A TCN-Driven Framework for Energy Consumption Prediction: The Role of Comprehensive Feature Engineering

The need for accurate energy consumption forecasts is paramount for efficient energy management, cost savings, and grid reliability. Addressing this, we present a forecasting methodology centered on Temporal Convolutional Networks (TCNs). TCNs excel at modeling temporal relationships in timeseries, effectively learning both short-term dynamics and longterm trends, making them well-suited for energy data. Our approach incorporates a multi-source feature engineering strategy, combining weather forecasts, historical consumption patterns, and economic factors. Extensive evaluations confirmed the TCN model's superior performance against established LightGBM and LSTM models, evidenced by an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.926 and MAE of 62.24 kWh. Crucially, the model showed excellent generalization to unseen data. Hyperparameter tuning via Optuna systematically optimized the TCN architecture, boosting its predictive capabilities over various time scales. These findings affirm the power of TCNs for demanding energy forecasting tasks involving largescale, high-frequency data, showcasing their notable scalability, computational speed, and predictive precision as a valuable asset for energy planning.