Combustible lower limit prediction method for mixed combustible gas

Abstract

The invention belongs to the technical field of combustion science and chemical kinetics simulation, and provides a method for predicting the lower combustible limit of mixed combustible gas. The method comprises the following steps: firstly, constructing and verifying a laminar flame velocity mechanism covering a single component and multiple components and the applicability to multi-component fuel; secondly, on the basis that flame stretching and radiation effects are considered in a traditional model, a dynamic instability factor is further introduced, so that the prediction precision of the lower combustible limit of the low-Lewis-number combustible gas is improved; finally, a single-component data set is generated in a unified mode, multiple regression algorithms are adopted for training and comparison, a prediction model is determined in a preferential mode, the Lunchaterow principle is combined, generalization prediction of the multi-component combustible lower limit is achieved, and finally comparison verification is conducted with existing experimental data. According to the system and the method, the dependence on a large-scale training data set is remarkably reduced, the calculation efficiency is improved on the premise of ensuring the prediction precision, and the system and the method have obvious engineering application value and are particularly suitable for the safety-critical field.

Description

Technical Field

[0001] This invention belongs to the field of combustion science and chemical kinetics simulation technology, and relates to a method for predicting the lower flammability limit of mixed combustible gases. Background Technology

[0002] Accurate prediction of the lower flammability limit (LFL) of multi-component combustible gases under a wide range of operating conditions is of great significance for energy security, combustion system optimization, and accident prevention. Currently widely used LFL prediction methods are mainly based on thermal explosion theory and chain reaction theory, and have derived two categories of criteria: calculating adiabatic flame temperature and limiting laminar combustion velocity (LBV). Although these methods perform well under certain conditions, their prediction accuracy is heavily dependent on the selected threshold, and a unified standard has not yet been established. They are particularly prone to underestimating the LFL of low Lewis number (Le) combustible gases (such as hydrogen and hydrogen-rich mixtures).

[0003] In recent years, numerical simulation methods based on offset premixed flame models have been widely used for LFL prediction. These models comprehensively consider the effects of flame stretching and radiative heat loss, and can describe the extinction behavior of non-adiabatic stretched flames relatively well. However, existing models usually neglect the effect of thermal diffusion instability on the combustion behavior of low-Le fuels, leading to significant biases in the LFL prediction of hydrogen and hydrogen-rich mixtures.

[0004] For multi-component mixtures, the Le Chatelier principle, as a classic empirical formula, is widely used for LFL calculations of mixed fuel gases. Meanwhile, machine learning has shown strong potential in predicting combustion characteristics, particularly adept at extracting complex nonlinear patterns from data. However, when it comes to multi-component wide-condition prediction, existing machine learning methods typically rely on large-scale training datasets. For example, Paykani points out in his review "Machine Learning Applications for Predicting Fuel Ignition and Flame Property Prediction" that in order for a model to have good generalization ability under multi-component and wide-condition conditions (including fuel type, mixture ratio, temperature, pressure, etc.), tens of thousands of experimental or simulation data are often required for training. In this patent, with the increase in the number of components and the combined changes in proportion, pressure, and temperature, if relying entirely on machine learning, the scale of training data could expand to millions or even higher, leading to a significant increase in data acquisition and computational costs. To address this bottleneck, this invention proposes an innovative strategy: first, constructing a high-precision single-component wide-condition LFL prediction model, and then extending it to multi-component prediction by incorporating the Le Chatelier principle, thereby significantly reducing the dependence on multi-component training data while maintaining prediction accuracy. Summary of the Invention

[0005] The purpose of this invention is to address the above-mentioned problems by proposing a method for predicting the lower flammability limit of mixed combustible gases.

[0006] The technical solution of this invention:

[0007] A method for predicting the lower flammability limit of mixed combustible gases includes the following steps:

[0008] Step (1): Select or construct a laminar flame velocity mechanism that can predict single-component and multi-component combustible gases as the basis for counterbalanced premixed flame simulation; within a preset temperature and pressure range (e.g., 250K~500K, 0.05~20 atm), verify the laminar flame velocity of single-component and multi-component combustible gases using the selected or constructed laminar flame velocity mechanism.

