Studies On Computational Intelligence And Its Application In Thermal System

Thermal power plant whose production process is complicated, require high control quality as well as economical and secure productions. While the capacity, parameters and complexity of modern thermal generation equipments are becoming much higher, the characteristic of some equipment could not be described easily with data models or could be changed in various working situations. Traditional optimization methods are inefficient in the design of generator set control strategy in most cases. Inspired by mechanism of biological evolution and some physical phenomena, people put forward computational intelligence, involving many new methods, to solve complex optimization problems. Because of its advantages such as efficient optimization performance and without having to know special information about the problem, etc., computational intelligence has aroused widespread concern and been used in many fields. Consequently, it has theoretical significance and practical value to investigate computational intelligence and use it in thermal process of power plant.In this dissertation, after BP network standard algorithm with improved momentum term was studied, RPROP algorithm was applied to weights adjustment of BP network. OLS algorithm was introduced and simulated based on the research of RBF standard algorithm. The PID-CMAC compound control strategy was designed based on CMAC network. And a new dynamic recursive neural network structure, HIOCDRN network, was constructed via improved Elman network.A coupling algorithm based on chaos strategy and RPROP algorithm which was presented in this dissertation, has solved the local minimum problem to some extent. After researched the standard PSO algorithm, a new PSO algorithm with dynamic variable interval and restart strategy were applied to main steam temperature control system PID parameter optimization. Also a new identification method for thermal process object was put forward based on PSO and RBF network, which ccould greatly improve the identification speed and the generalization ability.An adaptive genetic algorithm was given using advanced fitness functions. And a fuzzy calculation was introduced to fuzzy quantum genetic algorithm in order to overcome the disadvantage of immobile rotated angle and enhances calculation precision and convergence rate. This fuzzy quantum genetic algorithm was applied to thermal process identification, while identification software was compiled to prove great universality.A single neuron controller was designed and used to make DCS control algorithm modules in which undisturbed switching from manual mode to automatic mode was considered. This dissertation applied neural network technology to internal model control and predictive control with corresponding solution. The disturbance observer based on PID neural network inverse model, cooperating with PID-CMAC compound control strategy, improved the control quality and anti-interference ability. After applying neural network and genetic algorithm to combustion optimization, the nitrogen oxide emission was reduced significantly.The main innovations of this dissertation are as fellows:1. Advanced a new style dynamic recursive Neural Networks (HIOCDRN);2. Advanced RPROP and chaos optimization coupling algorithm in the learning process of Neural Networks;3. Improved the PSO algorithm and designed dynamic model identification method which based on the combination of the PSO algorithm and Neural Networks;4. Advanced fuzzy quantum genetic algorithm (FQGA);5. Designed disturbance observer based on PID Neural Network inverse model.