Marine disasters seriously affect the economic and social development of China's coastal cities. Therefore, predicting the losses caused by marine disasters can assist earthquake prevention and disaster reduction departments in optimally preventing and reducing economic losses. However, due to the substandard state of the current marine disaster loss system, it is difficult to predict the potential losses that a marine disaster would cause accurately. The purpose of this paper is to propose an effective method for predicting the direct economic loss by marine disasters. A power buffer operator is used to process the original data before modeling, reducing the impact of system shock disturbance and the randomness of the data. This allows the disaster loss data to reflect the characteristics of the system better. In addition, the background value of the GM(1,1) model is reconstructed based on the Boolean formula and Newton combination interpolation, and the time-weighted least squares method is used to improve the GM(1,1) model to predict the direct economic loss by marine disasters. It is demonstrated that the improved GM(1,1) model can predict the direct economic loss caused by marine disasters. Compared with the traditional GM(1,1) model, the new grey prediction model greatly improves the prediction accuracy of the model. The original prediction error of 40.74% dropped to 6.64%, allowing for the accurate prediction of direct economic loss by marine disasters.