神经网络预测控制局部优化初值确定方法

为解决局部优化算法初值选取不当造成神经网络预测控制性能下降的问题,本文提出了一种动态确定初值的方法.在每次优化时通过逆网络将初值选在输出误差最小点,通过修正目标性能函数中的权重因子来确保初值与当前控制量之间存在极值,并在理论上进行了证明.以BP神经网络预测控制为例,采用牛顿拉夫逊算法实现滚动优化,对所提方法进行了仿真实验,结果表明能够解决初值问题,提高控制系统的可靠性.

Method to determine initial value of local optimization for neural network predictive control

To deal with the performance degradation caused by improper initial values in neural network local optimization predictive control, we propose a method to dynamically determine the initial values. In each optimization cycle the minimum output error point is selected by calculating the inverse neural network. The existence of the minimal value of the objective function between this point and the current control point can be ensured and proved through modifying the weighting factor. Finally, a simulation experiment is carried out to verify the proposed method using a back propagation (BP) neural network as the predictive model, and the Newton-Raphson algorithm is employed as the receding horizon optimization strategy. The results show that the initial value problem can be solved to improve the reliability of the control system.