The Improved Fuzzy Neural Network In Chemical Process Modeling And Control

To meet the request of real-time control in industry, a modified fuzzy neural networks (FN7N) is proposed and introduced to the intelligent process control field. In this paper, the multilayer feed-forward neural networks is employed to express the Takagi-Sugeno fuzzy inference system. A hybrid algorithm is proposed according to the character of FNN structure, using the least squared method to optimize the linear parameters and BFGS method based on the gradient descent to tune the nonlinear parameters. This method greatly enhances the convergence speed and makes it possible for the industrial on-line application. To avoid the system instability induced by the improper value of q, an adaptive tuning method is adopted. In the application of identification and control of nonlinear system, the FNN has been tested to not only succeed in solving the problem of slow convergence speed of the traditional neural networks, but also enhance the robustness and adaptability of fuzzy control during its industrial application. To verify the proposed schema, we applied it to the continued stirred tank reactor and the results demonstrate its validity.