一类基于Levenberg—Marquardt算法的神经PID控制研究

传统的比例积分控制器具有一定的局限性，尤其是当被控对象会有非线性、不确定性和时变特性，常规的PID控制器往往难以发挥作用，甚至会失稳．利用神经网络进行复杂过程的PID控制可以很好地解决上述问题．Levenberg—Marquadt（LM）算法是梯度下降法与高斯一牛顿法的结合，就训练次数与精度而言，它明显优于共轭梯度法及变学习率的BP算法，适用于PID控制．得到了在线自适应神经网络PID控制算法，该算法改善了传统BP算法，实现了现有PID控制器控制方法．

Research into neural network system PID controller based on Levenberg - Marquardt algorithm

Traditional PID controller has certain limitation, and regular PID controller is difficult to effect and is even not stable especially when the objects which are controlled by regular PID controller are nonlinear, uncertain and time -varying. Above problems are easily solved by using neural network to control complex process. In this paper, BP neural network based on Levenberg -. Marquardt algorithm is used to process PID control. LM algorithm is the combination of the steepest decent algorithm with Gauss- Newton algorithm. With regard to the number of training and accuracy, LM algorithm is better than conjugate gradient algorithm and variable learning rate back propagation algorithm and can be applied to PID controller. Finally, online self-adaptive neural network PID controller algorithm is obtained, and this algorithm improves the traditional BP algorithm and realizes the controlling method of current PID controller.