Boiler combustion intelligent optimization control system based on machine learning

Abstract

The invention discloses a boiler combustion intelligent optimization control system based on machine learning, and relates to the field of boiler combustion control, and the system comprises a collection module which is used for collecting fuel characteristics, hearth temperature, flue gas components, air supply amount and auxiliary machine operation parameters in a boiler combustion process, and recording the collected parameters; the extraction module is used for synchronously receiving the original acquisition parameters, preprocessing the parameters to extract core characteristic parameters representing the combustion state and standardizing the core characteristic parameters; according to the method, the future combustion state is predicted based on the time sequence coupling model, the combustion working condition can be comprehensively evaluated by combining the comprehensive evaluation index calculated by the multi-dimensional coefficient, the model continuously iteratively optimizes parameters through self-supervised learning, the combustion state is optimal and the adjustment action is milder through a dual-target optimization adjustment scheme, and the adjustment efficiency is improved. And the combustion heat efficiency and the operation stability of the boiler are greatly improved integrally.

Description

Technical Field

[0001] This invention relates to the field of boiler combustion control technology, specifically to a machine learning-based intelligent optimization control system for boiler combustion. Background Technology

[0002] Boiler combustion optimization control aims to monitor parameters such as furnace temperature, flue gas oxygen content, fuel and air volume in real time, and dynamically adjust the air-fuel ratio and air distribution strategy to improve combustion efficiency, reduce energy consumption, and reduce emissions of pollutants such as nitrogen oxides while ensuring the safe and stable operation and output of the boiler.

[0003] The invention patent application with application number 202511170516.5 discloses a combustion optimization method and control method, system, equipment and medium for coal-fired boilers based on machine learning. The application aims to solve the problems that "relying solely on hot multi-condition test measurements is labor-intensive and easily deviates from the optimal operating conditions, making it impossible to obtain the lowest fly ash carbon content; existing prediction methods have the defects of easily generating local optima and low model generalization ability".

[0004] However, existing boiler combustion control technologies are unable to fully realize the synchronous acquisition, processing and analysis of multiple parameters, nor can they accurately assess and dynamically predict the combustion status in real time. At the same time, they lack optimal control strategies that take into account both combustion performance and adjustment actions, thus making it difficult to ensure the stable, safe, efficient and intelligent operation of the boiler.

[0005] To address this, we propose a machine learning-based intelligent optimization control system for boiler combustion. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a machine learning-based intelligent optimization control system for boiler combustion, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a machine learning-based intelligent optimization control system for boiler combustion, comprising: The system comprises the following modules: a data acquisition module, an extraction module, and a feedback module. The data acquisition module collects and records data on fuel characteristics, furnace temperature, flue gas composition, air supply, and auxiliary equipment operating parameters during boiler combustion. The extraction module receives the raw data acquisition parameters, preprocesses them to extract core characteristic parameters representing the combustion state, and standardizes these core characteristic parameters. The modeling module receives the standardized core characteristic parameters, constructs a boiler combustion state prediction model based on these parameters, updates the model parameters through autonomous learning, and outputs real-time combustion state assessment results. The decision-making module combines the combustion state assessment results with preset combustion control targets to generate optimal adjustment commands for fuel supply, air ratio, and auxiliary equipment operating parameters. The adjustment module receives the optimal adjustment commands and drives the boiler fuel supply mechanism, air supply mechanism, and associated auxiliary equipment to perform corresponding actions. The feedback module monitors the actual combustion parameters and combustion conditions after boiler combustion adjustment in real time, captures adjustment deviation data, and feeds it back to the modeling module. The acquisition module is interactively connected to the extraction module via a local area network. The extraction module is interactively connected to the modeling module via a local area network. The modeling module is interactively connected to the decision-making module and the feedback module via a local area network. The decision-making module and the feedback module are interactively connected to the adjustment module via a local area network.

