A deep geopressure disaster risk map construction method based on a generative adversarial network

Abstract

The present application relates to the technical field of ground pressure disaster analysis, in particular to a deep ground pressure disaster risk map construction method based on a generative adversarial network; comprising: obtaining geological information, real-time monitoring and historical disaster data, constructing a geological disaster knowledge graph and performing spatio-temporal fusion to generate a multi-channel spatio-temporal data field; inputting the data field into a pre-trained spatio-temporal generative adversarial network to generate a risk precursor feature field representing the evolution of ground pressure disasters; simulating engineering intervention through a digital perturbation model to generate an counterfactual risk feature field; evaluating the intervention effect based on a risk reduction index and optimizing the selection of an interpretable intervention scheme in combination with the knowledge graph; the generator adopts a 3D CNN encoder-decoder structure and embeds a convolutional long short-term memory network module; the discriminator adopts a hybrid structure composed of a three-dimensional convolutional network and a recurrent network; spatio-temporal fusion generates an embedding vector through a graph neural network and splices data. The present application significantly improves the prevention and control accuracy of deep ground pressure disasters.

Images