Modeling of Electric Energy Consumption in Humid Zones Using Selected Meteorological Variables Through XGBoost, ANFIS, and RNN Approaches

This paper presents the results of forecasting electricity consumption in a humid zone, taking Togo as a case study. Consumption data were analyzed with the following input variables: Temperature (T), Relative Humidity (R), Precipitation (P), Wind Speed (W), and Sunshine Duration (S). XGBoost, ANFIS, and RNN are explored as modeling methods, with performance evaluated using R², MAE, MSE, and RMSE. A correlation analysis was conducted among all variables. The findings reveal correlations of 83% between relative humidity and precipitation; 73% between power consumption and precipitation; and 67% between power consumption and relative humidity. In contrast, only 8% correlation is observed between power consumption and temperature, and 4% between wind speed and sunshine duration. With respect to modeling, ANFIS metrics are found to be unsatisfactory. Its best performance yields R² = 41.3498% under the TRPWS configuration. XGBoost provides moderate results, with R² = 51.39% for the TRPWS configuration, representing its most acceptable model. By comparison, RNN delivers superior outcomes, with the majority of R² values exceeding 71%. The lowest performance, obtained with the PWS configuration, records RMSE = 909.7192, MAE = 567.9969, MSE = 827,589.1092, and R² = 71.45%. The highest performance, obtained with the TRPWS configuration, yields RMSE = 846.1036, MAE = 490.9964, MSE = 715,891.3167, and R² = 75.31%. Furthermore, residual analysis confirms that the distribution of errors aligns well with the Gaussian normal law. It is therefore concluded that RNN is well-suited for predicting electricity consumption in humid zones using the considered meteorological variables.