Development of an LSTM-based Model for Energy Consumption Prediction with Data Pre-analysis

Electricity demand is increasing rapidly due to growth in development. Based on this trend, it is important to plan energy usage efficiently to eliminate energy waste and thus reduce carbon emissions. Towards more accurate energy consumption predictions, this study focuses on the time series data analysis and Long Short-Term Memory model in predicting energy consumption. The initial data analysis techniques adopted could be used to detect energy usage patterns and to gain a better understanding of the data. Such data analysis is important since it is crucial to understand the data before selecting an appropriate model to make predictions. The data analysis technique used was the augmented Dicky-Fuller test and the ETS Decomposition. Based on the nature and pattern of the data that have been analyzed, the LSTM method was adopted in generating energy consumption predictions. To determine the quality of prediction results, the accuracy-test methods used on the generated predictions were the Mean Absolute Percentage Error (MAPE), Mean Square Error (MSE) and R-square methods. The accuracy test results of this study showed that for all the datasets used, the highest MAPE value was 7.68%, while the MSE value was 10.23%, and thus proved that the LSTM model is highly accurate in making predictions.