Fuzzy Neural Network Predictive Control And Initial Value Problem Based On L-M Algorithms

With the development of science and technology as well as the increasing complexity of industrial products,traditional PID control has been difficult to meet the requirements of control accuracy and speed in industrial production process.Under the circumstances,the predictive control method with good control performance has been widely used.However,most of the actual production objects are strongly non-linear,which makes it difficult to establish an accurate prediction model,and the rolling optimization solution of control variables is also extremely complex.In order to solve the above problems,the modeling method of nonlinear systems in predictive control and the local optimization algorithm are studied in this paper.Specific research contents are as follows:(1)One-step and multi-step prediction models are established by using BP,RBF and fuzzy neural network respectively,and an adaptive fuzzy neural network that is automatically generated by rules is proposed by improving the fuzzy neural network.The improved fuzzy neural network with relatively high prediction accuracy is selected finally to build the prediction model.(2)Using L-M algorithm for rolling optimization,a two-step predictive controller of fuzzy neural network based on L-M algorithm is established,and the initial value of L-M algor ithm is selected as the control quantity at the last time according to the relevant literature.The simulation results show that the controller can not effectively track the reference output,which leads to the decline of control performance.(3)The causes of the initial value problem are analyzed in detail,and the feasibility of using the optimal performance point as the initial point is demonstrated.Two methods for dynamic determination of the initial value are proposed,namely,the inverse neural network method and GA algor ithm,and then the weight factor in the objective function is dynamically adjusted to ensure that there is an extreme value between the optimal performance point and the minimum control point.On this basis,the structure and algorithm of the improved predictive controller are analyzed in detail,and the stability of the system is verified.The simulation results show that the IBP-LM-FNN and GA-LM-FNN neural network predictive controller designed in this paper have good control performance and can solve the initial value selection problem of the traditional L-M algorithm.(4)The IBP-LM-FNN predictive controller is applied to the pH neutralization process.The simulation results show that the proposed method has faster response speed,shorter adjustment time and stronger robustness than the traditional linear MPC and the gradient-based fuzzy predictive controller.