Accurate PV Power Forecasting Using a Lightweight LSTM Model: A Case Study

This study presents a significant advancement in short-term photovoltaic (PV) power forecasting through the development and validation of a simple yet highly effective LSTM-based model tailored to address the operational demands of renewable energy integration. By harnessing a meticulously curated dataset and employing rigorous feature selection, the model achieved exceptional performance metrics an R² score of ۰.۹۲۱۲ and an RMSE of ۰.۰۶۵۰ on unseen data outperforming benchmark models such as CatBoost and GBR. These outcomes affirm the model's capacity to capture temporal dependencies in PV generation data while maintaining computational efficiency, making it well suited for real-time energy management applications. However, limitations such as dependence on high-quality input data and untested resilience under extreme weather conditions suggest areas for refinement. Future research could enhance the model by incorporating probabilistic forecasting, lightweight attention mechanisms, or transfer learning to improve adaptability across diverse geographic and climatic contexts. Ultimately, this work contributes a robust, practical tool to the evolving landscape of smart grid technologies, supporting the global transition toward sustainable energy systems with improved forecasting precision and scalability.

کلیدواژه ها:

Photovoltaic power forecasting ، LSTM ، short‑term prediction ، Renewable energy ، Time Series ، Feature Selection ، grid stability ، computational efficiency