Meteorological Variables Forecasting System Using Machine Learning and Open-Source Software

The techniques for forecasting meteorological variables are highly studied since prior knowledge of them allows for the efficient management of renewable energies, and also for other applications of science such as agriculture, health, engineering, energy, etc. In this research, the design, implementation, and comparison of forecasting models for meteorological variables have been performed using different Machine Learning techniques as part of Python open-source software. The techniques implemented include multiple linear regression, polynomial regression, random forest, decision tree, XGBoost, and multilayer perceptron neural network (MLP). To identify the best technique, the mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute error (MAE), and coefficient of determination (${\mathrm{R}}^2$) are used as evaluation metrics. The most efficient techniques depend on the variable to be forecasting, however, it is noted that for most of them, random forest and XGBoost techniques present better performance. For temperature, the best performing technique was Random Forest with an ${\mathrm{R}}^2$ of 0.8631, MAE of 0.4728 °C, MAPE of 2.73%, and RMSE of 0.6621 °C; for relative humidity, was Random Forest with an ${\mathrm{R}}^2$ of 0.8583, MAE of 2.1380RH, MAPE of 2.50% and RMSE of 2.9003 RH; for solar radiation, was Random Forest with an ${\mathrm{R}}^2$ of 0.7333, MAE of 65.8105 $\mathrm{W}/{\mathrm{m}}^2$, and RMSE of 105.9141 $\mathrm{W}/{\mathrm{m}}^2$; and for wind speed, was Random Forest with an ${\mathrm{R}}^2$ of 0.3660, MAE of 0.1097 $\mathrm{m}/\mathrm{s}$, and RMSE of 0.2136 $\mathrm{m}/\mathrm{s}$.