Research On Model Predictive Control Methods For Nonlinear Time Series

Generalized predictive control(GPC)models are generally suitable for the study of linear time series predictive control.In real life,nonlinear time series prediction is more widely used.Therefore,Two model predictive control methods for nonlinear time series are proposed.One is nonlinear time series predictive control based on neural network,the other is nonlinear time series predictive control based on external influences.The main research contents are as follows.Because of the lack of control amount in the neural network model or the unknown control amount in the actual situation,a neural network predictive control model based on PID control is proposed.A predictive control model of RBF neural network with a control amount is constructed.The parameters of the PID controller are adjusted through errors to update the control amount.Experiments show that the predictive control model proposed in this paper has better predictive performance.Aiming at the limitation that GPC model can only perform linear time series prediction,a generalized predictive control model based on neural network is proposed.The GPC model is extended to an input layer composed of multiple cells and input to the next fully connected layer as an output prediction.Adjust the input control amount through the error reverse transmission.Through the LSTM and the fully connected layer,a regulator of the control amount is formed.Experimental results show that compared with GPC model,the proposed model has higher prediction accuracy.Aiming at the problem that time series are easily affected by external factors,a nonlinear time series predictive control model based on external influences is proposed.Establish a GPC model for external influences to obtain the control amount of external influences.Then a state-space model with a control amount for the time series is established.And then an improved EM-PSO algorithm is used to optimize the parameters of the state-space model.Finally,the optimized parameters and external control variables are substituted into the state space model for prediction.Experimental results show that compared with other models,the proposed model has higher prediction accuracy.