Hydrogen-doped natural gas combustion efficiency prediction method and system based on transfer learning

Abstract

The invention relates to the technical field of energy, and discloses a hydrogen-doped natural gas combustion efficiency prediction method and system based on transfer learning, and the method comprises the steps: respectively obtaining a fuel characteristic coefficient and a combustion condition coefficient through a fuel characteristic model and a combustion condition model; a matching degree coefficient is obtained through a matching model by combining combustion pressure fluctuation and core flame signal intensity, and meanwhile a combustion product coefficient is obtained through a product model based on the combustion chamber temperature, the exhaust gas temperature and the carbon monoxide concentration. And then the matching degree coefficient and the combustion product coefficient are integrated, the predicted combustion efficiency is calculated through the prediction model, and finally based on the prediction efficiency and the current jet flow momentum, the target fuel and air jet flow momentum are calculated through the optimization model. The method can adapt to changes of the hydrogen mixing proportion, multi-source dynamic information is fused, the problems that a traditional model is poor in adaptability and low in precision under the hydrogen mixing working condition are solved, and precise prediction of combustion efficiency and real-time optimization regulation and control of operation parameters are achieved.

Description

Technical Field

[0001] This invention belongs to the field of energy technology, and in particular relates to a method and system for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning. Background Technology

[0002] As the global energy structure transitions towards a low-carbon model, the importance of hydrogen energy as a clean secondary energy source is becoming increasingly prominent. Hydrogen-blended natural gas technology, by transporting and utilizing hydrogen mixed with natural gas, has become an effective way to achieve low-carbon operation of existing gas infrastructure. However, hydrogen and natural gas have significantly different combustion characteristics. Hydrogen has a higher flame propagation speed, a wider combustion limit, and a shorter flameout distance. These characteristics cause traditional natural gas combustion control methods to gradually fail under hydrogen-blended conditions. Most existing combustion efficiency prediction models are based on pure natural gas; when the hydrogen blending ratio changes, the model prediction accuracy drops sharply, failing to meet the actual operational control requirements.

[0003] Currently, industrial boilers and combustion devices generally rely on empirical formulas based on excess air coefficients and flue gas temperature to estimate efficiency. This method fails to adequately account for the dynamic characteristics of hydrogen-blended fuels. On the other hand, while advanced numerical simulation techniques can accurately describe the combustion process, their high computational cost makes online application difficult. In recent years, transfer learning has demonstrated its advantages in modeling complex industrial processes. By transferring source domain knowledge, it accelerates model learning under new operating conditions, providing a new solution for multivariable and highly nonlinear processes such as hydrogen-blended natural gas combustion.

[0004] Existing technologies suffer from three main shortcomings: first, they are poorly adaptable to changes in fuel characteristics caused by hydrogen blending; second, they fail to effectively integrate dynamic features such as pressure fluctuations and flame signals; and third, they lack the ability to optimize and control based on real-time conditions. Therefore, there is an urgent need to develop a combustion efficiency prediction and optimization method that can adapt to changes in hydrogen ratios, integrate multi-source information, and is suitable for online applications, in order to support the large-scale and safe application of hydrogen-blended natural gas technology. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning, in order to solve the above-mentioned problems.

[0006] This invention is implemented as follows: a method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning. The method includes: obtaining fuel characteristic coefficients through a fuel characteristic model based on the hydrogen blending ratio and the calorific value of the blended fuel; obtaining combustion condition coefficients through a combustion condition model based on the excess air coefficient, air preheating temperature, and burner pressure drop; deriving a matching degree coefficient through a matching model based on combustion pressure fluctuations and core flame signal intensity under the current fuel characteristic group coefficients and combustion condition group coefficients; obtaining combustion product coefficients through a product model based on combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration; calculating the predicted combustion efficiency of hydrogen-blended natural gas through a prediction model based on the matching degree coefficient and the combustion product group coefficients; and calculating the target fuel and air jet flow rate through an optimization model based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate.

[0007] A further technical solution involves the following steps for calculating the target fuel and air jet flow rate based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate using an optimization model: Based on the difference between the predicted combustion efficiency of hydrogen-blended natural gas and the system's highest combustion efficiency, and combined with a momentum adjustment coefficient, the current fuel and air jet flow rate is positively compensated to obtain the target fuel and air jet flow rate; wherein the target fuel and air jet flow rate is positively correlated with the difference.

[0008] A further technical solution involves calculating the predicted combustion efficiency of hydrogen-blended natural gas using a prediction model based on the matching degree coefficient and the combustion product group coefficient. The specific steps are as follows: adding the lowest combustion efficiency of the system to an adjustment amount to obtain the predicted combustion efficiency of the hydrogen-blended natural gas; wherein, the adjustment amount is determined by the difference between the highest and lowest combustion efficiencies of the system, together with the combustion product coefficient and the matching degree coefficient; the predicted combustion efficiency of hydrogen-blended natural gas is positively correlated with both the combustion product coefficient and the matching degree coefficient.

[0009] A further technical solution involves obtaining the combustion product coefficients based on the combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration using a product model, with the following specific steps:

[0010] The combustion chamber temperature index is obtained by comparing the difference between the current combustion chamber temperature and the minimum allowable combustion chamber temperature with the difference between the optimal combustion chamber temperature and the minimum allowable combustion chamber temperature, and then applying min and max functions with upper limits of 1 and lower limits of 0. Similarly, the combustion chamber temperature index is obtained by comparing the difference between the current exhaust gas temperature and the theoretical minimum exhaust gas temperature with the difference between the maximum allowable exhaust gas temperature and the theoretical minimum exhaust gas temperature, and then applying min and max functions with upper limits of 1 and lower limits of 0. The carbon monoxide concentration index is obtained by comparing the current carbon monoxide concentration in the flue gas with the maximum allowable carbon monoxide concentration, and then applying min functions with upper limits of 1. The combustion product coefficient is obtained by comprehensively processing the combustion chamber temperature index, exhaust gas temperature index, and carbon monoxide concentration index. The combustion product coefficient is positively correlated with the combustion chamber temperature index and negatively correlated with the exhaust gas temperature index and the carbon monoxide concentration index.

[0011] A further technical solution involves the following steps for deriving the matching coefficient based on combustion pressure fluctuations and core flame signal intensity using a matching model under the current fuel characteristic group coefficients and combustion condition group coefficients: The root mean square value of the current combustion pressure fluctuation is compared to the maximum allowable pressure fluctuation value of the system, and then a min function is used to limit the amplitude to an upper limit of 1 to obtain the combustion pressure fluctuation index; the absolute value of the difference between the current core flame signal intensity and the optimal flame signal intensity is compared to the allowable deviation range, and then a min function is used to limit the amplitude to an upper limit of 1 to obtain the core flame signal intensity index; a theoretical expected value is obtained based on the fuel characteristic coefficients and combustion condition coefficients; the actual stability performance value determined by the combustion pressure fluctuation index and the core flame signal intensity index is compared with the theoretical expected value to obtain the matching coefficient; wherein, the matching coefficient characterizes the degree of agreement between the actual stability performance and the theoretical expected value.

