Method for predicting conversion efficiency of CO2-H2 underground biochemical synthetic natural gas based on machine learning
By constructing a machine learning-based proxy model for the conversion efficiency of CO2-H2 underground biochemical synthesis natural gas, the problems of slow prediction speed and inaccurate results in existing technologies have been solved, achieving faster and more accurate conversion efficiency prediction, which can comprehensively reflect the influence of reservoir environmental factors.
Patent Information
- Application Number
- CN202511109680.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-08
- Publication Date
- 2025-11-21
AI Technical Summary
Existing technologies are slow, inaccurate, or incomplete in predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas, making it difficult to reflect the degree and speed of conversion. Furthermore, traditional methods are costly to calculate.
A machine learning-based approach was adopted to construct a surrogate model for the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas using geological, microbiological, and engineering parameters. Input samples were obtained through Latin hypercube sampling and various mapping methods. The average time-dependent conversion rate of CO2-H2 was calculated by combining the results of biogeochemical simulations. A regression algorithm was used to construct a surrogate model for prediction.
It enables faster and more accurate prediction of conversion efficiency, comprehensively reflects the influence of reservoir environmental factors, reduces computational workload, and improves the comprehensiveness and accuracy of prediction.
Smart Images
Figure CN120998342A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of oil and gas field development, and particularly relates to a CO2-H2 underground biochemical synthesis natural gas conversion efficiency prediction method based on machine learning. BACKGROUND
[0002] With the increasingly serious global climate change problem, reducing CO2 emissions has become an urgent task in the energy and environmental fields. At present, carbon capture and storage technology is widely researched and applied, aiming to inject CO2 into underground geological reservoirs for long-term storage to prevent its release into the atmosphere. However, pure geological storage has problems such as high cost, complex risk management, and long-term safety uncertainty, prompting researchers to explore more value-added carbon utilization paths.
[0003] Under this background, the technology of using methanogenic bacteria in depleted oil and gas reservoirs to biochemically synthesize natural gas (CH4 as the main component of mixed gas) from CO2-H2 has gradually attracted attention. This technology not only realizes the resourceization and energyization of CO2, but also directly generates usable natural gas in the underground deep environment, while having the potential for large-scale underground energy storage. With the advantages of green environmental protection, strong sustainability, great economic potential, and multi-functionality, the CO2-H2 underground biochemical synthesis natural gas technology has gradually become one of the important directions for promoting low-carbon energy transformation.
[0004] However, the biological catalyst methanogenic bacteria are easily affected by the complex environment of the oil and gas reservoir, and microorganisms such as acetogenic bacteria and sulfate-reducing bacteria will also compete for and consume the injected CO2 and H2, further reducing the conversion efficiency of the technology. Therefore, accurate prediction of the conversion efficiency before the implementation of the technology can effectively guide the early site selection and optimization of operation parameters, thereby improving the overall performance of the technology. Traditional prediction methods mostly rely on numerical simulation, but when considering the complex interactions between gas, water, rock, and microorganisms, the computational cost of numerical simulation increases significantly. In addition, traditional methods usually use CO2 conversion rate or CH4 release rate as indicators to represent the conversion efficiency, which is difficult to reflect both the conversion degree and the conversion speed. Therefore, there is an urgent need to develop a new method for more comprehensive, accurate, and rapid prediction of the conversion efficiency of CO2-H2 underground biochemical synthesis natural gas. SUMMARY
[0005] The present application proposes a CO2-H2 underground biochemical synthesis natural gas conversion efficiency prediction method based on machine learning to solve the problems of slow prediction speed, inaccurate or incomplete results of existing methods.
[0006] The technical solution adopted by the present application to solve the above technical problems includes the following contents:
[0007] A machine learning-based method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas includes the following steps:
[0008] S1. Select geological parameters, microbial parameters, and engineering parameters as input feature variables for the machine learning model;
[0009] S2. Based on the range of values of the input feature variables and the Latin hypercube sampling method, obtain the input samples;
[0010] S3. Using the biogeochemical simulation results under different input sample conditions, calculate the average time-dependent conversion rate of CO2-H2, and use it as the output sample of the machine learning model;
[0011] S4. Based on the normalized input samples and the unprocessed output samples, construct a machine learning proxy model for the average time-efficiency conversion rate of CO2-H2;
[0012] S5. Based on the geological and microbiological parameters of the target depleted oil and gas reservoir, and combined with different combinations of engineering parameters, the corresponding average time-dependent CO2-H2 conversion rate is predicted using a surrogate model.
