遗传算法改进神经网络的电力系统谐波检测法

提出了将遗传算法与神经网络结合起来用于电力系统谐波幅值与相位测量的方法。根据电力系统谐波的特点,构造了用于谐波检测的神经网络模型,阐述了网络训练样本的形成方法。借助Matlab提供的遗传算法与神经网络算法工具箱,先用遗传算法对前馈神经网络进行全局训练,再用BP(back-propagation)算法进行精确训练。仿真结果验证了该方法的有效性和可靠性。通过与传统BP算法测量谐波的方法相比较,该方法具有训练速度快,不易陷入局部极值,测量精度高的优点。最后用未训练的样本检测训练好的神经网络,验证了该网络同时也具有较好的泛化能力。

A harmonic measuring approach based on genetic algorithm improved back-propagation neural network

Based on a genetic algorithm(GA) and an artificial neural network(ANN),an approach for measuring the magnitude and phase of harmonics was proposed.The neural network model was developed according to the requirements of measuring harmonics.First,a GA was used to train the network by updating the weights to minimize the error between the network output and the desired output.Next,a back-propagation(BP) algorithm was used to train further the ANN to increase model accuracy.Simulation experiments were performed using Matlab's toolbox.The results show the validity and reliability of the proposed harmonic measuring approach.Compared to a traditional BP algorithm,the proposed method has faster training speed,does not easily move into a local extremum,and has higher precision and better generalization ability.