Abstract
With the increasing demand for clean energy, there is a growing requirement for accurate load forecasting due to increased electricity usage. This research implements the attention-based mechanism using a convolutional neural network (CNN) based on a bidirectional LSTM model (CNN-BiLSTM) for short-term electrical load prediction. The main objective is to improve forecasting accuracy by incorporating attention processes into the CNN-BiLSTM architecture. The models are trained and tested using historical electricity data. The Attention-based CNN-BiLSTM model is compared with the BiLSTM and CNN-BiLSTM models. The results indicate that the Attention-based CNN-BiLSTM model is better than others, showing higher accuracy, lower root mean square error, and stronger correlation coefficients. The results indicate that incorporating CNN layers and attention processes into the BiLSTM model greatly improves load prediction accuracy, especially for 3-day and 7-day electricity forecasts.

This publication has 10 references indexed in Scilit: