Comparative Analysis of Short-Term Load Forecasting Techniques: Conventional Deep Learning Models Versus Proposed LSTM with Attention Layers

This research article explores a comparison of both conventional deep learning models and proposed a LSTM model equipped with attention layers for the short-term load forecasting (STLF). The study evaluates the accuracy and efficiency of both models using the power system data of Delhi city, India. The forecasts with the proposed LSTM model with attention layers outperform the conventional time-series models results better accuracy with a 1.9% Mean Squared Error (MSE) which is lower as compared to 2.35% for the best-performing conventional model. The proposed model also exhibits better interpretability, as the attention mechanism allows for the identification of significant input features that contribute to the forecasting performance. The proposed model is a promising technique for STLF in real-time applications. Overall, the study highlights the potential of using LSTM model with attention layers for accurate and efficient STLF in power systems.