基于深度聚类的居民用户电力负载模式识别

Recognition of User Power Load Pattern Based on Deep Clustering

电力用户负载模式的识别有利于引导用户参与需求侧管理,提高能源的利用效率。面对电力负荷数据日趋海量的情况,传统K-Means算法无法有效处理,且存在无法准确衡量高维电力数据距离、对噪声和异常数据敏感等问题,文章提出一种基于自编码器(auto-encoder,AE)与K-Means算法结合的方法。首先利用自编码器来提取出电力数据潜在的特征,然后利用K-Means算法对提取出的抽象特征进行聚类,最后利用自编码器与聚类的联合误差优化整个模型,使聚类效果更好。通过英国电力网络领导的低碳项目中伦敦家庭的电力数据集来验证方法效果,证明该方法能提供准确的聚类结果,有效分析出不同类型的负载模式。

The identification of power user load pattern is helpful to guide users to participate in demand side management and improve the efficiency of energy utilization. In the face of the increasing amount of power load data, the traditional K-Means algorithm has no effective processing method, and there are some problems, such as unable to accurately measure the distance of high-dimensional power data, sensitive to noise and abnormal data, this paper proposes a method based on the combination of auto-encoder(AE) and K-Means algorithm. Firstly, the potential features of power data are extracted by auto-encoder, and then the extracted abstract features are clustered by K-Means algorithm. Finally, the joint error of auto-encoder and clustering is used to optimize the whole model to make the clustering effect better. The effectiveness of the method is verified by the power data set of London households in the low-carbon project led by the British power network. It is proved that the method can provide accurate clustering results and effectively analyze different types of load patterns.