Improved Biogeography-Based Optimization Algorithm And Its Application For The Soft Sensor Of Methanol Synthesis Conversion

This dissertation researches the process of methanol synthesis, and the soft sensor modeling for the methanol synthesis conversion is established. Improved biogeography-based optimization algorithm is applied in BP neural network for the soft sensor modeling. Biogeography-Based optimization (BBO) is a new type algorithm inspired by biogeography. The unique mechanism injects new ideas to the field of optimization algorithm research, and many research achievements about biogeography also provide profound theoretical foundation for the algorithm. In BBO, problem solutions are represented as habitats, it mainly uses the biogeography-based migration operator which designed based on the probability to share the information among solutions.Firstly, the design principle, process, migration and mutation operators are detailed in this dissertation. The dissertation also discusses the difference of the BBO and the tradition optimization algorithm, and it details the evolution strategy (ES) algorithm and particle swarm optimization (PSO) algorithm. Meanwhile, two improved algorithms that combine with ES and PSO are proposed, called them BBOES and BBOPSO. The two kinds of improved algorithm carry out the corresponding improvements for the mutation operation of basic BBO algorithm. Another two algorithms and fourteen typical benchmark functions are adopted to verify the performance of the two improved algorithm.Then, for a methanol synthesis unit, introduces the methanol synthesis process in this dissertation. In order to define the important parameters of methanol synthesis process, the research object and the target, it researches the methanol synthesis process and kinetics. The proposed improved algorithms are used for the prediction of crude methanol conversion rate for the methanol synthesis reactor with BP neural network. The simulation results indicate that soft sensor modeling method based on improved algorithms has higher training efficiency and stronger generalization than the other three compared methods.