Particle Swarm Optimization And Its Application On Process Industry

As modern process industries becoming finer and facing ever-tighter performance specification, there are increasing applications of intelligent algorithms in process industry, and engineers'concerns on the performance and accuracy of intelligent algorithm are also rising. In this work, one of the intelligent algorithms, which names particle swarm optimization (PSO) algorithm, is improved to overcome its disadvantages on higher-dimension optimization problems and constraint optimization problems, applications of PSO algorithm on process industry, such as statistical modeling and dynamic optimization, are studied in depth. The main contributions of the present work are as follows:(1) In order to overcome the disadvantages of classical PSO algorithm, a variant kinds of improved PSO algorithm are discussed, and an cooperative PSO algorithm with improved evolutionary status estimate and evolutionary operators is presented to solve the high-dimension optimization problems in process statistical modeling and dynamic optimization. The proposed algorithm enhances the global search ability of particle on searching space and deduces the requirement of particle number, and thus improves the efficiency of optimization algorithm. All these algorithms are further tested on benchmark unimodal and multimodal optimization problems, and sufficient result discussions are present.(2) In order to avoid the costly and time-consuming measurement of melt index in laboratory, a black-box modeling scheme to predict melt index (MI) in industry propylene polymerization process based on PSO algorithm is present. Research proposed in this paper applies statistical learning algorithm, including fuzzy neural network (FNN), radical basis function (RBF) neural network and evolutional PSO algorithm, to learn unknown relationships between MI and process variables from history data record. Furthermore, online correction strategy (OCS) is introduced to ensure fast online computation as well as easier updating and maintenance of model in presence of variations in process behavior. The proposed statistical model is checked on its prediction precision and training efficiency. A detailed comparison on efficiency and precision between the global PSO, best-neighbor PSO, and OCS is also carried out.(3) In order to apply PSO algorithm on solving constrained optimization problems, an improved PSO algorithm is proposed. Comparisons of the proposed algorithm and other strategies of PSO algorithm on constrained optimization problem are discussed, and the algorithm is further applied on solving dynamic optimization problem which is of great importance in process control.