Analisis Perbandingan Metode Mamdani Fuzzy Inference System dengan Regresi Berganda dalam Estimasi Konsumsi Energi Listrik di Provinsi Sumatera Utara

Abstract

The continuously increasing demand for electricity presents a significant challenge in energy supply, particularly because electricity cannot be stored on a large scale. Therefore, accurate estimation methods are essential to avoid shortages or surpluses in power capacity, both of which can lead to substantial losses. This study aims to compare the accuracy of electricity consumption estimation in North Sumatra Province using the Mamdani Fuzzy Inference System (FIS) and multiple linear regression methods, both in their standard forms and after being modified with the Walk Forward Validation (WFV) approach and basic assumptions (AD). The independent variables used include Gross Regional Domestic Product (GRDP), population, and number of customers. The results show that all methods applied (fuzzy AD, fuzzy WFV, standard fuzzy, regression AD, regression WFV, and standard regression) demonstrated very high accuracy, with MAPE values below 10%. The regression WFV method achieved the highest accuracy with a MAPE of 0.83% and MSE of 11,137.51, followed by regression AD, standard regression, fuzzy AD, standard fuzzy, and fuzzy WFV. The analysis of the influence of the independent variables reveals that GRDP and the number of customers have a positive impact, while population has a negative impact on electricity consumption. This study also confirms that the Mamdani FIS can be applied to linear data, provided that appropriate approaches such as walk forward validation and basic assumptions are used. The findings recommend regression WFV as the most suitable method for predicting electricity consumption, particularly for linear data, while Mamdani FIS can be employed for both linear and non-linear data with the appropriate logic framework.