[0008] Furthermore, the acquisition module incorporates a multi-source sensing unit, a spatiotemporal synchronization unit, and a data archiving unit. The multi-source sensing unit includes in-situ sensing components distributed in each combustion zone of the furnace, fuel supply link sensing components, air supply system sensing components, and auxiliary machine operation sensing components. Each sensing component collects fuel characteristics, furnace temperature, flue gas composition, air supply and auxiliary machine operation parameters at the corresponding location according to a preset sampling frequency. The spatiotemporal synchronization unit adds a unified furnace combustion time stamp and spatial location code to all collected parameters. The time stamp is synchronized with the boiler's combustion cycle, and the spatial location code corresponds one-to-one with the furnace partition and equipment link location of the parameter collection. The data archiving unit stores the collected parameters in a structured manner and marks them with traceable indexes according to the order of time stamps and the hierarchical relationship of spatial location codes. Among them, fuel characteristics include the calorific value, volatile matter content, ash content, and particle size distribution parameters of the fuel fed into the furnace.

[0009] Furthermore, the extraction module preprocesses the original collected parameters, including time alignment, outlier removal, missing value completion, and noise filtering. After processing, the core feature parameters characterizing the combustion state are extracted. The timing alignment operation is based on the timing stamp of the acquired parameters, aligning all parameter sequences to a unified sampling time node; The outlier removal operation is based on the baseline fluctuation range of the operating condition range to which the parameter belongs, and parameter values ​​that exceed the preset fluctuation threshold are identified as outliers and removed. The missing value completion operation targets the missing values ​​generated after removing outliers and the missing values ​​in the original collected data. First, it determines the operating condition range and combustion cycle node to which the missing value belongs. Then, it retrieves the baseline change trend curve of the corresponding parameter within the operating condition range. Combining the effective values ​​of the parameters at two adjacent effective sampling times before and after the missing value, it calculates the missing value according to the slope of the baseline change trend curve within the time range and the missing duration, thus completing the missing value completion of the entire parameter sequence. The noise filtering operation adopts an operating condition adaptive variable order filtering method, that is, according to the reference noise characteristics of the operating condition range to which the parameter belongs, the corresponding filtering order and cutoff frequency are matched. The core feature parameters extracted by the extraction module include the average temperature and temperature gradient of each combustion zone in the furnace, the oxygen content of the flue gas, the exhaust gas temperature, the fuel feed rate, the ratio of primary air to secondary air, the furnace negative pressure, the auxiliary machine operating power, the combustion thermal efficiency, and the excess air coefficient.

[0010] Furthermore, the standardization processing of the core feature parameters by the extraction module follows the following rules: For any core feature parameter, the preprocessed measured value at the k-th sampling time Its standardized value ; In the formula: m is The operating condition interval number to which it belongs; This is the lower limit of the core feature parameter corresponding to the working condition interval numbered m; This is the upper limit of the core feature parameter corresponding to the working condition interval numbered m; This represents the change of the core feature parameter at time k relative to the previous valid sampling time. This is the baseline allowable fluctuation range of the core characteristic parameter within the operating condition interval numbered m; The extraction module determines the core feature parameters whose standardized values ​​exceed the preset valid range as invalid features and removes them, retaining only the valid core feature parameters and outputting them to the modeling module.

[0011] Furthermore, the boiler combustion state prediction model constructed by the modeling module is a time-coupled prediction model. The model takes the time sequence of core feature parameters after standardized processing at multiple consecutive sampling times as input and the boiler combustion state dimension parameters at multiple time steps within a preset future duration as output. The combustion state dimension parameters output by the model include combustion efficiency coefficient, combustion stability coefficient, and pollutant emission control coefficient. The modeling module calculates the real-time combustion state assessment result based on the combustion state dimension parameters, and the calculation formula is as follows: ; Where: I is the comprehensive evaluation index of combustion state; E is the combustion efficiency coefficient; S is the combustion stability coefficient; P is the pollutant emission control coefficient; The modeling module uses the measured combustion state dimension parameters corresponding to the actual boiler combustion conditions collected by the feedback module as the true labels for self-supervised learning, and takes the maximization of the comprehensive evaluation index of combustion state as the optimization objective. The network weights and bias parameters of the model are iteratively updated through self-supervised autonomous learning.

