Fusing CNN and LSTM Networks for Residential Electricity Load Forecasting

This study proposes a CNN - LSTM hybrid model for high-precision residential electricity demand forecasting. Utilizing four years of minute-level consumption data from Paris households. The architecture combines convolutional neural networks (CNN) to capture spatial characteristics and long short-term memory (LSTM) networks to model temporal correlations. Key variables, including active/reactive power, were selected through correlation analysis and feature engineering. The presented model demonstrates an RMSE value of 0.612 kW for predictions made 60 minutes in advance, outperforming traditional methods with a 53.6% reduction in mean squared error (MSE). The model provides real-time support for peak load management and smart grid optimization, advancing the application of deep learning in energy systems.