Industrial heating furnace thermal efficiency optimization method and system

Abstract

This invention relates to the field of energy-saving technology for heating furnaces, and particularly to a method and system for optimizing the thermal efficiency of industrial heating furnaces. The method includes: acquiring current operating condition data of the industrial heating furnace, including oxygen content in each furnace chamber and non-adjustable operating parameters; inputting this data into an XGBoost model to obtain predicted thermal efficiency values; fixing the non-adjustable parameters, using the oxygen content in each furnace chamber as an evolutionary variable, calling the model to determine fitness, and using a differential evolutionary algorithm to find the optimal oxygen content combination; and mapping this combination to a damper control strategy to guide on-site operation when the optimization termination condition is met. This invention predicts thermal efficiency using an XGBoost model and combines it with a differential evolutionary algorithm to globally optimize the oxygen content in each furnace chamber. This allows for the rapid acquisition of the optimal oxygen content combination that meets the expected thermal efficiency and operating procedures, and its conversion into a damper adjustment strategy. This overcomes the limitations of manual adjustment and static models, effectively improving the thermal efficiency of the heating furnace and reducing fuel consumption and heat loss.

Description

Technical Field

[0001] This invention relates to the field of energy-saving technology for heating furnaces, and in particular to a method and system for optimizing the thermal efficiency of industrial heating furnaces. Background Technology

[0002] Currently, in the field of energy-saving optimization for industrial heating furnaces, big data-based operation optimization techniques are commonly used to assist human decision-making in achieving economical and clean operation. However, existing optimization schemes based on static models or purely data-driven approaches have significant drawbacks: Firstly, purely data-driven models lack physical mechanism constraints, resulting in poor generalization ability when faced with fluctuations in fuel gas calorific value or varying operating conditions, making it difficult to accurately represent the true relationship between thermal efficiency and operating parameters. Secondly, for complex coupled scenarios such as multiple furnaces sharing a flue gas system, existing technologies often fail to effectively decouple unadjustable and adjustable operating conditions. Single adjustments can easily lead to system conflicts, and the numerical recommendations output by the algorithm are difficult to directly translate into physical control strategies for on-site air intake adjustment actuators, resulting in optimization results that cannot be accurately implemented in actual production.

[0003] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for optimizing the thermal efficiency of industrial heating furnaces.

[0005] In a first aspect, the present invention provides a method for optimizing the thermal efficiency of an industrial heating furnace, the technical solution of which is as follows:

[0006] Acquire the current operating condition data of the industrial heating furnace, which includes the current oxygen content of each furnace chamber and non-adjustable operating parameters;

[0007] The current operating condition data is input into the XGBoost model used to predict the thermal efficiency of the heating furnace to obtain the current thermal efficiency prediction value. The XGBoost model is a model trained based on steady-state time period data determined by a dynamically moving time window and combined features constructed by combining flue gas heat loss with temperature difference and nonlinear relationship with oxygen content.

[0008] Under the premise of fixing the non-adjustable operating parameters, the oxygen content of each furnace is used as the evolution variable. The XGBoost model is called to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness. Multidimensional collaborative optimization is performed through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace.

[0009] Determine whether the optimal combination meets the preset optimization termination condition. If the optimization termination condition is met, then map the optimal combination to the control strategy of the air intake volume adjustment actuator to guide on-site operation.

[0010] The beneficial effects of the industrial heating furnace thermal efficiency optimization method of the present invention are as follows:

[0011] The method of this invention predicts thermal efficiency using the XGBoost model and combines it with the differential evolution algorithm to globally optimize the oxygen content of each furnace. This allows for the rapid acquisition of the optimal oxygen content combination that meets the expected thermal efficiency and operating procedures, which is then converted into a damper adjustment strategy. This overcomes the limitations of manual adjustment and static models, effectively improving the thermal efficiency of the heating furnace and reducing fuel consumption and heat loss.

[0012] Based on the above scheme, the method for optimizing the thermal efficiency of an industrial heating furnace according to the present invention can be further improved as follows.

[0013] In one alternative approach, the training process of the XGBoost model includes:

[0014] Obtain the pre-processed historical operating data of the industrial heating furnace;

[0015] Based on the load fluctuation threshold or parameter variance, steady-state period data that reaches steady-state operating conditions are selected from the historical operating data;

[0016] Obtain temperature and oxygen content parameters related to the flue gas heat loss of the industrial heating furnace;

[0017] Based on the fact that flue gas heat loss is positively correlated with temperature difference and nonlinearly related to oxygen content, combined with the temperature parameter and the oxygen content parameter, a combination of features characterizing the magnitude of flue gas loss is determined.

[0018] The thermal efficiency corresponding to the steady-state period data is calculated based on the inverse equilibrium method and used as training labels;

[0019] The initial model is trained based on the steady-state period data, the combined features, and the training labels to obtain the XGBoost model, so that the XGBoost model includes historical fluctuation errors and dynamically matches the current operating conditions.

[0020] Among the above-mentioned optional methods, by filtering steady-state period data based on load fluctuations or parameter variance, noise interference during non-steady-state transition processes is eliminated, making the model training samples more representative of the actual thermal characteristics of the equipment. At the same time, this filtering mechanism enables the model to use data within a dynamic time window range of the most recent period that has reached steady state, thereby more accurately and dynamically matching the current actual operating conditions when applied online, and improving the robustness of the proxy model.

[0021] In one alternative approach, based on the positive correlation between flue gas heat loss and temperature difference and the non-linear relationship with oxygen content, combined with the temperature parameter and the oxygen content parameter, a combination of characteristics representing the magnitude of flue gas loss is determined, including:

[0022] The exhaust gas temperature and ambient temperature are obtained as the temperature parameters, and the oxygen volume fraction in the flue gas is obtained as the oxygen content parameter.

[0023] Calculate the temperature difference between the exhaust gas temperature and the ambient temperature;

[0024] The combined characteristics are obtained by multiplying the temperature difference with a correction term that includes the volume fraction of oxygen in the flue gas, and then multiplying by the exhaust heat loss coefficient.

[0025] The combined features are used as input features for training the XGBoost model;

[0026] The calculation structure of the correction term is as follows: ;in, The volume fraction of oxygen in the flue gas; This is the correction factor for flue gas heat loss.

[0027] Among the above-mentioned optional methods, by introducing a combination feature construction method that conforms to thermodynamic principles, the nonlinear coupling relationship between flue gas temperature difference and oxygen content is explicitly embedded into the model input. This overcomes the defect that pure data statistical correlation fails when fuel calorific value fluctuates or operating conditions deviate, and significantly enhances the model's physical interpretability and predictive generalization ability for the key indicator of flue gas heat loss.

[0028] In one alternative approach, under the premise of fixing the unadjustable operating parameters, the oxygen content of each furnace is used as the evolutionary variable. The XGBoost model is called to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness. A multi-dimensional collaborative optimization is performed using a differential evolution algorithm to obtain the optimal combination of oxygen contents for each furnace, including:

[0029] The current furnace load, fuel flow rate, and flue gas temperature are used as the parameters of the non-adjustable operating conditions.

[0030] The oxygen content of each furnace is used as the evolutionary variable, and the population is initialized within a preset range of oxygen content values. Each individual in the population is composed of the oxygen content values ​​of each furnace.

[0031] The following population evolution steps are executed iteratively until the preset termination condition of the differential evolution algorithm is met: In each generation of population evolution, for each individual in the population, its oxygen content value is concatenated with the unadjustable operating parameters to form an input feature vector, which is then input into the XGBoost model to obtain the predicted thermal efficiency as the fitness of that individual; based on the fitness, mutation, crossover, and selection operations are performed to generate the next generation of population.

[0032] When the termination condition of the differential evolution algorithm is met, the optimal combination is output.

[0033] Among the above-mentioned optional methods, by fixing the non-adjustable operating parameters such as the current load and fuel flow rate, and only performing multi-dimensional collaborative optimization on the oxygen content of each furnace, the external disturbances and internal control variables are effectively decoupled. This avoids deviations in the optimization direction caused by operating condition fluctuations in coupled scenarios such as multiple furnaces sharing a flue gas system, and ensures the feasibility and safety of the optimization results under the current actual production conditions.

[0034] In one alternative approach, performing mutation, crossover, and selection operations based on the fitness includes:

[0035] Based on the fitness of each individual, the individual with the highest fitness in the current population is determined as the current optimal individual;

[0036] For each individual currently being processed, a differential mutation strategy is used to generate a mutation vector. The differential mutation strategy is as follows: based on the current best individual, the differential vectors of two random individuals weighted by a random scaling factor are superimposed to obtain the mutation vector.

[0037] The mutation vector is mixed with the currently processed individual in each furnace oxygen content dimension through a binomial crossover operation with crossover probability control, forming a test vector for evaluating whether to replace the current individual.

[0038] Boundary repair is performed on oxygen content components that exceed the specified oxygen content range;

[0039] The predicted thermal efficiency corresponding to the test vector is calculated by calling the XGBoost model and used as the fitness of the test vector;

[0040] If the fitness of the experimental vector is not lower than the fitness of the individual currently being processed, then the experimental vector will be used as the individual at the corresponding position in the next generation population.

