Least squares support vector machines generalized inverse control for a class of multi-input and multi-output nonlinear systems
多输入多输出非线性系统的最小二乘支持向量机广义逆控制

Considering the deficiency of neural network inverse control method, for a class of multi-input and multioutput (MIMO) nonlinear systems with unknown model, when soft-sensing functions for immeasurable states are available, we propose a new identification and control strategy based on the generalized inverse control of least squares support vector machines (LSSVM). The generalized inverse converts the controlled nonlinear system into a pseudo linear system with expected pole placement. In place of the neural network, LSSVM is employed to fit the static nonlinear mapping of the generalized inverse system. The identification of state variables is combined with the identification of LSSVM inverse model. Meanwhile, the soft-sensing is implemented through LSSVM training and fitting. Simulation is performed on a two-motor variable-frequency speed-regulating system. Results show that the proposed control strategy is feasible and efficient.