Demand-side Resources Classification Model Based on Time-frequency Domain Feature Extraction and IAGA Feature Selection

Abstract

The demand-side resources classification plays an important role in demand-side resources aggregation and incentive policy formulations. Considering the data of demand-side resources has the characteristics of large scale, high dimension, and low value density, both feature extraction and feature selection are two key steps of classification. While the conventional methods have yet to be improved due to the absence of frequency domain feature and the deficiency of feature selection, which lead to the low accuracy. To this end, this paper proposes a demand-side resources classification model based on time-frequency domain feature extraction and improved adaptive genetic algorithm (IAGA). Firstly, this paper extracts the power consumption characteristics, ratio characteristics and statistical characteristics of demand response periods, and uses the Hilbert-Huang Transform (HHT) to decompose the data and extract the frequency domain features. Then the paper proposes IAGA to select the time-frequency domain features optimally. Finally, in terms of classifier selection methods, this paper compares three traditional classifiers: Support Vector Machines (SVM), Artificial Neural Network (ANN) and K-Nearest Neighbor (KNN). The experimental results show that compared with the traditional feature selection method, the IAGA-SVM classification method with 8 time-frequency domain features can effectively improve the accuracy of demand-side resources classification (82.12%).