Study On Swarm Intelligent Algorithm And Its Application To Biochemical Process Control

The intelligence algorithms and its application to the optimization of biochemicalprocess have been discussed in the thesis. According to history and actuality, becauseof the complex mechanism, nonlinear, variability of biochemical process and lack ofsensor, its automation level is not mature enough compared with other industrialprocesses. With the development and application of intelligence control technologysuch as intelligence algorithms, neural network, fuzzy system and support vectormachines, many up to date methods have been applied to biotechnological processes.Firstly, particle swarm optimization (PSO) is an evolutionary search techniquemotivated by the behavior of social organisms. In the thesis, quantum oscillator modelof PSO algorithm has been established and a method of parameter control has beenprovided. And therefore a new swarm intelligence algorithm (QOPSO) has beendesigned. The empirical results on benchmark function shows that QOSPOoutperforms the classical PSO. At the same time an adaptive approach intoQuantum-Oscillator-based Particle Swarm Optimization (QOPSO) algorithm isintroduced. In comparison with the classical PSO algorithm, the benchmark functionsare used to test the performance of the Adaptive QOPSO (AQOPSO). The vastnumber of experiment results show that AQOPSO has much stronger global searchingability compared to QOPSO.Secondly, we employ the combination of QOPSO and RBF neural network to theglutamic acid production process. An optimization architecture based on both ofgenetic algorithm (GA) and Quantum-Oscillator-based Particle Swarm Optimization(QOPSO) algorithm has been also established. So the yield can be increased while theresource wastage can be decreased under the condition of production demand. Thesimulation results show that the QOPSO can converge more rapidly than GA, whichmeans QOPSO is more applicable to biochemical process than genetic algorithm.Finally, the system of estimation and prediction to the biochemical process hasbeen set up by RBF neural network, TSK fuzzy system and SVM methods, It hasbeen testified through simulation that the SVM is able to predict the biochemicalvariables more precisely than RBF and TSK. Therefore, our future work should focuson combining QOPSO and SVM to develop more effective models. So mainbiochemical variables could hopefully be estimated, predicted and monitored inbiochemical production process.