[0013] Furthermore, the geological parameters in the input feature variables in step S1 include temperature, salinity, and carbonate mineral content; the microbiological parameters include maximum biomass capacity, optimal specific growth rate of methanogens, optimal specific growth rate of acetic acid bacteria, and optimal specific growth rate of sulfate-reducing bacteria; and the engineering parameters include the CO2-H2 ratio and the total CO2-H2 pressure.
[0014] Furthermore, the specific method for obtaining input samples in step S2 is as follows: first, a numerical sequence between 0 and 1 is generated using the Latin hypercube sampling method, and then the sequence is mapped to the value range of the input feature variables, wherein the maximum biomass capacity adopts a logarithmic mapping method, and other input feature variables adopt a linear mapping method.
[0015] Furthermore, the biogeochemical simulation in step S3 uses the software PHREEQC, and the microbial reproduction kinetics model used is as follows:
[0016] 1) Microbial specific growth rate calculation model:
[0017]
[0018] Where: μ gr The specific growth rate of microorganisms, μ opt For the optimal specific growth rate of microorganisms, C A and C D K represents the concentrations of electron acceptor and electron donor, respectively. A and K Dλ represents the half-saturation constant of the electron acceptor and electron donor, respectively. T , λ pH , λ S These are dimensionless influencing factors related to temperature, pH, and salinity, respectively.
[0019] 2) Substrate consumption calculation model:
[0020]
[0021] In the formula: r s denoted as substrate consumption rate, Y as yield coefficient, and N as number of microorganisms;
[0022] 3) Biomass evolution model:
[0023] r bio =-r s F x Y-dN
[0024] In the formula: r bio For the rate of biomass change, F X denoted as biomass capacity factor, and d as attenuation coefficient.
[0025] Furthermore, the method for calculating the average time-dependent CO2-H2 conversion rate in step S3 is as follows:
[0026] R = (R CO2 +R H2 ) / 2
[0027] In the formula: R is the average time-dependent conversion rate of CO2-H2, R CO2 R H2 The time-dependent conversion rates of CO2 and H2 are respectively, calculated based on the trapezoidal integral concept and the principle of equal area:
[0028]
[0029] In the formula: C 90 C 180 C 270 C 360 The effective CO2 conversion rates at 90 days, 180 days, 270 days, and 360 days, respectively, H 90 H 180 H 270 H 360 The effective conversion rates of H2 at 90 days, 180 days, 270 days, and 360 days, respectively.
[0030] Furthermore, in step S4, when constructing the machine learning proxy model for the average time-efficiency conversion rate of CO2-H2, the machine learning algorithm used is a regression algorithm, including but not limited to random forest regression algorithm, least squares boosting regression algorithm, and feedforward neural network algorithm.
[0031] Furthermore, in step S4, when constructing the machine learning proxy model for the average time-efficiency conversion rate of CO2-H2, the sample dataset is divided into a training set, a validation set, and a test set, with proportions of 70%, 15%, and 15%, respectively. The proxy model is trained using the training set, the model hyperparameters are optimized using the validation set, and the predictive performance indicators of the model are calculated based on the test set. These performance indicators include, but are not limited to, the coefficient of determination, root mean square error, and mean absolute error.
[0032] Furthermore, before predicting the target oil and gas reservoir using geological parameters, microbiological parameters, and engineering parameters in step S5, normalization processing is required based on the input sample dataset.
[0033] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention is the first to use machine learning algorithms to construct a proxy model for the conversion efficiency of CO2-H2 underground biochemical synthesis natural gas. Compared with the traditional single numerical simulation method, the prediction of conversion efficiency is faster and the results are more accurate; (2) The input feature variables cover a total of 9 items including geological parameters, microbial parameters and engineering parameters, which comprehensively reflects the influence of factors such as microbial competition and engineering operations in the reservoir environment on the conversion efficiency, and is closer to the actual situation; (3) By comprehensively adopting the Latin hypercube sampling method and multiple mapping methods, it can more comprehensively and uniformly cover the conversion efficiency. (3) Cover the input variable space, reduce the number of samples while improving the representativeness of the sample; (4) Use the average time-effect conversion rate of CO2-H2 as the conversion efficiency evaluation index. This index reflects both the degree of conversion and the time effect. Compared with the traditional CO2 conversion rate or CH4 release rate, it has stronger comprehensiveness and accuracy; (5) Based on the idea of trapezoidal integral and the principle of equal area, the average time-effect conversion rate is calculated using the effective conversion rates of CO2 and H2 at four time points (90 days, 180 days, 270 days and 360 days). There is no need to calculate for each time point, which significantly reduces the computational workload of obtaining output samples. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the overall process of the present invention;
[0035] Figure 2 This is a schematic diagram for calculating the CO2 time-efficiency conversion rate based on the trapezoidal integral concept and the principle of equal area.