[0012] A further technical solution involves obtaining the fuel characteristic coefficient based on the hydrogen blending ratio and the calorific value of the mixed fuel through a fuel characteristic model. The specific steps are as follows: The current hydrogen blending ratio is compared with the maximum allowable hydrogen blending ratio of the system to obtain a hydrogen blending ratio index; the current calorific value of the mixed fuel is compared with the lower heating value of the benchmark natural gas to obtain a mixed fuel calorific value index; the hydrogen blending ratio index and the mixed fuel calorific value index are then combined to obtain the fuel characteristic coefficient; wherein the fuel characteristic coefficient is positively correlated with both the hydrogen blending ratio index and the mixed fuel calorific value index.

[0013] A further technical solution, the specific steps for obtaining the combustion condition coefficient based on the excess air coefficient, air preheating temperature, and burner pressure drop through the combustion condition model are as follows: The absolute value of the difference between the current excess air coefficient and the optimal excess air coefficient is compared with the allowable fluctuation range of the excess air coefficient; then, a min function is used to limit the amplitude to an upper limit of 1 to obtain the excess air coefficient index; the difference between the current air preheating temperature and the system's minimum allowable air preheating temperature is compared with the difference between the system's maximum allowable air preheating temperature and the system's minimum allowable air preheating temperature; the differences are then sequentially... After limiting the upper limit to 1 and the lower limit to 0 using the min and max functions, the air preheating temperature index is obtained. The absolute value of the difference between the current burner pressure drop and the design optimal burner pressure drop is compared with the allowable fluctuation range of pressure drop, and then the upper limit of the min function is limited to 1 to obtain the burner pressure drop index. The excess air coefficient index, the air preheating temperature index, and the burner pressure drop index are comprehensively processed to obtain the combustion condition coefficient. The combustion condition coefficient is positively correlated with the air preheating temperature index and negatively correlated with the excess air coefficient index and the burner pressure drop index.

[0014] A hydrogen-blended natural gas combustion efficiency prediction system based on transfer learning includes: a fuel characteristic evaluation module, which obtains fuel characteristic coefficients based on the hydrogen blending ratio and the calorific value of the blended fuel through a fuel characteristic model; a combustion condition evaluation module, which obtains combustion condition coefficients based on the excess air coefficient, air preheating temperature, and burner pressure drop through a combustion condition model; a matching degree evaluation module, which derives a matching degree coefficient based on combustion pressure fluctuations and core flame signal intensity under the current fuel characteristic group coefficients and combustion condition group coefficients through a matching model; a product evaluation module, which obtains combustion product coefficients based on combustion chamber temperature, flue gas temperature, and carbon monoxide concentration through a product model; a prediction calculation module, which calculates the predicted hydrogen-blended natural gas combustion efficiency based on the matching degree coefficients and combustion product group coefficients through a prediction model; and an optimization module, which calculates the target fuel and air jet flow rate based on the predicted hydrogen-blended natural gas combustion efficiency and the current fuel and air jet flow rate through an optimization model.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0016] 1. By introducing a transfer learning mechanism, key parameters such as momentum adjustment coefficient, optimal excess air coefficient, and fuel calorific value benchmark are fine-tuned and calibrated online, enabling the model to dynamically adapt to changes in hydrogen blending ratio and operating condition drift such as equipment aging, thus overcoming the problem of rapid failure of traditional fixed parameter models under hydrogen blending conditions.

[0017] 2. A multi-level, multi-dimensional impact factor evaluation system was constructed. Through the synergistic effect of four modules—fuel characteristics, combustion conditions, dynamic stability (pressure fluctuations, flame signals), and combustion products (temperature, emissions)—the static and dynamic characteristics of hydrogen-blended combustion were deeply integrated, significantly improving the accuracy and reliability of combustion efficiency prediction.

[0018] 3. By innovatively coupling efficiency prediction results with jet momentum optimization, a complete closed-loop control chain from state perception and efficiency prediction to parameter optimization has been established. It can dynamically adjust the mixing momentum of fuel and air based on real-time predicted efficiency, realize online real-time optimization of the combustion process, and effectively improve the overall energy efficiency of the system.

[0019] 4. All core coefficients are normalized and limited, and the output range is stable between 0 and 1. The model structure is clear and the computational efficiency is high. This ensures the physical meaning and comparability of the prediction results, and also meets the real-time requirements of online industrial applications. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0021] Figure 1 The flowchart illustrates a method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning, as provided in this invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0023] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0024] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning. The method includes:

[0025] Based on the hydrogen blending ratio and the calorific value of the mixed fuel, the fuel characteristic coefficient is obtained through a fuel characteristic model.

[0026] The fuel characteristic coefficient can be understood as a quantitative index used to characterize the combustion characteristics of hydrogen-blended natural gas fuel. It can be achieved by analyzing the relationship between the hydrogen blending ratio and the calorific value of a baseline fuel, for example, by establishing a fuel characteristic model through experimental data fitting, or by simulating the combustion reaction process under different hydrogen blending ratios to extract fuel characteristic parameters.

[0027] Based on the excess air coefficient, air preheating temperature, and burner pressure drop, the combustion condition coefficient is obtained through a combustion condition model.

[0028] Fuel characteristic models can also be extended by incorporating other fuel physicochemical properties, such as density and viscosity, thereby improving their adaptability to dynamic changes in fuel.

[0029] The combustion condition coefficient is a comprehensive index reflecting the operating conditions of the combustion environment. It can be obtained by monitoring key parameters of the burner's operating status, such as collecting the ratio of airflow to fuel flow at the burner inlet, and combining this with real-time measurements of air preheating temperature and internal pressure drop data to construct a multivariate regression model for calculation. As a preferred implementation method, external environmental variables, such as atmospheric pressure or humidity, can also be incorporated into the acquisition of the combustion condition coefficient to enhance the model's adaptability to complex operating conditions.

[0030] Under the current fuel characteristic group coefficient and combustion condition group coefficient, the matching degree coefficient is obtained through the matching model based on combustion pressure fluctuation and core flame signal intensity;

[0031] The matching degree coefficient characterizes the degree of coordination between pressure fluctuations and flame signal intensity during combustion. It is obtained by analyzing the time-domain signal collected by the pressure sensor within the combustion chamber and processing it in conjunction with the light intensity signal output by the flame detector. For example, correlation analysis can be performed between the spectral characteristics of the pressure fluctuation signal and the intensity distribution of the flame signal to generate a matching degree evaluation index. Furthermore, combustion noise signals can be incorporated as auxiliary inputs into the calculation of the matching degree coefficient to further improve the accuracy of combustion stability assessment.

[0032] Combustion product coefficients are obtained through a product model based on combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration.

[0033] The combustion product coefficient is an indicator used to quantify the completeness and thermodynamic perfection of combustion. It can be obtained by analyzing the temperature distribution and chemical composition of combustion products, for example, by calculating the ratio of oxygen to carbon dioxide concentration in the flue gas based on the deviation between the combustion chamber outlet temperature and the theoretical combustion temperature. Specifically, the combustion product coefficient can also be obtained by combining historical combustion efficiency data and using machine learning algorithms for online correction to improve prediction accuracy.

[0034] Based on the matching degree coefficient and combustion product group coefficient, the predicted combustion efficiency of hydrogen-blended natural gas is calculated by the prediction model.

[0035] Based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate, the target fuel and air jet flow rate is calculated through an optimized model.

