基于LM算法的直流电机数学模型辨识

Parameter Identification of DC Motor Based on BP-LM

针对在PWM调速系统中直流电机速度的变化会受到负载转矩与控制电压两者的影响,该文提出一种基于LM(Levenberg-Marquardt)算法的直流电机关于转矩电压与转速数学模型辨识新方法。以直流电机的负载转矩和控制电压为输入,电机转速为输出构建BP神经网络模型,利用Matlab神经网络工具箱进行训练优化神经网络并给出权值矩阵。标准BP神经网络采用最速下降法来修正权值和阈值,虽然最速下降法可以使权值和阈值得到一个稳定的解,但其收敛速度慢,网络易陷入局部极小,学习过程会发生振荡,为了克服其缺点而采用LM算法更新权值和阈值。验证结果证明了该方法的有效性,实际样本数据与神经网络模型预测输出曲线拟合良好,网络模型预测输出误差达到最佳为1.0932×10^(-3)。

A new method of mathematical model identification of DC motor based on LM algorithm was proposed owing to the speed of DC motor will be affected by both load torque and control voltage in PWM speed regulation system.The BP neural network model was constructed,which took the load torque and control voltage of DC motor as the input and the speed as the output.The neural network was trained and optimized by Matlab neural network toolbox,and the weight matrices were given finally.The steepest descent back propagation(SDBP)is used to modify the weight and threshold in the application of standard BP neural network which convergence speed is slow,the network is easy to fall into local minima,and the learning process oscillates frequently,although SDBP can get a stable solution.In order to overcome its shortcomings,LM algorithm is presented.The effectiveness of this method has been verified,including the fine goodness of model fit between actual sample data and prediction output,and the minimum error is 1.0932×10^(-3).