Brain Emotional Learning Networks Based Interanl Model Control Ststems With Applications

As a model-based control method,internal model control(IMC)is widely used in practical industrial processes.However,the IMC method requires high accuracy for the model.Additionally,to problems of nonlinearity and time-varying in industrial processes,conventional modeling methods hardly lead to good results.For this reason,internal model control based on neural networks has attracted much attention.However,the establishment of process models using conventional neural networks requires a large number of open-loop experiments to acquire data,and conventional neural network models are difficult to update online quickly when process models change.Motivated by these observation,this thesis proposes an improved brain emotion learning network(BELN)and applies it to the internal model control system.The main research contents and achievements are presented as follows:1.In view of the analysis and improvement of the existing brain emotional learning models,the proposed brain emotional learning network not only simulates the actual information processing of the human brain,but also introduces "anxiety" and "confidence".Affective coefficients are employed to simulate changes in human emotional information.Through simulation experiments,the improved brain emotional learning network demonstrates a better effect performance on learning speeds.2.The brain emotional learning network is used to model the process and the controller of internal model control systems.Due to the fast learning speed of the network,a satisfactory offline modeling performance can be obtained by a small amount of learning sample data.In addition,when the controlled process model slightly changes,the brain emotional learning network model can track it online in time.3.Taking an industrial process—temperature control process of the co-current tube heat exchanger as an application example,an internal model control system for the heat exchanger has been established.Compared with the controllability of conventional neural network internal model control,the effectiveness of the proposed method is explicitly verified.