Time-series and deep learning approaches for renewable energy forecasting in Dhaka: a comparative study of ARIMA, SARIMA, and LSTM models
Mohammad Liton Hossain, S. M. Nasif Shams, Saeed Ullah
Abstract
Accurate forecasting of renewable energy generation is critical for sustainable energy planning in rapidly urbanizing cities like Dhaka. This study conducts a comprehensive comparative analysis of classical time-series models ARIMA and SARIMA and a deep learning model LSTM for long-term combined renewable energy (CED) forecasting. Feature engineering was employed to enhance model inputs and both statistical and recursive deep learning approaches were utilized to predict combined renewable energy (CED) density over a two-year horizon (2024–2025). The analysis used 10 years of hourly wind and solar energy data (2014–2023), comprising 87,648 data points, sourced from the NASA POWER database. The ARIMA(2,1,2) and SARIMA(2,1,2)(1,1,1,24) models captured basic temporal patterns but struggled with non-linear dynamics, achieving R 2 scores of − 0.0008 and − 0.1104, respectively. In contrast, the LSTM model achieved superior performance (R 2 = 0.9860) using a fixed test set and demonstrated strong generalizability under time-series cross-validation (Avg R 2 = 0.9847) by effectively learning complex temporal dependencies. The LSTM-based long-term forecasts demonstrated realistic seasonal trends and robust temporal consistency. These findings underscore the transformative potential of deep learning in enhancing renewable energy forecasting accuracy in developing urban regions, providing critical insights for future energy policy and infrastructure development in Dhaka City.