Deep learning improved algorithm for cold source disaster prediction

The invention discloses a deep learning improved algorithm for cold source disaster prediction, and the algorithm comprises the steps: carrying out the statistical classification of related factors affecting cold source disaster, carrying out the normalization of multi-source heterogeneous data through a data normalization function, and obtaining grouped sample data; establishing a deep belief network, adopting a deep belief network improved algorithm based on the momentum learning rate, carrying out the repeated iterative training of model parameters in combination with the multiple sets of sample data, and acquiring a determined mapping relation between input and output of the deep belief network; and pre-estimating the disaster level of the cold source disasters. According to the method, the improved algorithm of deep learning is adopted, so that the network learning efficiency is effectively improved, the error convergence rate of prediction is reduced, and the problem of prediction of ocean disasters wit