Multi-Task Power Load Forecasting in Urban Cities Considering User Category and Data Characteristics

Accurate electricity consumption forecasting is essential for efficient energy management and planning, particularly in urban environments. Urban power load forecasting involves multiple prediction tasks that vary in time scale, target entities, and data characteristics. This paper proposes a multi-task power load forecasting framework that systematically addresses these diverse forecasting needs by employing task-specific models. Specifically, mid-term electricity consumption forecasting (on a monthly scale) and short-term load forecasting (on an hourly scale) are considered separately, as their data characteristics and prediction objectives differ. Furthermore, electricity consumption is categorized into three distinct user groups—agency purchasing, market-based users, and residential/agricultural users—in addition to overall city-wide consumption. To achieve accurate forecasting, SARIMA, Random Forest (RF), and XGBoost models are applied to user-specific electricity demand prediction, while deep learning models, including Long Short-Term Memory (LSTM) and Temporal Convolutional Network-Gated Recurrent Unit (TCN-GRU), are used for overall city-wide load forecasting. Model selection is conducted systematically to determine the most suitable approach for each task. Experimental results demonstrate that this task-oriented modeling strategy significantly improves forecasting accuracy and reliability, emphasizing the importance of selecting appropriate models based on prediction tasks and data characteristics.