[0041] Among the above-mentioned optional methods, the differential mutation strategy based on the current best individual combined with the boundary repair mechanism not only ensures the algorithm's global search capability and convergence speed in the multi-dimensional space of oxygen content in multiple furnaces, but also ensures that the candidate solutions generated by each generation of evolution are strictly within the safe operating range allowed by the process, thus avoiding the risks of invalid search and illegal operation.

[0042] In one alternative approach, determining whether the optimal combination satisfies a preset optimization termination condition includes:

[0043] The optimal combination is input into the XGBoost model to obtain the corresponding predicted thermal efficiency;

[0044] Determine whether the predicted thermal efficiency reaches the preset expected thermal efficiency value;

[0045] Determine whether the oxygen content in each furnace is within the preset safe range allowed by the process.

[0046] If the predicted thermal efficiency reaches the expected thermal efficiency value and the oxygen content in each furnace is within the safe range, then the optimization termination condition is determined to be met.

[0047] Among the above-mentioned optional methods, the dual verification mechanism, which simultaneously verifies the thermal efficiency improvement target and process safety constraints, ensures that the output optimization strategy not only has energy-saving benefits in theory, but also has feasibility and compliance in actual production, preventing the sacrifice of equipment safety or violation of operating procedures in pursuit of ultimate efficiency.

[0048] In one alternative approach, when multiple heating furnaces share the same flue gas system, the industrial heating furnace thermal efficiency optimization method is applied to the multiple heating furnaces as a whole to avoid data confusion and adjustment conflicts between the heating furnaces; when the flue gas systems of the multiple heating furnaces are independent, the industrial heating furnace thermal efficiency optimization method is applied to each heating furnace separately.

[0049] Among the above-mentioned optional methods, the overall optimization or independent optimization strategy is adaptively selected according to the physical topology of the flue gas system. This fundamentally solves the problem of single-furnace independent optimization failure caused by airflow coupling when multiple furnaces share a flue, and ensures the coordination and effectiveness of the optimization strategy in complex equipment-level application scenarios.

[0050] Secondly, this invention provides an industrial heating furnace thermal efficiency optimization system, the technical solution of which is as follows:

[0051] The industrial heating furnace thermal efficiency optimization system includes:

[0052] The operating condition data acquisition module is used to acquire the current operating condition data of the industrial heating furnace, which includes the current oxygen content of each furnace chamber and non-adjustable operating condition parameters.

[0053] The thermal efficiency prediction module is used to input the current operating condition data into the XGBoost model for predicting the thermal efficiency of the heating furnace, and obtain the current thermal efficiency prediction value. The XGBoost model is a model trained based on steady-state time period data determined by a dynamically moving time window and combined features constructed by combining flue gas heat loss with temperature difference and nonlinear relationship with oxygen content.

[0054] The operating condition decoupling and collaborative optimization module is used to, under the premise of fixing the unadjustable operating condition parameters, use the oxygen content of each furnace as the evolution variable, call the XGBoost model to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness, and perform multi-dimensional collaborative optimization through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace.

[0055] The condition judgment and strategy mapping module is used to determine whether the optimal combination meets the preset optimization termination condition. If the optimization termination condition is met, the optimal combination is mapped to the control strategy of the air intake volume adjustment actuator to guide the on-site operation.

[0056] The beneficial effects of the industrial heating furnace thermal efficiency optimization system of the present invention are as follows:

[0057] The system of this invention predicts thermal efficiency using the XGBoost model and combines it with the differential evolution algorithm to globally optimize the oxygen content of each furnace. It can quickly obtain the optimal oxygen content combination that meets the expected thermal efficiency and operating procedures, and convert it into a damper adjustment strategy. This overcomes the limitations of manual adjustment and static models, effectively improves the thermal efficiency of the heating furnace, and reduces fuel consumption and heat loss.

[0058] Thirdly, the technical solution of an electronic device according to the present invention is as follows:

[0059] The invention includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the industrial heating furnace thermal efficiency optimization method of the present invention.

[0060] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows:

[0061] The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the industrial heating furnace thermal efficiency optimization method of the present invention.

[0062] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0063] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0064] Figure 1 This is a schematic flowchart illustrating an embodiment of a method for optimizing the thermal efficiency of an industrial heating furnace according to the present invention.

[0065] Figure 2 This is a schematic diagram of the optimization solution process of the differential evolution algorithm of this invention;

[0066] Figure 3 This is a schematic diagram of the overall process of offline modeling and online optimization of the present invention;

[0067] Figure 4 This is a schematic diagram of an embodiment of an industrial heating furnace thermal efficiency optimization system according to the present invention;

[0068] Figure 5 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0069] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0070] Example 1:

[0071] Figure 1 The diagram illustrates a flowchart of an embodiment of an industrial heating furnace thermal efficiency optimization method provided by the present invention. This method is applied to the online operation control stage of an industrial heating furnace, aiming to optimize and guide on-site operations in real time through a data-driven and mechanism-based approach, and is executed by the industrial heating furnace thermal efficiency optimization system. Figure 1 As shown, it includes the following steps:

[0072] Step S201: Obtain the current operating condition data of the industrial heating furnace, which includes the current oxygen content of each furnace chamber and non-adjustable operating parameters.

[0073] Specifically, during the normal operation of the heating furnace, the monitoring data of the field sensors are collected in real time through the distributed control system (DCS) or the data acquisition and monitoring control system (SCADA) to obtain the current operating condition data.

[0074] The current operating condition data includes, but is not limited to: the oxygen volume fraction in the flue gas of each furnace (as an adjustable variable to be optimized), the real-time load of each furnace, fuel gas flow rate, raw material inlet and outlet temperatures, flue gas temperature, and ambient temperature. The non-adjustable operating parameters refer to variables that cannot be directly adjusted by the operator within the current optimization cycle or are constrained by upstream processes and remain relatively stable, such as the real-time load of each furnace, fuel gas flow rate, raw material inlet and outlet temperatures, and flue gas temperature. By distinguishing between non-adjustable operating parameters and adjustable oxygen content, a data foundation is laid for subsequent precise optimization under fixed external disturbances.

[0075] Step S202: Input the current operating condition data into the XGBoost model used to predict the thermal efficiency of the heating furnace to obtain the current thermal efficiency prediction value; the XGBoost model is a model trained based on steady-state time period data determined by a dynamically moving time window and combined features constructed by combining flue gas heat loss with temperature difference and oxygen content in a positive correlation.

[0076] Specifically, the XGBoost model is a proxy model that is pre-trained offline and deployed on edge computing nodes or servers. This model is not a general black-box model, but a specialized model that incorporates industrial thermal mechanisms. Its training data has been filtered through a dynamically moving time window to remove transient noise, and the input features include combined features characterizing the physical mechanisms of flue gas heat loss (the construction details of these features will be described in detail in subsequent embodiments).

[0077] During the online optimization phase, the real-time data vector obtained in step S201, containing the current oxygen content and unadjustable operating parameters, is input into the model, which can then quickly output the corresponding predicted thermal efficiency value. This prediction method based on a pre-trained surrogate model avoids the computational delay caused by solving complex heat balance equations online in real time, meeting the optimization response requirements of second-level or minute-level in industrial settings.

[0078] Step S203: Under the premise of fixing the non-adjustable operating parameters, the oxygen content of each furnace is used as the evolution variable. The XGBoost model is called to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness. Multidimensional collaborative optimization is performed through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace.

[0079] Specifically, during the optimization process, the current unadjustable operating parameters such as load and fuel flow rate are kept constant, while only the oxygen content of each furnace is set as a variable dimension. The differential evolution algorithm initializes the population within a preset safe range for oxygen content, with each individual representing a candidate combination of oxygen contents from each furnace. During iteration, the algorithm concatenates the oxygen content value of each individual with the fixed unadjustable operating parameters to form a complete input feature vector, calls the aforementioned XGBoost model to calculate the predicted thermal efficiency, and uses this predicted value as the fitness function to evaluate the individual's performance. Through mutation, crossover, and selection operations, the algorithm globally searches the solution space of the multi-furnace coupling, ultimately converging to obtain the optimal combination of oxygen contents from each furnace that maximizes the predicted thermal efficiency. This decoupled operating condition optimization strategy effectively eliminates the interference of external load fluctuations on the optimization direction, ensuring the feasibility of the recommended strategy under the current actual operating conditions.

[0080] Step S204: Determine whether the optimal combination meets the preset optimization termination condition. If the optimization termination condition is met, map the optimal combination to the control strategy of the air intake volume regulating actuator to guide on-site operation.

[0081] Specifically, the optimization termination condition typically includes a dual mechanism of verifying the expected thermal efficiency improvement and verifying safety constraints. The termination condition is determined to be met when the optimal combination output by the differential evolution algorithm can achieve the expected thermal efficiency improvement value as predicted by the model, and the oxygen content in each furnace is within the safe range allowed by the process operation procedure.

