(204r) Stochastic Procedures for the Optimal Sensor Location in Chemical Plants

Process information is the foundation upon which other plant activities (monitoring, control, optimization, planning and scheduling, fault diagnosis, etc.) are based. To fulfill information requirements, regarding its quality and availability, it is essential to install an appropriate sensor network (SN) in the plant, and also data reconciliation and SN maintenance tasks should be executed during process operation.

 The SN designer should decide to measure each process variable or not. These decisions are mathematically formulated in terms of binary variables. A combinatorial optimization problem results that is usually multimodal and involves many binary variables. Its solution has been addressed using tree search algorithms, MILP techniques and stochastic procedures.

 Regarding the stochastic solution schemes, techniques based on Genetic Algorithms (GAs) were proposed initially. Benqlilou et al. (2004) applied the GA toolbox of MATLAB program to solve the design and retrofit of reliable SNs, and Gerkens and Heyen (2005) presented two ways of parallelizing the classic GA to reduce the solution time. A hybrid procedure was developed by Carnero et al. (2005a) to minimize the instrumentation cost subject to precision constraints on key variables, which makes use of a structured population in the form of neighborhoods and a local optimizer of the best current solutions. Furthermore, Carnero et al. (2005b) developed a new design strategy within the framework of Tabu Search, which uses a Strategic Oscillation Technique around the feasibility boundary (SOTS).

 Recently a metaheuristic approach based on another population based methodology, the Estimation of Distribution Algorithms (EDAs), and SOTS was presented (Carnero, et al, 2013). Application results of that procedure demonstrated that the combination of EDAs and SOTS advantages has a synergistic effect on the solution of the SN design problem. The proposed solution scheme makes use of the Population Based Incremental Learning Algorithm developed by Baluja (1994), which assumes independent relationships among variables.

 According to the model complexity, EDAs can be broadly divided into univariate, bivariate or multivariate approaches (Hauschild and Pelikan, 2011). To handle variable interdependencies, the second and third classes of EDAs require complex learning algorithms and significant additional computational resources. In this work, the methodology  presented by Carnero et al. (2013) is extended to the utilization of probabilistic models that capture variable interdependencies, such as, the marginal product factorization model (Santana et al., 2010). Furthermore, a comparative performance study is conducted to evaluate the benefits of increasing the complexity of the distribution model. Case studies of incremental size, extracted from the literature, are used as application examples.

 References

 Baluja S, Caruana R. Removing the genetics from the standard genetic algorithm. Technical Report CMU-CS-95-141; Carnegie Mellon University, 1995.

Benqlilou C, Graells M, Musulin E, Puigjaner L. Design and retrofit of reliable sensor networks. Ind. Eng. Chem. Res. 2004; 43: 8026-36.

Carnero M, Hernández J, Sánchez M, Bandoni A. On the solution of the instrumentation selection problem. Ind. Eng. Chem. Res 2005a; 44: 358-67.

Carnero M, Hernández J, Sánchez M. Optimal sensor network design and upgrade using tabu search. Comp. Aided Chem. Eng. 2005b; 20: 1447-52.

Carnero M, Hernández J, Sánchez M.A new metaheuristic based approach for the design of sensor networks. Comp. Chem. Eng.  2013; 55: 83-96.

Gerkens C, Heyen G. Use of parallel computers in rational design of redundant sensor networks. Comput.  Chem. Eng. 2005; 29: 1379-87.

Hauschild M, Pelikan M. An introduction and survey of estimation of distribution algorithms. Swarm and Evol. Comp. 2011,1:111–28.

Santana R, Larrañaga P, Lozano, J. Learning Factorizations in Estimation of Distribution Algorithms Using Affinity Propagation. Evol. Comp. 2010; 18: 515–546.