[0036] First, based on the hydrogen blending ratio and the calorific value of the blended fuel, a fuel characteristic coefficient is obtained through a fuel characteristic model. This coefficient quantifies the impact of hydrogen blending on fuel energy characteristics and reactivity, thus providing a basic input for subsequent analysis. Further, based on the excess air coefficient, air preheating temperature, and burner pressure drop, a combustion condition coefficient is obtained through a combustion condition model. This coefficient integrates key operating parameters of the combustion environment, ensuring the model's applicability under different operating conditions. Building upon the current fuel characteristic group coefficients and combustion condition group coefficients, and combining combustion pressure fluctuations and core flame signal intensity, a matching degree coefficient is derived through a matching model. This coefficient integrates dynamic pressure and flame signal characteristics during combustion, enhancing the real-time perception of the combustion state. Specifically, based on the combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration, a combustion product coefficient is obtained through a product model. This coefficient indirectly assesses combustion completeness by analyzing product temperature and emission concentration. Therefore, based on the matching degree coefficient and combustion product group coefficients, a prediction model calculates the predicted combustion efficiency of hydrogen-blended natural gas, integrating combustion stability and product information to output a comprehensive prediction result. Finally, based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rates, the target fuel and air jet flow rates are calculated through an optimized model. This transforms the prediction results into control commands, guiding the system to dynamically adjust its operating parameters. These technical features work together to form a coherent processing flow, jointly addressing the multivariate and highly nonlinear challenges brought about by changes in the hydrogen blending ratio, and achieving accurate prediction and closed-loop optimization of combustion efficiency.

[0037] like Figure 1 As shown, in a preferred embodiment of the present invention, the specific steps for calculating the target fuel and air jet flow rate based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate through an optimization model are as follows:

[0038] Based on the difference between the predicted hydrogen-blended natural gas combustion efficiency and the system's highest combustion efficiency, and in conjunction with the momentum adjustment coefficient, the current fuel and air jet flow rate is positively compensated to obtain the target fuel and air jet flow rate.

[0039] The target fuel and air jet flow rate are positively correlated with the difference.

[0040] The momentum adjustment coefficient is a parameter used to regulate the magnitude of change in the momentum of the target fuel and air jet. It can be achieved through methods such as statistical analysis of historical operating data, calibration using numerical simulation results, or setting based on expert experience, with the aim of ensuring the smoothness and accuracy of the momentum adjustment process. The system's maximum combustion efficiency can be understood as the theoretically optimal combustion efficiency value achievable under current equipment and operating conditions. It can be obtained through methods such as design parameter calculation, extraction of historical best operating data, or experimental calibration, with the aim of providing a clear optimization target benchmark. The predicted combustion efficiency of hydrogen-blended natural gas refers to the combustion efficiency value predicted based on current operating conditions and model predictions. It can be achieved through machine learning model prediction, physical model calculation, or hybrid modeling techniques, with the aim of reflecting the current combustion state in real time.

[0041] The optimized model in this example is: ,in, The target fuel and air jet flow rate, Given the current fuel and air jet flow rate, This is the momentum adjustment factor. For the system's highest combustion efficiency, To predict the combustion efficiency of hydrogen-blended natural gas.

[0042] This scheme achieves precise calculation of target momentum by constructing a dynamic adjustment formula to address the real-time adaptability issue of combustion optimization under hydrogen blending conditions. The core lies in using the difference between the predicted combustion efficiency and the system's highest combustion efficiency as the adjustment basis. This difference directly quantifies the optimization space between the current operating state and the optimal state, thus guiding the direction of momentum adjustment. The adjustment mechanism based on the product of the current momentum and the difference ensures that the deviation between the target momentum and the actual efficiency changes proportionally, avoiding over-adjustment or under-adjustment problems caused by fixed parameters. The initial value of the momentum adjustment coefficient is determined through historical data or simulation, providing a reliable starting point for the calculation. The transfer learning online fine-tuning mechanism dynamically updates the coefficient according to the actual efficiency change trend. This is particularly effective against abrupt changes in combustion characteristics caused by fluctuations in the hydrogen blending ratio, enabling the optimization model to continuously absorb new operating condition knowledge. The system's highest combustion efficiency serves as a benchmark reference, anchoring the optimization target, while the real-time input of the predicted combustion efficiency ensures that the adjustment is closely related to the current combustion state. The combination of these two ensures that the momentum adjustment is both geared towards the optimal target and aligned with actual operation.

[0043] Building upon this foundation, the aforementioned scheme deeply integrates transfer learning into the momentum optimization process, enabling the target momentum calculation to possess self-learning capabilities and effectively enhancing its adaptability to the dynamic characteristics of hydrogen-blended natural gas combustion. This approach not only solves the problem of dynamic fluctuations in the combustion characteristics of hydrogen-blended natural gas due to variations in the hydrogen ratio, but also overcomes the shortcomings of existing optimization methods in lacking real-time response to actual efficiency changes, significantly improving the accuracy of combustion efficiency prediction and the timeliness of control.

[0044] like Figure 1 As shown, in a preferred embodiment of the present invention, the specific steps for calculating the predicted combustion efficiency of hydrogen-blended natural gas based on the matching degree coefficient and the combustion product group coefficient using a prediction model are as follows:

[0045] The system's lowest combustion efficiency is added to an adjustment factor to obtain the predicted hydrogen-blended natural gas combustion efficiency;

[0046] The adjustment amount is determined by the difference between the highest and lowest combustion efficiency of the system, together with the combustion product coefficient and the matching degree coefficient.

[0047] The predicted combustion efficiency of hydrogen-blended natural gas is positively correlated with both the combustion product coefficient and the matching degree coefficient.

[0048] Predicted combustion efficiency of hydrogen-blended natural gas refers to an indicator used to quantitatively evaluate the energy conversion effect of the combustion process through mathematical modeling. It can be achieved by fusing a normalization framework with dynamic parameters, aiming to ensure that the predicted value remains within a physically feasible range while providing a standardized efficiency scale to support cross-condition comparisons and optimization decisions. The system's minimum combustion efficiency can be understood as the lower limit of energy conversion under the most unfavorable operating conditions. It can be obtained through statistical analysis of historical operating data and dynamically updated using transfer learning techniques to adapt to efficiency drift caused by equipment aging or changes in fuel characteristics. The system's maximum combustion efficiency refers to the optimal energy conversion performance of the equipment under ideal operating conditions. It can be determined based on design specifications or historical best records and maintained in real-time accuracy through online calibration, aiming to eliminate prediction bias caused by changes in equipment condition. The combustion product coefficient is a key parameter characterizing the completeness and thermodynamic perfection of the combustion process. It can be generated through multi-dimensional combustion product characteristic analysis, aiming to quantify the effectiveness of converting chemical energy into thermal energy. The combustion stability matching coefficient can be understood as a quantitative indicator of the dynamic adaptability between fuel characteristics and combustion conditions. It can be obtained through comprehensive analysis of dynamic characteristics such as pressure fluctuations and flame signals, with the aim of capturing the fluctuations in combustion characteristics caused by changes in the hydrogen blending ratio.

[0049] The prediction model in this example is: ,in, To predict the combustion efficiency of hydrogen-blended natural gas, the predicted combustion efficiency output range is 0-1. A higher predicted combustion efficiency value indicates higher combustion efficiency. The system's lowest combustion efficiency is determined based on the lowest efficiency value from historical data and updated through transfer learning. To achieve the system's highest combustion efficiency, based on design values ​​or historical best values, and calibrated online, The combustion product coefficient, This is the combustion stability matching coefficient.

