Steady-state Optimization And Control Based On Neural Networks

With the development of industrial process automation, not only the safe operation is needed, but also the high yield , high quality and low costs are required. Steady-state optimal control of industrial processes is the assurance of high economical benefits. The conventional steady-state optimization method is on the relied precise mathematical model. But for the complex industrial processes, it is very difficult to get the precise mathematical model. More attention is paid to explore the modeling and optimization methods which need not know the process accurately, and only need the expert knowledge or process data. Based on the neural network, the modeling methods for the steady-state process are studied in this dissertation, then several optimization methods, such as genetic algorithm, SQP algorithm and Hopfield neural network, are discussed carefully. These methods can solve the optimization problems of steady-state process whose mathematical model is hard to obtain. The main results of this dissertation are summarized as follows: ? BP neural network is an efficient model, but it is easy to be trapped in local minima and it抯 convergence speed is slow. So a new improved algorithm weight balance algorithm is proposed to exert all weights? influences on neural network. Simulation results show it is a effective algorithm and then this algorithm is applied to model on industrial manufacture process system. ? A new modified algorithm of GA is presented where select strategy and combined local search algorithms are obtained. The simulation results show that the new method can accelerate the convergence speed of GA. ? A practical chemical process is optimized by NN-GA method. And the results show the optimization method is greatly better than orthogonal experiment ones. ? NN-SQP method is proposed and applied to the optimal control of ~1I1 ~Z / / ~) industrial manufacture process. ? With respect to iterative decompose and coordination strategy of large-scale systems, Hopfield hierarchical optimization method based BP neural network model is presented and the neural network preferences are also studied.. Finally, based on the summarization of the obtained research results, future research problems are pointed out and discussed.