Artificial Intelligence Applications in Petroleum Exploration and Production

Recent advances in computer and data sciences have made artificial intelligence techniques a useful tool in tackling the problems in petroleum exploration and production. Intelligent exploration and production have become a hot topic both in academia and oil and gas companies. This Special Issue aims to solicit recent progress and best practices in the application of artificial intelligence techniques in petroleum exploration and production. This Special Issue comprises 11 research articles.

Li et al. [1] developed a method of intelligent stuck pipe type recognition using digital twins and a knowledge graph model. The results show that the stuck pipe type can be identified according to the degree of stuck pipe, the changing trend of the characteristic parameters of stuck pipe, and the knowledge graph of stuck pipe types, and the method can accurately identify the stuck pipe type and provide a basis for taking targeted deconstruction measures.

Zang et al. [2] established an intelligent rate of penetration (ROP) prediction model through the random forest algorithm and conducted drilling parameters optimization for horizontal wells based on a multi-objective genetic algorithm to improve ROP and reduce drill string drag. The results show that the ROP of the horizontal section of the new well increases by 10.3%, and the drag reduces by 4.5% on average compared with the adjacent well.

Fang et al. [3] developed a novel cementing quality evaluation method based on the convolutional neural network. The authors proposed a multi-scale perceptual convolutional neural network with kernels of different sizes that can extract and fuse information of different scales in variable density logging. The results show that the new model is more stable and more suitable than other convolutional neural networks for the identification of cementing quality.

Li et al. [4] developed a machine learning-assisted prediction method of Oil production and CO₂ storage effect in CO₂-water-alternating-gas injection (CO₂-WAG). It demonstrated that the proposed machine learning method can rapidly predict CO₂-WAG performance with high accuracy by the average absolute prediction deviations of 1.10%, 3.04%, and 2.24% for cumulative oil production, CO₂ storage amount, and CO₂ storage efficiency with high computational efficiency under conditions of various injection parameters.

Yin et al. [5] proposed an identification method of the gas kick type in a fractured formation by combining the dynamic time warping (DTW) and the wellbore formation, coupled with a two-phase flow model. The field example results show that the method can identify the type of gas kick, based on the real-time surface measurement parameters, and provide a basis for taking targeted well control measures.

Ji et al. [6] proposed a deep-learning-based model by using long short-term memory (LSTM) and an artificial neural network (ANN) to generate synthetic compressional wave velocity and conducted a case study of the Ulleung Basin gas hydrate in the Republic of Korea. The results implied that the proposed method can accurately estimate missing well-logging data, contributing to the reservoir characterization of gas-hydrate-bearing sediments.

Liu et al. [7] combined the powerful timing information-based mining ability of the LSTM with the nonlinear fitting ability of FNN (fully connected neural network), and established a rate of penetration prediction method for ultra-deep wells based on LSTM–FNN. The results show that the average relative error and R2 of the LSTM–FNN model on the data of well 1 and well 2 is better than the FNN and LSTM models, with the accuracy of adjacent wells reduced by only 5%.

Zhu et al. [8] proposed a hybrid neural network model for predicting the bottomhole pressure in managed pressure drilling by combining the different advantages of back propagation (BP), LSTM, and a one-dimensional convolutional neural network (1DCNN) model. The results show that the relative error of the best model is about 70% lower than the optimal single intelligent model.

Pratama and Latiff [9] conducted automated geological feature detection in 3D seismic data using semi-supervised learning. It was demonstrated that the proposed convolutional neural network (CNN)-based model is highly accurate and consistent with the previous manual interpretation in both cases with the synthetic data and the real seismic investigation from the A Field in the Malay Basin.

Zhu et al. [10] developed an intelligent prediction of stuck pipe using the combined data-driven and knowledge-driven model. The results show that the proposed model can predict stuck pipe events with an F1 of 0.98 and a FAR (false alarm rate) of 1%.

Syahputra et al. [11] studied pay-zone determination using enhanced workflow and neural networks. Unsupervised learning of self-organizing maps (SOM) is applied to delineate hydrocarbons from given AVO properties for detecting hydrocarbons. The proposed method validates a promising performance in defining probable hydrocarbons using real seismic data and it enables the early detection of hydrocarbon content during the preliminary stage of exploration when no well is accessible.