摘要
针对配电网线路负荷短期预测存在精度不佳的问题,提出一种基于聚类及趋势指标的长短期神经网络负荷预测方法。首先利用K-Means聚类方法将线路下特性变化相似的配电网台区负荷数据进行聚类重构;此外,按类分别计算负荷历史同期数据的趋势变化指标,并且作为负荷预测模型的输入特征;然后,建立能够传递时间序列信息的长短期神经网络预测模型,通过模型学习训练每类负荷的历史数据及变化趋势,并对测试集进行预测,将每类负荷的预测结果进行叠加得到线路总负荷的预测结果。以湖南某线路负荷数据为基础,预测未来一天96个点的线路负荷数据,经验证,所提方法能够深入挖掘配电负荷的特性规律和变化趋势,提升配网线路负荷的短期预测精度。
Aiming at the problem of poor accuracy in short-term load forecasting of distribution network lines, this paper proposes a long short-term neural network load forecasting method based on clustering and trend indicators. First, the K-Means clustering method is used to classify the distribution network stations with similar load characteristics. Then, on the basis of each cluster of load data the seasonal trend change index from historical load data is calculated as the input feature of load prediction model. Besides, the long short-term neural network model that can transmit time series information is established to train and predict each cluster load. Finally, the prediction results of each cluster load are superimposed to obtain the total line load prediction results. Based on the load data of a certain line in Hunan, the line load changes at 96 points in the feature are predicted. It is verified that the method proposed can deeply excavate the characteristics and changing trends of power load, improving the short-term prediction accuracy of distribution network line load.
作者
邓威
郭钇秀
李勇
周王峰
乔学博
罗威成
DENG Wei;GUO Yixiu;LI Yong;ZHOU Wangfeng;QIAO Xuebo;LUO Weicheng(State Grid Hunan Electric Power Company Limited Research Institute,Changsha 410007,China;Hunan University,Changsha 410082,China)
出处
《湖南电力》
2021年第4期27-33,共7页
Hunan Electric Power
基金
国网湖南省电力有限公司电力科学研究院科技项目(B316A5200013)。
关键词
短期负荷预测
K均值聚类
长短期神经网络
趋势指标
智能配电网
short-term load prediction
K-means cluster
long short-term neural network
trend indicators
smart distribution grid
