Design On Steady-state Optimization Based On Fuzzy Neural Network

Steady-state optimization is a technology, for making larger benefit with less investment based on present technological conditions, by adjusting operation parameters after sufficient understanding of process object. Because of industry complexity, non-linearity and uncertain, building model and global optimization would be important procedures and difficulties.The paper uses fuzzy neural network to build system model. Because fuzzy rules decide the structure of fuzzy inference layer, how to get fuzzy rules is the key of building system model. Density clustering analysis is presented to solve the problem. In this process, data would be divided into many parts. Each part is a rule. The method can better reflect connections between input and output, and avoid impact from operators. At the same time, initial values of membership function'centers and breadths are got which can cut down much computation. The fuzzy reasoning and neural network structure were improved according to density clustering's characters which made fuzzy neural network reasonable. For fastening training speed, momentum back propagation method is introduced. The method can solve the problem of slow rate of convergence, by influencing parameters'corrections using the last.The improved simulated annealing algorithm is presented to get optimal parameters fast. Base on global optimization of SA, Powell method is introduced into local optimal parameters search. The improved algorithm greatly advances local search efficiency and can faster get global data.We made the oxidation decomposition reaction process as our research object for verifying effectiveness of the optimization method based on fuzzy neural network, in which the phenol/acetone are produced by decomposing cumene hydroperoxide. The fuzzy rules are extracted using density clustering analyzing data of production process. Then we established fuzzy neural network and got optimal sets of reaction process. By analyzing the result shows that the method is feasible and it can be use on optimization research of complex industrial systems.