Artificial Neural Networks in Distribution Management Systems: Improving FLISR Efficiency

Failures in electrical systems present significant obstacles to ensuring a dependable and continuous power supply, potentially leading to power outages, equipment damage, and interruptions to critical services. This emphasizes the significance of executing efficient fault detection, prevention, and restoration techniques. This study investigates the application of Artificial Neural Networks (ANNs) in 11kV Distribution Management Systems (DMS) to enhance the efficiency of Fault Location, Isolation, and Service Restoration (FLISR) processes. A comprehensive methodology is proposed, integrating power flow equations, FLISR technology, and a real-time database, implemented through MATLAB/Simulink and optimized using the Levenberg-Marquardt algorithm. An 11kV distribution network is modeled, and fault data are collected through simulations and real-world scenarios. A three-layer ANN architecture is designed and trained using the fault dataset to detect, classify, and locate faults. The network's performance is evaluated through regression analysis and validation curves, demonstrating a high correlation between predicted and actual values. The optimal validation performance is achieved at the 258th epoch, with a Mean Squared Error of 49.5591. The results indicate the model's efficacy in simulating various fault conditions and accurately predicting fault effects, with slight underestimation and good generalization to unseen data. The implementation of this ANN-based approach in real-world power distribution systems has the potential to significantly reduce downtime and improve overall system reliability by enabling rapid fault detection and localization.