Research On Lm Optimization Algorithm And Neural Network Predictive Control In Nonlinear System

With the progress of the time, the level of science and technology has been rapidly promoted, the industrial process is more complicated, and the higher control precision of controller is required. In such a severe industrial environment, the control ability of PID control and other traditional methods is slightly weak, the predictive control shows a strong industrial adaptability and has gradually established its position in the industrial control. Predictive control can be modified according to the current situation of dynamic prediction model, the advantage of its control effect and the precision is more prominent for controlling the object with multiple input and output, nonlinear, time-varying and large time delay, and so on. In this paper, the nonlinear system is applied as the research object of predictive control, and its related background, structure theory and industrial application are described and analyzed.Firstly, the research background and current situation of predictive control are analyzed, the basic principle, stability and robustness of predictive control are introduced, and several typical methods of building prediction models are listed. For the nonlinear controlled-object, related improved methods are proposed to solve the problems that neural network predictive control model is not accurate and the rolling optimization is difficult to be solved. Specific contents is as follows:BP neural network is selected as the model identification method for its ability of approximation of any object in strongly nonlinear circumstance. Based on the problems that BP neural network is easy to fall into local minimum and slow convergence speed, LM-MEA optimization algorithm is proposed by combining mind evolutionary algorithm and LM algorithm. In LM-MEA algorithm, the advantages of mind evolutionary algorithm in simulating the processes of human mind and rapidly solving and high computing accuracy are applied to improve the disadvantages of LM algorithm depending on initial value. The simulation results of standard test functions demonstrate the better optimization performance of LM-MEA optimization algorithm. The proposed method is used to establish model for the nonlinear object, simulation results show that the BP neural network based on LM-MEA optimization algorithm has higher precision, anti-interference ability and adaptability.In rolling optimization, the LM algorithm and the particle swarm optimization algorithm and are combined and the LM-PSO algorithm is proposed. LM-PSO algorithm retains the fast global convergence of PSO algorithm and the high accuracy of LM in the near local minimum value, and overcomes the trend of falling into local extreme of PSO algorithm and shortcoming of over depending on initial value of LM algorithm. The simulation results of standard test functions demonstrate the good optimization of LM-PSO algorithm. The nonlinear system is chosen as the research object, and the BP neural network is used as the model identification method. LM-PSO algorithm is applied to obtain the optimal control of the objective function in rolling optimization and compared with the other algorithms. The simulation results demonstrate that the predictive control method based on LM-PSO-BP neural network has improved its integrated control performance.Finally, the predictive control method based on LM-MEA-BP neural network for prediction model and LM-PSO algorithm for receding horizontal optimization strategy is applied to the control solution of the reaction concentration of CSTR. The experimental results show that the method has good tracking control effect and anti-interference ability.