基于时间序列的大型光伏电站发电短期预测

Short-term forecasting method of large-scale photovoltaic power station based on time series

环境变化条件下,短期精确预测光伏(photovoltaic,PV)发电量对于确保电网运行、调度和电网能源管理至关重要,尤其是在未记录太阳辐射测量或天气参数历史值的位置。通过搭建基于人工神经网络(双向长期短期记忆)的方法和统计方法(季节性自回归综合移动平均值)模型,对比分析大型光伏发电量的时间序列预测结果,考虑预测时间范围变化对所有算法的影响。当前工作中使用的数据为从20 MW并网光伏电站获取的3640 h运行数据。人工神经网络和所提统计模型可用于提前1 h准确预测光伏电站的发电量,对光伏系统与智能电网的集成起重要指导作用。

Short-term accurate prediction of photo-voltaic(PV)power generation is essential to ensure grid operation,dispatch and grid energy management in changing environmental conditions,especially in locations where solar radiation measurements or historical values of weather parameters are not recorded.By building an artificial neural network(two-way long-term short-term memory)method and a statistical method(seasonal auto-regressive comprehensive moving average)model,the time series forecast results of large-scale photo-voltaic power generation are compared and analyzed,the changes in the forecast time range for all algorithms impact are considered.The data used in the current work is 3640 h operating data obtained from 20 MW grid-connected photo-voltaic power plants.The artificial neural network and the proposed statistical model can be used to accurately predict the power generation of photo-voltaic power plants 1 h in advance,and play an important guiding role in the integration of photo-voltaic systems and smart grids.