Regional Daily Electricity Consumption Forecasting with Meta-Learning and Attention Mechanism

With the rapid development of China's economy and society, the installation capacity and frequency of power users' capacity expansion have been continuously increasing, leading to more significant non-stationary fluctuations in electricity consumption. Traditional static forecasting models struggle to accurately predict daily electricity consumption, especially when accounting for the impact of capacity expansion. To address this issue, this paper proposes a regional daily electricity consumption forecasting method under the influence of capacity expansion, based on meta-learning and efficient channel attention (ECA). First, a meta-learning-based prediction model training framework and a meta-task dataset are established. The model-agnostic meta-learning (MAML) method is employed to optimize the initialization parameters, enabling the model to acquire prior knowledge. This knowledge is then used to fine-tune the model with small samples after capacity expansion. Subsequently, a combined convolutional neural network (CNN) and ECA-based electricity consumption prediction model is developed, which allows for the deep extraction of external influencing factors at multiple time scales and accurate electricity consumption prediction. Finally, simulation results based on case studies show that the proposed model can dynamically adapt to the capacity changes brought about by capacity expansion. The mean absolute percentage error (MAPE) is reduced by an average of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{2. 3 1 \%}$</tex>.