[0082] Subsequently, based on the correspondence between oxygen content and damper opening (such as valve characteristic curves or calibration tables), the abstract optimal oxygen content value is converted into the opening adjustment amount or speed setting value of specific air intake regulating actuators (such as duct dampers, fan frequency converters, etc.).

[0083] This step establishes a link between digital space optimization and physical space execution, enabling operators to make precise adjustments based on clearly quantified control strategies, and truly achieve online closed-loop optimization of the heating furnace's thermal efficiency.

[0084] Example 2:

[0085] In one alternative approach, the training process for an XGBoost model includes the following steps.

[0086] First, the historical operating data of the pre-processed industrial heating furnace is obtained.

[0087] Specifically, this embodiment refers to the offline construction stage of the XGBoost model used in Embodiment 1 above. First, various raw operating data continuously recorded by the industrial heating furnace during actual production through the distributed control system (DCS) or the supervisory control system (SCADA) are acquired, including but not limited to the load of each furnace chamber, fuel gas flow rate, flue gas temperature, ambient temperature, flue gas oxygen content, and furnace surface temperature.

[0088] After acquiring the raw historical data, it is first preprocessed, including removing outliers caused by sensor malfunctions (such as jump points exceeding the physical range or infinity values), and aligning the timestamps of each sensor channel to ensure the synchronization of multi-source data. For missing values ​​caused by short-term communication interruptions, median imputation is used, that is, the median of the feature across the entire dataset is used to replace the missing value. Considering that various data collected in industrial fields have significant differences in physical dimensions, for example, the unit of flue gas temperature is degrees Celsius (°C), while the unit of fuel flow rate may be cubic meters per hour (m³ / h). 3 The oxygen content is expressed as a volume percentage (%), while the volume percentage is expressed as a percentage of the total volume (h). If these raw values ​​are directly input into the model, features with larger values ​​will dominate the update direction of the loss function during gradient descent, resulting in the model not learning sufficiently for features with small values ​​but high sensitivity (such as oxygen content), thus affecting prediction accuracy. To eliminate this influence of units, this embodiment performs Z-score standardization on the preprocessed features of each type, and the calculation formula is as follows:

[0089] ;

[0090] Where x is the original feature value; This is the mean of the feature across the dataset; denoted as σ0, where σ0 is the standard deviation; z is the standardized eigenvalue.

[0091] Understandably, this processing transforms all input features into a standard normal distribution with a mean of 0 and a standard deviation of 1. This allows the model to treat each physical variable fairly, focusing on uncovering its intrinsic nonlinear relationship with thermal efficiency. It also prevents features with large values ​​from dominating the update direction of the loss function, thereby improving the model's learning ability and prediction accuracy for key sensitive parameters such as oxygen content. The preprocessed historical data obtained after these preprocessing and standardization steps will serve as the foundation for subsequent steady-state screening and model training.

[0092] Furthermore, based on the load fluctuation threshold or parameter variance, steady-state period data that reaches steady-state operating conditions are selected from the historical operating data.

[0093] Specifically, since industrial heating furnaces frequently experience transitional processes such as load increase, load decrease, or fuel switching during actual operation, the data from these non-steady-state periods are often accompanied by strong thermal inertia and hysteresis effects. If used directly for training, the model will learn incorrect dynamic response relationships rather than the true thermal conversion characteristics.

[0094] Therefore, this embodiment cleanses historical data by setting strict steady-state determination criteria. For example, the load fluctuation threshold method is used, that is, when the absolute value of the rate of change of the heating furnace load within a continuous preset time period (such as 30 minutes) is lower than a set percentage (such as 2%), the time period is determined to be a steady-state period; or the parameter variance method is used, which calculates the variance of key process parameters (such as flue gas temperature and fuel flow rate) within a sliding time window, and when the variance value is lower than a preset threshold, it is considered to be in a steady state. The steady-state time period data selected in this way can effectively eliminate noise interference from the transition process and ensure that the model training samples accurately represent the energy conversion law of the equipment under thermal equilibrium.

[0095] It should be noted that the selection of steady-state period data mentioned above is not completed all at once using a fixed historical data interval, but rather through a dynamic, moving time window mechanism for continuous updates. During each model training or update, the system uses the current time as the endpoint and extracts historical data for a preset duration (e.g., 6 hours) as a decision window. Steady-state determination is performed on the data within this window, and steady-state segments are extracted for training. As production progresses, this window slides forward, with newly generated operational data continuously added to the tail of the window, while the oldest historical data is removed from the head. This dynamic window mechanism ensures that the training samples always cover the equipment operating characteristics of the most recent period, enabling the XGBoost model to continuously adapt to operating condition drift caused by factors such as equipment aging, ash accumulation, fuel calorific value fluctuations, or changes in environmental conditions. This avoids prediction bias caused by using outdated historical data and ensures that the prediction accuracy during the online optimization phase always matches the current actual operating conditions.

[0096] Furthermore, temperature parameters and oxygen content parameters related to the flue gas heat loss of the industrial heating furnace are obtained.

[0097] Specifically, after selecting steady-state data, physical parameters directly related to flue gas heat loss are extracted. These include temperature parameters such as flue gas temperature and ambient temperature: flue gas temperature reflects the amount of sensible heat carried by the flue gas as it leaves the furnace and is the main driver of flue gas heat loss; ambient temperature serves as a reference temperature for calculating the actual temperature difference between the flue gas and the surrounding environment. The oxygen content parameter is the volume fraction of oxygen in the flue gas, whose value directly reflects the excess air coefficient during combustion, thus affecting the volumetric flow rate of the flue gas and the magnitude of flue gas heat loss.

[0098] It should be noted that the selection of these parameters is not an arbitrary feature engineering, but is based on the basic principles of industrial thermodynamics, namely, the essence of flue gas heat loss is the extra heat carried by high-temperature flue gas relative to the environment. This heat is proportional to the temperature difference and has a non-linear relationship with the flue gas volume (determined by the excess air coefficient).

[0099] Furthermore, based on the fact that flue gas heat loss is positively correlated with temperature difference and nonlinearly related to oxygen content, combined with the temperature parameter and the oxygen content parameter, a combination of characteristics representing the magnitude of flue gas loss is determined.

[0100] Specifically, this embodiment does not directly input the original parameters such as exhaust temperature, ambient temperature, and oxygen content as independent features into the model, but rather integrates them into a combined feature with clear physical semantics based on the thermal balance mechanism.

[0101] The construction logic of this combined feature is as follows: the heat loss of flue gas is positively correlated with the temperature difference between the flue gas temperature and the ambient temperature, and at the same time, it has a non-linear relationship with the excess air volume reflected by the oxygen content in the flue gas. Therefore, the temperature difference is multiplied by a correction term including oxygen content, and then multiplied by the flue gas heat loss coefficient to obtain a physical quantity that directly characterizes the amount of heat carried away by the flue gas.

[0102] The detailed construction of this combined feature will be further explained in Example 3. Using it as input features for the XGBoost model is equivalent to embedding physical constraints that conform to the laws of thermodynamics into the data-driven model, enabling the model to output predicted values ​​that conform to the energy conservation trend even under conditions not covered by the training samples.

[0103] Furthermore, the thermal efficiency corresponding to the steady-state period data is calculated based on the inverse equilibrium method and used as the training label.

[0104] Specifically, the thermal efficiency of the heating furnace is determined by subtracting the proportion of each heat loss to the supplied energy, and the calculation formula is as follows:

[0105] ;

[0106] in, Overall thermal efficiency (%) The percentage of heat lost due to smoke exhaust relative to the energy supplied (%). The percentage of heat lost due to incomplete combustion relative to the energy supplied (%). The percentage of heat lost due to surface heat dissipation relative to the supplied energy (%).

[0107] Understandably, this formula indicates that to improve thermal efficiency... Essentially, the goal is to reduce exhaust gas losses, incomplete combustion losses, and heat dissipation losses. Compared to the positive balance method, which requires precise measurement of fuel calorific value and effective heat absorption (often difficult to obtain in real time in industrial settings), the parameters required for the inverse balance method, such as exhaust gas temperature, flue gas oxygen content, and furnace surface temperature, can be directly collected by on-site sensors. Therefore, it is more suitable for online calculations in industrial scenarios. By substituting the steady-state data into the above formula, the corresponding thermal efficiency value can be calculated for each steady-state sample. This value serves as the label benchmark for training the XGBoost model, giving the training label a clear physical meaning. This allows the model to learn the mapping relationship between operating parameters and thermal efficiency that conforms to the law of conservation of energy, ensuring that the prediction results have physical interpretability.

[0108] Furthermore, the initial model is trained based on the steady-state period data, the combined features, and the training labels to obtain the XGBoost model, so that the XGBoost model includes historical fluctuation errors and dynamically matches the current operating conditions.

[0109] Specifically, the standardized steady-state time period data and mechanism combination features mentioned above are used as input, and the thermal efficiency calculated by the inverse equilibrium method is used as the label to train the initial XGBoost model.

