Metodología para la evaluación de desempeño de plantas solares fotovoltaicas a través del uso de la ciencia de datos

Abstract

The continuous evaluation of solar photovoltaic plants is essential for their operation. Their main variables must be monitored to verify that the electrical energy is delivered under optimal operating and efficiency conditions. This research presents a methodology based on data science to evaluate solar photovoltaic plants. This methodology was applied to the data set of a solar plant of the US National Renewable Energy Laboratory, analyzing the data to obtain temporal curves of irradiance and energy, as well as the leading performance indicators. Also, this study used the K-Means algorithm to generate clusters within the data set and the K-NN algorithm to create class prediction models of the energy and PR indicator. Clusters grouping the generated power values and the PR values were obtained. The energy class classification model had an accuracy of 91.67%, while the PR indicator class classification model had an accuracy of 83.33%. Since the average fouling rate in the monthly and annual scales was above 90%, while those of the PR were around 70%, a study is recommended to determine the origin of the losses in the plant. It is also suggested that a model be developed to determine the impact of ambient temperature, PV module temperature, and wind speed on  of electric power production.