Shale gas carbon dioxide displacement well selection method and system based on interpretable graph neural network

Abstract

The invention focuses on the technical field of oil and gas development, and provides a shale gas carbon dioxide displacement well selection method and system based on an interpretable graph neural network. According to the method, dynamic topology modeling is innovatively adopted, well group production time sequence data is mapped into a topological graph containing hypervariable attributes by means of the characteristic that node association is strengthened through a graph neural network, and a production dynamic rule is precisely fitted by fusing the Fourier graph neural network and Bayesian optimization. And meanwhile, a GNNExplainer interpretable module is embedded, and a pressure channeling thermodynamic diagram is generated by quantifying inter-well weight coefficients, so that the pressure channeling degree visual analysis is realized. And finally, a multi-dimensional decision model is constructed based on the pressure channeling strength and the production indexes to complete well selection. According to the scheme, through well pattern diagram structure dynamic modeling, the problem of quantization of nonlinear factors such as crack closing is solved, the dominant displacement channel can be accurately recognized without high-cost monitoring, an intelligent well selection scheme is provided for large-scale application of the CCUS technology, and low-carbon efficient development of shale gas reservoirs is assisted.

Images