Abstract: With the rapid development of the power Internet of Things, clarification of user load status are significant for improving the quality of power supply service, power price decision-making, demand-side response, and paid precise services. The classification and identification of industrial power equipment based on non-intrusive methods is faced with the challenges of poor prior knowledge of industrial loads and low sampling frequency of the used data. Using the power quality monitoring data of large industrial users, this paper proposes a classification and identification method of power equipment for industrial users based on the level of harmonic emission. Firstly, a detection method is proposed for industrial equipment switching events based on bilateral cumulative sum algorithm. Secondly, the typical characteristics of each event is extracted to construct a feature matrix for industrial user equipment, and a feature screening method with high contribution rate is proposed to reduce the number of features. A classification and identification method of power equipment for industrial users based on k-means algorithm and silhouette coefficient is proposed. The 14-day power quality monitoring data of a rolling mill user under the 10 kV voltage level in Shanghai area was used to verify the proposed method. The result proves that the proposed method has high accuracy in classification of industrial user equipment with unknown equipment conditions, and has significant practical and promotion values.

Key words: industrial equipment classification, harmonic emission level, bilateral CUSUM algorithm, high contribution rate feature, silhouette coefficient, k-means algorithm