[0050] This predictive model ensures the engineering applicability of the results by limiting the predicted efficiency to the minimum and maximum efficiency boundaries of the actual system operation, thus avoiding predicted values ​​exceeding the physically feasible range. The design of a 0-1 range for the predicted hydrogen-blended natural gas combustion efficiency output provides a standardized scale for efficiency values, facilitating comparison and optimization decisions under different operating conditions. The system's minimum combustion efficiency is based on the lowest efficiency value from historical data and updated through transfer learning. This allows the model to dynamically adapt to equipment aging or changes in fuel characteristics, continuously calibrating the lower efficiency limit and enhancing the model's robustness in long-term operation. The system's maximum combustion efficiency is based on the design value or historical best value and calibrated online, ensuring that the upper efficiency limit always reflects the current optimal performance of the system and effectively eliminating prediction bias caused by equipment state drift. The combustion product coefficient, as a comprehensive characterization of the completeness and thermodynamic perfection of the combustion process, is multiplied by the combustion stability matching coefficient, allowing the predicted efficiency to simultaneously integrate the dual dimensions of combustion product quality (such as heat loss and emission characteristics) and combustion process stability (such as pressure fluctuations and flame uniformity). In particular, the predicted combustion efficiency of hydrogen-blended natural gas is generated based on the combustion product coefficient and the combustion stability matching coefficient. This is because the former directly quantifies the effectiveness of converting chemical energy into thermal energy, while the latter characterizes the dynamic matching state between fuel and operating conditions. The combination of the two can accurately capture the fluctuations in combustion characteristics caused by changes in the hydrogen blending ratio, overcoming the shortcomings of traditional single-parameter models that cannot adapt to the dynamic characteristics of hydrogen-blended operating conditions, thereby significantly improving the accuracy and adaptability of the prediction.

[0051] The above technical solution enables precise quantification of the combustion efficiency of hydrogen-blended natural gas, laying a reliable foundation for optimized control. The model not only dynamically adapts to equipment aging or changes in fuel characteristics but also effectively eliminates prediction biases caused by equipment state drift, significantly improving prediction accuracy and real-time performance, and providing a reliable basis for combustion efficiency optimization.

[0052] like Figure 1 As shown, in a preferred embodiment of the present invention, the specific steps for obtaining the combustion product coefficients based on the combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration using a product model are as follows:

[0053] The ratio of the difference between the current combustion chamber temperature and the minimum allowable combustion chamber temperature is compared with the difference between the optimal combustion chamber temperature and the minimum allowable combustion chamber temperature. The min and max functions are used to limit the range, with the upper limit set to 1 and the lower limit set to 0, to obtain the combustion chamber temperature index.

[0054] The combustion chamber temperature index is the result of comparing the difference between the current combustion chamber temperature and the minimum permissible combustion chamber temperature with the difference between the optimal combustion chamber temperature and the minimum permissible combustion chamber temperature. Its purpose is to assess the impact of temperature changes on the combustion process using a relative scale. In practical applications, real-time data can be collected using temperature sensors and calculated in conjunction with preset temperature thresholds. The combustion chamber temperature index reflects the rate and intensity of the conversion of fuel chemical energy into thermal energy. A higher combustion chamber temperature indicates that the fuel burns more rapidly and concentratedly in the core reaction zone, and the released heat is not diluted or carried away by excessive excess air or cooling surfaces.

[0055] The difference between the current exhaust temperature and the theoretical minimum exhaust temperature (which can be the ambient temperature) is compared with the difference between the maximum allowable exhaust temperature and the theoretical minimum exhaust temperature. The difference is then limited by the min and max functions, with the upper limit set to 1 and the lower limit set to 0, to obtain the exhaust temperature index.

[0056] The flue gas temperature index can be understood as the ratio of the difference between the current flue gas temperature and the theoretical minimum flue gas temperature to the difference between the allowable maximum flue gas temperature and the theoretical minimum flue gas temperature. Its purpose is to quantify flue gas heat loss. In practice, real-time flue gas temperature can be obtained through temperature detection devices installed in the flue gas ducts, and normalized by combining this with ambient temperature and system design parameters. The flue gas temperature index reflects the degree to which the heat generated by combustion is effectively utilized by the system. Its core is to quantify flue gas heat loss; the higher the flue gas temperature, the more heat is carried away by the flue gas and released into the environment, resulting in lower system energy utilization efficiency. The theoretical minimum flue gas temperature can be obtained either directly from the ambient temperature or by adding the ambient temperature to the minimum heat transfer temperature difference. The minimum heat transfer temperature difference is a value determined based on the heat exchanger design and performance, typically between 20-50 degrees Celsius, and can be obtained from equipment manuals or design drawings.

[0057] The carbon monoxide concentration in the current flue gas is compared with the maximum allowable carbon monoxide concentration (which can be determined by environmental standards). The carbon monoxide concentration index is obtained by using a min function with an upper limit of 1.

[0058] The carbon monoxide (CO) concentration index is the result of comparing the current CO concentration in flue gas with the maximum permissible CO concentration. Its purpose is to assess the loss due to incomplete combustion. This index can be calculated by detecting flue gas components using a gas analyzer and combining this with concentration limits in environmental standards. The CO concentration index reflects whether the fuel has been fully and completely oxidized; a higher CO concentration in the flue gas indicates that more fuel has not been completely burned to produce carbon dioxide. Its core function is to quantify the heat loss due to incomplete combustion.

[0059] The combustion chamber temperature index, exhaust gas temperature index, and carbon monoxide concentration index are comprehensively processed to obtain the combustion product coefficient; wherein, the combustion product coefficient is positively correlated with the combustion chamber temperature index and negatively correlated with the exhaust gas temperature index and the carbon monoxide concentration index.

[0060] The above-mentioned scheme achieves precise quantification of combustion product coefficients in hydrogen-blended natural gas combustion by constructing a product model that integrates multi-source information. First, by standardizing three dynamic parameters—combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration—corresponding normalized indices are generated. These indices not only independently reflect the influence of their respective parameters on the combustion process but also synthesize combustion product coefficients using specific formulas, thus overcoming the insufficient adaptability of traditional methods that rely on only a single parameter.

[0061] Specifically, this involves importing the combustion chamber temperature index, exhaust gas temperature index, and carbon monoxide concentration index into the formula. Obtain the combustion product coefficient The combustion product coefficient ranges from 0 to 1. It is used to quantitatively characterize the "completeness" and "thermodynamic perfection" of the combustion process. A larger combustion product coefficient indicates more complete combustion and more efficient heat release and utilization. The combustion chamber temperature index. The exhaust gas temperature index, This is the carbon monoxide concentration index.

[0062] The above technical solution not only achieves effective fusion of multi-source information during the combustion of hydrogen-blended natural gas, but also establishes a method for calculating combustion product coefficients that can adapt to changes in the hydrogen ratio. This method significantly improves prediction accuracy while maintaining computational efficiency, providing reliable technical support for online applications.

[0063] like Figure 1 As shown, in a preferred embodiment of the present invention, the specific steps for deriving the matching degree coefficient based on combustion pressure fluctuation and core flame signal intensity using a matching model under the current fuel characteristic group coefficient and combustion condition group coefficient are as follows:

[0064] The current root mean square value of combustion pressure fluctuation is compared with the maximum allowable pressure fluctuation value of the system, and then the upper limit of the amplitude is limited to 1 using the min function to obtain the combustion pressure fluctuation index.

