Abstract: In the circumstance where both the delay model and the controlled object model were unknown in networked control systems, a class of networked predictive control systems based on improved BP network were studied. For the problem of obtaining the hidden layer nodes number of BP network, a rapid calculation method was proposed. For the problem of avoiding local optimum of BP network, a hybrid learning method was proposed. A off-line BP network model was proposed for coping with the problem of the delay prediction based on the above algorithms. For the problem of the linear and nonlinear control of the networked control systems, a single step prediction controller based on on-line BP network was designed. The simulation results show that the performance of the networked predictive control systems is improved in the running speed, the delay prediction, the linear and nonlinear controls.

Key words: networked control system, BP network, delay prediction, genetic algorithm, linear controlled object, nonlinear controlled object