基于灰色关联分析和K均值聚类的短期负荷预测

在基于相似日的短期电力负荷预测技术中,相似日的选取影响着负荷预测精度,提出一种基于灰色关联分析(grey relation analysis,GRA)和K均值(K-means)聚类选取相似日的短期负荷预测模型。首先,采用灰色关联分析方法选取相似日粗集,再对相似日粗集的外部因素使用K均值聚类。然后,计算待预测日与聚类中心的欧氏距离,将距离最小一类作为最终相似日集合。最后,利用最终相似日集合训练长短期记忆(long-short term memory, LSTM)神经网络,进行负荷预测。与未采用相似日的LSTM模型和采用传统的灰色关联分析的LSTM模型相比,所提方法的平均绝对百分比误差(mean absolute percentage error,MAPE)分别降低了0.911%、0.637%。算例分析表明,采用GRA-K-means选取相似日可以有效提升短期电力负荷的预测精度。

Short-Term Load Forecasting Based on Similar-Day Selection with GRA-K-means

As the selection of similar days affecting the accuracy of load forecasting, a new short-term load forecasting model using GRA-K-means to select similar days is proposed in this paper. Firstly, GRA is used to select similar days to obtain rough sets of similar days. Secondly, K-means is used to cluster the external factors of rough sets of similar days, and then calculate the Euclidean distance between the predicated days and cluster center. The minimum distance is regarded as the final similar day set. Finally, the final similar day set is applied to train LSTM forecast model of neutral network to load forecast. Compared with LSTM model without similar days or LSTM model using traditional GRA, the MAPE in this paper is partly reduced by 0.911% and 0.637%. The analysis of the results shows that the model using GRA-K-means to select similar days can effectively improve the accuracy of short-term load forecasting.