[0009] Step (2): In the offset premixed flame prediction LFL model, the flame stretching effect (the effect of the flow velocity gradient on the flame propagation rate) and radiative heat loss (characterized by adding a radiative heat transfer term to the energy equation to represent energy dissipation at high temperatures) are introduced by commonly used combustion theories, and Lewis number correction is introduced (corrected according to the difference in mass diffusion and thermal diffusion of different components) to obtain the corrected offset premixed flame prediction LFL model; by introducing Lewis number correction, the prediction accuracy of the offset premixed flame prediction LFL model for the lower flammability limit (LFL) of combustible gases with a Lewis number less than 1 is improved; based on the laminar flame velocity mechanism selected or constructed in step (1), the offset premixed flame prediction LFL model for single-component and multi-component combustible gases is verified under preset temperature and pressure conditions, and the corrected offset premixed flame prediction LFL model is used to generate single-component training dataset 1 and unknown working condition verification dataset 2;

[0010] Step 1: Introduce the RADCAL radiation heat loss model into the code of the anti-hybridized flame prediction LFL model on the Canterta computing platform.

[0011] Step 2: Under the set volume fraction conditions of multiple combustible gases, calculate the elongation limit of the counter-flame and determine whether extinction occurs; when extinction occurs, record the corresponding volume fraction of combustible gas; as the volume fraction of combustible gas gradually decreases, the range that the flame can sustain continuously shrinks until it is completely extinguished; through iterative calculation, determine the inflection point where the range that the flame can sustain shrinks to disappear, that is, obtain the lower limit of extinction of the combustible gas under the current conditions;

[0012] Step 3: Calculate the Lewis number of the mixed combustible gas under the condition of stoichiometry, i.e., equivalence ratio of 1; for gases with a Lewis number less than 1, introduce a kinetic instability correction factor, and multiply the initial lower flammability limit by the Lewis number to obtain the final lower flammability limit.

[0013] Step 4: Within the preset temperature and pressure range, verify the lower flammability limits of single and multi-component combustible gases using the modified offset premixed flame prediction LFL model.

[0014] Step 5: Generate baseline operating conditions using the modified offset premixed flame prediction LFL model: increasing temperature and pressure; and use the Latin hypercube sampling method (LHS) to generate additional supplementary training conditions to enhance the coverage of operating conditions and data diversity. Finally, obtain a single-component training dataset 1 that meets the training requirements for modeling analysis. At the same time, use the Latin hypercube sampling method to generate a validation dataset 2 of 100 unknown operating conditions for testing the generalization ability of the modified offset premixed flame prediction LFL model.

[0015] Step (3): Based on the dataset generated in step (2), data preprocessing and feature engineering are performed first; then, multiple regression algorithms are used for training and comparison, and the best prediction algorithm model is selected. The Le Chatelier principle is combined to achieve generalized prediction of the lower limit of flammability of multiple components; finally, the reliability and generalization ability of the selected prediction algorithm model are confirmed by comparing and verifying with existing experimental data.

[0016] Step 1: Preprocess the single-component training dataset 1 and the unknown working condition verification dataset 2 generated in step (2), and use the mixed combustible gas types and their volume fractions, temperature and pressure as input variables for the subsequent machine learning algorithm model (where temperature and pressure are normalized), and use the lower combustible limit as the output response variable.

[0017] Step 2: Use the five-fold cross-validation method to divide the single-component training dataset 1, where 80% of the data is used as the training set and 20% of the data is used as the test set;

[0018] Step 3: Using eight regression algorithms—Linear Regression (LR), Support Vector Regression (SVR), XGBoost Regression (XGBR), Random Forest Regression (RFR), LightGBM (LGBM), K-Nearest Neighbor Regression (KNN), Gradient Boosting Regression (GBR), and Gaussian Process Regression (GPR)—the hyperparameters of these eight regression algorithms are optimized within the Optuna framework. The algorithms are then trained, and the prediction results are compared.

