Multi-wave seismic oil and gas reservoir prediction method under unsupervised and supervised learning

Abstract

The invention discloses a multi-wave seismic oil and gas reservoir prediction method under unsupervised and supervised learning. The multi-wave seismic oil and gas reservoir prediction method comprises the steps of: firstly, generating various longitudinal and transverse wave seismic attributes through convolution and dimensionality raising by utilizing different convolution kernels; secondly, utilizing a clustering analysis method to conduct unsupervised learning, performing dimensionality reduction through carrying out clustering analysis on the longitudinal and transverse wave seismic attributes, and calculating multi-wave seismic aggregate attributes which can highlight oil and gas reservoir features by adopting an aggregation method based on the dimensionality reduction result; and finally, regarding the aggregate attributes after dimensionality reduction as a learning set of a support vector machine, and conducting prediction of seismic oil and gas reservoirs from known to unknown. By applying the multi-wave seismic oil and gas reservoir prediction method to actual oil and gas reservoir prediction, the result shows that the predicted seismic oil and gas reservoir boundaries are clearer, and the prediction result is basically consistent with the actual situation.

Description

technical field [0001] The invention relates to a multi-wave seismic oil and gas reservoir prediction method under non-supervised and supervised learning. Background technique [0002] Using seismic attributes to describe the characteristics of oil and gas reservoirs is one of the main means to solve petroleum geophysical exploration. Many methods have been used by people, including cluster analysis method, multi-attribute fusion technology and neural network. These data mining techniques are more supported by large-scale training samples. When the training samples are insufficient, the generalization ability of the model is severely limited, and it is easy to fall into the state of over-learning or under-learning. Support vector machines (SVR) are pattern classifiers based on structured risk minimization. In the state of small samples, it has higher generalization ability, and the robustness of the model is ideal. However, the difficulty of support vector machine lies in...

AI Technical Summary

Problems solved by technology

However, the difficulty of support vector machines lies in parameter optimization

Images