[0012] Furthermore, the decision module takes maximizing the comprehensive evaluation index of combustion state and minimizing the adjustment action as dual optimization objectives, constructs an optimization solution space for adjustment commands, and generates optimal adjustment commands for fuel supply, air ratio and auxiliary machine operating parameters. The decision-making module calculates the optimal adjustment increment for each type of controllable variable among fuel supply, air ratio, and auxiliary machine operating parameters. The formula is: ; In the formula: j is the number of the control quantity to be adjusted; , For control variable j, specify the upper and lower limits of its operation. A comprehensive evaluation index for the preset target combustion state; This is a comprehensive evaluation index of the combustion state at the current moment; Let j be the normalized sensitivity coefficient of the control quantity with respect to the combustion state, and its value range is [0,1]. The change in the comprehensive evaluation index of combustion state when the control quantity j, calculated by the boiler combustion state prediction model, is adjusted by a preset unit step size; The maximum absolute value of the change in the comprehensive evaluation index of combustion state when each of the control variables to be adjusted individually by a preset unit step size; The decision module superimposes the optimal adjustment increment of all control variables to be adjusted onto the current operating value of the corresponding control variable to obtain the optimal adjustment command, and the value of the optimal adjustment command does not exceed the range defined by the upper and lower operating limits of the corresponding control variable.

[0013] Furthermore, the adjustment module integrates an instruction parsing unit, a partition execution unit, and an action locking unit; The instruction parsing unit decomposes the optimal adjustment instruction into branch execution instructions for the corresponding fuel supply mechanism, air supply mechanism and each associated auxiliary machine, and the branch execution instructions are matched with the control protocol of the corresponding actuator; The partition execution unit executes the corresponding fuel supply and air supply adjustment commands for different combustion zones of the furnace. The action interlocking unit monitors the operating status of the actuator in real time. When the operating parameters of the actuator exceed the preset safety threshold, it immediately triggers the interlocking protection, stops the adjustment action of the corresponding mechanism, and maintains the current operating status.

[0014] Furthermore, the feedback module maintains a synchronized sampling frequency and time stamp with the acquisition module, collecting the actual combustion parameters and combustion conditions of the boiler after adjustment in real time. It also calculates the deviation between the predicted and actual combustion state values ​​and the deviation between the target and actual values ​​of the adjustment command in real time, obtaining two types of adjustment deviation data. When the adjustment deviation data exceeds the preset deviation threshold, it is immediately fed back to the modeling module to trigger the autonomous learning and updating of the model parameters. When the adjustment deviation data does not exceed the preset deviation threshold, it is fed back to the modeling module at a preset period for periodic iterative optimization of the model parameters.

[0015] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: In this invention, the system can accurately collect multi-dimensional parameters of the entire boiler combustion process and complete structured traceable storage. At the same time, it retains effective parameter features by combining adaptive filtering based on operating conditions. The extracted core features are more in line with the actual combustion state. The standardized processing takes into account the characteristics of the operating condition range and the parameter change trend, making the feature input more accurate. In addition, it predicts the future combustion state based on the time-series coupling model. The comprehensive evaluation index calculated by multi-dimensional coefficients can comprehensively evaluate the combustion condition. The model continuously iterates and optimizes the parameters through self-supervised learning. The dual-objective optimization adjustment scheme optimizes the combustion state and makes the adjustment action more gentle. Overall, it greatly improves the boiler combustion thermal efficiency and operational stability, effectively reduces flue gas pollutant emissions, and reduces auxiliary machine energy consumption. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0017] Figure 1 This is a schematic diagram of a machine learning-based intelligent optimization control system for boiler combustion. Figure 2 This is a schematic diagram of the system operation logic in this invention; Figure 3 This is a simplified diagram of the main structure of the boiler in this invention; The labels in the diagram represent: 1. Furnace; 2. Burner; 3. Air preheater; 4. Water-cooled wall; 5. Superheater. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be further described below with reference to embodiments.