[0019] Step 4: Test the above eight regression algorithms on the unknown working condition verification dataset 2, and select the best performing prediction model based on the results of the training set and the test set.

[0020] Step 5: Use the optimal prediction model to obtain the lower flammability limit of each individual combustible gas under different conditions, and substitute the lower flammability limit of each individual combustible gas under different conditions into Le Chatelier's principle formula:

[0021]

[0022] in: This refers to the mole fraction of a single combustible gas. The lower flammability limit value obtained by the optimal prediction model for a single combustible gas is used to extend the prediction of the lower flammability limit of a multi-component combustible gas system by substituting the prediction result of a single combustible gas into the mixing rule of Le Chatelier's principle formula.

[0023] Step 6: Collect experimental data on the lower flammability limit of multiple components;

[0024] Step 7: Compare and verify the multi-component prediction results with the collected multi-component lower flammability limit experimental data.

[0025] The beneficial effects of this invention are:

[0026] (1) Improve prediction accuracy: By introducing Lewis number correction into the traditional offset premixed flame model, the model can more accurately reflect the actual flame propagation characteristics, especially for gases with Lewis number less than 1 (such as hydrogen and hydrogen-rich mixtures), which significantly improves the prediction accuracy of LFL.

[0027] (2) Rapid prediction capability: Based on the single-component prediction model and combined with Le Chatelier's principle, it can quickly and accurately predict the lower limit of LFL of any type and proportion of mixed combustible gas within a set temperature and pressure range.

[0028] (3) Reduce data costs: The present invention adopts the method of "single-component model prediction + Le Chatelier principle extension", which avoids the dependence on large-scale multi-component experimental or training data, and maintains high prediction accuracy while reducing data acquisition and calculation costs.

[0029] (4) High scalability: By adding new single-component prediction models, it can be quickly extended to new fuel systems, with good flexibility and portability. Attached Figure Description

[0030] Figure 1 This is a technical roadmap for the multi-component, wide-range operating condition flammability lower limit prediction method proposed in this invention.

[0031] Figure 2 The diagram shows the verification results of the laminar flame velocity of the CO–C2 mechanism constructed in this invention. Among them, (a) is the verification of the laminar flame velocity of ammonia / methane, (b) is the verification of the laminar flame velocity of ammonia / hydrogen / carbon monoxide, (c) is the verification of the laminar flame velocity of methanol / ammonia, (d) is the verification of the laminar flame velocity of ethanol, (e) is the verification of the laminar flame velocity of hydrogen / methane / dimethyl ether, and (f) is the laminar flame velocity of ethane, ethylene, and acetylene.

[0032] Figure 3 This is a diagram showing the tensile extinguishing characteristics of a pure ammonia non-insulated flame (including radiative heat loss) calculated under normal temperature and pressure conditions according to the present invention.

[0033] Figure 4 The figures show the verification results of the lower flammability limit of hydrogen under various operating conditions based on the improved counter-mixed flame model of this invention. Among them, (a) is the verification of the lower flammability limit of hydrogen with temperature, (b) is the verification of the lower flammability limit of hydrogen with pressure, (c) is the verification of the lower flammability limit of methane / hydrogen with hydrogen ratio, (d) is the verification of the lower flammability limit of ammonia / hydrogen with hydrogen ratio, (e) is the verification of the lower flammability limit of carbon monoxide / hydrogen with hydrogen ratio, and (f) is the verification of the lower flammability limit of hydrogen / methanol / ammonia with hydrogen ratio.

[0034] Figure 5 The performance comparison chart of the single-component flammability lower limit prediction model constructed in this invention is shown (left: training and test datasets; right: unknown dataset), where (a) is a comparison of the prediction performance of different regression methods on the training set and test set, and (b) is a comparison of the validation performance of different regression methods on the unknown dataset.

[0035] Figure 6 This is a distribution diagram of the experimental data on the lower limit of flammability of multiple components collected in this invention.

