Explainable Artificial Intelligence for Prediction of Non-Technical Losses in Electricity Distribution Networks

There is a growing concern about the high degree of non-technical losses (NTL) in developing countries especially sub-saharan Africa. Whereas several studies have employed artificial intelligence (AI) to analyze NTL, a major drawback in these studies is the focus on customer data only without considering the possible contribution of electricity distribution staff to NTL. This study introduces a novel approach to NTL reduction by analyzing a combined dataset of staff operational processes and customer consumption data. A deep-learning architecture called non-technical losses convolutional neural network (NTLCONVNET) was developed which consists of a series of three one-dimensional convolutional neural networks (1D-CNN) with different depths combined with several fully connected layers. Furthermore, limited or no research has studied the decision rationale influencing how AI models interpret the significance of features in predicting NTL. To achieve the explainability of the model, SHapley Additive exPlanations (SHAP) kernel and tree-based explainers were used for the deep and ensemble learning models respectively to determine the relative importance of the variables and how they contribute to the overall model prediction. A novel ranking framework was used to compute the holistic ranking of the variables across multiple models. The finding suggests that the staff-related variables omitted in the extant literature are significant predictors of NTL. The NTLCONVNET was compared with 5 ensemble learning algorithms and the results show that the NTLCONVNET significantly surpasses all other models, scoring 0.844, 0.838, 0.836 and 0.836 on weighted average Precision, Recall, f1 and accuracy respectively. This study suggests a policy outcome of introducing human resource metrics into NTL reduction strategies.