基于改进BP神经网络的光伏发电系统输出功率短期预测模型

A short-term prediction model to forecast output power of photovoltaic system based on improved BP neural network

随着光伏发电系统的大规模应用,其输出功率预测技术可以有效地缓解该类随机能源对电力系统的不利影响。提出了一种基于改进BP神经网络的光伏发电系统输出功率短期预测模型,利用输出功率的历史值、过往及预测日气象信息,对输出功率进行直接预测。通过对影响输出功率各项因素的分析,得出了预测模型输入变量选择的理论依据;为了提高模型在各种天气条件下的预测精度,提出了相似日选择算法和训练样本确定方法;针对传统BP学习算法易陷入局部极小点、收敛速度慢等缺陷,利用增加动量项和可变学习率相结合的方法对其进行了改进。最后通过预测结果分析,验证了所提模型和算法的有效性。

With the increasing application of large-scale photovoltaic （PV） system, output power forecasting technique can mitigate the adverse effects of such random energy on power systems. A short-term prediction model based on improved back- propagation （BP） artificial neural network is proposed to forecast output power of PV system directly, which uses historical output data and meteorology information, meteorology information of forecast day. Factors that affect PV system output are analyzed to obtain theoretical basis for the selection of input variables; similar day selection algorithm and training samples determination method are proposed to improve forecasting accuracy in different weather types; traditional BP algorithm is improved to avoid local minima and speed convergence by combining increasing momentum method with varying learning rate method; and the results validate the effectiveness of the proposed model and algorithm.