摘要
针对传统误差反向传播(BP)神经网络算法在短期光伏功率预测模型的训练过程中易陷入局部极值点而无法达到全局最优,初始网络的权重随机给定导致模型预测精度低、迭代效率低等问题,提出一种基于改进BP神经网络的短期光伏功率预测模型。所提模型采用莱文贝格-马夸特(L-M)算法代替传统的梯度下降法进行训练,提高传统BP神经网络训练的迭代速度,并利用遗传算法优化网络模型的初始权重,进一步提高模型预测精度。仿真结果表明,所提模型对短期光伏发电的预测精度明显优于传统算法,并且大大提高了稳定性和迭代效率。
In response to the problem of traditional back propagation(BP)neural network is prone to fall into local extreme points during the training process of short-term photovoltaic power prediction models,which prevent it from reaching the global optimization,and the weights of the initial network are randomly given,resulting in low prediction accuracy and iteration efficiency of the model,this paper proposes a short-term photovoltaic power prediction model based on improved BP neural network.The proposed model replaces the traditional gradient descent method with the Levenberg-Marquardt(L-M)algorithm in training,which enhances the iteration speed of the traditional BP neural network training.Additionally,genetic algorithm is employed to optimize the initial weights of the network model,which further improves the prediction accuracy of the model.Simulation results demonstrate that the proposed model significantly outperforms the traditional algorithm in terms of short-term photovoltaic power prediction accuracy,while also greatly improving stability and iteration efficiency.
作者
何之倬
马一凡
俞志云
HE Zhizhuo;MA Yifan;YU Zhiyun(State Grid Shanghai Qingpu Electric Power Company,Shanghai 201799,China)
出处
《微型电脑应用》
2025年第6期204-207,共4页
Microcomputer Applications
