Research On Application Of Bounded Derivative Neural Network For Nonlinear Predictive Control

Learning in neural networks is an optimization process, that is, a neural network adjusts the weights on its concrete error information. When establishing input and output of networks model, the most important guidance information is the derivative relation between input and output. Only when we establish such derivative relation, we can get accurate numerical parallelism between input and output.Currently, the optimization algorithm about artificial neural networks learning is a method of training networks only according to the error of output data, without importing the information of models' derivative effectively, which results in lacking the ability of generalization and practicability. Using the error of output data as criterion to training networks is interpolation for data of samples. It is not surprising to get big error in non-sample data for such interpolation is non-smooth. Hence, in training networks, besides the information offered by sample data, we should study the information of model's derivative and transform it into constraint conditions to use in the optimization learning of networks. In this article, the neural network learning method with constraints is studied and applied in the nonlinear model predictive control.The main content of the article and its innovations are as following:(1) Two different kinds of way were used to train the neural network with constraints. The advantages and disadvantages of them are also analyzed, through the training results of penalty neural network and bounded derivative neural network.(2) The bounded derivative neural network is used to model the static part of the model, which is combined with the dynamic matrix control algorithm to schedule the gain of the predictive model. Through this kind of way, the quality of control is improved.(3) Though changing the structure of the bounded derivative neural network, the static neural network is transformed to be the dynamic neural network. The advantage of this neural network is that it can still keep the ability of training the neural network with constraints, which means that the bounded derivative neural network can model not only the static part but also the dynamic part of the model.