基于优化BP神经网络光伏出力短期预测研究

Prediction of photovoltaic output based on improved neural network

摘要:

太阳能拥有丰富的资源，而且分布广泛，现已被广泛应用到各种应用中，光伏发电已是一种可靠可行，可扩展的重要可再生能源利用的方式，因此对光伏出力进行精准的预测意义重大。从宁夏市某光伏发电站获得了一年的光伏发电数据与气象等因素，选取四月至五月的数据进行研究预测。针对BP神经网络的收敛时间长，容易陷入局部极小值等缺点。建立单一BP神经网络预测模型，基于遗传算法(GA)优化BP神经网络的GA-BP预测模型与基于狼群算法(WPA)优化的BP神经网络的WPA-BP预测模型。选择平均相对误差作为误差评估指标，结果表明，三种预测模型均能对光伏电站的发电功率进行预测，但是单一的BP神经网络模型误差较大，晴天时，误差为5.1%，经遗传算法改进后的预测误差为4.9%，较单一模型提高了0.2%精度，而WPA-BP预测模型误差为4.4%，预测精度高于前者。同时多云天和雨天的时，均为WPA-BP模型的预测误差小，稳定性高，具有一定的研究价值。

Abstract:

Solar energy is rich in resources and widely distributed, and has been widely used in various applications. Photovoltaic power generation has become a reliable, feasible and extensible and important way to use renewable energy. Therefore, accurate prediction of photovoltaic output is of great significance. A year's photovoltaic power generation data and meteorological factors were obtained from a photovoltaic power station in Ningxia, and the data from April to May were selected for research and prediction. The convergence time of BP neural network is long and it is easy to fall into local minimum. A single BP neural network prediction model was established, gA-BP prediction model based on genetic algorithm (GA) optimization of BP neural network and WPA-BP prediction model based on Wolf pack algorithm (WPA) optimization of BP neural network were established. Average relative error is chosen as the error evaluation index, the result shows that the three kinds of prediction model can forecast photovoltaic power station of power, but a single BP neural network model of the error is bigger, sunny days, the error is 5.1%, the genetic algorithm improved the prediction error is 4.9%, increased by 0.2% than that of single model accuracy, and WPA - BP prediction model error is 4.4%, the prediction accuracy is higher than the former. At the same time, when it is cloudy and rainy, the PREDICTION error of WPA-BP model is small and its stability is high, so it has certain research value.