Generalized Predictive Control Based On RBF Neural Networks For Time Delay System

In industrial processes, time-delay is very common. With time-delay, it lead to the bad influence to the controlled system. So, it is always difficult to find appropriate modeling and controlling strategies in the controlled field. Because neural networks have learning ability and adaptation, they have been used in the research of controlling the process.Predicting model can predict the future outputs of time delay systems, and the strategy overcomes the influence of time delay, so predictive control is suitable for such system with time delay. Because feedforward neural networks have the ability to apply nonlinear mapping, radial basis function (RBF) neural networks are chosen to be process models. They have the merits of smaller calculation, fast convergence and without local infinitesimal values, and so on. But it is sensitive to initial values of hidden neurons and difficult to define the appropriate number of neurons. So, supervised algorithm and self-adaptation algorithm are applied to improve the performance of the networks, In the simulation, it is proven that it can overcome the problems mentioned above, and is provided with fast convergence and good generalization. Based on such model and recursive strategy, multi-steps predictive model is constructed to predict outputs of the process. Considering there may be model error and uncertain factors in the process, compensated strategy is designed to improve the precision of prediction. In this paper, generalized predictive control (GPC) algorithm is applied, which is introduced new concept of cost horizon that is suit for the process of unknown time delay. To test the performance of the controller, the simulation of continuous stirred tank reactor (CSTR) is made. Meanwhile, neural-networks-based PID controller is given to contrast with the control strategy mentioned above. It can be proven that the actual output of system can track desired output, and GPC based on neural networks is provided with good adaptation, robust and ability of anti-interference.