基于神经网络的吕梁市光伏电站发电量预测研究
赵红梅¹，杨洁²，贾景伟²
（1. 吕梁市乡村振兴局，山西 吕梁 033000；2. 吕梁学院物理系，山西 吕梁 033000）
摘要： 依据吕梁市光伏电站的历史发电数据和历史气象数据，使用BP 神经网络建立了光伏电站发电量预测模型。模
型一的输入变量为天气类型、最高温度、最低温度和前一日的发电量，模型二的输入变量为天气类型、最高温度、最低
温度和相似日的发电量。使用预测模型预测了2021 年5 月10 日至16 日连续7 天的发电量。其中模型一的平均绝对百分
误差为28.89%，模型二的平均绝对百分误差为16.39%。通过对比发现，使用相似日发电量作为神经网络模型的输入变
量可显著提高预测精度。
关键词： 光伏电站；发电量预测；神经网络；吕梁市
中图分类号： TK51     文献标志码： A     文章编号： 2095-0802-(2022)07-0021-03
Power Generation Forecasting of Photovoltaic Power Station in Lvliang City Based on
Neural Network
ZHAO Hongmei¹, YANG Jie², JIA Jingwei²
(1. Lvliang Rural Revitalization Bureau, Lvliang 033000, Shanxi, China; 2. Department of Physics, Lvliang University, Lvliang
033000, Shanxi, China)
Abstract: Based on the historical power generation data and historical meteorological data of Lvliang photovoltaic power station,
the power generation prediction model of photovoltaic power station is established by using BP neural network. The input variables
of model 1 are weather type, maximum temperature, minimum temperature and power generation of the previous day, and the
input variables of model 2 are weather type, maximum temperature, minimum temperature and power generation of similar days.
The prediction model was used to predict the power generation for 7 consecutive days from May 10 to 16, 2021. The average
absolute percentage error of model 1 is 28.89%, and the average absolute percentage error of model 2 is 16.39%. Through
comparison, it is found that using similar daily power generation as the input variable of neural network model can significantly
improve the prediction accuracy.

Key words: photovoltaic power station; power generation forecasting; neural network; Lvliang City