Abstract: Aiming at the problem that the early warning method based on risk matrix and considering settlement rules is affected by massive cooperative operation data, which leads to poor early warning effect, the risk early warning of man-machine cooperative operation of the hydropower station under deep data fusion is proposed. The closed-loop mode of digitalized cooperative operation is constructed to ensure the stability of risk warning process of digitalized man-machine cooperative operation of hydropower station operation. The maximum confidence gain is calculated, and the features are normalized and combined, so as to build a full-dimensional data deep fusion model for hydropower stations to avoid data loss. Deep fusion data are obtained through sensor nodes, and multidimensional time series are established based on space-time in space and time to provide effective data support for the safe operation of field workers. Based on the results of intelligent perception and recognition of potential risks, a risk early-warning process for man-machine collaborative operation of hydropower stations based on deep data fusion is designed, so that workers can deal with on-site emergencies in time. The risk warning probability of this method is closer to the actual value, which indicates that the warning effect of this method is better.