[0110] In this embodiment, the input feature vector consists of 19 process reference numbers, including original parameters such as oxygen content, fuel gas flow rate, flue gas temperature, ambient temperature, and furnace load of each furnace, as well as the aforementioned mechanism combination features. The training samples are 360 ​​minutes-level historical data points after steady-state screening and standardization. The above data is randomly divided into training and test sets at a ratio of 80% and 20%, respectively. The training set is used for model parameter learning, and the test set is used for final performance verification. The hyperparameters of the XGBoost model (such as maximum tree depth, learning rate, subsampling ratio, etc.) are tuned through cross-validation. After training, the model needs to be verified: the predicted thermal efficiency of the model on the test set is compared with the actual thermal efficiency. When the goodness of fit R... 2 When the accuracy reaches 0.85 or higher, the model is deployed as an online predictive agent model.

[0111] It should be noted that this training strategy based on steady-state data not only enables the model to master the baseline performance curve under ideal operating conditions, but more importantly, since steady state itself is a range that allows for small fluctuations rather than an absolutely static point, the model naturally internalizes the reasonable fluctuation errors present in historical data during the training process. When this model is applied to the online optimization stage of Example 1, even if there are slight disturbances in the current operating conditions, the model can still output stable and reliable thermal efficiency predictions thanks to its internalized fluctuation tolerance, thereby achieving dynamic and accurate matching to the current actual operating conditions.

[0112] Example 3:

[0113] Based on the steady-state data screening and standardization completed in Example 2, this example further performs physical enhancement processing on the features input to the XGBoost model.

[0114] In one alternative approach, the determination of combined characteristics representing the magnitude of smoke exhaust heat loss, based on the positive correlation between smoke exhaust heat loss and temperature difference and the non-linear relationship with oxygen content, and combining the temperature parameter and the oxygen content parameter, includes the following steps.

[0115] First, the exhaust gas temperature and ambient temperature are obtained as the temperature parameters, and the oxygen volume fraction in the flue gas is obtained as the oxygen content parameter.

[0116] In this embodiment, the flue gas temperature and ambient temperature are extracted from the steady-state data as temperature parameters, and the oxygen volume fraction in the flue gas is extracted as an oxygen content parameter.

[0117] It should be noted that in the thermal process of industrial heating furnaces, flue gas heat loss usually accounts for the largest proportion of all heat losses. Its magnitude is not linearly determined by a single variable, but rather depends on the nonlinear coupling between the flue gas temperature and the ambient temperature, and the flue gas volumetric flow rate (which is closely related to the excess air coefficient, i.e., oxygen content). If raw sensor readings such as flue gas temperature, ambient temperature, and oxygen content are directly input into a machine learning model as independent features, the model needs to rely on massive amounts of data to discover the complex interactions between these variables. This not only increases the training difficulty, but also makes it easy for pure statistical correlation to fail when fuel calorific value fluctuates or equipment aging causes operating condition deviations, resulting in a significant decrease in prediction accuracy.

[0118] To overcome this deficiency, this embodiment constructs a combined feature according to the heat balance mechanism that shows a positive correlation between flue gas heat loss and temperature difference and oxygen content, so that the feature itself carries a clear physical semantics.

[0119] Furthermore, the temperature difference between the exhaust gas temperature and the ambient temperature is calculated.

[0120] Specifically, the temperature difference is obtained by subtracting the ambient temperature from the flue gas temperature. This temperature difference directly reflects the amount of sensible heat carried by the flue gas relative to the environment. The larger the temperature difference, the more heat the flue gas carries away, and the lower the thermal efficiency. This temperature difference is the core driving factor in calculating flue gas heat loss.

[0121] Furthermore, the temperature difference is multiplied by a correction term that includes the volume fraction of oxygen in the flue gas, and then multiplied by the exhaust heat loss coefficient to obtain the combined characteristics.

[0122] Specifically, based on the inverse balance method principle for calculating the thermal efficiency of industrial heating furnaces, the estimation of flue gas heat loss follows the physical formula below:

[0123] ;

[0124] in, T represents the flue gas heat loss; k is the flue gas heat loss coefficient, the value of which is related to the fuel type and furnace structure; flue T represents the exhaust gas temperature. env Ambient temperature; The formula represents the correction factor for flue gas heat loss, used to characterize the differences in specific heat capacity of combustion products from different fuels; O2 is the volume fraction of oxygen in the flue gas. This formula clearly reveals the product relationship between flue gas heat loss and the correction terms for temperature difference and oxygen content.

[0125] It should be noted that the combined features are used as input features for training the XGBoost model.

[0126] Specifically, the combined feature constructed in this embodiment is the physical quantity obtained by multiplying the temperature difference between the flue gas temperature and the ambient temperature by a correction term including the oxygen volume fraction in the flue gas, and then multiplying by the flue gas heat loss coefficient. For example, the entire right side of the formula can be directly input into the XGBoost model as a combined feature; or given k and When it is a constant, As a core composite feature, this feature is no longer an abstract mathematical transformation but a direct physical entity representing the amount of heat carried away by the flue gas under the current operating conditions. Using it as input to train the XGBoost model is equivalent to embedding a thermodynamically compliant "white box" anchor point into a data-driven black-box model. Even in extreme operating conditions not covered by the training samples or changes in fuel quality encountered in actual operation, the fundamental physical laws of thermal balance remain unchanged. This composite feature can still guide the model to output predicted values ​​that conform to the energy conservation trend, thus significantly enhancing the model's physical interpretability and cross-condition generalization ability for the key indicator of flue gas heat loss, effectively avoiding the prediction distortion problem of pure data models under unsteady or variable operating conditions.

[0127] It should be noted that the calculation structure of the correction term is as follows: ;in, The volume fraction of oxygen in the flue gas; This is the correction factor for flue gas heat loss.

[0128] Specifically, The term reflects the amplifying effect of the excess air coefficient on flue gas volume, while This constitutes a dynamic correction factor for the temperature difference. When the oxygen content increases, the excess air coefficient increases, the flue gas volume increases, and the exhaust heat loss increases accordingly; the correction term accurately describes the amplification effect of oxygen content on exhaust heat loss through this nonlinear relationship.

[0129] Example 4:

[0130] It should be noted that this embodiment refers to the online optimization stage of Embodiment 1, and elaborates on the differential evolution optimization process in step S203. The core of this embodiment lies in constructing an optimization environment with decoupled operating conditions to ensure the reliability of the optimization results under conditions of multi-furnace coupling or external disturbances.

[0131] In one alternative approach, under the premise of fixing the unadjustable operating parameters, the oxygen content of each furnace is used as the evolutionary variable. The XGBoost model is called to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness. A multi-dimensional collaborative optimization is performed using a differential evolution algorithm to obtain the optimal combination of oxygen contents for each furnace, including:

[0132] Specifically, the current furnace load, fuel flow rate, and flue gas temperature are used as the non-adjustable operating condition parameters.

[0133] It should be noted that these parameters are defined as "non-adjustable" because within a minute-level time window of a single optimization calculation, they remain relatively stable due to constraints from upstream production plans, raw material supply, or equipment thermal inertia, or they belong to process baseline values ​​that operators should not frequently and significantly adjust in pursuit of instantaneous thermal efficiency. For example, in the actual operation of a reaction heater in a petrochemical plant, the raw material processing volume is determined by the plant's overall material balance, while the fuel gas flow rate mainly follows changes in load demand; both can be considered constant boundary conditions in the short term.

[0134] Furthermore, based on this, the oxygen content of each furnace is used as the evolutionary variable, and the population is initialized within a preset oxygen content range. Each individual in the population is composed of the oxygen content value of each furnace.

[0135] For example, in a heating furnace containing two chambers, each individual is a two-dimensional vector [O 2_1 O 2_2 ], representing the set values ​​of flue gas oxygen content for furnace No. 1 and furnace No. 2, respectively.

[0136] It should be noted that the initialization of the population is not random across the entire domain, but is strictly limited to a safe range allowed by the process operation procedures (such as between 2% and 4.5%). This range ensures complete combustion to avoid CO exceeding the standard, and also prevents a surge in exhaust heat loss caused by an excessively high excess air coefficient. This results in an initial set of candidate solutions covering the current feasible operating space.

[0137] Further, we move into the core step of iterative optimization. The following population evolution steps are executed iteratively until the preset termination condition of the differential evolution algorithm is met: In each generation of population evolution, for each individual in the population, its oxygen content value is concatenated with the unadjustable operating parameters to form an input feature vector, which is then input into the XGBoost model to obtain the predicted thermal efficiency as the fitness of that individual.

[0138] Understandably, this step reflects a dynamic and static data flow logic: although oxygen content varies greatly within the population, background parameters such as load and fuel flow remain constant within the current generation. For example, if the load of Boiler 1 is currently 80% and the fuel flow rate is 1200 Nm³... 3 Given an exhaust gas temperature of 320℃ and an oxygen content of [3.0%, 3.5%], the system will concatenate these values ​​into a complete 19-dimensional feature vector [3.0%, 3.5%, 80%, 1200, 320, ...] for any individual in the population. This vector will then be input into the XGBoost model. Based on the steady-state mechanism knowledge trained in Examples 2 and 3, the model quickly outputs the predicted thermal efficiency value for this specific operating condition combination. This value serves as the fitness function for evaluating the quality of this oxygen content combination.