[0065] The combustion pressure fluctuation index can be understood as a quantitative indicator measuring the relative severity of pressure fluctuations during combustion. It is calculated by statistically analyzing real-time collected pressure data to obtain the root mean square value, and then combining this with the system's maximum allowable pressure fluctuation value. The purpose of introducing this index is to provide a standardized evaluation method, making combustion pressure fluctuations under different hydrogen blending ratios comparable. The combustion pressure fluctuation index reflects the intensity and stability of the combustion flame; the smaller the index, the smaller the pressure fluctuation and the more ideal the combustion state. The system's maximum allowable pressure fluctuation value is determined by combustion stability experiments and updated online.

[0066] The absolute value of the difference between the current core flame signal intensity and the optimal flame signal intensity is compared with the allowable deviation range, and then the upper limit of the amplitude is limited to 1 using the min function to obtain the core flame signal intensity index.

[0067] The core flame signal intensity index is a quantitative parameter reflecting the degree of deviation from flame stability. It is calculated by real-time monitoring of the flame signal intensity and comparing it with the optimal value. This index is designed to capture dynamic changes in flame state, especially adapting to the rapid fluctuations in flame characteristics during hydrogen-blended natural gas combustion. The core flame signal intensity index reflects the degree of pressure oscillation during combustion and is a direct indicator of combustion stability. The smaller the core flame signal intensity index, the more uniform the heat distribution during combustion, and the better the flame stability. The optimal flame signal intensity is determined experimentally under standard pure natural gas conditions and different load points, using the average signal intensity measured by the flame detector at the point of highest combustion efficiency and lowest CO emissions as the initial optimal value under that load. The allowable deviation range is typically the larger of the difference between the optimal flame signal intensity and the historical minimum flame signal intensity, or the difference between the historical maximum flame signal intensity and the optimal flame signal intensity.

[0068] The product of the fuel characteristic coefficient and the combustion condition coefficient is processed, and the product and the minimum constant are limited to the lower limit of the minimum constant by the min function to obtain the theoretical expected value.

[0069] The theoretical expected value refers to the theoretical prediction of the combustion stability level after comprehensively considering fuel characteristics and operating conditions. It can be achieved through various mathematical models, such as using machine learning algorithms to establish a nonlinear mapping relationship between fuel characteristics and operating parameters, or using physical models to simulate the kinetic characteristics of the combustion process. The theoretical expected value refers to the level of combustion stability that should theoretically be achieved under current fuel characteristics and operating conditions. The larger the theoretical expected value, the more stable the expected combustion. The product and the minimum constant are limited to a minimum constant using a min function to avoid a theoretical expected value of 0.

[0070] The actual stability performance value determined based on the combustion pressure fluctuation index and the core flame signal intensity index is compared with the theoretical expected value to obtain the matching coefficient; wherein, the matching coefficient characterizes the degree of agreement between the actual stability performance and the theoretical expected value.

[0071] This scheme achieves precise quantification of the combustion stability of hydrogen-blended natural gas by constructing a matching coefficient calculation mechanism. First, combustion stability is characterized from two dimensions: pressure characteristics and flame characteristics, using the combustion pressure fluctuation index and the core flame signal intensity index, respectively. These two indices are independent yet complementary, together forming a complete description of the combustion state. Second, theoretical expected values ​​provide a benchmark reference, enabling the matching coefficient to accurately reflect the deviation between the actual combustion state and the ideal state. Finally, by dynamically integrating pressure fluctuations, flame signal intensity, and theoretical expected values, a complete matching evaluation system is formed, effectively solving the problem of combustion characteristic fluctuations caused by changes in the hydrogen blending ratio.

[0072] Specifically, the product of 1 and the difference between the combustion pressure fluctuation index and 1 and the difference between the core flame signal intensity index is processed. The product result is then compared with the theoretical expected value, and a min function is used to limit the amplitude to 1 to obtain the matching coefficient. The matching degree coefficient output ranges from 0 to 1. The larger the matching degree coefficient, the more complete and stable the combustion.

[0073] The above technical solution enables real-time and accurate assessment of the combustion stability of hydrogen-blended natural gas, improving the adaptability and accuracy of the combustion efficiency prediction model. In particular, this solution can promptly capture changes in combustion characteristics when the hydrogen blending ratio changes, ensuring the reliability of the prediction model.

[0074] like Figure 1 As shown, in a preferred embodiment of the present invention, the specific steps for obtaining the fuel characteristic coefficient based on the hydrogen blending ratio and the calorific value of the mixed fuel through a fuel characteristic model are as follows:

[0075] The hydrogen blending ratio index is obtained by comparing the current hydrogen blending ratio with the maximum allowable hydrogen blending ratio of the system.

[0076] The hydrogen blending ratio index is the normalized ratio between the current hydrogen blending ratio and the maximum allowable hydrogen blending ratio of the system. It can be achieved by real-time monitoring of hydrogen flow rate combined with dynamic adjustments to the equipment's safety boundaries. Its purpose is to accurately reflect the impact of hydrogen content on combustion stability. The maximum allowable hydrogen blending ratio of the system is determined initially based on the burner manufacturer's safety design specifications, material hydrogen embrittlement resistance test reports, or combustion stability experiments (by gradually increasing the hydrogen ratio until flashover or flameout occurs).

[0077] The calorific value of the current blended fuel is compared with the lower heating value of the benchmark natural gas to obtain the blended fuel calorific value index.

[0078] The blended fuel calorific value index is the normalized ratio between the current calorific value of the blended fuel and the lower heating value (LHC) of the benchmark natural gas. It can be achieved by dynamically calibrating the benchmark value using real-time fuel composition data and a transfer learning algorithm. The aim is to address the nonlinear fluctuations in calorific value caused by hydrogen blending. The LHC of the benchmark natural gas is taken as the design value for pure natural gas and calibrated online using transfer learning.

[0079] The hydrogen blending ratio index and the calorific value index of the mixed fuel are comprehensively processed to obtain the fuel characteristic coefficient; wherein the fuel characteristic coefficient is positively correlated with both the hydrogen blending ratio index and the calorific value index of the mixed fuel.

[0080] The fuel characteristic coefficient is a dimensionless index obtained by comprehensively processing the hydrogen blending ratio index and the calorific value index of the blended fuel. It can be implemented by combining a mathematical operation module with a logical judgment unit. Its purpose is to comprehensively characterize the quality or potential of fuel in terms of efficient and stable combustion.

[0081] The comprehensive processing can be achieved by multiplying the hydrogen blending ratio index and the calorific value index of the mixed fuel, and then successively applying the min and max functions with an upper limit of 1 and a lower limit of 0 to obtain the fuel characteristic coefficient. ;

[0082] First, the hydrogen blending ratio is calculated by comparing the current hydrogen blending ratio with the maximum allowable hydrogen blending ratio of the system to obtain the hydrogen blending ratio index. This normalization method based on safety boundaries can accurately reflect the potential impact of hydrogen content on combustion stability. The maximum allowable hydrogen blending ratio of the system is updated online through transfer learning, ensuring that the model can integrate historical safety data in real time and avoiding the risk of failure of fixed thresholds when equipment ages or operating conditions change abruptly. Second, the calorific value of the blended fuel is calculated by comparing the current calorific value of the mixed fuel with the lower heating value of the benchmark natural gas to obtain the calorific value index of the mixed fuel. The benchmark calorific value is calibrated online through transfer learning, solving the problem of nonlinear fluctuations in calorific value caused by hydrogen blending, and ensuring that the calorific value comparison is always anchored to the real-time calibrated benchmark. Next, the hydrogen blending ratio index and the calorific value index of the mixed fuel are multiplied and the amplitude is limited to obtain the fuel characteristic coefficient. This dual-dimensional multiplication mechanism retains the positive contribution of the hydrogen ratio to combustion activity and integrates the constraint effect of calorific value on energy release. At the same time, the amplitude limiting operation of the min and max functions forces the coefficient to be within the range of 0-1 to prevent numerical overflow under abnormal operating conditions. Finally, the fuel characteristic coefficient is defined as the quality or potential of fuel in terms of efficient and stable combustion. Its increase indicates that the fuel is evolving towards high activity and high stability, providing accurate input basis for subsequent matching models and fundamentally improving the model's adaptability to hydrogen-blended conditions.

