Short-Term Photovoltaic Power Generation Prediction Using Long Short-Term Memory Network Fusion Attention Mechanism

View at publisher Summarise with VIRAT
Authors: Xiangcheng Tong
Cited by: 0
References: 12

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper proposes a meteorology-driven attention Bidirectional Long Short-Term Memory Network (BiLSTM-MDA) to improve short-term photovoltaic (PV) power forecasting under dynamic weather conditions. Unlike existing models that inadequately distinguish historical information and lack physical constraints in attention mechanisms, the proposed method integrates meteorological prior weights derived from solar irradiance variation rates and a cloud cover abrupt change index to guide attention allocation across time steps. A BiLSTM encoder-decoder framework with a sliding window multi-step prediction strategy enables forecasts from 15 minutes to 2 hours ahead. Experiments on PV plant datasets with diverse climatic conditions show that BiLSTM-MDA outperforms LSTM, Transformer, and Attention-LSTM models. Notably, under abrupt weather transitions, MAE is reduced by up to 21.2 %, and error accumulation in multi-step prediction is effectively suppressed. The results demonstrate that embedding interpretable meteorological priors enhances accuracy, robustness, and generalization for renewable energy integration.

Fields of Study & Subjects

Solar Radiation and Photovoltaics Energy Load and Power Forecasting Photovoltaic System Optimization Techniques

Publication details

Journal / source
International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
Pages
497–501
Publication date
16 January 2026
Language
EN