[0020] Example: This embodiment presents a machine learning-based intelligent optimization control system for boiler combustion, such as... Figure 1 As shown, it includes: The data acquisition module is used to collect data on fuel characteristics, furnace temperature, flue gas composition, air supply, and auxiliary equipment operating parameters during the boiler combustion process, and to record the collected parameters. The acquisition module has a built-in multi-source sensing unit, a spatiotemporal synchronization unit, and a data archiving unit. The multi-source sensing unit includes in-situ sensing components distributed in each combustion zone of the furnace, fuel supply link sensing components, air supply system sensing components, and auxiliary machine operation sensing components. Each sensing component collects fuel characteristics, furnace temperature, flue gas composition, air supply and auxiliary machine operation parameters at the corresponding location according to a preset sampling frequency. The spatiotemporal synchronization unit adds a unified furnace combustion time stamp and spatial location code to all collected parameters. The time stamp is synchronized with the boiler's combustion cycle, and the spatial location code corresponds one-to-one with the furnace partition and equipment link location of the parameter collection. The data archiving unit stores the collected parameters in a structured manner and marks them with traceable indexes according to the order of time stamps and the hierarchical relationship of spatial location codes; Among them, fuel characteristics include the calorific value, volatile matter content, ash content and particle size distribution parameters of the fuel fed into the furnace; The extraction module is used to synchronously receive the raw acquisition parameters, preprocess the parameters to extract the core characteristic parameters that characterize the combustion state, and standardize the core characteristic parameters. The extraction module preprocesses the raw collected parameters, including time alignment, outlier removal, missing value completion, and noise filtering. After processing, the core feature parameters characterizing the combustion state are extracted. The timing alignment operation aligns all parameter sequences to a unified sampling time node based on the timing stamps of the acquired parameters. The outlier removal operation is based on the baseline fluctuation range of the operating condition range to which the parameter belongs. Parameter values ​​that exceed the preset fluctuation threshold are identified as outliers and removed. The missing value completion operation targets the missing values ​​generated after outlier removal and the missing values ​​in the original collected data. First, the operating condition range and combustion cycle node to which the missing value belongs are determined. The baseline change trend curve of the corresponding parameter in the operating condition range is retrieved. Combined with the effective values ​​of the parameter at two adjacent effective sampling times before and after the missing value, the missing value is calculated according to the slope of the baseline change trend curve in the time range and the missing duration, thus completing the missing value completion of the entire parameter sequence. The noise filtering operation adopts the operating condition adaptive variable order filtering method, that is, according to the reference noise characteristics of the operating condition range to which the parameter belongs, the corresponding filter order and cutoff frequency are matched. The boiler steady-state operating condition is matched with high-order low-pass filter parameters, and the boiler transition operating condition is matched with low-order adaptive filter parameters. The completed parameter sequence is filtered to remove random noise while retaining the effective parameter change characteristics corresponding to the sudden change of boiler combustion conditions. Among them, the core feature parameters extracted by the extraction module include the average temperature and temperature gradient of each combustion zone in the furnace, the oxygen content of the flue gas, the exhaust temperature, the fuel feed rate, the ratio of primary air to secondary air, the furnace negative pressure, the auxiliary machine operating power, the combustion thermal efficiency, and the excess air coefficient. The standardization processing of core feature parameters by the extraction module follows the following rules: For any core feature parameter, the preprocessed measured value at the k-th sampling time Its standardized value ; In the formula: m is The operating condition interval number to which it belongs; This is the lower limit of the core feature parameter corresponding to the working condition interval numbered m; This is the upper limit of the core feature parameter corresponding to the working condition interval numbered m; This represents the change of the core feature parameter at time k relative to the previous valid sampling time. This is the baseline allowable fluctuation range of the core characteristic parameter within the operating condition interval numbered m; The above formula completes basic normalization based on the upper and lower limits of the parameter benchmark in different operating condition ranges, which solves the problem that traditional static normalization cannot adapt to the multi-operating condition operation of boilers. At the same time, it introduces an exponential correction term that combines the parameter change and the allowable fluctuation range of the operating condition benchmark to reflect the real-time dynamic change characteristics of the parameters. Then, invalid features that exceed the effective range are removed, so that the core feature parameters of the output fit the actual combustion process of the boiler, thereby providing more adaptable feature inputs for subsequent prediction models. The extraction module determines the core feature parameters whose standardized values ​​exceed the preset valid range as invalid features and removes them, retaining only the valid core feature parameters and outputting them to the modeling module; Among them, the operating condition intervals for each number and their corresponding baseline allowable fluctuation ranges are all preset, and the operating condition intervals are divided based on steady-state operating conditions and transitional operating conditions. The modeling module receives the standardized core feature parameters, builds a boiler combustion state prediction model based on the core feature parameters, updates the model parameters through autonomous learning, and outputs real-time combustion state assessment results. The boiler combustion state prediction model constructed by the modeling module is a time-coupled prediction model. The model takes the time sequence of core feature parameters after standardization of multiple consecutive sampling times as input and the boiler combustion state dimension parameters of multiple time steps within a preset future duration as output. The combustion state dimension parameters output by the model include combustion efficiency coefficient, combustion stability coefficient, and pollutant emission control coefficient. The combustion efficiency coefficient is the normalized value of the ratio of the boiler's real-time thermal efficiency to the rated operating condition design thermal efficiency. The combustion stability coefficient is the inverse normalized value of the real-time fluctuation of