Machine Learning with Meteorological Variables for the Prediction of the Electric Field in East Lima, Peru

Abstract

Environmental pollution and its effects on global warming and climate change are a key concern for all life on our planet. That is why meteorological variables such as maximum temperature, solar radiation, and ultraviolet levels were analyzed in this study, with a sample of 19564 readings. The data was collected using the Vantage Pro2 weather station, which was synchronized with the time and dates of the electric field measurements made by an EFM-100 sensor. The Machine Learning analysis was applied with the Regression Learner App, from which the linear regression model, regression tree, support vector machine, Gaussian process regression, and ensembles of tree algorithms were trained. The most optimal model for the prediction of the maximum temperature associated with the electric field was the Gaussian Process Regression with an RMSE of 1.3436. Likewise, for the meteorological variable of solar radiation, the optimal model was Regression Tree Medium with an RMSE of 1.3820 and for the meteorological variable of UV level, the most optimal model was Gaussian Process Regression (Rational quadratic) with an RMSE of 1.3410. Gaussian Process Regression allowed for the estimation and prediction of the meteorological variables and it was found that in the winter season at low temperatures the negative electric field is associated with high variability in its behavior; while at high temperatures they are associated with positive electric fields with low variability.