CO2 injection in unconventional oil and gas reservoirs is a key technology for enhancing oil and gas recovery while enabling CO2 storage. Its application is becoming increasingly prevalent in the development of such reservoirs within the context of “dual carbon” goals. Accurate monitoring and evaluation of CO2 uptake across different layers are essential for guiding the optimization and adjustment of oil and gas reservoir development schemes. Pulsed neutron oxygen activation logging is employed as a dynamic monitoring technology for oxygen-containing fluid injection in a complex string structure, reflecting the dynamic behavior of injected CO2 by recording variations in activated oxygen spectrum peaks. However, activation spectrum peaks in CO2 injection often exhibit single-peak tailing and double-peak overlapping, influenced by statistical fluctuations in activated gamma-ray count rates, fluid properties, multi-layer string structure, and other factors. As a result, accurately determining the transit time of activation spectra and evaluating the amount of CO2 uptake in each small layer becomes challenging.
To minimize errors in transit time calculation caused by overlapping peak separation and activation peak boundary selection, the morphological characteristics of activation peaks under varying influences were analyzed meticulously. The primary factors affecting oxygen activation logging data were string structure and flow rate differences. From a morphological perspective, peak types were classified into four categories: symmetric single peaks, asymmetric single peaks, partially overlapping double peaks, and severely overlapping double peaks. Asymmetric single peaks, characterized by tailing phenomena, occurred under conditions of significant fluid flow velocity differences and dispersed arrival times at the probe. Conversely, overlapping double peaks appeared when multiple flows from the tubing and the annulus produced superimposed signals, with similar flow rates and identical directions. Usually, the water flow was faster than that in the tubing-casing annulus, resulting in narrower and taller peaks for tubing flow.
Due to the randomness and uncertainty of neutron emission from neutron source, oxygen activation reactions, and the detector technology, the counting rate in the time spectrum under ideal conditions conformed to the normal distribution (also termed Gaussian distribution). Compared with the measured oxygen activation spectrum peak, the Gaussian function exhibited a high degree of morphological similarity. The Gaussian function was used to fit the oxygen activation spectrum peak, and the peak position, peak width, and peak height information were derived from its parameters, subsequently enabling the determination of the transit time. Furthermore, overlapping peaks generated by the tubing flow signal and the tubing-casing annulus flow signal could also be effectively separated using multiple Gaussian functions, enabling precise analysis of multiple downhole flow characteristics.
The spectral signal, characterized by multiple Gaussian peak functions, represented a typical nonlinear model. While the peak width and peak position of each characteristic peak exhibited nonlinear behavior, the peak height remained a linear parameter within this framework. Therefore, the Nelder-Mead (NM) algorithm was used to optimize the nonlinear parameters, with linear parameters being directly calculated by linear regression in each iteration. This approach reduced the dimension of the solution vector and enhanced operational efficiency. Despite the NM algorithm’s advantages of requiring no prior guidance and exhibiting rapid convergence, as a direct optimization algorithm, its results were greatly affected by the initial solution. To address this, the Ant Colony Optimization (ACO) algorithm was introduced. In ACO optimization, ants migrated towards spectral bands containing local maxima based on predefined movement rules, with iteration terminated once all ants halted. All ants were distributed within spectral bands containing local maxima. Through the preliminary optimization of the spectral lines, a reasonable initial solution was provided for the NM algorithm, thereby improving the stability of the transit time calculation results and enabling high-precision quantitative computation of the transit time in the oxygen activation injection profile logging. Compared with the traditional methods involving manual peak boundary determination combined with weighted average or Gaussian function fitting methods, this approach offered higher fitting efficiency, reduced human intervention, and lower calculation error.
Through a comparative analysis of pulse neutron oxygen activation data processing and interpretation in well X (CO2 injection well) of the M oilfield, the established ACO-NM optimization model could effectively realize the bimodal separation of overlapping peaks in tubing and casing spaces. Transit times were obtained via automatic peak fitting, enabling the quantitative calculation of CO2 flow in different spaces of the complex string structure. To validate the algorithm’s accuracy, comparative analysis was conducted between the ACO-NM hybrid optimization and the traditional least squares method. Taking surface metered injection volumes as the evaluation standard, relative errors were quantified. The least squares method exhibited errors of 9.59% (tubing) and 9.29% (annulus), while the ACO-NM hybrid optimization algorithm yielded relative errors of 1.87% and 3.31% in the tubing and annulus, respectively. Compared with the traditional least squares method, the calculation results of the optimization algorithm were closer to the surface metered injection volumes. A relative error below 5% was observed between the injected fluid flow calculated by the ACO-NM hybrid optimization algorithm and the actual injection volume at the wellhead. This indicated an improvement in calculation accuracy over the traditional least squares method, which met the needs of dynamic monitoring and evaluation of CO2 injection in the field. The proposed ACO-NM hybrid optimization calculation method in the dynamic monitoring of CO2 injection provides crucial technical support for oilfield development and carbon dioxide storage. The application of this method enables enhanced operational efficiency and economic viability of CO2 injection, improved oil and gas recovery, and more precise and efficient resource development.