the furnace combustion condition. The pollutant emission control coefficient is the inverse normalized value of the ratio of the real-time emission concentration of boiler flue gas pollutants to the emission standard limit. The values ​​of the three types of coefficients are all in the range of [0,1]. The closer the value is to 1, the better the combustion performance of the corresponding dimension. The modeling module calculates real-time combustion state assessment results based on combustion state dimension parameters, and the calculation formula is as follows: ; Where: I is the comprehensive evaluation index of combustion state; E is the combustion efficiency coefficient; S is the combustion stability coefficient; P is the pollutant emission control coefficient; The above formula integrates the product of three types of coefficients—combustion efficiency, stability, and pollutant emission control—in the form of a cube root, reflecting the synergistic relationship between the three core objectives of boiler combustion. At the same time, through the correction term of the sum of squares of the deviations of the three types of coefficients, it avoids the situation where one dimension is optimal while other dimensions are unbalanced, forcing balanced development of each dimension, thereby more scientifically and accurately reflecting the comprehensive performance of boiler combustion. The modeling module uses the measured combustion state dimension parameters corresponding to the actual boiler combustion conditions collected by the feedback module as the true labels for self-supervised learning, and takes the maximization of the comprehensive evaluation index of combustion state as the optimization objective. The network weights and bias parameters of the model are iteratively updated through self-supervised autonomous learning. The decision-making module is used to combine the combustion state assessment results and generate optimal adjustment instructions for fuel supply, air ratio and auxiliary machine operating parameters based on preset combustion control targets. The decision-making module takes maximizing the comprehensive evaluation index of combustion state and minimizing the adjustment action as dual optimization objectives, constructs the optimization solution space of adjustment commands, and generates the optimal adjustment commands for fuel supply, air ratio and auxiliary machine operating parameters. The decision module calculates the optimal adjustment increment for each type of controllable variable among fuel supply, air ratio, and auxiliary machine operating parameters. The formula is: ; In the formula: j is the number of the control quantity to be adjusted; , For control variable j, specify the upper and lower limits of its operation. A comprehensive evaluation index for the preset target combustion state; This is a comprehensive evaluation index of the combustion state at the current moment; Let j be the normalized sensitivity coefficient of the control quantity with respect to the combustion state, and its value range is [0,1]. The change in the comprehensive evaluation index of combustion state when the control quantity j, calculated by the boiler combustion state prediction model, is adjusted by a preset unit step size; The maximum absolute value of the change in the comprehensive evaluation index of combustion state when each of the control variables to be adjusted individually by a preset unit step size; The above formula uses the difference between the upper and lower limits of the control quantity as the base, and combines the difference between the target and the current evaluation index to match the adjustment base amplitude. It introduces a normalized sensitivity coefficient to distinguish the degree of influence of different control quantities on the combustion state, and applies the saturation characteristics of the hyperbolic tangent function for nonlinear correction. At the same time, it combines the maximum change of all control quantities for normalization, which avoids the problem of adjustment overshoot and makes the adjustment increment of different control quantities comparable. It accurately calculates the adjustment increment while taking into account the improvement of combustion state and the minimization of adjustment action. The decision module superimposes the optimal adjustment increment of all control variables to be adjusted onto the current operating value of the corresponding control variable to obtain the optimal adjustment command, and the value of the optimal adjustment command does not exceed the range defined by the upper and lower operating limits of the corresponding control variable. The adjustment module is used to receive optimal adjustment commands and drive the boiler fuel supply mechanism, air supply mechanism and related auxiliary machines to perform corresponding actions. The adjustment module integrates an instruction parsing unit, a partition execution unit, and an action locking unit; The instruction parsing unit breaks down the optimal adjustment instruction into branch execution instructions for the corresponding fuel supply mechanism, air supply mechanism and each associated auxiliary machine. The branch execution instructions are matched with the control protocols of the corresponding actuators. The zone execution unit executes the corresponding fuel supply and air supply adjustment commands for different combustion zones in the furnace. The action interlocking unit monitors the operating status of the actuator in real time. When the operating parameters of the actuator exceed the preset safety threshold, it immediately triggers the interlocking protection, stops the adjustment action of the corresponding mechanism, and maintains the current operating status. The feedback module is used to monitor the actual combustion parameters and combustion conditions of the boiler after combustion adjustment in real time, capture adjustment deviation data and feed it back to the modeling module; The feedback module maintains a synchronized sampling frequency and time stamp with the acquisition module, collecting the actual combustion parameters and combustion conditions of the boiler after adjustment in real time. It also calculates the deviation between the predicted and actual values ​​of the combustion state and the deviation between the target value and the actual value of the adjustment command in real time, obtaining two types of adjustment deviation data. When the adjustment deviation data exceeds the preset deviation threshold, it is immediately fed back to the modeling module to trigger the autonomous learning and updating of the model parameters. When the adjustment deviation data does not exceed the preset deviation threshold, it is fed back to the modeling module at a preset period for periodic iterative optimization of the model parameters. Each module in the system is equipped with a unified timing coordination controller. The timing coordination controller synchronously triggers the corresponding operations of each module according to the boiler's combustion cycle, so that the data transmission and processing of each module are consistent in timing. The data acquisition module is interconnected with the extraction module via a local area network. The extraction module is interconnected with the modeling module via a local area network. The modeling module is interconnected with the decision-making module and the feedback module via a local area network. The decision-making module and the feedback module are interconnected with the adjustment module via a local area network.

