Predicción del nivel de agua del embalse, basado en redes neuronales, para la mejora de la planificación de producción de energía en la Central Hidroeléctrica Agoyán.

Abstract

The objective of this research was to predict the future water flow of the Pastaza River in order to determine the reservoir level of the dam, for the purpose of planning the production of electrical energy at the Agoyán hydroelectric power plant. Historical data of the water flow, its level and active power was used ranging from the year 2005 until 2016. The records were uploaded to the database manager PostgreSQL . A nonprobabilistic sample was chosen to which a data filter was applied, a division by climactic seasons and a logarithmic transformation. A software platform was created using the Python language, which allowed for the training of artificial neural networks based on the NeuroLab library and the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm for flow prediction and dam modeling, as well as the tools for the testing of effectiveness of the models. The system was tested for twenty-two (22) days obtaining a RMSE average of 9.96 cubic meters per second, with an accuracy percentage of 45.45% for the flow predictor and a RMSE average of 0.2735 meters above the sea level with an accuracy percentage of 86.36%; in a threshold of eight (8) hours of prediction. it was concluded that the forecast threshold reached doesn’t yet affect the production indices, for which the implementation of weather stations at the basins of the Chambo and Patate rivers is recommend.