[0139] Furthermore, mutation, crossover, and selection operations are performed based on the fitness to generate the next generation population.

[0140] The differential evolution algorithm uses the differences between individuals in the population to guide the search direction, continuously exploring a better oxygen content ratio by simulating the biological evolution mechanism. When the termination condition of the differential evolution algorithm is met (such as reaching the maximum number of iterations of 100 generations or fitness convergence accuracy), the optimal combination is output.

[0141] It should be noted that the operating condition decoupling strategy adopted in this embodiment has significant technical advantages. In industrial settings, especially in scenarios where multiple furnaces share a flue gas system, the operating conditions of each furnace are often coupled and fluctuate dynamically over time. If external variables such as load and fuel flow are not fixed during the optimization process, the algorithm is prone to misinterpreting the natural fluctuations in thermal efficiency caused by upstream disturbances as the effect of oxygen content adjustment, thus leading to deviation or even divergence in the optimization direction. By "freezing" the non-adjustable operating condition parameters as constants, this method effectively constructs a quasi-steady-state virtual experimental environment, ensuring that the fitness changes in the XGBoost model output are caused only by changes in the oxygen content combination. This purification of causal relationships not only ensures the convergence stability of the differential evolution algorithm in complex multidimensional spaces but also ensures that the final output optimal oxygen content combination is a feasible solution that truly fits the current actual production conditions, rather than a mathematical extreme value detached from the operating condition context.

[0142] Example 5:

[0143] In one alternative approach, performing mutation, crossover, and selection operations based on the fitness includes:

[0144] Based on the fitness of each individual, the individual with the highest fitness in the current population is determined as the current optimal individual.

[0145] Specifically, at the start of each generation of population evolution, the system first traverses all current individuals and identifies the individual with the highest predicted thermal efficiency as the baseline anchor point for evolution. This optimal individual-based guidance mechanism (i.e., the best1bin strategy) can fully utilize the prior knowledge accumulated during the historical search process, causing subsequently generated candidate solutions to tend to cluster towards known high-quality regions, thereby significantly improving the convergence speed of the algorithm on the complex nonlinear response surface of furnace thermal efficiency.

[0146] Furthermore, for each individual currently being processed, a differential mutation strategy is used to generate a mutation vector. The differential mutation strategy is as follows: based on the current best individual, the differential vectors of two random individuals weighted by a random scaling factor are superimposed to obtain the mutation vector.

[0147] In this embodiment, the scaling factor F is not a fixed constant, but is randomly selected from the interval (0.5, 1.0) in each iteration. This dynamic selection strategy has clear engineering considerations: when F is close to 0.5, the difference perturbation amplitude is small, which is conducive to fine-tuning near the current optimal solution and improving the optimization accuracy; when F is close to 1.0, the perturbation amplitude increases, which helps to escape the local extremum trap and explore a wider oxygen content ratio space. For thermal systems with multi-peak characteristics, such as heating furnaces, a fixed scaling factor can easily lead to premature convergence or oscillatory divergence of the algorithm, while the random perturbation in the (0.5, 1.0) interval, while maintaining search stability, gives the algorithm the ability to adaptively explore, enabling it to better adapt to the drift of the objective function shape caused by fuel calorific value fluctuations or load changes.

[0148] Furthermore, through a binomial crossover operation controlled by crossover probability, the mutation vector is mixed with the currently processed individual in each furnace oxygen content dimension to form a test vector for evaluating whether to replace the current individual.

[0149] The crossover probability (CR) is set to 0.7, which determines the proportion of components inherited by the experimental vector from the mutated vector. In the scenario of optimizing oxygen content in a heating furnace, CR=0.7 means that the newly generated experimental vector has a 70% probability of retaining the superior gene segments carried by the mutated vector, while having a 30% probability of retaining some characteristics of the original individual to maintain population diversity. This ratio is a balance point derived from extensive field debugging experience: too high a CR will cause the population to converge too quickly, losing its ability to adapt to sudden changes in operating conditions; too low a CR will cause the evolutionary process to degenerate into a random walk, making it difficult to effectively approach the global optimum. A crossover probability of 0.7 ensures that the algorithm can quickly absorb the improvement information brought by the mutation operation, while retaining sufficient genetic diversity to cope with the uncertainties of the industrial field.

[0150] Furthermore, boundary repair is performed on oxygen content components that exceed the stated oxygen content range.

[0151] It should be noted that after differential mutation and crossover operations, the oxygen content values ​​of some furnaces in the generated experimental vectors may exceed the safe range allowed by the process. For example, the algorithm may calculate an oxygen content of 1.5% or 6.0% in pursuit of theoretically optimal thermal efficiency, but in actual production, this corresponds to the risk of incomplete combustion leading to excessive CO and explosion, and the risk of excessive air coefficient leading to a surge in flue gas heat loss or even flameout, respectively.

[0152] Therefore, in this embodiment, components below 2% are forcibly corrected to 2%, and components above 4.5% are forcibly corrected to 4.5%, ensuring that all test vectors entering the evaluation stage are strictly within the process safety range of [2%, 4.5%]. This boundary range is determined comprehensively based on the nameplate parameters of the heating furnace equipment, fuel characteristics, and operating procedures. It constitutes a rigid firewall between digital optimization and physical safety, fundamentally eliminating the possibility of the algorithm outputting illegal or dangerous operating instructions.

[0153] Furthermore, the XGBoost model is invoked to calculate the predicted thermal efficiency corresponding to the test vector as the fitness of the test vector.

[0154] Understandably, the test vector, after boundary repair, is concatenated with fixed, unadjustable operating parameters and input into a pre-trained XGBoost model to quickly obtain the predicted thermal efficiency under the candidate operating condition. Then, a greedy selection process is performed: if the fitness of the test vector is not lower than the fitness of the currently processed individual, the test vector is selected as the individual at the corresponding position in the next generation population; otherwise, the original individual is retained for the next generation.

[0155] It should be noted that this selection mechanism ensures that the overall quality of the population remains monotonically constant with each generation, thus avoiding performance regression during the evolutionary process.

[0156] In this embodiment, within the aforementioned iterative framework, the population size is set to 50, and the maximum number of iterations is set to 100. These two parameters represent a trade-off between the real-time requirements of industrial settings and the accuracy of optimization. A population of 50 individuals is sufficient to cover the main characteristic regions of the two-dimensional to four-dimensional oxygen content solution space, providing enough statistical samples to support the effectiveness of differential evolution; while the upper limit of 100 generations ensures that the entire optimization calculation can be completed within seconds, meeting the minute-level response requirements of online control of the heating furnace. In actual operation, if the algorithm converges to a satisfactory solution before reaching 100 generations, it can be terminated early to further conserve computational resources.

[0157] Through the synergistic operation of mutation, crossover, boundary repair and selection, the differential evolution algorithm can not only search for the global optimum mathematically efficiently, but also ensure that each step of evolution conforms to the safety specifications and process constraints of industrial production, truly realizing the deep integration of algorithm logic and industrial mechanism.

[0158] To more clearly demonstrate the overall process of the Differential Evolution (DE) algorithm described in this embodiment, the following is a detailed explanation. Figure 2 The mutation, crossover, boundary repair, fitness calculation, and selection operations of this embodiment are described in detail.

[0159] At the start of each generation of population evolution, it is first determined whether the preset termination condition of the differential evolution algorithm is met (i.e., the aforementioned "preset termination condition of the differential evolution algorithm"). If the termination condition is met (e.g., reaching the maximum number of iterations of 100 generations or fitness convergence accuracy), the best individual in the current population is directly output as the optimization result, and the entire optimization process ends. If the termination condition is not met, the following mutation, crossover, and other operations are continued.

[0160] First, population mutation is performed (corresponding to the aforementioned "generating mutation vectors using differential mutation strategy"). For each individual in the current population (referred to as the target individual), two distinct indices, both different from the target individual, are randomly selected from the current population; their corresponding individuals are denoted as... and Calculate the difference vector. Then select the individual with the highest fitness in the current population. Based on this, a difference vector weighted by a random scaling factor F is superimposed to generate a mutation vector. The scaling factor F is randomly selected from the interval (0.5, 1.0) in each iteration.

[0161] Next, a crossover operation is performed (corresponding to the aforementioned "binomial crossover operation controlled by crossover probability"). The mutation vector V and the target individual X are binomially cross-mixed along the oxygen content dimension of each furnace to form the experimental vector U. For each dimension j (i.e., the oxygen content component of each furnace), a uniformly random number in the interval [0,1] is generated. ,like ,but ,otherwise Meanwhile, to ensure that the experimental vector is not exactly equal to the target individual, a dimension is randomly selected to force the inheritance of the mutation vector value. The crossover probability CR is set to 0.7.

