The power industry is a basic industry in the national economy and a key industry for China to achieve the "dual carbon goals". Accurate forecasting of power demand is the primary basic work for the development of the national power master plan, coal power withdrawal, and renewable energy investment decisions. Therefore, using the modeling idea driven by multi-modal information fusion to construct a new integrated forecasting model of power demand based on CNN-LSTM (Convolution Neural Network, Long Short-term Memory) in a multi-source heterogeneous data environment. Firstly, CNN is used to extract implicit features from power demand numerical time series data and text data (including policy texts, news reports, and forum comments); Secondly, series feature and text feature are organically fused by series fusion method; Finally, the fused features are input into the LSTM model for prediction. The experimental results show that, on the one hand, the proposed multi-modal information fusion prediction model is superior to the widely-used single prediction model (e.g. ARIMA, CNN, and LSSVM) and combined prediction model (e.g. EEMD-ARIMA and EEMD-LSSVM) in terms of level accuracy and directional accuracy; on the other hand, it proves that the organic fusion of time series data and text data can effectively improve forecasting performance. The forecast results show that due to the influence of multiple factors such as China’s economic restructuring and energy system transformation, China’s power demand growth will gradually slow down or even show a downward trend in the next two years. This finding provides an important decision-making reference for the low-carbon transformation of China’s power system.

power demand,forecasting,multimodal information fusion,feature fusion,CNN-LSTM