利用最佳相似日的光伏电站短期出力预测

Short-term forecasting for photovoltaic power generation based on optimal similar set

提出一种基于最佳相似日的光伏电站短期出力预测方法。该方法利用密度指标确定初始聚类中心优化K-means聚类算法,采用加权欧式距离法获得历史样本的出力水平相似日集,采用相关系数法获得历史样本的曲线形状相似日集,确定预测日出力水平相似日集和曲线形状相似日集,选取两集合的交集样本作为最佳相似日。建立BP神经网络出力预测模型,采用光伏电站的实测数据训练预测模型,对比不同类型天气的预测结果与实测数据,表明论文的预测方法具有较高的预测精度。

This paper proposes a short-term power forecasting method for photovoltaie generation based on optimal similar days. Density index is adopted to choose the initial clustering centers to op- timize K-means algorithm. Weighted Euclidean distance is used to get power generation level set of historical samples. Con＇elation coeffi- cient is used to get power generation shape set of historical samples. The intersection set of power level and power shape is the optimal similar set. Forecast model of BP neural network is constructed and trained by operation data from photovaltaic power station. Forecast re- sults reveal that the proposed method has high accuracy by compar- ing the forecast results and operation data in different type weathers.