基于支持向量机回归和RBF神经网络的PID整定

针对工业控制领域中非线性系统控制,在基于梯度下降法的RBF网络PID整定的基础上,对整定算法作出改进,控制目标不再是使当前跟踪误差最小,而是使当前跟踪误差和下一时刻跟踪误差的平方和最小。实现过程为:先由RBF神经网络在线辨识被控对象离散模型,得到被控对象的Jacobian信息,采用梯度下降法对PID控制器参数进行初步整定;然后,将系统跟踪误差和PID参数输入支持向量机模型,通过回归预测系统下一时刻的误差,改进的整定算法利用预测误差信息对参数进行再整定。仿真结果表明,引入支持向量机回归优化的RBF神经网络PID整定收敛速度更快,精度更高,跟踪性能优于RBF神经网络PID整定。

Self Tuning PID Based on Support Vector Machine Regression and RBF Neural Network

Based on the PID parameters self-tuning model using RBF neural network, a method of improving the adaptivecapacity of the control system by introducing support vector machine regression into it was proposed. According to the cur.rent PID parameters and error information, the system error at the next moment was predicted. Based on the predicted er.ror information, the controlling algorithm was improved and then the response speed and tracking accuracy of the systemwere enhanced.