Carbon Emission Prediction for Power Systems Based on Multimodal Big Data Fusion
Zitian Liu, Yanhong Deng, Yingbin Bao, Guangyu Chen, Tao Liu, Hongrui Wang · 2025
Driven by the goal of “double carbon”, the power industry, as a key area of carbon emissions, is becoming a core support for green and low-carbon transformation. However, the current carbon emission modeling mostly relies on single-modal data, which is difficult to comprehensively portray the complex carbon emission influencing factors during the operation of the power system. To this end, this paper proposes a multimodal big data fusion-based power system carbon emission prediction method, which integrates heterogeneous information from multiple sources, such as structured time series data, meteorological information, image monitoring, policy text, and power grid topology, to construct an end-to-end prediction model. In this study, a multi-channel deep learning architecture is designed to extract each modal feature using CNN, LSTM, BERT and graph neural network (GNN), respectively, and introduce Modal Attention Mechanism (MAM) for dynamic weight fusion. The experiments are conducted based on the actual data of typical regional power grids, and compared with ARIMA, LSTM, and unimodal neural networks under multi-timescale and multi-region conditions, and the results show that the proposed model significantly outperforms the traditional methods in terms of MAE, RMSE, MAPE, and $\mathbf{R}^{\mathbf{2}}$.