[0083] The above technical solution solves the core problem of insufficient adaptability of fuel characteristics under changes in hydrogen ratio, significantly improves the accuracy of fuel characteristic coefficient calculation, and thus enhances the robustness of combustion efficiency prediction and online control capability.

[0084] like Figure 1 As shown, in a preferred embodiment of the present invention, the specific steps for obtaining the combustion condition coefficient based on the excess air coefficient, air preheating temperature, and burner pressure drop through the combustion condition model are as follows:

[0085] The excess air coefficient index is obtained by comparing the absolute value of the difference between the current excess air coefficient and the optimal excess air coefficient with the allowable fluctuation range of the excess air coefficient, and then using the min function to limit the upper limit to 1.

[0086] The excess air coefficient index is a quantitative indicator that measures the accuracy of the air-fuel ratio. It can be calculated by the ratio of the absolute value of the difference between the current excess air coefficient and the optimal excess air coefficient to the allowable fluctuation range. Its purpose is to capture the degree to which the air-fuel ratio deviates from the ideal state. The higher the air-fuel ratio, the closer it is to the optimal combustion range, and the more complete the chemical energy release. The optimal excess air coefficient is obtained through pre-training under pure natural gas conditions and is adaptively fine-tuned online through transfer learning to adapt to the impact of changes in the hydrogen blending ratio on the optimal point. The allowable fluctuation range of the excess air coefficient is determined based on historical data statistics and updated online. Alternatively, the initial range can be the design value. During online updates, the absolute deviation of the excess air coefficient from the optimal value in the most recent M samples is calculated, and the 90th percentile of this sequence is taken as the new allowable fluctuation range.

[0087] The difference between the current air preheating temperature and the minimum allowable air preheating temperature of the system is compared with the difference between the maximum allowable air preheating temperature and the minimum allowable air preheating temperature of the system. The difference is then limited by the min and max functions, with the upper limit set to 1 and the lower limit set to 0, to obtain the air preheating temperature index.

[0088] The air preheating temperature index is a quantitative indicator reflecting the sensible heat carried by the air entering the furnace. It can be calculated as the ratio of the current air preheating temperature to the system's allowable temperature range. The purpose is to standardize the contribution of preheating temperature to combustion. A higher air preheating temperature index indicates a higher preheating temperature, which in turn increases flame temperature and reduces exhaust heat loss. The system's allowable temperature range is determined by the heat exchanger design and material temperature limits, ensuring that the index dynamically responds to changes in calorific value after hydrogen doping within safe boundaries.

[0089] The absolute value of the difference between the current burner pressure drop and the optimal burner pressure drop is compared with the allowable fluctuation range of the pressure drop. Then, the upper limit of the amplitude is limited to 1 using the min function to obtain the burner pressure drop index.

[0090] The burner pressure drop index is a quantitative indicator of fuel-air mixture intensity. It can be calculated as the ratio of the absolute value of the difference between the current pressure drop and the design optimum pressure drop to the allowable fluctuation range. Its purpose is to assess the degree to which the mixture intensity deviates from the ideal state. A smaller burner pressure drop index indicates more thorough turbulent mixing. The design optimum burner pressure drop is provided by the burner manufacturer. The allowable fluctuation range of the pressure drop is adjusted online, taking into account the changes in turbulent mixing requirements caused by the high flame propagation velocity of hydrogen.

[0091] The excess air coefficient index, air preheating temperature index, and burner pressure drop index are comprehensively processed to obtain the combustion condition coefficient; wherein, the combustion condition coefficient is positively correlated with the air preheating temperature index and negatively correlated with the excess air coefficient index and the burner pressure drop index.

[0092] The comprehensive processing specifically involves: incorporating the excess air coefficient index, air preheating temperature index, and burner pressure drop index into the formula. Obtain the combustion condition coefficient The combustion condition coefficient ranges from 0 to 1. It reflects the degree to which current operating conditions support efficient combustion. A higher combustion condition coefficient indicates a more precise air-fuel ratio, more complete air preheating, and a more intense and uniform mixing, resulting in a more ideal high-temperature stable flame. The excess air coefficient index. The air preheating temperature index. This refers to the burner pressure drop index.

[0093] By constructing a dynamically adjustable combustion condition model, the problem of operating parameter deviations caused by changes in the hydrogen ratio in hydrogen-blended natural gas combustion was solved. When obtaining the excess air coefficient index, the ratio of the absolute value of the difference between the current excess air coefficient and the optimal excess air coefficient to the allowable fluctuation range is processed and limited, effectively capturing the degree to which the air-fuel ratio deviates from the ideal state. The optimal excess air coefficient is fine-tuned online through transfer learning, enabling the calculation to adapt to the deviation of the optimal point caused by changes in the hydrogen blending ratio, thus accurately reflecting the completeness of chemical energy release. When obtaining the air preheating temperature index, the ratio of the current air preheating temperature to the system's allowable temperature range is processed and limited, standardizing the contribution of preheating temperature to combustion. The system's allowable temperature range is determined by the heat exchanger design and material temperature resistance limits, ensuring that the index dynamically responds to the impact of changes in calorific value after hydrogen blending within safe boundaries. When obtaining the burner pressure drop index, the ratio of the absolute value of the difference between the current pressure drop and the design optimal pressure drop to the allowable fluctuation range is processed and limited, enabling the assessment of the degree to which the mixing intensity deviates from the ideal state. The pressure drop fluctuation range is adjustable online, taking into account the changes in turbulent mixing requirements caused by the high flame propagation velocity of hydrogen, enabling the index to dynamically adapt to changes in mixing characteristics under hydrogen-blended conditions. When calculating the combustion condition coefficient, three indices are combined using a specific formula. The excess air coefficient index and burner pressure drop index participate in the form of (1-index), while the air preheating temperature index participates directly, reflecting the comprehensive impact of air-fuel ratio accuracy, mixing adequacy, and preheating temperature on combustion support. The formula design ensures that the combustion condition coefficient intuitively represents the operating condition quality within the 0-1 range, providing reliable input for subsequent predictions. Simultaneously, the inverse correlation accurately characterizes the negative impact of parameter deviations on combustion efficiency.

[0094] Through the above technical solution, the combustion condition coefficient can accurately quantify the degree to which operating conditions support efficient combustion, thus providing a reliable basis for the matching coefficient and combustion efficiency prediction. This method not only solves the problem of dynamic shift in operating conditions caused by changes in the hydrogen blending ratio, which traditional methods cannot handle, but also achieves accurate modeling of the hydrogen-blended natural gas combustion process by dynamically adjusting key parameters, significantly improving the accuracy of combustion efficiency prediction and online application capabilities.

