[4Xin2-73] Improvement of the performance and explainability for electricity demand forecasting model through time-series XAI
Keywords:Deep Learning, Time series forecasting, XAI, Energy management
Due to the drastic changes in the energy environment driven by liberalization of electric market and the action towards a carbon-neutral society, efficient management of energy such as electricity and heat, is crucial for large-scale factories. The energy demand in factories is influenced by various factors such as equipment operation and weather conditions, causing it to fluctuate constantly. Therefore, there is a strong need for accurate methods to forecast future demand based on these conditions.
In recent years, various time-series forecasting methods in the field of deep learning have been proposed, which offer the potential for improved accuracy. However, deep learning models are black box models, lacking explainablity in understanding the reasoning behind the output.
In this study, focusing on electricity demand forecasting in factories, we attempted to construct a model that balances both explainability and performance using XAI techniques applicable to deep learning time-series models. Specifically, we employed the global interpretability method called TIME to understand the overall behavior of the model and achieve both explainability and performance. The results showed that by reconstructing the prediction model using the features extracted by TIME, we were able to achieve a balance between explainability and performance improvement.
In recent years, various time-series forecasting methods in the field of deep learning have been proposed, which offer the potential for improved accuracy. However, deep learning models are black box models, lacking explainablity in understanding the reasoning behind the output.
In this study, focusing on electricity demand forecasting in factories, we attempted to construct a model that balances both explainability and performance using XAI techniques applicable to deep learning time-series models. Specifically, we employed the global interpretability method called TIME to understand the overall behavior of the model and achieve both explainability and performance. The results showed that by reconstructing the prediction model using the features extracted by TIME, we were able to achieve a balance between explainability and performance improvement.
Please log in with your participant account.
» Participant Log In