Deep Belief Network-Based Anomaly Recognition Method of Power Supply Service Work Orders

Abstract

The recognition of abnormal data from the power supply service work orders is a key issue to improve the processing efficiency of power supply work orders as well as the quality of power supply services. However, the existing abnormal work orders recognition mainly relies on manual screening methods, and the efficiency and accuracy of recognition depend on expert experience. Given this background, an anomaly recognition method of power supply service work orders based on deep belief network is proposed. First, a pre-processing method for power supply work orders based on non-negative matrix factorization (NMF) is proposed to implement initial recognition of abnormal work orders from historical data, and reduce noise impact during the collection process of work orders. Second, with the crucial factors and characteristics that affect the number of power supply work orders considered, the multi-dimensional feature-based anomaly characteristic indices of work orders are presented to reflect features of abnormal work orders comprehensively. Then, an anomaly recognition model of power supply work orders based on deep belief network (DBN) is constructed. Finally, case studies on actual power supply work orders of power supply company in Zhejiang province, China are performed for verifying the effectiveness of the proposed method. The results show that the proposed method can accurately and efficiently recognize abnormal work orders.