Multi-layer nonlinear local receptive field extreme learning machine method for logging gas analysis

With China′s increasing energy demand and the complex drilling environment, it is of great significance to carry out high-precision detection of alkane gas concentration to improve oil and gas exploration efficiency. Spectral logging technology has become a research hotspot in the process of oil exploration with the advantages of quick and accurate recording results. In this article, a multi-layer nonlinear local receptive field extreme learning machine (NM-LRF-ELM) model is proposed for resolving nonlinear problems caused by saturation absorption, noise interference, and baseline drift. The model converts one-dimensional spectral data into two-dimensional matrix format and realizes nonlinear feature extraction between input and hidden layer by using local receptive field data processing. Meanwhile, an improved T-sigmoid activation function is introduced and the dropout layer is added after the fully connected layer to reduce the overfitting risk of the model. The feature extraction and quantitative analysis of the model show an integrated structure and directly outputs the predicted value of quantitative analysis. In this article, the infrared spectra of 407 mixed alkane gas samples from two groups are collected as an experimental data set for quantitative analysis. The experimental results show that the training time of this model is reduced by more than 90% compared with the sliding window model and the gray Wolf model, and the prediction accuracy of the model is still lower than the system error under the nonlinear interference of the homolog. Therefore, the proposed method is helpful in reducing the nonlinear interference of unknown gas and improve the infrared spectrum detection accuracy of target gas under the condition of complex field environment changes. © 2024 Science Press. All rights reserved.