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. In order to solve the problem that the online least support vector machine (SVM) is computationally expensive to update the model every sampling period, an adaptive PID control algorithm of online robust least square support vector machine (LSSVM) is proposed. The LSSVM The robustness of the proposed method combined the prediction error of the samples and the prior knowledge of the process gives the complexity criterion of the control model and effectively improves the accuracy, speed and sparsity of the model. Combined with the predictive control theory, the online robust LSSVM algorithm is applied to PID nonlinear control.The simulation results show that the robust cost function combined with robust Huber function and ε-insensitive function can effectively model the local nonlinear region of the system and adaptively recognize the change of the system operating point , Not only has higher control precision, but also has strong robustness and modeling speed, which can adapt to the control of time-varying parameter objects.