[0036] Figure 7 This is a comparison and verification diagram of the predicted lower flammability limit of the multi-component system in this invention with experimental data. Detailed Implementation

[0037] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0038] As attached Figure 1 As shown, a method for predicting the lower flammability limit of mixed combustible gases is proposed, which can rapidly predict the lower flammability limit of a single component or any number of components of ten combustible gases, namely H2, NH3, CH4, CH3OH, CO, C2H5OH, CH3OCH3, C2H6, C2H4, and C2H2, within a temperature range of 250K-500K and a pressure range of 0.05-2MPa.

[0039] Includes the following steps:

[0040] Step (1): Construct a laminar flame velocity prediction mechanism applicable to single-component and multi-component combustible gases in the C0–C2 range, as the basis for offset premixed flame simulation. This laminar flame velocity prediction mechanism shows high prediction accuracy for laminar combustion velocities of single-component and multi-component gases in the temperature range of 250K-500K and the pressure range of 0.05-2MPa.

[0041] Step 1: Building upon existing combustion mechanisms such as NH3 / CH4 / CO / H2, NH3 / CH3OH, NH3 / C2H5OH, and NH3 / DME, and adhering to the principles of component coverage and reaction pathway integrity, this step further incorporates the C2H6, C2H4, and C2H2 mechanisms constructed by Metcalfe WK et al. in their paper "A Hierarchical and Comparative Kinetic Modeling Study of C1 − C2Hydrocarbon and Oxygenated Fuels." By merging repetitive reactions, standardizing rate parameters, and optimizing non-critical reactions, a comprehensive chemical kinetic mechanism consisting of 110 substances and 682 reactions is ultimately formed to describe the combustion reaction processes of the aforementioned ten combustible gases.

[0042] Step 2: Under conditions of temperature 298K~473K and pressure 0.1~10atm, the constructed laminar flame velocity prediction mechanism was verified for single-component and multi-component laminar flame velocities. The verification results are as follows: Figure 2 As shown;

[0043] Step (2): In the traditional offset premixed flame prediction LFL model, which includes the flame stretching effect (the influence of the flow velocity gradient on the flame propagation rate) and radiative heat loss (characterized by adding a radiative heat transfer term to the energy equation to represent energy dissipation at high temperatures) commonly used in combustion science, a Lewis number correction is further introduced (based on the differences in mass diffusion and thermal diffusion of different components). By introducing the Lewis number correction, the offset premixed flame prediction LFL model can improve the prediction accuracy of the lower flammability limit (LFL) of low Lewis number gases (such as hydrogen and hydrogen-rich mixtures). Based on the laminar flame velocity prediction mechanism established in step (1), the LFL of single and multi-component combustible gases in the temperature range of 250K-500K and the pressure range of 0.05-2MPa were verified. A large-scale single-component training dataset 1 and an unknown working condition verification dataset 2 were generated using the corrected offset premixed flame prediction LFL model. The specific content and method steps include:

[0044] Step 1: Introduce the RADCAL radiation heat loss model into the code of the anti-hybrid flame prediction LFL model on the Canterta computing platform to fully consider the radiation effects of H2O and CO2.

[0045] Step 2: Under the set volume fraction conditions of multiple combustible gases, calculate the elongation limit of the counter-flame and determine whether extinction occurs; when extinction occurs, record the corresponding volume fraction of combustible gas; as the volume fraction of combustible gas gradually decreases, the range that the flame can sustain continuously shrinks until it is completely extinguished; through iterative calculation, determine the inflection point where the range that the flame can sustain shrinks to disappearance, that is, obtain the lower limit of extinction of the combustible gas under the current conditions. The calculation process is as follows: Figure 3 As shown;

[0046] Step 3: Calculate the Lewis number of the mixed combustible gas under the condition of stoichiometry, i.e., equivalence ratio of 1; for gases with a Lewis number less than 1, introduce a kinetic instability correction factor, and multiply the initial lower flammability limit by the Lewis number to obtain the final lower flammability limit.

