The division of oil and gas accumulation assemblage in Sichuan Basin and the construction of favorable accumulation assemblage prediction model

Abstract

With the rapid development of modern technology and economy, mineral resources such as oil and natural gas have attracted more and more attention. Oil and gas accumulation is a key link in China’s national economy and major national safety production. The prediction of favorable hydrocarbon accumulation assemblages has always been a matter of great concern to people. However, the current oil and gas exploration methods have undergone tremendous changes and the prospecting environment has become increasingly complex, and the exploration methods have gradually developed in a vectorized direction. The exploration focus has also shifted to complex deep deposits, overburdens and concealed deposits. With the increasing abundance of mineral resources, the challenges faced are also increasing, so it is particularly important to develop new exploration technologies and new technical means. With the continuous development of big data, artificial intelligence, machine learning, swarm intelligence and other technologies have good application prospects in various aspects. In this paper, machine learning technology was used for oil and gas accumulation combination classification and favorable accumulation combination prediction. The prediction model was established by using machine learning method, and the prediction data of the model was compared with the conventional prediction method, so that the prediction effect of the model was verified. The test results showed that the average relative errors of oil wells No. 1, No. 2, No. 3, and No. 4 in the proposed model were 23.33%, 24.93%, 26.38%, and 22.60%, respectively; the average relative errors of No. 1, No. 2, No. 3 and No. 4 in the conventional prediction model were 26.19%, 32.37%, 36.16, and 27.87%, respectively. Therefore, it can be concluded that the prediction effect of the prediction model using the machine learning method was significantly better than that of the traditional prediction model.