Enhancing Mitigation Algorithm For Energy Theft In Low Income Households Using Nature Inspired Artificial Intelligence

This research presents an innovative approach to energy theft detection in low-income households through the development of a hybrid algorithm that integrates Particle Swarm Optimisation (PSO), Ant Colony Optimisation (ACO), and Neural Network classification. The study addresses the critical challenge of energy theft in the Nigerian energy sector by creating a sophisticated synthetic dataset that simulates realistic consumption patterns and theft scenarios. This dataset provided a robust foundation for algorithm development and testing, incorporating temporal variations, householdspecific behaviours, and systematic anomalies. The hybrid PSO-ACO-Neural Network algorithm demonstrated exceptional performance, achieving perfect accuracy, precision, recall, and F1 scores in theft detection. Comparative analysis with a Support Vector Machine (SVM) benchmark highlighted the superior capabilities of the proposed hybrid approach, particularly in minimising false positives. The algorithm's optimal performance was achieved with minimal computational resources, requiring only four particles and four ants, making it particularly suitable for deployment in resource-constrained environments. The research identifies several limitations, including reliance on synthetic data and potential overfitting, whilst providing comprehensive recommendations for real-world implementation. These include pilot programme deployment, continuous learning mechanism integration, and strategic partnerships with electricity distribution companies. This study contributes significantly to the field of energy theft detection, offering a promising foundation for practical applications in low-income communities and future research directions in utility security.