Research On Model Predictive Control Based On Recurrent Neural Network

Model predictive control is an advanced computer control algorithm. Its special rolling optimization strategy can restrain the influence on the control performance effectively, such as model uncertainty, external disturbance, time delay et al. As a result, model predictive control is widely used in industrial process control. But with the development of the rolling optimization strategy, making the online calculation quantity complex, affecting the control performance. Recurrent neural network is an intelligent real-time optimal method. It has the ability of paralleling and distributing the problems of optimization, and solving the problems of predictive optimization quickly under different condition. Therefore, the recurrent neural network is applied to the optimal results of model predictive control.The main content and work of this thesis can be described briefly in the following aspects:1. On the issues of inequality constraints for the model predictive control, increasing one step equality constraint, researches on the mixed constraints of model predictive control. In order to reduce the difficulty of calculation, this paper puts forward a simplified dual recurrent neural network model predictive control algorithm.2. Considering the effect of time-delay and disturbance on the performance of the control system respectively. For the purpose of solving the predictive control optimal problems of time-delay system, putting forward a general neural network dynamic optimal methods. In consideration of model predictive control with time-delay and disturbance, a two-layer projection neural network optimization robust model predictive control algorithm is proposed. Special examples and comparison with the reference documents show the advantage and effectiveness of the algorithm. Moreover, this method can be used in other time-delay systems.3. Model predictive control algorithm based on recurrent neural network is applying to solve the multi-agent system flocking control is studied. Based on the consistency protocol of multi-agent system, giving a thought to the input influence on the agent system control, a projection about neural network optimization distributing model predictive control algorithm is proposed for solving the problems of multi-agent system flocking control in this paper. The simulations reveal that the method is more effective than other methods.