Hybrid Deep Learning Models for Energy Consumption Forecasting: A CNN-LSTM Approach for Large-Scale Datasets

Long-term electricity consumption forecasting is essential with the rise of smart grids and advanced metering infrastructures for optimized energy management. Categorizing energy consumption into historical time series enables accurate predictions and deeper insights into energy trends. Deep learning has driven innovative forecasting models, particularly in smart grids leveraging data-driven approaches. This study applies convolutional neural networks (CNNs), gated recurrent units (GRUs), and long short-term memory (LSTM) networks to analyze complex energy consumption patterns. Hybrid CNN-GRU and CNN-LSTM models are proposed to enhance forecasting accuracy by capturing spatial and temporal correlations. A comparative analysis is conducted against standalone GRU and LSTM models using historical data from American Electric Power (AEP) and Dominion Virginia Power (DOM). Performance is assessed using symmetric Mean Absolute Percentage Error (sMAPE), Loss, and Root Mean Square Error (RMSE). The findings demonstrate that hybridCNN-LSTMmodels improve forecasting accuracy, enabling proactive strategies like load shedding for efficient energy management as compared to other models. This research contributes to both academia and industry by enhancing smart grid reliability and operational efficiency.