The stability analysis and improvement of SPIDNN control system

SPIDNN (single-output proportional integral derivative neural network) control is the dynamic integration of neural network with PID. PID neural network control system was only applied to the system which has PID prior knowledge. For improving and expanding its using range, the improved algorithm was proposed on the basis of analyzing of the structure, the algorithm and its stability based on the different applied condition. The improved algorithm includes the improvement of output function, adoption of the attachment momentum method, the dynamic change of the step length and the elimination of the static error. And carried on the simulation with Matlab to the improved algorithm, the result indicates that the improved algorithm has very good effect in the aspect of the speed of convergence. The validity of the algorithm has been verified experimentally. Therefore, no matter that having the PID prior knowledge or not, the improved PID neural network control system can realize the systematic stability.