基于在线鲁棒LSSVM的自适应PID算法

Adaptive PID Algorithm Based on Online Robust LSSVM

为了解决在线最小支持向量机在每个采样周期更新模型带来计算量大的问题,提出了在线鲁棒最小二乘支持向量机(LSSVM)的自适应PID控制算法.通过2步加权策略提高LSSVM的鲁棒性,把样本预测误差与过程先验知识相结合给出控制模型复杂度准则,有效地提高了模型的精度、速度以及稀疏性;结合预测控制思想,把在线鲁棒LSSVM算法用于PID非线性控制.仿真结果表明:鲁棒Huber函数与ε-不敏感函数相结合的鲁棒代价函数,能够有效地对系统局部非线性区域进行建模,随系统工作点变化而自适应地辨识,不仅有较高的控制精度,而且具有较强的鲁棒性和建模速度,能够适应时变参数对象的控制.

An adaptive PID algorithm based on online robust least square support vector machines （LSSVM） was presented to solve the problem of large calculation resulted from the online least support vector machines updating the model in every sampling period. A two-step weights principle was used to improve robustness in LSSVM regression. The predictive error and prior process knowledge were combined to control model complexity. The precision and speed of the LSSVM were improved effectively. Based on the idea of predictive control, the online robust LSSVM modeling algorithm was applied to PID. The results show that the robust cost function combining robust Huber function with ε-insensitive loss function can be used to model for the partial nonlinear region of the system and adaptively identify the model of system with the changing of working point. This algorithm adapts to the control of time-varying parameters object and achieves higher control precision, stronger robustness and modeling speed.