Implementation Of Neural Network Predictive Control In Optimized Control Platform

Predictive control has the advantages of low model accuracy requirements,ability to handle multiple variables and handle constraints.Industrial production processes are generally non-linear systems,and predictive control algorithms based on linear models are difficult to meet the control requirements of non-linear,time-varying,and large inertia objects.In view of this situation,this paper proposes a neural network predictive control algorithm.This paper introduces the basic principles,network structure and training process of long short-term memory networks.Combining its advantages in modeling time series data with a recursive multi-step model,we establish the prediction model.Golden optimization method is used to design a rolling optimization method.Combined with feedback correction,the design of neural network predictive control algorithm is completed.This paper presents a multi-model neural network predictive control structure based on recursive Bayesian probability weighting method.This paper introduces an independently developed optimization control platform,and explains its hardware architecture,software design scheme,and platform advantages.From the perspective of module generation,implementation principles,interface design,and implementation of non-disruptive switching,the advanced algorithm container of the optimized control platform are explained in detail.This paper verifies the robustness and anti-disturbance of single-model neural network predictive control through simulation.The control effects of multi-model and single-model neural network predictive control in two cases of step response and variable condition tracking are simulated,and the effectiveness of the multi-model control scheme for the control of complex systems is verified.This paper embeds single-model and multi-model neural network predictive control algorithms into the advanced algorithm container of the optimized control platform.Real-time simulation by building a logical configuration verified the openness and real-time nature of the advanced algorithm container.