为挖掘复杂环境因素对电力负荷预测效果的影响，提高电力负荷预测精确度，提出了一种基于k-shape时间序列聚类与STL季节趋势分解算法相结合的负荷曲线聚类预测模型(k-shape-seasonal and trend decomposition using loess-gradient boosting decision tree，k-shape-STL-GBDT)。首先分析用户用电时序特征，利用k-shape时间序列聚类算法根据负荷曲线划分用户聚类，其次，使用STL算法将不同簇的负荷数据划分为季节项、趋势项与随机项。然后，结合温度、湿度等影响因素搭建预测模型，以麻省大学smart*可再生能源项目的公开数据集为例进行分析，并与多种主流聚类分解预测模型进行对比。结果表明新提出的模型框架MAPE减少了4%以上，针对短期负荷预测表现出了较好的性能与预测精度。
User Short-Term Power Load Forecasting Model Based on k-shape_STL
To excavate the influence of complex environmental factors on power load forecas-ting,improve the accuracy of power load forecasting,a load curve clustering forecasting model based on time series clustering and seasonal trend decomposition algorithm(k-shape-STL-GBDT)is proposed.Firstly,the time series characteristics of users'electricity con-sumption are analyzed,and the k-shape algorithm is used to divide the user clusters accord-ing to the load curve.Secondly,the STL algorithm is used to divide the load data of different clusters into seasonal items,trend items and resid items.Then,we build a prediction model combining the influencing factors such as temperature and humidity and take the public data set of the UMass smart* renewable energy project as an example,and compare it with a va-riety of mainstream clustering decomposition prediction models.The results show that the newly proposed model framework MAPE reduces by more than 4%,and shows better per-formance and prediction accuracy for short-term load forecasting.
NETL
NSTL
万方数据
