Classification Method of Customer Based on Load Curve Image Information

The classification of customer is an essential basis for demand-side-oriented power system operation and control research. In order to accurately identify the customer behavior, this paper proposes a classification method of customer based on load curve image information. Inspired by the application of deep learning theory in image processing, the one-dimensional load series is transformed into a two-dimensional recurrence plot as the input sample, then convolutional neural network (CNN) and support vector machine (SVM) algorithms are combined to construct a load feature extraction and customer classification model, so that the machine vision technique can work to extract curve feature without temporal information loss of the load data. SVM is introduced as the classifier in the CNN model, and the optimal classification hyperplane is built in the high-dimensional feature space by adjusting the classification threshold and hyperplane normal vector which enables the customer classification to accurately reflect the daily load curve feature. Simulations are performed using practical load data of a city in China, and the results show that the load curve image information based method has satisfactory performance in terms of the effectiveness of customer classification and the accuracy of feature extraction.