation of the j-th variable estimate |
Acronyms
CV | Coefficient of Variation |
EDA | Estimation of Distribution Algorithm |
GA | Genetic Algorithm |
HGA | Hybrid Genetic Algorithm |
PBIL | Population Based Incremental Learning Algorithm |
PR | Path Relinking |
SN | Sensor Network |
SNDP | Sensor Network Design Problem |
SO | Strategic Oscillation |
SS | Scatter Search |
TS | Tabu Search |
© 2009 by the authors; licensee Molecular Diversity Preservation International, Basel, Switzerland. This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).
Share and Cite
Carnero, M.; Hernández, J.L.; Sánchez, M.C. Design of Sensor Networks for Chemical Plants Based on Meta-Heuristics. Algorithms 2009, 2, 259-281. https://doi.org/10.3390/a2010259
Carnero M, Hernández JL, Sánchez MC. Design of Sensor Networks for Chemical Plants Based on Meta-Heuristics. Algorithms. 2009; 2(1):259-281. https://doi.org/10.3390/a2010259
Chicago/Turabian StyleCarnero, Mercedes, José L. Hernández, and Mabel C. Sánchez. 2009. "Design of Sensor Networks for Chemical Plants Based on Meta-Heuristics" Algorithms 2, no. 1: 259-281. https://doi.org/10.3390/a2010259
APA StyleCarnero, M., Hernández, J. L., & Sánchez, M. C. (2009). Design of Sensor Networks for Chemical Plants Based on Meta-Heuristics. Algorithms, 2(1), 259-281. https://doi.org/10.3390/a2010259