Joint Forecasting of Residential Energy Consumption and Solar Generation Using Advanced AI Architectures

This paper presents a machine learning–based Smart Home Energy Management System designed to predict both household electricity consumption and rooftop solar power generation. It incorporates a structured pipeline for data preprocessing and feature extraction, enriched with contextual variables such as weather conditions and calendar information. sequence-tosequence learning using convolutional neural networks (CNNs) and long short-term memory (LSTM) architectures is implemented. For comparative evaluation, a robust tree-based model serves as a baseline. Model performance is evaluated using standard metrics, including MAE, RMSE, and 𝑅 2 along with peak oriented metrics to assess ramp rate and peak fidelity. Additionally, a lightweight web front-end based is developed to provide real-time inference and interactive visualization for decision support. The results show that LSTM achieved the highest accuracy in global metrics and was therefore adopted as the system’s forecasting backbone.