Recurrent neural networks for predicting inflows in hydroelectric reservoirs
Publication date
2026-07-01
Document type
Konferenzbeitrag
Author
Organisational unit
Scopus ID
Conference
26th International Conference on Computational Science and Its Applications (ICCSA 2026) ; Braga, Portugal ; June 30 – July 3, 2026
Publisher
Springer Nature Switzerland
Book title
Computational Science and Its Applications – ICCSA 2026 Workshops
First page
272
Last page
284
Peer-reviewed
✅
Part of the university bibliography
✅
Language
English
Keyword
Climate AI
Data Analysis
Hydropower
Inflow Prediction
Neural Networks
Reservoir Management
Sustainability
Time-series
dtec.bw
Abstract
Hydropower plays a crucial role in many countries’ electricity systems, but its operation often conflicts with ecological preservation and agricultural water needs. Reservoirs must simultaneously deliver reliable energy, maintain minimum ecological flows to protect downstream ecosystems, and guarantee sufficient water for irrigation, making inflow forecasting critical for sustainable operation. In this work, we study neural network-based prediction of inflows to the Miranda hydropower reservoir in Brazil, aiming to provide accurate, data-driven forecasts that can support environmentally conscious and socially responsible decision-making. Using publicly available hydrological data, we perform a systematic exploratory analysis and pre-processing of hourly and daily time-series, including treatment of missing values, negative measurements, and outliers, as well as normalization and feature engineering. We then compare several recurrent neural network architectures (RNN, LSTM, BiLSTM, and GRU) under different training and model configurations. Our results show that a single-layer GRU model trained on the daily, normalized inflow series with a one-day lag achieves the best predictive performance. The selected model is used to produce 7-day ahead forecasts, which closely follow short-term dynamics while exposing limitations in anticipating abrupt peaks, thus highlighting the remaining challenges in forecasting extreme inflow events for hydropower management.
Version
Published version
Access right on openHSU
Metadata only access