Experimental study of SPSA approach to intelligent control systems

Simultaneous perturbation stochastic approximation (SPSA) approach is a general approximate method to estimate the gradient of system performance function. The neural network-based SPSA does not need a priori knowledge of the plant. A direct adaptive SPSA control system with a diagonal recurrent neural network as controller was examined by simulation. To improve the system performance, a conventional PID controller was used as compensator to form a hybrid scheme. Applying the SPSA approach to a fuzzy neural network-based control (FNNC) system, a four-layer neural network architecture was proposed to implement the hybrid SPSA FNNC scheme. Simulation results are presented