Abstract: Regional short-term power load forecasting is essential for developing regional energy-saving optimization strategies and establishing smart buildings. The popularity of smart meters can collect a large number of effective load data, and meteorological factors also have an important impact on regional power load. This paper presents a CNN-GRU short term power load forecasting approach incorporating an attention mechanism. The goal is to identify the characteristic relationships between regional power load, time, and meteorological data by building a deep neural network model, thereby enabling short term predictions of regional power load. The integration of the attention mechanism strengthens network model’s ability to filter data features, helping to improve prediction accuracy. Ultimately, It is determined whether the suggested approach is effective through a dataset of regional power load and meteorological data.