A fast sparse least squares support vector machine hysteresis model for piezoelectric actuator

Abstract The inherent nonlinearities of piezoelectric actuator (PEA), especially hysteresis, greatly reduce the tracking performance of PEA. With a lot of computing resources consumed in the predicting process, the hysteresis modeling method of PEA based on the least-squares support vector machine (LSSVM) cannot be used for hysteresis compensation at high frequency. To solve this problem, a sequential selection approximate algorithm is proposed to obtain a fast sparse LSSVM (SLSSVM) hysteresis model. The SLSSVM model’s support vectors are only 6.8% of the original LSSVM model, by which the modeling speed and calculation speed are greatly improved. The experimental results show that the SLSSVM model improves the tracking accuracy when used in hybrid control system, especially for high frequency trajectories.