Improved tri-training method for identifying user abnormal behavior based on adaptive golden jackal algorithm

• Published in: AIP advances Vol. 13; no. 3; pp. 035030 - 035030-7
• Main Authors: Wang, Kun, Gao, Jinggeng, Kang, Xiaohua, Li, Huan
• Format: Journal Article
• Language: English
• Published: Melville American Institute of Physics 01.03.2023; AIP Publishing LLC
• Subjects: Adaptive algorithms; Electricity consumption; Identification methods; Local optimization; Training
• ISSN: 2158-3226; 2158-3226
• DOI: 10.1063/5.0147299

Identification of abnormal user behavior helps reduce non-technical losses and regulatory operating costs for power marketing departments. Therefore, this paper proposes an adaptive golden jackal algorithm optimization improved tri-training method to identify user abnormal behavior. First, this paper constructs multiple weak learners based on the abnormal behavior data of users, combined with the method of sampling and putting back, and uses the filtering method to select the tri-training base model. Second, aiming at the problem that the traditional optimization algorithm has a slow convergence speed and is easy to fall into local optimization, the adaptive golden jackal algorithm is used to realize the parameter optimization of tri-training. Based on the electricity consumption data of a certain place in the past five years, it is found that the model can provide stable identification results: accuracy = 0.987, f1-score = 0.973.