An Attention-Integrated BiLSTM Approach for High-Precision Short-Term Load Forecasting
Syed Ahmad Saleem Bokhari, Syed Zarak Shah, Afnan Anwar Khan, Haider Saleem, Muhammad Abbas, Amjad Rehman Khan
Abstract
Modern power systems have an important issue regarding the trend of more significant demand, variability, and grid digitalisation, which requires precise ultra-short-term power load prediction to resolve the real-time operational stability and dispatch optimisation. Forecasts play a vital role in the economic efficiency of the electricity networks and the improvement of the safety of these systems. This paper introduced a new deep learning model, AC-BiLSTM, which combines BiLSTM networks and an Attention mechanism to deal with the non-linear and inter-temporal properties of the load data. The model used BiLSTM as a bidirectional temporal dependency-acknowledging means, whereas the attention mechanism helps increase focus toward the historically significant load patterns and minimise information loss during training. A fully connected layer subsequently produces the final load predictions. The suggested architecture has been tested against the real-like loads of data.