Cross-Domain Transfer Learning Techniques for Power Consumption Forcasting

Le résumé fourni par la source

The aim of this study is to improve power consumption prediction accuracy using LSTM-based transfer learning and compare it with XGBoost. High-accuracy machine learning models are needed for accurate predictions of power consumption in order to optimize energy consumption, cost reduction, and stability of the grid.A historical power consumption data set from diverse domains were employed. The study employs a two-group experimental design:Group 1 leverages the LSTM-transfer learning model, which draws on pre-trained knowledge for enhanced forecasting accuracy.Group 2 uses the XGBoost optimized method for time-series prediction.Model performance was assessed based on accuracy, precision, recall, F1-score, and mean absolute error (MAE). Validity test was conducted using G Power with 80% power, 0.05 p-value, and 95% confidence level.LSTM worked better than XGBoost with higher accuracy, lower variability (SD), and improved forecasting capability as indicated by its higher F1-score and MAE.The findings establish that transfer learning using LSTM is better than XGBoost in forecasting electricity consumption since it is able to learn sequence patterns and transfer across domains. Upcoming research needs to focus on enhancing LSTM for real-time forecasting and examining hybrid models for increased accuracy and scalability.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.