Teminal Bias Neural Network Control And Its Application

In recent years,the market competition at home and abroad is becoming more and more intense,which makes the production enterprises have higher requirements for the control performance of the production process of industrial products,and the actual industrial production process usually has the characteristics of strong nonlinearity,uncertainty,coupling and time-varying.These characteristics make the control method based on linear system theory no longer applicable.Therefore,the nonlinear control method of production process has become a research focus in the field of process control.In the existing nonlinear control methods,the neural network control method has been widely studied because the neural network has strong nonlinear approximation ability and self-learning ability.However,the proposed neural network control method has great limitations in practical engineering applications,which is mainly manifested in the effective determination of the initial value of the network parameters.When the initial setting of the network parameters is not reasonable,these methods are difficult to obtain the ideal control effect.Therefore,taking the neural network control method of model reference as an example,this paper studies the effective determination of network parameters,mainly including the following aspects:First,the principle of parameter adjustment of model reference neural network control method is studied,the limitation of the method is analyzed,and the reasons for the limitation are pointed out,which lays the foundation for the research of the later chapters.Secondly,based on the model reference neural network control method,a neural network control method with terminal bias is proposed.Starting from the objective function of training network parameters,the objective function of end deviation is constructed respectively from mean variance and maximum correlation entropy.At the same time,in order to avoid the shortcomings of the traditional BP algorithm and the LM algorithm being easily trapped in local extremum and difficulty in selecting initial value,this method uses a differential evolution algorithm with global search ability to train parameters.In order to verify the effectiveness of the proposed method,a multi input and multi output system is selected to carry out simulation experiments.The simulation results show that the end point biased neural network control method has good control effect.Finally,the end point bias neural network control method is applied to the liquid level control of single/double tank.From the application process,the end bias control is more convenient than the model reference neural network control.The reason is that the off-line parameters of this method can be directly used for real-time control.From the control effect,the terminal bias control method has good control accuracy.The integrated simulation and the actual control results show that the end point biased neural network control method proposed in this paper can be used for system control with the characteristics of nonlinearity,uncertainty and coupling.