[0021] In this embodiment, the acquisition module collects fuel characteristics, furnace temperature, flue gas composition, air supply, and auxiliary machine operating parameters during the boiler combustion process and records the collected parameters. The extraction module simultaneously receives the raw collected parameters, preprocesses them to extract core characteristic parameters representing the combustion state, and standardizes the core characteristic parameters. The modeling module then receives the standardized core characteristic parameters, constructs a boiler combustion state prediction model based on the core characteristic parameters, and updates the model parameters through autonomous learning, outputting real-time combustion state evaluation results. The decision module further combines the combustion state evaluation results and generates optimal adjustment instructions for fuel supply, air ratio, and auxiliary machine operating parameters according to preset combustion control targets. The adjustment module then receives the optimal adjustment instructions and drives the boiler fuel supply mechanism, air supply mechanism, and associated auxiliary machines to perform corresponding actions. Finally, the feedback module monitors the actual combustion parameters and combustion conditions after boiler combustion adjustment in real time, captures adjustment deviation data, and feeds it back to the modeling module.

[0022] In the above embodiments, the system captures key parameters of the entire combustion process, accurately extracts and standardizes core features, and relies on a time-series coupling model to achieve accurate prediction and comprehensive evaluation of the combustion state. Based on this, the optimal adjustment scheme is generated and executed precisely. In practical applications, the boiler combustion efficiency, stability and pollutant control level can be improved simultaneously, the adjustment range can be reduced, the operating risks of the actuator can be avoided, and the overall operating efficiency of the combustion condition can be improved.

