Artificial intelligence has shown great performance and considerable potential in petroleum and natural gas prospect evaluation and prediction (PPEP). Multiple machine learning (ML) and deep learning (DL) methods are applied in petroleum geology, development geology and geophysical exploration. The PPEP work in petroleum geology field which aims to provide a probability distribution for petroleum prospects with small data. The PPEP work in development field utilize seismic data as the fundamental data to predict petroleum reservoir features and favourable geological facies with the constrain of well log data. While PPEP work in geophysical exploration focus on seismic or well log facies recognition and prediction using multiple intelligent seismic denoising and well log augmentation methods. The popular AI algorithms are almost all used in all kinds of PPEP work without a unique choice and standard procedure. The decisive factors for AI application in PPEP are data quality and model interpretability. The current data size and precision used in PPEP work are hard to fulfill the requirement for AI model training, which need further augmentation and multimodal fusion. The interpretability is significant for evaluating the reliability and rationality of these data-driven models. At present, the dual-driven model driven by both of data and physical constrain may be a possible way to further decrease the requirement of data quality and increase the model interpretability. In the future, an artificial generative model which can determine petroleum prospects in multiple dimensions with a large amount of data may be the final goal for the utilization of AI in the petroleum industry.
Wang et al. (Wed,) studied this question.