[0162] The boundary conditions are then processed (corresponding to the aforementioned "boundary repair of oxygen content components exceeding the oxygen content value range"). Boundary repair is performed on the components in the test vector U that exceed the preset oxygen content value range (e.g., 2% to 4.5%): those below the lower limit are set as the lower limit value, and those above the upper limit are set as the upper limit value.

[0163] Then, the objective function is calculated (corresponding to the aforementioned "calling the XGBoost model to calculate the predicted thermal efficiency corresponding to the test vector as the fitness"). The repaired test vector U is concatenated with the current fixed, unadjustable operating parameters to form a complete input feature vector. The trained XGBoost model is then called, and the output predicted thermal efficiency is used as the fitness of the test vector.

[0164] Finally, a selection operation is performed (corresponding to the aforementioned "if the fitness of the experimental vector is not lower than that of the original individual, then it is used as the individual at the corresponding position in the next generation population"). If the fitness of the experimental vector U is not lower than that of the current target individual X, then U replaces X in the next generation population; otherwise, X remains unchanged. After completing the selection operation, the system returns to determine whether the preset termination condition of the differential evolution algorithm is met, and begins the next iteration. The optimal combination is output when the preset termination condition of the differential evolution algorithm is met.

[0165] It should be noted that the "termination condition of differential evolution algorithm" and the "optimization termination condition" mentioned in this embodiment are two different concepts, and they are at different stages and play different roles in the optimization process.

[0166] The termination condition of the differential evolution algorithm (e.g., reaching the maximum number of iterations of 100 generations or fitness convergence accuracy) is used to control the iteration process of the differential evolution algorithm itself: when the condition is met, the algorithm stops iterating and outputs the currently found optimal combination of oxygen content.

[0167] The optimization termination condition involves verifying the safety and effectiveness of the optimal combination from an engineering application perspective after the differential evolution algorithm outputs the best combination. Specifically, this includes checking whether the optimal combination achieves the expected thermal efficiency value and whether the oxygen content in each furnace chamber meets the safety range specified in the operating procedures. If the optimization termination condition is met, the optimal combination is output as the damper control strategy; otherwise, the differential evolution optimization is re-executed.

[0168] Therefore, the termination condition of the differential evolution algorithm is the convergence condition within the algorithm, determining "when to stop the optimization"; the optimization termination condition is the engineering acceptance condition, determining "whether the optimization result is usable". These two conditions are independent of each other and are executed sequentially, together forming a complete closed loop for this invention from global optimization to secure deployment.

[0169] Example 6:

[0170] It should be noted that this embodiment is a further refinement of step S204 in Embodiment 1, forming the final safety and efficiency defense line between the output of the differential evolution algorithm and on-site physical execution. Although the boundary repair mechanism described in Embodiment 5 ensures that individuals in the evolution process are always within the basic safety domain, after the algorithm converges, the final output global optimal combination still needs to be independently double-checked. This is because the theoretical extreme point obtained by the mathematical model optimization may not fully fit the complex industrial production reality at the current moment, or the marginal benefits it brings may not be sufficient to support the adjustment costs of on-site operations.

[0171] In one alternative approach, determining whether the optimal combination satisfies a preset optimization termination condition includes:

[0172] The optimal combination is input into the XGBoost model to obtain the corresponding predicted thermal efficiency.

[0173] Specifically, the system inputs the optimal combination of oxygen content in each furnace output by the differential evolution algorithm back into the XGBoost model to obtain the accurate predicted thermal efficiency value for that combination. The purpose of this step is that although the differential evolution algorithm has already evaluated the candidate combinations through the fitness function during the iteration process, the optimal combination that finally converges still needs to undergo an independent accurate prediction to obtain its corresponding thermal efficiency value, providing a quantitative benchmark for subsequent verification.

[0174] Furthermore, it is determined whether the predicted thermal efficiency reaches the preset expected thermal efficiency value.

[0175] Specifically, the expected thermal efficiency value is not a fixed constant, but a threshold that can be dynamically set according to the production plan. For example, it can be set as the improvement relative to the current baseline thermal efficiency (e.g., an improvement of more than 0.5%), or as an absolute target value (e.g., more than 92%). The purpose of this check is to avoid "ineffective optimization": under certain extreme operating conditions, the algorithm may find a theoretically slightly better combination of oxygen content than the current situation, but the resulting improvement in thermal efficiency is negligible (e.g., only 0.02%). Considering the mechanical wear, operational lag, and possible short-term fluctuations in operating conditions caused by the dampers on site, such a small theoretical benefit has no practical value in engineering. Only when the predicted benefit exceeds the preset expected threshold does the system consider the optimization strategy necessary to implement.

[0176] Furthermore, it is determined whether the oxygen content in each furnace is within the preset safe range allowed by the process.

[0177] Specifically, although Example 5 underwent boundary repair during the iteration process, the final verification stage defined a more stringent and comprehensive safety range, directly derived from the furnace's equipment nameplate parameters, burner design characteristic curves, and the company's safe operating procedures. For example, for a specific model of gas-fired furnace, the operating procedures might explicitly stipulate that at 80% load, the flue gas oxygen content must not be lower than 2.5% to prevent incomplete combustion from generating carbon monoxide accumulation that could lead to an explosion risk, while it must not be higher than 4.0% to avoid excessive heat loss from the exhaust gas, which could cause furnace temperature runaway or flameout. This safety range is a rigid constraint of the physical world, taking precedence over the optimization objectives of any mathematical model. If the oxygen content in any furnace chamber in the optimal combination output by the algorithm exceeds this dynamic safety window, even if its predicted thermal efficiency is high, the system will immediately determine that the termination condition is not met, thereby intercepting the output of the dangerous command.

[0178] Furthermore, if the predicted thermal efficiency reaches the expected thermal efficiency value and the oxygen content in each furnace is within the safe range, then the optimization termination condition is determined to be met.

[0179] Specifically, only when both of the above checks pass—that is, the predicted thermal efficiency reaches the expected value and the oxygen content in each furnace is within the safe range—is the optimization termination condition finally met, and the optimal combination is then released to the subsequent strategy mapping module. This dual-check mechanism fundamentally solves the common defect in pure data-driven optimization methods that rely solely on indicators to draw conclusions, ensuring that every output control strategy is efficient, safe, compliant, and has practical operational value. In this way, the present invention effectively prevents the sacrifice of equipment safety or violation of process discipline in pursuit of ultimate theoretical efficiency, realizing the reliable implementation and closed-loop application of industrial heating furnace energy-saving optimization technology in complex production sites.

[0180] Example 7:

[0181] In one alternative approach, when multiple heating furnaces share the same flue gas system, the industrial heating furnace thermal efficiency optimization method is applied to the multiple heating furnaces as a whole to avoid data confusion and adjustment conflicts between the heating furnaces; when the flue gas systems of the multiple heating furnaces are independent, the industrial heating furnace thermal efficiency optimization method is applied to each heating furnace separately.

[0182] Specifically, this embodiment provides an adaptive scenario adaptation mechanism for complex device-level topologies in industrial settings. This is a key prerequisite for ensuring that the optimization strategies in the aforementioned embodiments can be safely implemented in actual production.

[0183] In process industries such as petrochemicals and chemicals, heating furnaces are configured in various ways, with a shared flue gas system across multiple furnaces being a typical and challenging architecture to control. In this architecture, a single heating furnace typically uses a common chimney for exhaust, and the air supply ducts use a unified main intake pipe, with distribution only occurring before reaching each furnace via branch pipes and independent baffles. From a fluid dynamics and combustion control perspective, this physical connection results in strong coupling between furnaces: when operators or the control system adjust the damper opening of one furnace to change its air intake, it not only alters the excess air coefficient of that furnace but also causes fluctuations in the main intake pipe pressure, passively affecting the actual air intake and combustion state of other parallel furnaces. If the traditional single-furnace independent optimization mode is still used in such scenarios, the optimization algorithms for each furnace will treat the adjustments of other furnaces as uncontrollable external disturbances, leading to conflicting oxygen content setpoints, system-level airflow oscillations, abnormal flue gas temperature fluctuations, and even the risk of flameout, causing severe data confusion and adjustment conflicts.

[0184] To overcome the aforementioned coupling interference, this embodiment stipulates that when multiple heating furnaces are detected sharing the same flue gas system, these heating furnaces must be treated as a unified controlled object, and the optimization method of this invention must be applied as a whole. In practical terms, this means that during the differential evolution algorithm optimization process described in Embodiment 4, the oxygen content of a single furnace is no longer treated as an independent optimization variable; instead, the oxygen content of all furnaces sharing the same flue gas system is concatenated into a joint evolution variable vector.

[0185] For example, for a shared system containing three furnaces, each individual in the population will be represented by a three-dimensional oxygen content vector [O 2_1 O 2_2 O 2_3The system consists of three independent one-dimensional variables, rather than three separate ones. Furthermore, the training and prediction of the XGBoost model must be based on the combined operating data of the entire system, enabling the model to internalize the nonlinear mapping relationship of inter-furnace airflow coupling. Through this holistic collaborative optimization, the algorithm outputs a globally coordinated optimal combination of oxygen contents under the current main pipe pressure and load conditions. This fundamentally eliminates the conflict risks caused by independent decision-making in a single furnace, ensuring the dynamic balance and operational stability of the multi-furnace system in the pursuit of maximizing thermal efficiency.