[0036] Figure 3 This is a comparison chart of the actual and predicted values of the average time-efficiency CO2-H2 conversion rate under the training set;
[0037] Figure 4 This is a comparison chart of the actual and predicted values of the average time-efficiency CO2-H2 conversion rate under the validation set;
[0038] Figure 5 This is a comparison chart of the actual and predicted values of the average time-dependent conversion rate of CO2-H2 under the test set. Detailed Implementation
[0039] The present invention will be described in detail below with reference to the embodiments. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0040] A machine learning-based method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas includes the following steps ( Figure 1 ):
[0041] S1. Select geological parameters, microbial parameters, and engineering parameters as input feature variables for the machine learning model;
[0042] S2. Based on the range of values of the input feature variables and the Latin hypercube sampling method, obtain the input samples;
[0043] S3. Using the biogeochemical simulation results under different input sample conditions, calculate the average time-dependent conversion rate of CO2-H2, and use it as the output sample of the machine learning model;
[0044] S4. Based on the normalized input samples and the unprocessed output samples, construct a machine learning proxy model for the average time-efficiency conversion rate of CO2-H2;
[0045] S5. Based on the geological and microbiological parameters of the target depleted oil and gas reservoir, and combined with different combinations of engineering parameters, the corresponding average time-dependent CO2-H2 conversion rate is predicted using a surrogate model.
[0046] The geological parameters in the input feature variables in step S1 include temperature, salinity, and carbonate mineral content; the microbiological parameters include maximum biomass capacity, optimal specific growth rate of methanogens, optimal specific growth rate of acetic acid bacteria, and optimal specific growth rate of sulfate-reducing bacteria; and the engineering parameters include the CO2-H2 ratio and total CO2-H2 pressure.
[0047] The specific method for obtaining input samples in step S2 is as follows: First, a numerical sequence between 0 and 1 is generated using the Latin hypercube sampling method. Then, the sequence is mapped to the value range of the input feature variables. The maximum biomass capacity is mapped using a logarithmic mapping method, while other input feature variables are mapped using a linear mapping method.
[0048] The biogeochemical simulation in step S3 uses the software PHREEQC, and the microbial reproduction kinetics model used is as follows:
[0049] 1) Microbial specific growth rate calculation model:
[0050]
[0051] Where: μ gr The specific growth rate of microorganisms, μ opt For the optimal specific growth rate of microorganisms, C A and C D K represents the concentrations of electron acceptor and electron donor, respectively. A and K D λ represents the half-saturation constant of the electron acceptor and electron donor, respectively. T , λ pH , λ S These are dimensionless influencing factors related to temperature, pH, and salinity, respectively.
[0052] 2) Substrate consumption calculation model:
[0053]
[0054] In the formula: r s denoted as substrate consumption rate, Y as yield coefficient, and N as number of microorganisms;
[0055] 3) Biomass evolution model:
[0056] r bio =-r s F X Y-dN
[0057] In the formula: r bio For the rate of biomass change, F X denoted as biomass capacity factor, and d as attenuation coefficient.
[0058] The calculation method for the average time-dependent conversion rate of CO2-H2 in step S3 is as follows:
[0059] R = (R CO2 +R H2 ) / 2
[0060] In the formula: R is the average time-dependent conversion rate of CO2-H2, R CO2 R H2 The time-dependent conversion rates of CO2 and H2 are respectively, calculated based on the trapezoidal integral concept and the principle of equal area:
[0061]
[0062] In the formula: C 90C 180 C 270 C 360 The effective CO2 conversion rates at 90 days, 180 days, 270 days, and 360 days, respectively, H 90 H 180 H 270 H 360 The effective conversion rates of H2 at 90 days, 180 days, 270 days, and 360 days, respectively.
[0063] In step S4, when constructing the machine learning proxy model for the average time-efficiency conversion rate of CO2-H2, the machine learning algorithm used is a regression algorithm, including but not limited to random forest regression algorithm, least squares boosting regression algorithm and feedforward neural network algorithm.
[0064] In step S4, when constructing the machine learning proxy model for the average time-efficiency conversion rate of CO2-H2, the sample dataset is divided into a training set, a validation set, and a test set, with proportions of 70%, 15%, and 15%, respectively. The proxy model is trained using the training set, the model hyperparameters are optimized using the validation set, and the model's predictive performance metrics are calculated based on the test set. These performance metrics include, but are not limited to, the coefficient of determination, root mean square error, and mean absolute error.
