Improving Deep Learning Forecasting Model Based on LSTM for Türkiye’s Hydro-Electricity Generation

Abstract

Electricity production in hydraulic power plants depends on the amount of water coming into the basin. This varies depending on precipitation such as snow and rain during the year, but when looking at the years, production is shaped according to the years when meteorological data are similar to each other. LSTM (Long Short-Term Memory) plays an important role in hydropower forecasting, as it is a special artificial neural network designed to model complex relationships on time series data, which is affected by various meteorological factors such as precipitation, temperature, and hydrological data such as water level, such as hydroelectric power production. Therefore, in this study, a forecast system based on the LSTM network model which is one of the deep learning methods was proposed for monthly hydropower-based electricity production forecast in Türkiye. The developed deep learning-based hydropower forecast model provides future production planning based on time series based on actual hydropower production data. Using real production data and LSTM learning models of different structures, monthly hydraulic electricity production forecasts for the next year were made and the models' performances were examined. As a result of this study, RMSE 32.4245 and MAPE 16.03% values and 200-layer LSTM model trained with 12-year data with 144 monthly data points containing hydroelectric generation information was obtained as the best model, and the performance values of the model showed that it was the correct forecasting model. The overall efficiency parameters of the found LSTM model were checked with NSE 0.5398 and KGE 0.8413 values. The performance of the model was found to be a high-accuracy model within acceptable limits and with a correlation value of R2 0.9035 to be very close to reality. The results obtained from this study have shown that deep learning models developed based on many years of production data give successful results in hydroelectric production prediction and can be used as a basis for electricity production planning.