The Research And Application Of T-S RBF Neural Network In Electrolyte Components Modeling

Copper and sulphuric acid concentration of electrolyte directly affect cathode copper quality in the process of electrolyzation refining. It is an important routine to measure Cu and sulphuric acid concentration for liquid purifying attemper and quality control. Circulation and purifying working procedure is a major means to adjust concentration. However, up to date, there is no generally applicable method for on-line monitoring of these concentrations in the industry process. Enterprises accept the trend of sampling and assay, but it results in consuming several hours or more. Circulation and purifying process are difficult to carry out for exact, punctual controlling and optimizing.In recent years, advancement of fuzzy theory and neural network techniques give impulse to the development of nonlinear fuzzy system modeling. Combination of process knowledge with intelligent modeling technologies to model complex system has been a focus in the domain of metallurgy electrolyzation.The dissertation establishes mechanism-NN hybrid model of electrolyte's concentration, based on the mechanism model and T-S RBFNN compensation model.At first, basic method and respective characteristics in industry modeling are briefly introduced. Then, factors related with mechanism modeling of electrolyzation process was chiefly introduced, the model to guarantee the whole characteristic of hybrid model was obtained. During electrolyzation, a lot of basic data were gathered and a series of complex reactions occurred. Electrolyzation is a non-linear and time-lag process. Intelligent modeling selected T-S RBFNN data drive distinguishing modeling. T-S RBFNN compensation model was established in order to compensate error of mechanism model.On the basis of fuzzy neural network theory and application research, we bring a excellent T-S RBF network configuration.By optimizing network hidden nodes selection mechanism and increasing deletion strategy to dynamic parameter adjustment algorithm,we gain an improved dynamic parameter adjustment algorithm based on T-S RBF neural network.Training the smallest network by this algorithm, it can not only regulate dynamically hidden nodes number, but also adapt the data centre and the expansion constants of Gaussian Radial Basis Functions, and eventually increase generalization ability.At last, preprocessed data samples were employed for established hybrid predictive model with IDYN algorithm, and simulation results shows that this method is correct and validate from approximate curve and MSE.