The conventional cable logging method has low efficiency, because it requires drilling to be completed before logging, and the measurement results are easily affected by external environmental factors such as borehole environment, drilling fluid and mud intrusion, which cannot meet the actual drilling requirements. However, logging while drilling is to install the transmitter and receiver coils on the drill collar, meanwhile, by calculating the phase difference and amplitude attenuation of receiver coils, the spatial and temporal distribution characteristics of electromagnetic field are established, and the conductivity of different formations is detected, so as to obtain the formation material information(Zhu, Gao and Wang 2023, Li et al. 2022, Wu et al. 2020, Huang, Coope and Shen 1984, Bittar et al. 2009). Logging-while-drilling technology overcomes the shortcomings of traditional methods, can detect formation information while drilling(Li et al. 2005), improve operation efficiency and cut down the cost of oilfield development. It is the main means to evaluate formation oil and gas content, as well as the main technology of resistivity logging technology.
At present, electromagnetic wave logging while drilling generally adopts multi-transmitter-receiver coil system structure, and the measurement results with borehole compensation can be obtained by using symmetrical coil system arrangement with equal transmitter-receiver spacing, thus reducing the influence of external environmental factors on the measured formation resistivity parameters(Zhang et al. 2016). At the same time, because the transmitter-receiver spacing is one of the important parameters that affect the measured depth(Zhao, Dun and Yuan 2011, Li, Shen and Zhu 2016), in order to meet the borehole compensation function, the transmitting coils need to be symmetrically arranged, which makes the size of the coil system increase significantly, leading to the increase of the drill collar size and the risk of drilling tool sticking, At the same time, the manufacturing process is more complicated, which leads to the increase of the cost(Macune et al. 2006, Ma 2010).
Nowadays, the space for placing the coil system of drill collar is limited, the power supply of LWD instrument is limited(Xiao, Ju and Yang 2009). Reducing the number of transmitting coils is beneficial to shorten the length of drill collar, reduce the risk of drilling tool sticking and increase the downhole working time of LWD instrument. However, the reduction of the number of coils will lead to the measurement results not having borehole compensation function. At present, due to the artificial intelligence shows excellent learning and inversion ability for existing data, it provides a feasible solution to the above problems.
Artificial intelligence technology has the ability to establish complex mapping between nonlinear input and output data (Kumar and Afzal 2023, Liu et al. 2014). At present, reasonable inversion of logging data by artificial intelligence technology is of great significance in the field of petroleum exploration and development. The inversion methods mainly include the following: First, Nonlinear iterative method, which requires the construction and minimization of a quadratic cost function to reduce the difference between simulated and actual data. For example, Wang developed a fast forward solver based on Gauss Newton method in the inversion of cross-bedded formation (Wang et al. 2017). Pardo implemented a one-dimensional inversion in horizontal and high-angle wells based on Gauss Newton method (Pardo and Torres-Verdín 2015). Wang implemented fast inversion in anisotropic formations based on the regularized Leven Berg-Marquardt minimization method (Wang and Fan 2019). Second, the neural network inversion method, which applies the neural network architecture to the inversion of logging data, for example, Zhang used modular neural network for fast forward modeling to realize logging inversion.(Zhang 2000); Singh presented the effect of the learning parameters on the artificial neural networks inversion of isotropic formations(Singh, Tiwari and Singh 2013);Raj used a single layer feed-forward neural network to invert non-linear apparent resistivity in formations(Raj et al. 2014). Third, the deep learning technology, which emphasizes the complexity of the model architecture, usually with five to six layers or even more hidden layers, is able to deeply portray the rich intrinsic information of the data, thus making it easier to classify or invert. For example, Shahriari et al. investigated the use of deep learning to invert borehole resistivity measurements(Shahriari et al. 2020). The research shows that DNN has remarkable representation ability for complex structures through the design and establishment of appropriate number of neuron computing nodes and multi-layer operation structure(Zhu et al. 2018, Zhu et al. 2020). With this background, deep learning method has made remarkable achievements in the field of artificial intelligence (Janiesch, Zschech and Heinrich 2021, Schmidhuber 2015).
In this paper, the reasonable inversion of logging data is carried out by artificial intelligence. The combination of logging data and neural network is innovatively put forward to accurately inverse the logging data of the removed transmitter coil and realize the measurement results with borehole compensation. By reducing the number of transmitting coils, the length of the drill collar is shortened, resulting in a reduced risk of drill sticking, decreased power consumption, increased downhole working time for the drill collar, and a significant improvement in the efficiency of the LWD tool.