[0186] In contrast, when the flue gas systems of multiple heating furnaces are completely independent—that is, each furnace has its own independent blower, exhaust duct, and fuel supply system, and there is no fluid dynamic coupling between them—this embodiment allows for the application of industrial heating furnace thermal efficiency optimization methods to each heating furnace separately. Under this decoupled physical topology, the operating conditions of each furnace do not affect each other. By breaking them down into multiple independent low-dimensional optimization problems, not only can control accuracy equivalent to overall optimization be achieved, but the search dimension of the differential evolution algorithm and the inference computation of the XGBoost model can also be significantly reduced, thereby improving the system's real-time response speed and the utilization efficiency of computing resources.

[0187] In summary, this embodiment achieves precise alignment between control strategies and physical object characteristics by distinguishing between shared and independent flue gas system topologies and matching corresponding optimization granularities. This ensures both system security and coordination in complex coupled scenarios and computational efficiency in simple independent scenarios. It is an important adaptation step for this invention from theoretical algorithms to engineering applications.

[0188] In addition, to facilitate understanding of the overall logic of this invention from offline modeling to online optimization and then to on-site execution, please refer to the following... Figure 3 In summary, the process is clearly divided into two sequential phases: offline modeling and online optimization and secure execution, fully covering the entire technical chain from historical data collection to on-site operation execution.

[0189] The core objective of the offline modeling phase is to train an XGBoost proxy model that can accurately predict the thermal efficiency of the heating furnace. As the starting point of the entire process, the original historical operating data of the industrial heating furnace is first collected from each data monitoring point through a distributed control system or a data acquisition and monitoring control system. This includes core process parameters such as oxygen, fuel gas flow rate, flue gas temperature, ambient temperature, carbon monoxide concentration, raw material inlet and outlet temperatures, and raw material load.

[0190] The collected data is then preprocessed, including null imputation and standardization operations, specifically outlier removal, median missing value imputation, and Z-score standardization, to eliminate noise and dimensional differences in the original data and obtain clean and usable preprocessed data.

[0191] Next, the overall thermal efficiency is calculated. This step also includes null value filling and standardization operations. The inverse balance method is used to calculate the comprehensive thermal efficiency corresponding to each piece of data, the calculation results are preprocessed, and finally the target labels for model training are obtained.

[0192] Then it enters the XGBoost model construction process in the dashed box. First, feature engineering is performed to determine features and target variables. Temperature parameters and oxygen content parameters related to exhaust gas heat loss are extracted from the preprocessed data, and combined features characterizing the magnitude of exhaust gas heat loss are constructed based on the heat balance mechanism. Next, model training is carried out: the feature set and thermal efficiency labels are split into a training set and a test set in an 8:2 ratio, and the XGBoost algorithm is used for supervised learning training. Finally, model evaluation is performed: the test set is used to evaluate the performance of the trained model, and the root mean square error and goodness of fit are output. The model is determined to be qualified when the goodness of fit is greater than or equal to 0.85, and the off-line modeling phase is completed.

[0193] After the off-line model passes the evaluation, it enters the on-line optimization and safety execution phase. The core objective of this phase is to use the off-line trained model to perform real-time global optimization and output a safe and feasible on-site operation strategy.

[0194] First, the differential evolution algorithm is executed to optimize and solve the oxygen content of each furnace. The current non-adjustable working condition parameters are fixed, the oxygen content of each furnace is taken as the evolution variable, the XGBoost model is called to calculate the predicted thermal efficiency as the fitness, and multi-dimensional collaborative global optimization is performed through mutation, crossover and selection operations. The three furnaces shown in the figure are application examples of the present invention in typical industrial scenarios, and the actual optimization dimension can be adjusted to two to four dimensions according to the number of furnaces.

[0195] After the optimization is completed, two strict verification links are executed in sequence. First, it is judged whether the calculated thermal efficiency improvement meets the expectation, and whether the predicted thermal efficiency corresponding to the optimal combination reaches the preset improvement target is verified. If the judgment is yes, it enters the second judgment node to verify whether the recommended oxygen value complies with the operating procedures, that is, to verify whether the oxygen content is within the safe range allowed by the process. Only when both verifications are passed does the process enter the final execution link, generate a damper adjustment operation optimization strategy, map the optimal oxygen content combination to a specific damper opening adjustment amount, and display it on the on-site operation interface to guide the operator to execute. If any verification link returns a negative judgment, the process automatically returns to the differential evolution algorithm step to perform global optimization again, until an optimal control strategy that meets both benefit requirements and safety requirements is output.

[0196] This flowchart clearly illustrates the core technical architecture of this invention, which includes mechanism-guided feature construction, data-driven model training, global intelligent optimization, dual security verification, and physical operation implementation. The offline modeling part is detailed in Examples 2 and 3, the online optimization part is detailed in Examples 4 and 5, the dual verification part is detailed in Example 6, and the multi-furnace scenario adaptation is supplemented in Example 7.

[0197] Figure 4 A schematic diagram of an embodiment of an industrial heating furnace thermal efficiency optimization system 500 provided by the present invention is shown. Figure 4 As shown, the system 500 includes:

[0198] The operating condition data acquisition module 510 is used to acquire the current operating condition data of the industrial heating furnace, which includes the current oxygen content of each furnace chamber and non-adjustable operating condition parameters.

[0199] Specifically, this module establishes a real-time communication connection with a distributed control system (DCS) or supervisory control system (SCADA) via a data interface at the industrial site, directly reading sensor tag data located at key parts of the heating furnace. For example, it obtains the flue gas temperature through thermocouple tag numbers, the fuel gas flow rate through flow meter tag numbers, and the flue gas oxygen content through zirconia analyzer tag numbers. This module not only handles real-time data acquisition but also performs initial data quality screening, removing outliers and filling in missing values ​​in the acquired raw signals to ensure that subsequent modules receive valid operating condition data reflecting the actual operating status of the equipment. This direct coupling with on-site physical monitoring points guarantees the real-time performance and accuracy of the optimization system's sensing end.

[0200] The thermal efficiency prediction module 520 is used to input the current operating condition data into the XGBoost model for predicting the thermal efficiency of the heating furnace to obtain the current thermal efficiency prediction value. The XGBoost model is a model trained based on steady-state time period data determined by a dynamically moving time window and combined features constructed by combining flue gas heat loss with temperature difference and nonlinear relationship with oxygen content.

[0201] Specifically, this module loads a pre-trained and validated XGBoost proxy model file offline. During runtime, it receives real-time data vectors from the operating condition data acquisition module and feeds them into the model inference engine. Because the model incorporates a combination of steady-state operating condition screening and thermal equilibrium mechanisms during training (as described in Examples 2 and 3), the predicted values ​​output by the thermal efficiency prediction module are not simple statistical regression results, but rather a reliable assessment of the equipment's thermal-physical properties. This module is typically deployed on edge computing nodes or local servers to meet millisecond-level or second-level online inference response requirements, providing a high-precision fitness evaluation benchmark for subsequent optimization.

[0202] The operating condition decoupling and collaborative optimization module 530 is used to, under the premise of fixing the unadjustable operating condition parameters, use the oxygen content of each furnace as the evolution variable, call the XGBoost model to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness, and perform multi-dimensional collaborative optimization through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace.

[0203] Specifically, this module is the core computing unit of the system, encapsulating a differential evolutionary algorithm engine and operating condition decoupling logic. When performing the optimization task, this module first locks the current unadjustable operating parameters such as load and fuel flow rate as a constant background. Then, within a preset safe range, it initializes a population containing multiple candidate oxygen content combinations. During the iteration process, this module repeatedly calls the thermal efficiency prediction module to obtain the fitness value of each candidate combination, and continuously updates the population state based on evolutionary operators such as mutation, crossover, selection, and boundary repair (as described in Example 5) until it converges to the global optimum. The design of this module achieves a deep binding between the optimization algorithm and process constraints, ensuring that under conditions of multi-furnace coupling or external disturbances, the output optimal combination is always a feasible solution adapted to the current actual production conditions.

[0204] The condition judgment and strategy mapping module 540 is used to determine whether the optimal combination meets the preset optimization termination condition. If the optimization termination condition is met, the optimal combination is mapped to the control strategy of the air intake volume adjustment actuator to guide the on-site operation.

[0205] Specifically, this module constitutes the safety valve and actuator interface at the system output. It first performs dual verification on the optimal combination output by the decoupling and collaborative optimization modules: on the one hand, it verifies whether the predicted thermal efficiency has reached the expected improvement target; on the other hand, it checks whether the oxygen content in each furnace is strictly within the safe range allowed by the process operation procedures (as described in Example 6). Only when both verifications pass, the module converts the abstract optimal oxygen content value into a specific damper opening adjustment or fan speed setting value based on a preset valve characteristic curve or calibration table, and sends it to the on-site airflow adjustment actuator (such as an electric damper, frequency converter, etc.) via the control bus. The existence of this module completely bridges the last mile from digital space optimization decision-making to precise physical space execution, enabling operators to directly perform safe and efficient adjustment operations based on clearly quantified control strategies.

