采用自适应自回归小波神经网络的单步预测控制

针对非线性系统的控制问题，提出一种基于神经网络辨识的单步预测控制算法。算法在自回归小波神经网络的基础上，利用混沌机制消除了神经网络易陷入局部极值的缺点．采用自适应性学习率，提高神经网络的收敛能力和速度．以该神经网络为预测模型，引入输出反馈和偏差校正克服预测误差，以此构造一步加权预测控制性能指标。然后采用Brent一维搜索方法求取控制律，Brent法无需任何相关的导数信息，需调整的参数少，使得Brent法适合实时控制．仿真研究说明了该非线性预测控制器的有效性。

One-step-ahead predictive control using adaptive self-recurrent wavelet neural networks

A one-step-ahead predictive control algorithm via neural network identification is proposed for the control of nonlinear systems. The algorithm eliminates the defect that neural networks are prone to be trapped in local minimum through utilizing chaos mechanism based on self-recurrent wavelet neural networks. Then adaptive learning ratio is adopted to enhance convergence ability and speed of neural networks. As the neural network being predictive model and the output feedback and deviation rectification being introduced to reduce predictive error, a one-step-ahead weighted predictive control performance index is formulated. Lastly, the control law is derived via Brent optimization method which is efficient and reliable in one dimension search without knowing any relative derivative information. The method has less parameters to choose, which is very suitable for real-time control. The simulation shows that the presented method is effective.