Aplicação de modelos de regressão espacial na análise das perdas não técnicas na rede elétrica de distribuição de baixa tensão, caso de estudo: Município da Cidade de Maputo

Abstract

In this study, non-technical losses (NTLs) in the Maputo City Municipality (MCM) were analyzed using Multiscale Geographically Weighted Regression (MGWR), which integrated technical and socioeconomic variables in a spatial context. Traditionally, studies on NTLs do not explicitly consider spatial location, which limits the understanding of the characteristics of the locations where losses occur. The results of the MGWR model demonstrated its superiority over the Geographically Weighted Regression (GWR) model, with a high coefficient of determination (R2 = 0.97) and a significantly lower Akaike information criterion (AIC = 0.173). This reflected the model's ability to capture complex spatial patterns at different scales and to adjust explanatory variables independently. The analysis of the residuals did not detect significant spatial autocorrelation, indicating the adequacy of the model for the data analyzed. The analysis revealed that variables such as Integrated Meters, Population Density and Split Meters have the greatest influence on NTPs. Integrated Meters, with 25% influence, were the most impactful variable, highlighting the vulnerability of this type of meter in densely populated urban areas, where control and monitoring are more difficult. Population Density, with 22% influence, was the second most important variable, indicating that intensive monitoring in high-density areas has a direct impact on reducing NTPs. Split Meters, with 20% influence, were also crucial, given that installing these devices inside homes increases their susceptibility to fraud