[0206] Through the close collaboration of the four modules mentioned above, this embodiment constructs a closed-loop optimization system integrating real-time perception, mechanism fusion prediction, operating condition decoupling optimization, and safety strategy mapping. It overcomes the limitations of manual adjustment and static models, effectively improving the thermal efficiency of the heating furnace and reducing fuel consumption and heat loss.

[0207] The parameters and steps for each module in the industrial heating furnace thermal efficiency optimization system 500 described above to achieve their respective functions can be found in the parameters and steps in the embodiments of the industrial heating furnace thermal efficiency optimization method above, and will not be repeated here.

[0208] Example 9:

[0209] like Figure 5 As shown, an electronic device 600 according to an embodiment of the present invention includes a processor 620 coupled to a memory 610. The memory 610 stores at least one computer program 630, which is loaded and executed by the processor 620 to enable the electronic device 600 to implement any of the above-mentioned methods for optimizing the thermal efficiency of industrial heating furnaces. Specifically:

[0210] The electronic device 600 can vary considerably due to differences in configuration or performance. It may include one or more processors 620 (Central Processing Units, CPUs) and one or more memories 610, wherein the one or more memories 610 store at least one computer program 630, which is loaded and executed by the one or more processors 620 to enable the electronic device 600 to implement any of the industrial heating furnace thermal efficiency optimization methods provided in the above embodiments. Of course, the electronic device 600 may also have wired or wireless network interfaces, keyboards, and input / output interfaces for input and output. The electronic device 600 may also include other components for implementing device functions, which will not be elaborated here.

[0211] Example 10:

[0212] An embodiment of the present invention provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above-described methods for optimizing the thermal efficiency of an industrial heating furnace.

[0213] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0214] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the above-described methods for optimizing the thermal efficiency of an industrial heating furnace.

[0215] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0216] Any combination of one or more computer-readable media can be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0217] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for optimizing the thermal efficiency of an industrial heating furnace, characterized in that, The method includes: Acquire the current operating condition data of the industrial heating furnace, which includes the current oxygen content of each furnace chamber and non-adjustable operating parameters; The current operating condition data is input into the XGBoost model used to predict the thermal efficiency of the heating furnace to obtain the current thermal efficiency prediction value. The XGBoost model is a model trained based on steady-state time period data determined by a dynamically moving time window and combined features constructed by combining flue gas heat loss with temperature difference and nonlinear relationship with oxygen content. Under the premise of fixing the non-adjustable operating parameters, the oxygen content of each furnace is used as the evolution variable. The XGBoost model is called to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness. Multidimensional collaborative optimization is performed through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace. Determine whether the optimal combination meets the preset optimization termination condition. If the optimization termination condition is met, then map the optimal combination to the control strategy of the air intake volume adjustment actuator to guide on-site operation.

2. The method according to claim 1, characterized in that, The training process of the XGBoost model includes: Obtain the pre-processed historical operating data of the industrial heating furnace; Based on the load fluctuation threshold or parameter variance, steady-state period data that reaches steady-state operating conditions are selected from the historical operating data; Obtain temperature and oxygen content parameters related to the flue gas heat loss of the industrial heating furnace; Based on the fact that flue gas heat loss is positively correlated with temperature difference and nonlinearly related to oxygen content, combined with the temperature parameter and the oxygen content parameter, a combination of features characterizing the magnitude of flue gas loss is determined. The thermal efficiency corresponding to the steady-state period data is calculated based on the inverse equilibrium method and used as training labels; The initial model is trained based on the steady-state period data, the combined features, and the training labels to obtain the XGBoost model, so that the XGBoost model includes historical fluctuation errors and dynamically matches the current operating conditions.

3. The method according to claim 2, characterized in that, The method for determining the combined characteristics characterizing the magnitude of smoke exhaust heat loss, based on the positive correlation between smoke exhaust heat loss and temperature difference and the non-linear relationship with oxygen content, and combining the temperature parameter and the oxygen content parameter, includes: The exhaust gas temperature and ambient temperature are obtained as the temperature parameters, and the oxygen volume fraction in the flue gas is obtained as the oxygen content parameter. Calculate the temperature difference between the exhaust gas temperature and the ambient temperature; The combined characteristics are obtained by multiplying the temperature difference with a correction term that includes the volume fraction of oxygen in the flue gas, and then multiplying by the exhaust heat loss coefficient. The combined features are used as input features for training the XGBoost model; The calculation structure of the correction term is as follows: ;in, The volume fraction of oxygen in the flue gas; This is the correction factor for flue gas heat loss.

4. The method according to claim 1, characterized in that, Under the premise of fixing the non-adjustable operating parameters, the oxygen content of each furnace is used as the evolutionary variable. The XGBoost model is called to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness. Multidimensional collaborative optimization is performed through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace, including: The current furnace load, fuel flow rate, and flue gas temperature are used as the parameters of the non-adjustable operating conditions. The oxygen content of each furnace is used as the evolutionary variable, and the population is initialized within a preset range of oxygen content values. Each individual in the population is composed of the oxygen content values ​​of each furnace. The following population evolution steps are executed iteratively until the preset termination condition of the differential evolution algorithm is met: In each generation of population evolution, for each individual in the population, its oxygen content value is concatenated with the unadjustable operating parameters to form an input feature vector, which is then input into the XGBoost model to obtain the predicted thermal efficiency as the fitness of that individual; based on the fitness, mutation, crossover, and selection operations are performed to generate the next generation of population. When the termination condition of the differential evolution algorithm is met, the optimal combination is output.

5. The method according to claim 4, characterized in that, The mutation, crossover, and selection operations based on the fitness include: Based on the fitness of each individual, the individual with the highest fitness in the current population is determined as the current optimal individual; For each individual currently being processed, a differential mutation strategy is used to generate a mutation vector. The differential mutation strategy is as follows: based on the current best individual, the differential vectors of two random individuals weighted by a random scaling factor are superimposed to obtain the mutation vector. The mutation vector is mixed with the currently processed individual in each furnace oxygen content dimension through a binomial crossover operation with crossover probability control, forming a test vector for evaluating whether to replace the current individual. Boundary repair is performed on oxygen content components that exceed the specified oxygen content range; The predicted thermal efficiency corresponding to the test vector is calculated by calling the XGBoost model and used as the fitness of the test vector; If the fitness of the experimental vector is not lower than the fitness of the individual currently being processed, then the experimental vector will be used as the individual at the corresponding position in the next generation population.

6. The method according to claim 1, characterized in that, The step of determining whether the optimal combination satisfies the preset optimization termination condition includes: The optimal combination is input into the XGBoost model to obtain the corresponding predicted thermal efficiency; Determine whether the predicted thermal efficiency reaches the preset expected thermal efficiency value; Determine whether the oxygen content in each furnace is within the preset safe range allowed by the process. If the predicted thermal efficiency reaches the expected thermal efficiency value and the oxygen content in each furnace is within the safe range, then the optimization termination condition is determined to be met.

7. The method according to claim 1, characterized in that, When multiple heating furnaces share the same flue gas system, the industrial heating furnace thermal efficiency optimization method is applied to the multiple heating furnaces as a whole to avoid data confusion and adjustment conflicts between the heating furnaces; when the flue gas systems of the multiple heating furnaces are independent, the industrial heating furnace thermal efficiency optimization method is applied to each heating furnace separately.

8. An industrial heating furnace thermal efficiency optimization system, characterized in that, include: The operating condition data acquisition module is used to acquire the current operating condition data of the industrial heating furnace, which includes the current oxygen content of each furnace chamber and non-adjustable operating condition parameters. The thermal efficiency prediction module is used to input the current operating condition data into the XGBoost model for predicting the thermal efficiency of the heating furnace, and obtain the current thermal efficiency prediction value. The XGBoost model is a model trained based on steady-state time period data determined by a dynamically moving time window and combined features constructed by combining flue gas heat loss with temperature difference and nonlinear relationship with oxygen content. The operating condition decoupling and collaborative optimization module is used to, under the premise of fixing the unadjustable operating condition parameters, use the oxygen content of each furnace as the evolution variable, call the XGBoost model to calculate the predicted thermal efficiency corresponding to each candidate oxygen content combination as the fitness, and perform multi-dimensional collaborative optimization through differential evolution algorithm to obtain the optimal combination of oxygen content in each furnace. The condition judgment and strategy mapping module is used to determine whether the optimal combination meets the preset optimization termination condition. If the optimization termination condition is met, the optimal combination is mapped to the control strategy of the air intake volume adjustment actuator to guide the on-site operation.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the industrial heating furnace thermal efficiency optimization method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer-readable storage medium to implement the industrial heating furnace thermal efficiency optimization method as described in any one of claims 1 to 7.

Images