[0047] Step 4: The lower flammability limits of the modified counter-mixed flame prediction LFL model are verified using single and multi-component combustible gases within a temperature range of 250K–500K and a pressure range of 0.05–20 atm. Figure 4 As shown;

[0048] Step 5: Generate baseline operating conditions using the improved offset premixed flame prediction LFL model: Temperature increases from 250K to 500K in increments of 50K, and pressure is set at 0.05, 0.1, 0.5, 1, 1.5, and 2MPa, resulting in 360 baseline operating conditions. An additional 740 operating conditions are generated using the Latin hypercube sampling (LHS) method to enhance the coverage and diversity of the operating conditions. In the end, a total of 1000 single-component data points (dataset 1) are obtained for modeling and analysis. At the same time, 100 independent operating condition data points (dataset 2) are generated using the LHS method for testing the model's generalization ability.

[0049] Step (3): Based on the dataset generated in step (2), data preprocessing and feature engineering are first performed; then, multiple regression algorithms are used for training and comparison, and the best prediction algorithm model is selected. The Le Chatelier principle is then used to achieve generalized prediction of the lower limit of flammability for multiple components; finally, the reliability and generalization ability of the selected prediction algorithm model are confirmed by comparing it with existing experimental data. Specific implementation steps include:

[0050] Step 1: Preprocess the single-component training dataset 1 and the unknown working condition verification dataset 2 generated in step (2), and use the mixed combustible gas types and their volume fractions, temperature and pressure as input variables for the subsequent machine learning algorithm model (where temperature and pressure are normalized), and use the lower combustible limit as the output response variable.

[0051] Step 2: Use the five-fold cross-validation method to divide the single-component training dataset 1, where 80% of the data is used as the training set and 20% of the data is used as the test set;

[0052] Step 3: Eight regression algorithms were employed, including Linear Regression (LR), Support Vector Regression (SVR), XGBoost Regression (XGBR), Random Forest Regression (RFR), LightGBM (LGBM), K-Nearest Neighbor Regression (KNN), Gradient Boosting Regression (GBR), and Gaussian Process Regression (GPR). After hyperparameter optimization of these algorithms within the Optuna framework, the model was trained, and the prediction results were compared. Figure 5 As shown in (a);

[0053] Step 4: Test the above eight models on the unknown dataset 2, such as... Figure 5 As shown in (b). Based on the results of the training and test sets, the best-performing prediction model is selected;

[0054] Step 5: Use the optimal prediction model to obtain the lower flammability limit of each individual combustible gas under different conditions, and substitute the lower flammability limit of each individual combustible gas under different conditions into Le Chatelier's principle formula:

[0055]

[0056] in: This refers to the mole fraction of a single combustible gas. The optimal prediction model for a single combustible gas yields the lower flammability limit (LFL). By substituting the prediction results for a single combustible gas into the mixing rules of Le Chatelier's principle, an extended prediction of the LFL for multi-component combustible gas systems is achieved. The combined results demonstrate that this method maintains the high accuracy of the single-component model while accurately predicting the LFL of mixed fuels without requiring large-scale multi-component training data.

[0057] Step 6: Collect existing experimental data on the lower flammability limits of multi-component flammability, and the data distribution is as follows: Figure 6 As shown;

[0058] Step 7: Compare and verify the multi-component prediction results with the collected experimental data. The results are as follows: Figure 7 As shown.

[0059] Ultimately, the novel machine learning ensemble model constructed in this invention can rapidly predict the lower flammability limits of the aforementioned ten combustible gases within a temperature range of 250K–500K and a pressure range of 0.05–2MPa, for either a single component or any combination of components. The model prediction results are compared with experimental data, and the results are as follows: Figure 7 As shown in the figure. The comparison results show that the error between the predicted value and the actual value of the constructed integrated model is small, and it can predict the actual lower flammability limit of any combination of components relatively accurately.

