A Self‐Attention Autoformer With Period–Trend Decoupling for Power Grid Call Volume Forecasting

ABSTRACT

Accurate forecasting of power grid customer service call volume is essential for optimising resource allocation in customer service centres and providing data support for grid operation monitoring and emergency response. However, call volume time series typically exhibit multiscale periodic dependencies, whereas abrupt surges caused by unexpected disturbances may significantly degrade forecasting performance. To address these challenges, this paper proposes a period–trend decoupled self‐attention Autoformer model for power grid call volume forecasting. Specifically, considering the pronounced intraday bimodal periodic pattern of call volume data, together with long‐term variations at weekly, monthly and even yearly scales, the seasonal–trend decomposition mechanism of Autoformer is introduced to decompose the original sequence into short‐term seasonal components and long‐term trend components. These two components are then modelled separately using a multihead self‐attention mechanism and a multilayer perceptron, enabling targeted learning of temporal features at different time scales. Furthermore, to mitigate the adverse influence of abnormal disturbances on model fitting, a square‐root transformation is applied to nonlinearly compress the scale of call volume data and suppress outlier‐induced deviations. Meanwhile, the L1 loss is employed instead of the mean squared error loss, which is more sensitive to outliers, thereby reducing the negative impact of abnormal samples during model training. Experimental results demonstrate that the proposed method achieves superior performance compared with several traditional forecasting methods and deep learning models on the power grid customer service call volume forecasting task. The proposed model can effectively characterise both seasonal and trend patterns in time series data while improving robustness against abrupt disturbances, providing reliable support for the optimisation of power grid customer service systems.