Neural Feedforward Compensation for Closed Cycle Diesel System

With the development of closed cycle diesel system (CCD), some problems have already arisen but they have not been completely resolved. The relatively slow response of the oxygen sensor and the cross-coupling effects existing in control units result in an unsatisfactory dynamic performance of CCD system whose operational circumstances are complicated because of its nonlinear nature, time delays and complex interactions if only the PID feedback controllers are used. In this paper, the neural feedforward compensation is firstly applied in CCD system to improve its dynamic performance. The neural network is used as the feedforward controller and combined with the corresponding feedback controller. The network is trained to reject the load changes gradually by minimizing the feedback controller output, which is utilized as an error signal in training. The results of simulation show that the dynamic characteristics of CCD control system are improved greatly by the feedforward actions of the networks once the training is completed. The training methods used on-line and offline, the unknown delay time discrimination and the adaptability of the neural controller are investigated.