Short-Term PV Power Forecasting Based on LSTM and Multi-Head Attention Mechanism

This paper presents a novel hybrid model for accurate short-term photovoltaic (PV) power forecasting, integrating Long Short-Term Memory (LSTM) network and multi-head attention mechanism. The model is trained and evaluated using a dataset from the Desert Knowledge Australia Solar Centre (DKASC), covering diverse weather conditions. A comprehensive data preprocessing workflow, including correlation analysis, normalization, outlier detection, missing value imputation, and sliding window mechanism, ensures robustness and generalization. Bayesian optimization is employed for hyperparameter tuning. Experimental results demonstrate superior performance compared to common models across various weather conditions.