With the continuous growth of power load in the mining industry, the problem of high energy consumption has become a key bottleneck restricting green and low-carbon development. Aiming to address the non-linear and non-stationary characteristics of the mining power load sequence and the lack of suitable prediction methods in existing research, this paper proposes a VMD-BiGRU-Attention regional power load prediction model for mines. Firstly, the original load data are modally decomposed by VMD (Variational Mode Decomposition) to extract multi-scale features to reduce noise interference; Then, the dynamic dependency relationship of load time series data is mined in both directions by BiGRU (Bidirectional Gated Recurrent Unit) network, and the key features are focused adaptively by the attention mechanism to improve the model’s ability to capture complex load changes. Finally, the analysis of the actual power consumption data prediction results shows that root mean square error(RMSE), mean squared error (MAE) and MAPE (mean absolute percentage error) of this paper’s method are at least reduced by 0.064, 0.011 and 0.004, and R2 is improved by at least 0.05. It provides a reliable data basis for the dynamic management of the mining power consumption load.