Continuous Self-Learning Control Under System Identification Based on Hopfield Neural Network

Considering the dynamic modeling of an unknown and time-varying complex dynamic system in the model-based control, this paper develops a data-driven system identification method based on the Hopfield neural network (abbreviated as HNN). The normalization and denormalization transformation are designed to preprocess the training samples, and by the sliding window to refresh the sampling continuously, an online iteration method is proposed to identify the dynamic parameters. Furthermore, a continuous self-learning control based on HNN is developed in coordination with the model-based prediction and inversion control. Simulation results on the manipulator show that the HNN could efficiently identify the dynamic model in real time. The identified model has an approximately isomorphic equivalent form to the dynamic mechanism model, which could be used for the qualitative and quantitative analysis of the control quality. Results also verify the dynamic adaptability of the HNN continuous self-learning control for the nonlinear unknown and time-varying system.