Neural Network's Application For Non-Linear System Identification

System identification is an important branch of control theory study, and it is also the foundation of control system design. In this field, the most difficult point is non-linear system identification. In recent years some achievements have been achieved in applying neural networks to the study of non-linear system identification, but its identification accuracy and efficiency is not ideal enough. On the ground of previous studies, the thesis comes up with a hybrid particle swarm optimization (HPSO) algorithm based on conjugate gradient algorithm, by combining particle swarm optimization algorithm and conjugate gradient algorithm together. Then HPSO is applied to the study of non-linear system identification based on BP networks, and the consequent simulation result proves the efficiency of hybrid particle swarm optimization algorithm. To overcome BP static networks' excess dependence on system' s order and delay in the process of identification, memory neural network and extended kalman filter learning algorithm are applied to the research of non-linear dynamics system identification, the use of which accelerates convergence speed and improves identification accuracy.Firstly, the thesis studies non-linear system identification based on BP neural networks by matlab language, and brings in genetic algorithm with good property of global search in order to overcome BP networks' invariable defects of slow convergence, local extreme minimum and bad identification inaccuracy. Genetic algorithm has improved identification effect to some extent, but copy, exchange and aberrance of genetic algorithm is a complicated process. Therefore, particle swarm optimization algorithm without copy, exchange and aberrance process is applied to train BP networks.Secondly, to improve particle swarm optimization algorithm' s...