考虑样本不平衡的并行化用户负荷类型辨识方法

Parallel Load Type Identification Algorithm Considering Sample Class Imbalance

电力物联网背景下的态势感知技术,是以电力系统运行数据为驱动,通过对数据信息的获取、理解、分析预测,实现正确的决策与行动。该文针对此背景下对于海量用户负荷类型辨识和用电行为态势感知,提出一种基于MapReduce的并行化长短期记忆网络(parallel long short-term memory,Par-LSTM)的负荷分类方法,该方法通过将负荷数据分割交给多个LSTM并行处理以提高分类速度。并且采用一种考虑样本合成倍率的改进Borderline-SMOTE方法(BorderlineSMOTE considering synthetic multiple,SMB-SMOTE)处理负荷训练样本类别不平衡现象。算例表明,对SMB-SMOTE方法处理样本不平衡后的负荷数据进行Par-LSTM分类,能够有效提高小类负荷样本的分类精度。

Situational awareness technology in the context of the power Internet of things is driven by the operation data of power system.It can achieve correct decision and action by acquiring,understanding,analyzing and predicting data information.In view of this background,a parallel long short-term memory network(Par-LSTM)load classification method based on MapReduce is proposed for mass user load identification and situational awareness of power consumption behavior.In this method,the load data are divided for different LSTMs to process in parallel.In addition,an improved Borderline-SMOTE method(SMB-SMOTE)considering sample composition ratio is adopted to process load training.The example shows that the load data classified by Par-LSTM after class imbalance is processed in SMB-SMOTE method can effectively improve the classification accuracy of small load samples.