[0065] Before predicting the target oil and gas reservoir using geological parameters, microbiological parameters, and engineering parameters in step S5, normalization processing is required based on the input sample dataset.
[0066] In one specific embodiment, temperature, salinity, carbonate mineral content, maximum biomass capacity, optimal specific growth rate of methanogens, optimal specific growth rate of acetic acid bacteria, optimal specific growth rate of sulfate-reducing bacteria, CO2-H2 ratio, and total CO2-H2 pressure were selected as input feature variables for the machine learning model. The value ranges of each variable are detailed in Table 1. Subsequently, a Latin hypercube sampling method was used to generate 3000×9 numerical samples between 0 and 1. The maximum biomass capacity was mapped logarithmically, while other variables were mapped linearly, mapping them to their respective actual value ranges to form the input samples. Next, based on these 3000 sets of input samples, biogeochemical simulations were conducted using PHREEQC software to calculate the effective CO2 and H2 conversion rates at 90, 180, 270, and 360 days. Then, based on the trapezoidal integral concept and the principle of equal area (…),… Figure 2 ) Calculate the time-efficiency conversion rate of CO2 and H2, and use the average time-efficiency conversion rate of CO2-H2 as the output sample of the machine learning model.
[0067] Table 1 Input feature variables and their value ranges
[0068] Parameter Value range Parameter Value range Temperature 30–90℃ Maximum biomass capacity 0.005 - 0.5 g / L Salinity 20 - 170 g / L Methanogen optimal specific growth rate 0.3 - 4.1 d -1 ]] Carbonate mineral content 0 - 5 wt% Acetogen optimal specific growth rate 0.4 - 1.9 d -1 ]] CO2-H2 ratio 1:5–1:3 Sulfate-reducing bacteria optimal specific growth rate 0.2 - 5.5 d -1 ]] CO2-H2 total pressure 50 - 150 atm
[0069] The input samples were normalized, while the output samples remained in their original state, to construct a machine learning surrogate model for the average time-of-use conversion rate of CO2-H2. The machine learning algorithm used was a feedforward neural network. The sample dataset was divided into training, validation, and test sets at 70%, 15%, and 15% ratios, corresponding to 2100, 450, and 450 data sets, respectively. The surrogate model was trained using the training set, and the model hyperparameters were optimized using the validation set. The final neural network structure consisted of two hidden layers: the first layer contained 20 neurons, and the second layer contained 10 neurons. After training, the model performance was evaluated on the test set, calculating the coefficient of determination, root mean square error, and mean absolute error. Comparisons between the actual and predicted values for the training, validation, and test sets are shown below. Figures 3 to 5 The results show that the data points are closely distributed near the diagonal, exhibiting a good fit. Specifically, the coefficient of determination, root mean square error, and mean absolute error of the test set are 0.9946, 0.0126, and 0.0192, respectively, indicating that the surrogate model has high prediction accuracy and good generalization ability, and can be effectively applied to the prediction of CO2-H2 underground biochemical synthesis of natural gas conversion efficiency.
[0070] The target gas reservoir parameters to be predicted are: temperature 60℃, salinity 30g / L, carbonate mineral content 2wt%, maximum biomass capacity 0.05g / L, optimal specific growth rate of methanogens 1.5d⁻¹, optimal specific growth rate of acetic acid bacteria 1.0d⁻¹, optimal specific growth rate of sulfate-reducing bacteria 1.2d⁻¹, CO₂-H₂ ratio 1:4, and total CO₂-H₂ pressure 130atm. After normalizing these parameters according to the maximum and minimum values in Table 1, they are input into the constructed surrogate model for prediction. The average calculation time is less than 0.1 seconds, and the average effective CO₂-H₂ conversion rate is 0.4653. Therefore, the average daily conversion rate is 0.4653 × 100% ÷ 360 = 0.129%, meaning the average daily CO₂-H₂ conversion rate over 360 days is approximately 0.129%.
[0071] This invention is the first to utilize machine learning algorithms to construct a proxy model for the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas. Compared to traditional single numerical simulation methods, it predicts conversion efficiency faster and yields more accurate results. The input feature variables cover a total of nine items, including geological parameters, microbial parameters, and engineering parameters, comprehensively reflecting the impact of factors such as microbial competition and engineering operations in the reservoir environment on conversion efficiency, thus more closely reflecting actual conditions. By comprehensively employing the Latin hypercube sampling method and multiple mapping methods, it can more comprehensively and uniformly cover the input variable space, improving the representativeness of the sampling while reducing the number of samples. The average time-dependent conversion rate of CO2-H2 is used as the conversion efficiency evaluation index. This index reflects both the degree of conversion and the time effect, and has stronger comprehensiveness and accuracy compared to the traditional CO2 conversion rate or CH4 release rate. Based on the trapezoidal integral concept and the principle of equal area, the average time-dependent conversion rate is calculated using the effective conversion rates of CO2 and H2 at four time points (90 days, 180 days, 270 days, and 360 days), eliminating the need for calculation at each time point and significantly reducing the computational workload of obtaining output samples.
