Artificial Affective Neural Networks With Applications In Predictive Control

Industrial processes are always of characteristics of nonlinearity and large time delays.Additionally,it is rather difficult to obtain large amounts of data to train a neural networks dynamic model.Therefore,it is undoubtedly of significance of academic researches and engineering applications to build a neural networks model and construct an effective control strategy in the case of a small amount of dynamic training data.In this context,this thesis carries out research work and obtains the following main results.1.In order to improve the modeling accuracy and learning efficiency of neural networks for small data samples,this thesis introduces an artificial affective neural networks prediction model emulating the human brain structure which is recognized as a non-fully connected neural network composed of two parts:the dorsal brain networks and the ventral brain networks.The cognitive and emotional data are processed through the dorsal and ventral networks,respectively.Two emotional factors,"emotional" and "confidence",are used to simulate the emotional changes in human learning,which can help improve the neural network's ability to recognize objects so as to enhance the recognition accuracy.Compared with ordinary BP networks,it shows that the artificial affective neural networks enjoy high recognition accuracy and learning speed with small amounts of data samples.2.To deal with nonlinearity and large time-delay problems of industrial processes,an Artificial Affective Neural Networks(AANNs)based predictive control scheme is proposed.The AANNs are used as the predictive model while the particle swarm optimization method is employed as the rolling optimization algorithm.Compared with the common BP networks based predictive control scheme,the proposed strategy demonstrates desired effectiveness.3.The proposed control scheme is applied to the reheat steam section of an industrial supercritical 650 MW once-through boiler.The control performance of the proposed scheme is verified by comparing with those of BP networks based.