Load prediction based on feature attention LSTM

This paper presents an innovative long short-term memory (LSTM) model augmented with feature attention mechanisms to enhance the precision of electricity load forecasting. This model leverages the robust time-series modeling capabilities of LSTM, coupled with the selective emphasis provided by attention mechanisms, to detect both immediate and extended dependencies along with pivotal load features within the dataset. The LSTM network within the model is tasked with capturing the sequential patterns inherent in load data. By integrating an attention mechanism, the model can automatically prioritize features that significantly affect prediction accuracy. It does so by allocating varying weights to these features, thereby generating enhanced load-influencing features for future predictions. This approach ensures that the model can effectively extract critical information from load data, leading to improved predictive accuracy. The Feature-Attention LSTM (FA-LSTM) model model has demonstrated a notable improvement in forecasting precision over traditional LSTM models, as evidenced by experimental results.