Explainable Multi-Horizon Deep Learning Framework for Household Energy Forecasting

Abstract

. The growing need for intelligent energy management systems has also highlighted the importance of accurate, interpretable models for household energy consumption forecasting. This paper presents an explainable multi-horizon deep learning model for forecasting household electricity consumption at hourly, daily, and weekly time horizons. The proposed model combines optimised preprocessing, appliance-level energy decomposition, and sequence-to-sequence recurrent models such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks. To improve the interpretability of the proposed model, SHAP feature attribution and permutation importance analysis are used to explain the effect of historical consumption patterns and appliance-level features on the forecasting results. The proposed model forecasts multi-horizon energy consumption while providing appliance-level information. The experimental results show high forecasting accuracy across various time horizons. The proposed system improves the interpretability and trustworthiness of deep learning models in energy management systems