[0095] In another embodiment, this application also discloses a hydrogen-blended natural gas combustion efficiency prediction system based on transfer learning, comprising:

[0096] The fuel characteristic evaluation module obtains fuel characteristic coefficients based on the hydrogen blending ratio and the calorific value of the blended fuel through a fuel characteristic model.

[0097] The combustion condition assessment module obtains combustion condition coefficients based on the excess air coefficient, air preheating temperature, and burner pressure drop through a combustion condition model.

[0098] The matching degree evaluation module, under the current fuel characteristic group coefficient and combustion condition group coefficient, derives the matching degree coefficient based on combustion pressure fluctuation and core flame signal intensity through the matching model;

[0099] The product evaluation module obtains combustion product coefficients based on combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration through a product model.

[0100] The prediction calculation module, based on the matching degree coefficient and combustion product group coefficient, calculates the predicted combustion efficiency of hydrogen-blended natural gas through a prediction model.

[0101] The optimization module calculates the target fuel and air jet flow rate based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate through an optimization model.

[0102] The core innovation of this embodiment lies in combining the fuel characteristic assessment module and the combustion condition assessment module with a modular architecture, and introducing a matching degree assessment module to integrate dynamic characteristics such as combustion pressure fluctuations and core flame signal intensity. This solves the problem of poor adaptability to changes in hydrogen blending ratio in existing technologies, while enhancing the real-time perception capability of combustion status. Furthermore, by generating control commands through an optimization module, the lack of real-time optimization capabilities in existing technologies is compensated for, forming a complete chain from parameter perception to optimization execution, achieving accurate prediction and closed-loop optimization.

[0103] In practical applications, the fuel characteristic coefficient can be understood as a quantitative indicator used to characterize the combustion characteristics of hydrogen-blended natural gas fuel. It can be achieved by analyzing the relationship between the hydrogen blending ratio and the calorific value of a baseline fuel, for example, by establishing a fuel characteristic model through experimental data fitting, or by simulating the combustion reaction process under different hydrogen blending ratios to extract fuel characteristic parameters. Furthermore, the fuel characteristic model can be extended by incorporating other fuel physicochemical properties, such as density and viscosity, thereby improving its adaptability to dynamic changes in the fuel.

[0104] The combustion condition coefficient is a comprehensive index reflecting the operating conditions of the combustion environment. It can be obtained by monitoring key parameters of the burner's operating status, such as collecting the ratio of airflow to fuel flow at the burner inlet, and combining this with real-time measurements of air preheating temperature and internal pressure drop data to construct a multivariate regression model for calculation. As a preferred implementation method, external environmental variables, such as atmospheric pressure or humidity, can also be incorporated into the acquisition of the combustion condition coefficient to enhance the model's adaptability to complex operating conditions.

[0105] The matching degree coefficient characterizes the degree of coordination between pressure fluctuations and flame signal intensity during combustion. It is obtained by analyzing the time-domain signal collected by the pressure sensor within the combustion chamber and processing it in conjunction with the light intensity signal output by the flame detector. For example, correlation analysis can be performed between the spectral characteristics of the pressure fluctuation signal and the intensity distribution of the flame signal to generate a matching degree evaluation index. Furthermore, combustion noise signals can be incorporated as auxiliary inputs into the calculation of the matching degree coefficient to further improve the accuracy of combustion stability assessment.

[0106] The combustion product coefficient is an indicator used to quantify the completeness and thermodynamic perfection of combustion. It can be obtained by analyzing the temperature distribution and chemical composition of combustion products, for example, by calculating the ratio of oxygen to carbon dioxide concentration in the flue gas based on the deviation between the combustion chamber outlet temperature and the theoretical combustion temperature. Specifically, the combustion product coefficient can also be obtained by combining historical combustion efficiency data and using machine learning algorithms for online correction to improve prediction accuracy.

[0107] The innovation of this application lies in constructing a multi-level model processing framework, realizing a complete prediction and control chain from fuel input to efficiency optimization. Compared with existing technologies, this embodiment not only solves the problem of insufficient adaptability of traditional models to dynamic changes in fuel characteristics, but also enhances the real-time perception capability of combustion status by integrating dynamic features such as combustion pressure fluctuations and flame signals. Simultaneously, by introducing optimization models to generate control commands, it compensates for the lack of real-time optimization capabilities in existing technologies, forming a coherent processing flow that can effectively cope with the multivariate and highly nonlinear challenges brought about by changes in hydrogen blending ratios, achieving accurate prediction and closed-loop optimization of combustion efficiency.

[0108] The working principle is as follows: First, based on the hydrogen blending ratio and the calorific value of the mixed fuel, a fuel characteristic coefficient is obtained through a fuel characteristic model. This coefficient quantifies the impact of hydrogen blending on fuel energy characteristics and reactivity, thus providing a basic input for subsequent analysis. Further, based on the excess air coefficient, air preheating temperature, and burner pressure drop, a combustion condition coefficient is obtained through a combustion condition model. This coefficient integrates key operating parameters of the combustion environment, ensuring the model's applicability under different operating conditions. Based on the current fuel characteristic group coefficient and combustion condition group coefficient, combined with combustion pressure fluctuations and core flame signal intensity, a matching degree coefficient is derived through a matching model. This coefficient integrates dynamic pressure and flame signal characteristics during combustion, enhancing the real-time perception of the combustion state. Specifically, based on the combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration, a combustion product coefficient is obtained through a product model. This coefficient indirectly assesses combustion completeness through the analysis of product temperature and emission concentration. Therefore, based on the matching degree coefficient and combustion product group coefficient, a prediction model calculates the predicted combustion efficiency of hydrogen-blended natural gas, integrating combustion stability and product information to output a comprehensive prediction result. Finally, based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rates, the target fuel and air jet flow rates are calculated through an optimized model. This transforms the prediction results into control commands, guiding the system to dynamically adjust its operating parameters. These technical features work together to form a coherent processing flow, jointly addressing the multivariate and highly nonlinear challenges brought about by changes in the hydrogen blending ratio, and achieving accurate prediction and closed-loop optimization of combustion efficiency.

[0109] The transfer learning and parameter update mechanism is as follows:

[0110] The initial model parameter set (such as the initial value of the optimal excess air coefficient, the initial value of the momentum adjustment coefficient, the reference calorific value, etc.) is obtained by training with historical steady-state high-efficiency operation data under pure natural gas or fixed low hydrogen blending ratio conditions.

[0111] The current online real-time operation data stream under the changing operating conditions of hydrogen-blended natural gas.

[0112] Transfer learning algorithms and processes:

[0113] Model fine-tuning is employed: This invention preferably uses a model fine-tuning method based on online gradient descent. The pre-trained source domain model is used as the initial state, and supervised learning continues on the target domain data.

[0114] Specific steps:

[0115] a. Triggering conditions: A fine-tuning is initiated every N (e.g., N=100) new running data samples accumulated, or when the deviation between the prediction efficiency and the actual inferred efficiency (e.g., through the inverse balancing method) continuously exceeds the threshold Δ.

[0116] b. Loss Function: The loss function is defined as the mean square error between the predicted and target values. For example, for fine-tuning the momentum adjustment coefficient, the loss function can be defined as: ,in, The actual efficiency is obtained through flue gas analyzer and heat balance calculation.