[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. A machine learning-based method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas, characterized in that, Includes the following steps: S1. Select geological parameters, microbial parameters, and engineering parameters as input feature variables for the machine learning model; S2. Based on the range of values of the input feature variables and the Latin hypercube sampling method, obtain the input samples; S3. Using the biogeochemical simulation results under different input sample conditions, calculate the average time-dependent conversion rate of CO2-H2, and use it as the output sample of the machine learning model; S4. Based on the normalized input samples and the unprocessed output samples, construct a machine learning proxy model for the average time-of-use conversion rate of CO2-H2; S5. Based on the geological and microbiological parameters of the target depleted oil and gas reservoir, and combined with different combinations of engineering parameters, the corresponding average time-dependent CO2-H2 conversion rate is predicted using a surrogate model.
2. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, The geological parameters in the input feature variables in step S1 include temperature, salinity, and carbonate mineral content; the microbiological parameters include maximum biomass capacity, optimal specific growth rate of methanogens, optimal specific growth rate of acetic acid bacteria, and optimal specific growth rate of sulfate-reducing bacteria; and the engineering parameters include CO2-H2 ratio and total CO2-H2 pressure.
3. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, The specific method for obtaining input samples in step S2 is as follows: First, a numerical sequence between 0 and 1 is generated using the Latin hypercube sampling method. Then, the sequence is mapped to the value range of the input feature variables. The maximum biomass capacity is mapped using a logarithmic mapping method, while other input feature variables are mapped using a linear mapping method.
4. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, The biogeochemical simulation in step S3 uses the software PHREEQC, and the microbial reproduction kinetics model used is as follows: 1) Microbial specific growth rate calculation model: Where: μ gr The specific growth rate of microorganisms, μ opt For the optimal specific growth rate of microorganisms, C A and C D K represents the concentrations of electron acceptor and electron donor, respectively. A and K D λ represents the half-saturation constant of the electron acceptor and electron donor, respectively. T , λ pH , λ S These are dimensionless influencing factors related to temperature, pH, and salinity, respectively. 2) Substrate consumption calculation model: In the formula: r s denoted as substrate consumption rate, Y as yield coefficient, and N as number of microorganisms; 3) Biomass evolution model: r bio =-r s F X Y-dN In the formula: r bio For the rate of biomass change, F X denoted as biomass capacity factor, and d as attenuation coefficient.
5. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, The method for calculating the average time-dependent conversion rate of CO2-H2 in step S4 is as follows: R=(R CO2 +R H2 ) / 2 In the formula: R is the average time-dependent conversion rate of CO2-H2, R CO2 R H2 The time-dependent conversion rates of CO2 and H2 are respectively, calculated based on the trapezoidal integral concept and the principle of equal area: In the formula: C 90 C 180 C 270 C 360 The effective CO2 conversion rates at 90 days, 180 days, 270 days, and 360 days, respectively, H 90 H 180 H 270 H 360 The effective conversion rates of H2 at 90 days, 180 days, 270 days, and 360 days, respectively.
6. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, In step S4, when constructing the machine learning proxy model for the average time-efficiency conversion rate of CO2-H2, the machine learning algorithm used is a regression algorithm, including but not limited to random forest regression algorithm, least squares boosting regression algorithm and feedforward neural network algorithm.
7. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, In step S4, when constructing the machine learning proxy model for the average time-efficiency conversion rate of CO2-H2, the sample dataset is divided into a training set, a validation set, and a test set, with proportions of 70%, 15%, and 15%, respectively. The proxy model is trained using the training set, the model hyperparameters are optimized using the validation set, and the predictive performance indicators of the model are calculated based on the test set. These performance indicators include, but are not limited to, the coefficient of determination, root mean square error, and mean absolute error.
8. The method for predicting the conversion efficiency of CO2-H2 underground biochemical synthesis of natural gas based on machine learning according to claim 1, characterized in that, Before predicting the target oil and gas reservoir using geological parameters, microbiological parameters, and engineering parameters in step S5, normalization processing is required based on the input sample dataset.