Machine Learning Model for Predicting Evaporation Losses in Reservoirs

This paper presents a modification in method Support Vector Regression applied in the prediction of evaporation losses in reservoirs. For this approach, a penalty constant was included in the training phase (adjustment) and in the test phase of the SVR method, being called a method correlated with SVR. For training and testing, the predictive variables are the mean temperature (oC), the average wind speed (m/s), sunshine hours (h/day) and the average relative humidity (%). However, the output variable was the evaporation value (mm/day). The data used correspond to the Manasgaon water reservoir, reservoir located in Anand Sagar, Shegaon, India. The evaluation took into account the error values and the correlation coefficient as a comparison criterion. In the end, the results were compared with the values obtained by the conventional SVR method and by artificial neural networks, obtained smaller values for the error if it were more efficient for this application.