Design Of Inverse Controller For Nonlinear Systems Based On Lazy Learning Algorithms

With the rapid development of modern industrial production,the complexity,non-linearity,variability of workplaces,strong coupling and high control performance of the actual process make it more difficult for us to grasp the relationship between variables in the controlled process through the mechanism model,and the cost also increases.Generally,nonlinear systems are time-varying and uncertain,and a single model structure can not accurately describe the global system,which adds many difficulties to the global identification process.Therefore,this paper takes the non-linear system as the research object,directly from the input and output data of the system,and based on the local learning theory,puts forward the on-line optimization control method of the non-linear system based on the instant learning algorithm.This paper compares several kinds of commonly used data-driven strategy methods,and finally chooses the real-time learning algorithm to achieve the task of online local modeling.Real-time learning algorithm is based on the idea of "similar input produces similar output".By improving the similarity index of the algorithm and data update search strategy,local modeling of unknown systems can be realized only by relying on system data.The real-time learning algorithm can make the identification and control of the non-linear system more precise and improve the operation speed.Aiming at the non-linear system with a large amount of input and output data available,a controller scheme is designed based on the idea of local model on-line identification and neural network inverse control.Based on the input and output data of the system and the optimized performance index,the output value of the current inverse controller is obtained.Once again,the forward model is established and the forward optimal controller is designed to achieve the tracking effect.The inverse controller of the non-linear system based on the instant learning algorithm designed in this paper can effectively identify and control the non-linear system,overcome the difficulty of the global modeling data redundancy,achieve the simultaneous improvement of prediction accuracy and operation speed,and meet various constraints,at the same time,achieve the performance indicators prescribed by the system.The controller can be widely used in industrial practice and in complex working conditions,reflecting good industrial application value.And this paper makes a summary and outlook of this research direction.