Prediction of Electricity Consumption Demand Based on Long-Short Term Memory Network

Long Short-Term Memory (LSTM) networks are widely recognized for their ability to capture and retain long-term dependencies within time series data, making them a valuable tool for dealing with complex relationships between elements over extended periods of time. This research proposes a KerasTuner-based LSTM network to predict future electricity consumption and maximum demand. Dataset used in this study is historical electricity consumption data of a plastic manufacturing plant in Malaysia, collected at 30-min intervals from 1st January 2017 to 31st December 2019. Both random selection and KerasTuner-based hyperparameter tuning were used to determine the best hyperparameters. The results demonstrated that the KerasTuner-based LSTM approach is effective in predicting future electricity consumption and captures the complex dependencies within the electricity consumption data. The evaluation metrics, training time, and limits of the future maximum demand indicated the effectiveness of the proposed model. This is proven when the proposed model outperformed other models and could improve the prediction accuracy while saving time. This research shows that the proposed model could serve as a valuable tool for predicting maximum electricity demand and could be applied in other industries to provide crucial insights for energy planning and management.