Anomaly Detection in Animal-Related Failures in Overhead Distribution Systems

Occurrences of faults in overhead distribution systems are a significant factor in distribution system reliability. Analysis of animal-related failures has a practical value since animals cause large number of faults in overhead distribution systems. The artificial immune system approach for self-nonself discrimination and its application to anomaly detection problems in engineering is showing great promise. This paper presents a method to apply anomaly detection in weekly animal-related failures in overhead distribution system. Detecting anomalous behavior in power distribution systems can be of much use for maintenance-related decision-making in electrical utility companies. A seminal contribution in anomaly detection is the V-detectors algorithm that can very effectively cover the nonself region of the feature space with a set of detectors. The detector set can be used to detect anomalous inputs. In this paper, procreating approaches to create an effective set of V-detectors is considered. The first phase of the algorithm generates an initial set of V-detectors. In subsequent phase, new detectors are grown from existing ones, by means of a mechanism called procreation. Procreating detectors can more effectively fill hard-to-reach interstices in the nonself region, resulting in better coverage. Simulation results indicate that this combined method detects more anomalies and displays better performance.