Research On PID Intelligent Control Algorithm And Generalization Ability Of Multivariable System

PID controllers are widely used in industrial control due to their simplicity,stability,and easy adjustment characteristics.However,traditional PID controller can't satisfy the control of complex systems such as multivariable,nonlinear and strong coupling.Therefore,this thesis integrates the idea of neuron network into PID control,and researches a PID controller based on neuron network according to the shortcomings of traditional PID controllers,and realizes the decoupling control of multi-variable,strong coupling system.In view of the shortcomings of sparrows search algorithm,such as long training time and easy to fall into local optimum in global search,this thesis incorporates the idea of DE/current-to-best/1differential mutation into the basic algorithm,so that when the population falls into local optimum,the search direction of the algorithm is changed,and the natural selection mechanism is incorporated to improve the convergence performance.Thus,a sparrow search algorithm(LDSSA)based on differential learning mutation of natural selection is developed,and this algorithm is used to optimize the initial weight value of the neural network,and further improve the decoupling ability and calculation accuracy of the neural network.In order to make the PID intelligent control algorithm proposed in this thesis have a certain generalization ability,the adaptive selection idea is integrated on the basis of the neural network PID,and the neural network is selected according to the different input and output characteristics of the controlled object.This algorithm can perform adaptive decoupling control for multi-variable and strongly coupled controlled objects within 5 inputs and 5 outputs.Several multi-variable,nonlinear,and strongly coupled control objects with different characteristics are used for simulation verification to prove that the algorithm has a certain generalization ability.At the same time,it is compared with the simulation results using the standard neural network algorithm.The result shows that the neural network PID decoupling control based on the LDSSA algorithm has a fast convergence speed and can quickly approach the given control target.At the end of this thesis,the above algorithm is used for decoupling control on a new type of controllable excitation linear magnetic levitation motor to verify the practical applicability of the algorithm.The result shows that the algorithm in this thesis can adaptively select a two-output neuron network structure for the controllable excitation magnetic levitation motor,which verifies that the algorithm has good adaptability and decoupling control capabilities when dealing with actual motor systems.Simultaneously,the simulation results show that the neural network using the optimized algorithm can approach the control target more quickly in the decoupling control of the motor.