Robust hybrid control based on PD and novel CMAC with improved architecture and learning scheme for electric load simulator

Considering the intrinsic nonlinear factors of electric load simulator and interference of surplus torque, new control strategy is required. This paper improves the architecture and learning scheme of cerebellar model articulation controller (CMAC) and proposes a novel CMAC-Proportional Derivative (PD) hybrid controller. The instruction torque and the output torque are regarded as stimulus signals of CMAC. A method of nonuniform quantization is proposed to fit the sinusoidal density of sampling distribution. Introducing quantitative distance and utilizing Gaussian weighting coefficient to distribute error, the approximation ability of CMAC is promoted for high-order differentiable input signals. A new learning scheme for CMAC is investigated to resolve its overlearning issue and restrain external disturbance as well. The results of dynamic simulation and experimental analysis indicate that the hybrid control algorithm can effectively restrain interference, smooth output error, and avoid overlearning of CMAC.