Regression Ensemble Techniques Bagging, Gradient Boosting and XGBoost to Predict the Power of Photovoltaic Systems in the High Andean Area

The production of electrical energy through the use of photovoltaic systems is very important, especially in the high Andean area and at altitudes above 3800 meters above sea level; knowing the phenomenon, managing it and being able to predict its behavior is very important to establish an electric energy management policy for the benefit of humanity. Ensemble techniques for regression such as Bagging, Gradient Boosting and XGBoost allow predictions of the power generated by photovoltaic systems based on predictor variables such as temperature, humidity, ultraviolet irradiance and illuminance. A photovoltaic monitoring system designed and implemented by the authors of this research has generated the dataset, it is possible to implement the same supervision system in other areas but only using the measurement of the predictor variables, without the need to implement the photovoltaic systems and determine the power thanks to the model generated by the ensemble techniques. The results show the lowest MSE using XGBoost with 4.1481, Bagging with Normalization has the second best MSE due to the simplicity of the technique. The lowest MAE is using XGBoost with 0.560257, the second lowest MAE is using Bagging with 0.577226. MAPE cannot be used because there are zero or close to zero values, and division by zero causes the MAPE value to tend to very high values. The best Score for the tests is XGBoost with 0.994991, very close are Bagging with Normalization and Gradient Boosting with Normalization.