[0060] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for predicting the lower flammability limit of mixed combustible gases, characterized in that, Includes the following steps: Step (1): Select or construct a laminar flame velocity mechanism that can predict single-component and multi-component combustible gases; within a preset temperature and pressure range, verify the laminar flame velocity of the selected or constructed laminar flame mechanism for single-component and multi-component combustible gases. Step (2): Introduce Lewis number correction into the offset premixed flame prediction LFL model to obtain the corrected offset premixed flame prediction LFL model; By introducing Lewis number correction, improve the prediction accuracy of the offset premixed flame prediction LFL model for the lower flammability limit of combustible gases with Lewis number less than 1; Based on the laminar flame velocity mechanism selected or constructed in step (1), verify the offset premixed flame prediction LFL model for single-component and multi-component combustible gases under preset temperature and pressure conditions, and use the corrected offset premixed flame prediction LFL model to generate single-component training dataset 1 and unknown working condition verification dataset 2; Step 1: Introduce the RADCAL radiation heat loss model into the code of the anti-hybridized flame prediction LFL model on the Canterta computing platform. Step 2: Under the set volume fraction conditions of multiple combustible gases, calculate the elongation limit of the counter-flame and determine whether extinction occurs; when extinction occurs, record the corresponding volume fraction of combustible gas; as the volume fraction of combustible gas gradually decreases, the range that the flame can sustain continuously shrinks until it is completely extinguished; through iterative calculation, determine the inflection point where the range that the flame can sustain shrinks to disappear, that is, obtain the lower limit of extinction of the combustible gas under the current conditions; Step 3: Calculate the Lewis number of the mixed combustible gas under the condition of stoichiometry, i.e., equivalence ratio of 1; for gases with a Lewis number less than 1, introduce a kinetic instability correction factor, and multiply the initial lower flammability limit by the Lewis number to obtain the final lower flammability limit. Step 4: Within the preset temperature and pressure range, verify the lower flammability limits of single and multi-component combustible gases using the modified offset premixed flame prediction LFL model. Step 5: Use the modified offset premixed flame prediction LFL model to generate baseline operating conditions: increasing temperature and pressure; In addition, a supplementary training dataset 1 was generated using the Latin hypercube sampling method to enhance the coverage of the training dataset and the diversity of the data. Finally, a single-component training dataset 1 was obtained to meet the training requirements for modeling and analysis. At the same time, a validation dataset 2 of 100 unknown training datasets was generated using the Latin hypercube sampling method for testing the generalization ability of the corrected hedging premixed flame prediction LFL model. Step (3): Based on the dataset generated in step (2), data preprocessing and feature engineering are performed first; Secondly, multiple regression algorithms were used for training and comparison, and the best prediction algorithm model was selected. The Le Chatelier principle was then combined to achieve generalized prediction of the lower limit of flammability of multiple components. Finally, the reliability and generalization ability of the selected prediction algorithm model were confirmed by comparing it with existing experimental data. Step 1: Preprocess the single-component training dataset 1 and the unknown working condition verification dataset 2 generated in step (2), and use the mixed combustible gas types and their volume fractions, temperature and pressure as input variables for the subsequent machine learning algorithm model, and use the lower limit of combustibility as the output response variable; Step 2: Use the five-fold cross-validation method to divide the single-component training dataset 1, where 80% of the data is used as the training set and 20% of the data is used as the test set; Step 3: Using eight regression algorithms—linear regression, support vector regression, XGBoost regression, random forest regression, LightGBM, K-nearest neighbor regression, gradient boosting regression, and Gaussian process regression—we optimized the hyperparameters of these eight algorithms within the Optuna framework, trained them, and compared the prediction results. Step 4: Test the above eight regression algorithms on the unknown working condition verification dataset 2, and select the best performing prediction model based on the results of the training set and the test set. Step 5: Use the optimal prediction model to obtain the lower flammability limit of each individual combustible gas under different conditions, and substitute the lower flammability limit of each individual combustible gas under different conditions into Le Chatelier's principle formula: ； in: This refers to the mole fraction of a single combustible gas. The lower flammability limit value obtained by the optimal prediction model for a single combustible gas is used to extend the prediction of the lower flammability limit of a multi-component combustible gas system by substituting the prediction result of a single combustible gas into the mixing rule of Le Chatelier's principle formula. Step 6: Collect experimental data on the lower flammability limit of multiple components; Step 7: Compare and verify the multi-component prediction results with the collected multi-component lower flammability limit experimental data.

Images