面向海量用户用电数据的集成负荷预测

Aggregated Load Forecasting Based on Massive Household Smart Meter Data

智能电表的普及为准确把握用户级负荷变化规律提供了数据基础,使从用户级负荷入手的短期负荷预测成为可能。在此背景下,提出了一种计及海量用户数据信息的集成负荷预测方法。考虑居民用户用电模式差异性,首先以用户负荷数据为样本进行聚类,其次分别建立不同用户群的负荷预测模型,最后将各用户群的负荷预测值汇总得到全局预测结果。使用多种经典聚类与预测方法相结合,以最佳聚类数为切入点,探讨了集成预测策略在不同预测时间尺度下对预测精度的提升效果,并通过实际算例证明该策略的有效性。

Smart meters provide data foundation to find out residential level consumption patterns, making short-time load forecasting enabled for residential customers. Considering the difference among massive individual residential customers, this paper proposes a cluster-based aggregated forecasting strategy, i.e. firstly clustering the households, then establishing forecasting models separately and finally aggregating the forecasts. This paper uses classical clustering and forecasting methods to determine the optimal cluster number in 1 h and 24 h ahead forecasting. Effectiveness of this strategy is verified with an actual data example.