Artificial immunity-based energy theft detection for advanced metering infrastructures

作     者：Fu, Jie Yang, Chengxi Liu, Yuxuan Zhang, Kunsan Li, Jiaqi Li, Beibei 

作者机构：State Grid Fujian Elect Power Co Ltd Fuzhou 350003 Fujian Peoples R China Sichuan Univ Sch Cyber Sci & Engn Chengdu 610207 Sichuan Peoples R China 

出 版 物：《INTERNATIONAL JOURNAL OF CRITICAL INFRASTRUCTURE PROTECTION》 (Int. J. Crit. Infrastruct. Prot.)

年 卷 期：2025年第48卷

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foun-dation of China Sichuan Science and Technology Program [2024YFHZ0086, 2025YFHZ0014] 

主　　题：Advanced metering infrastructure (AMI) Energy theft detection Artificial immunity Negative selection algorithm 

摘      要：Advanced Metering Infrastructure (AMI) is envisioned to enable smart energy management and consumption while ensuring the integrity of real energy consumption data. However, existing smart meters, gateways, and communication channels are usually weakly protected, often opening a huge door for data eavesdroppers who may be easily to further construct energy thefts. Although some energy theft detection schemes have already been reported in the literature, they often fail to take into account the dense data distribution characteristics of energy consumption data, resulting in compromised detection performance. To this end, we in this paper propose a novel arTificial IM mune based E nergy theft D etection (TIMED) scheme, which can effectively identify five types of energy thefts. Specifically, we first develop an energy consumption data pre-processing method, which can effectively reduce the dimensionality of raw energy consumption data to facilitate the data analyzing efficiency. Second, we design a center-distance-based energy theft detector generation method to create high- quality detectors with low elimination rates. Last, we devise a nonself-based hole repair method for energy theft detectors, which can further reduce the false negative alarms. Extensive experiments on areal public AMI dataset demonstrate that the proposed TIMED scheme is highly effective in identifying pulse attacks, scaling attacks, ramping attacks, random attacks, and smooth-curve attacks. The results show that TIMED outperforms many existing machine learning and traditional artificial immunity-based energy theft detection methods.