[Paper Review] Hydroelectric Generation Forecasting with Long Short Term Memory (LSTM) Based Deep Learning Model for Turkey

Mehmet Bulut|arXiv (Cornell University)|Sep 18, 2021

Energy Load and Power Forecasting18 references4 citations

TL;DR

This study proposes a deep learning model based on Long Short-Term Memory (LSTM) networks to forecast monthly hydroelectric power generation in Turkey using 10–12 years of historical production data. The 100-layer LSTM model achieved a MAPE of 0.1311 (13.1%) annually and 1.09% monthly average, demonstrating high accuracy in long-term hydropower forecasting when trained on at least 120 months of data.

ABSTRACT

Hydroelectricity is one of the renewable energy source, has been used for many years in Turkey. The production of hydraulic power plants based on water reservoirs varies based on different parameters. For this reason, the estimation of hydraulic production gains importance in terms of the planning of electricity generation. In this article, the estimation of Turkey's monthly hydroelectricity production has been made with the long-short-term memory (LSTM) network-based deep learning model. The designed deep learning model is based on hydraulic production time series and future production planning for many years. By using real production data and different LSTM deep learning models, their performance on the monthly forecast of hydraulic electricity generation of the next year has been examined. The obtained results showed that the use of time series based on real production data for many years and deep learning model together is successful in long-term prediction. In the study, it is seen that the 100-layer LSTM model, in which 120 months (10 years) hydroelectric generation time data are used according to the RMSE and MAPE values, are the highest model in terms of estimation accuracy, with a MAPE value of 0.1311 (13.1%) in the annual total and 1.09% as the monthly average distribution. In this model, the best results were obtained for the 100-layer LSTM model, in which the time data of 144 months (12 years) hydroelectric generation data are used, with a RMSE value of 29,689 annually and 2474.08 in monthly distribution. According to the results of the study, time data covering at least 120 months of production is recommended to create an acceptable hydropower forecasting model with LSTM.

Motivation & Objective

• To develop a deep learning model for accurate long-term forecasting of hydroelectric power generation in Turkey.
• To evaluate the performance of different LSTM architectures using varying lengths of historical hydroelectric production data.
• To identify the optimal model depth and input sequence length for minimizing forecasting error in annual and monthly predictions.
• To provide data-driven recommendations for hydropower planning based on model performance metrics like RMSE and MAPE.

Proposed method

• The study employs a Long Short-Term Memory (LSTM) recurrent neural network to model temporal dependencies in hydroelectric generation time series.
• Historical monthly hydroelectric production data from Turkey spanning up to 12 years (144 months) are used as input features.
• Multiple LSTM models with varying depths (e.g., 100 layers) and sequence lengths (120 to 144 months) are trained and compared.
• Model performance is evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) for both annual and monthly forecasts.
• The best-performing model is selected based on lowest RMSE and MAPE across test periods.
• Hyperparameter tuning focuses on sequence length and network depth to optimize long-term prediction accuracy.

Experimental results

Research questions

• RQ1 What is the optimal sequence length of historical hydroelectric production data for accurate long-term forecasting using LSTM models in Turkey?
• RQ2 How does the depth of the LSTM network influence forecasting accuracy for monthly and annual hydroelectric generation?
• RQ3 Can a deep learning model based on real production data achieve reliable long-term predictions of hydropower output in Turkey?
• RQ4 What are the minimum data requirements (in months) for building a high-accuracy LSTM-based hydroelectric forecasting model?

Key findings

• The 100-layer LSTM model trained on 144 months (12 years) of historical data achieved the lowest RMSE of 29,689 for annual forecasts and 2,474.08 for monthly forecasts.
• The same model recorded a MAPE of 0.1311 (13.1%) for annual total generation and 1.09% for monthly average distribution, indicating high forecasting accuracy.
• Using at least 120 months (10 years) of historical data significantly improves forecasting performance, with diminishing returns beyond 144 months.
• The 100-layer architecture outperformed shallower models in both RMSE and MAPE metrics, highlighting the benefit of deeper networks for long-term temporal modeling.
• The study confirms that LSTM models trained on long historical sequences can effectively capture seasonal and long-term trends in hydroelectric generation.
• The results support the use of deep learning with LSTM for reliable long-term hydropower planning in data-rich regions like Turkey.