[0117] c. Parameter Update: Update specific parameters using gradient descent, for example: ,in The learning rate is a small positive number (e.g., 0.001). This applies only to some key parameters (e.g., ...). Fine-tuning was performed on the optimal excess air coefficient, while fixing other model structures to avoid overfitting.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning, characterized in that, The method includes: Based on the hydrogen blending ratio and the calorific value of the mixed fuel, the fuel characteristic coefficient is obtained through a fuel characteristic model. Based on the excess air coefficient, air preheating temperature, and burner pressure drop, the combustion condition coefficient is obtained through a combustion condition model. Under the current fuel characteristic group coefficient and combustion condition group coefficient, the matching degree coefficient is obtained through the matching model based on combustion pressure fluctuation and core flame signal intensity; Combustion product coefficients are obtained through a product model based on combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration. Based on the matching degree coefficient and combustion product group coefficient, the predicted combustion efficiency of hydrogen-blended natural gas is calculated by the prediction model. Based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate, the target fuel and air jet flow rate is calculated through an optimized model.

2. The method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning according to claim 1, characterized in that, The specific steps for calculating the target fuel and air jet flow rate based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate through an optimization model are as follows: Based on the difference between the predicted hydrogen-blended natural gas combustion efficiency and the system's highest combustion efficiency, and in conjunction with the momentum adjustment coefficient, the current fuel and air jet flow rate is positively compensated to obtain the target fuel and air jet flow rate. The target fuel and air jet flow rate are positively correlated with the difference.

3. The method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning according to claim 2, characterized in that, The specific steps for calculating the predicted combustion efficiency of hydrogen-blended natural gas using a prediction model based on the matching degree coefficient and combustion product group coefficient are as follows: The system's lowest combustion efficiency is added to an adjustment factor to obtain the predicted hydrogen-blended natural gas combustion efficiency; The adjustment amount is determined by the difference between the highest and lowest combustion efficiency of the system, together with the combustion product coefficient and the matching degree coefficient. The predicted combustion efficiency of hydrogen-blended natural gas is positively correlated with both the combustion product coefficient and the matching degree coefficient.

4. The method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning according to claim 3, characterized in that, The specific steps for obtaining the combustion product coefficients based on the combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration using a product model are as follows: The ratio of the difference between the current combustion chamber temperature and the minimum allowable combustion chamber temperature is compared with the difference between the optimal combustion chamber temperature and the minimum allowable combustion chamber temperature. The min and max functions are used to limit the range, with the upper limit set to 1 and the lower limit set to 0, to obtain the combustion chamber temperature index. The difference between the current exhaust temperature and the theoretical minimum exhaust temperature is compared with the difference between the allowable maximum exhaust temperature and the theoretical minimum exhaust temperature. The difference is then limited by the min and max functions, with the upper limit set to 1 and the lower limit set to 0, to obtain the exhaust temperature index. The carbon monoxide concentration in the current flue gas is compared with the maximum allowable carbon monoxide concentration. The carbon monoxide concentration index is obtained by using the min function with an upper limit of 1. The combustion chamber temperature index, exhaust gas temperature index, and carbon monoxide concentration index are comprehensively processed to obtain the combustion product coefficient; wherein the combustion product coefficient is positively correlated with the combustion chamber temperature index and negatively correlated with the exhaust gas temperature index and the carbon monoxide concentration index.

5. The method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning according to claim 3, characterized in that, The specific steps for deriving the matching degree coefficient using a matching model based on combustion pressure fluctuations and core flame signal intensity under the current fuel characteristic group coefficient and combustion condition group coefficient are as follows: The current root mean square value of combustion pressure fluctuation is compared with the maximum allowable pressure fluctuation value of the system, and then the upper limit of the amplitude is limited to 1 using the min function to obtain the combustion pressure fluctuation index. The absolute value of the difference between the current core flame signal intensity and the optimal flame signal intensity is compared with the allowable deviation range, and then the upper limit of the amplitude is limited to 1 using the min function to obtain the core flame signal intensity index. The theoretical expected value is obtained based on fuel characteristic coefficient and combustion condition coefficient; The actual stability performance value determined based on the combustion pressure fluctuation index and the core flame signal intensity index is compared with the theoretical expected value to obtain the matching coefficient; wherein, the matching coefficient characterizes the degree of agreement between the actual stability performance and the theoretical expected value.

6. The method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning according to claim 5, characterized in that, The specific steps for obtaining fuel characteristic coefficients based on hydrogen blending ratio and calorific value of mixed fuel through fuel characteristic model are as follows: The hydrogen blending ratio index is obtained by comparing the current hydrogen blending ratio with the maximum allowable hydrogen blending ratio of the system. The calorific value of the current blended fuel is compared with the lower heating value of the benchmark natural gas to obtain the blended fuel calorific value index. The hydrogen blending ratio index and the calorific value index of the mixed fuel are combined to obtain the fuel characteristic coefficient; wherein the fuel characteristic coefficient is positively correlated with both the hydrogen blending ratio index and the calorific value index of the mixed fuel.

7. The method for predicting the combustion efficiency of hydrogen-blended natural gas based on transfer learning according to claim 5, characterized in that, The specific steps for obtaining the combustion condition coefficient based on the excess air coefficient, air preheating temperature, and burner pressure drop through the combustion condition model are as follows: The excess air coefficient index is obtained by comparing the absolute value of the difference between the current excess air coefficient and the optimal excess air coefficient with the allowable fluctuation range of the excess air coefficient, and then using the min function to limit the upper limit to 1. The difference between the current air preheating temperature and the minimum allowable air preheating temperature of the system is compared with the difference between the maximum allowable air preheating temperature and the minimum allowable air preheating temperature of the system. The difference is then limited by the min and max functions, with the upper limit set to 1 and the lower limit set to 0, to obtain the air preheating temperature index. The absolute value of the difference between the current burner pressure drop and the optimal burner pressure drop is compared with the allowable fluctuation range of the pressure drop. Then, the upper limit of the amplitude is limited to 1 using the min function to obtain the burner pressure drop index. The excess air coefficient index, air preheating temperature index, and burner pressure drop index are comprehensively processed to obtain the combustion condition coefficient; wherein, the combustion condition coefficient is positively correlated with the air preheating temperature index and negatively correlated with the excess air coefficient index and the burner pressure drop index.

8. A hydrogen-blended natural gas combustion efficiency prediction system based on transfer learning, characterized in that, include: The fuel characteristic evaluation module obtains fuel characteristic coefficients based on the hydrogen blending ratio and the calorific value of the blended fuel through a fuel characteristic model. The combustion condition assessment module obtains combustion condition coefficients based on the excess air coefficient, air preheating temperature, and burner pressure drop through a combustion condition model. The matching degree evaluation module, under the current fuel characteristic group coefficient and combustion condition group coefficient, derives the matching degree coefficient based on combustion pressure fluctuation and core flame signal intensity through the matching model; The product evaluation module obtains combustion product coefficients based on combustion chamber temperature, exhaust gas temperature, and carbon monoxide concentration through a product model. The prediction calculation module, based on the matching degree coefficient and combustion product group coefficient, calculates the predicted combustion efficiency of hydrogen-blended natural gas through a prediction model. The optimization module calculates the target fuel and air jet flow rate based on the predicted combustion efficiency of hydrogen-blended natural gas and the current fuel and air jet flow rate through an optimization model.

Images