[0023] See Figure 3 As shown in the figure, this figure further illustrates the application subject of the system and its basic structural assembly in this embodiment.

[0024] Application example: To improve the combustion thermal efficiency of the main coal-fired boiler, reduce flue gas pollutant emissions, and stabilize combustion conditions, the 300MW coal-fired power plant in XX District introduced this system, which is applied to the entire process combustion control of the boiler. The various modules of the system are interconnected through a local area network and complete various operations under the synchronous triggering of a unified timing and coordination controller, thereby realizing intelligent optimization of boiler combustion.

[0025] During system operation, the acquisition module first uses multi-source sensing units deployed in each combustion zone of the furnace, fuel supply link, air supply system and auxiliary equipment to collect fuel characteristics such as calorific value and volatile matter content of the coal entering the furnace, flue gas composition such as temperature of each zone of the furnace and oxygen content, and auxiliary equipment operating parameters such as air supply and power of forced and induced draft fans, according to a preset sampling frequency. The spatiotemporal synchronization unit adds a time stamp synchronized with the boiler combustion cycle and a spatial location code corresponding to the acquisition location to all acquired parameters. The data archiving unit completes the structured storage and traceable index marking of the parameters according to the time stamp and coding hierarchy.

[0026] After receiving the original collected parameters synchronously, the extraction module sequentially completes the preprocessing operations of time alignment, outlier removal, missing value completion, and noise filtering to extract core feature parameters such as the average temperature and temperature gradient of each combustion zone in the furnace, flue gas oxygen content, excess air coefficient, and combustion thermal efficiency. Subsequently, according to the boiler steady-state and transitional operating conditions, the core feature parameters are standardized, invalid features are removed, and a set of standardized valid core feature parameters is formed and output to the modeling module.

[0027] After receiving the parameter set, the modeling module uses the constructed temporal coupling prediction model, taking the time sequence of core feature parameters from multiple consecutive sampling times as input, to output the combustion efficiency coefficient, combustion stability coefficient, and pollutant emission control coefficient within a preset time period in the future. The calculated comprehensive evaluation index of the current boiler combustion state is 0.71. The model uses the measured combustion state dimension parameters collected by the feedback module as the true labels for self-supervised learning, and takes maximizing the comprehensive evaluation index of the combustion state as the optimization objective, continuously learning autonomously and updating the model parameters.

[0028] The decision-making module aims to maximize the comprehensive combustion state evaluation index and minimize adjustment actions. For the control variables such as fuel feed rate, primary and secondary air ratio, and auxiliary machine operating power, it calculates the optimal adjustment increment for each control variable. After superimposing the increments onto the current operating values ​​of each control variable, it generates the optimal adjustment command. All command values ​​are within the upper and lower limits of the corresponding control variable's operating range. Finally, it determines the adjustment requirements of slightly reducing the fuel feed rate by 2%, adjusting the primary air to secondary air ratio to 1:2.5, and slightly increasing the induced draft fan operating power by 1%.

[0029] The instruction parsing unit of the adjustment module decomposes the optimal adjustment instruction into branch execution instructions that match the control protocols of each actuator. The zone execution unit executes the corresponding fuel supply and air supply adjustment actions for different combustion zones in the furnace. The action interlocking unit monitors the operating status of each actuator in real time throughout the process. The operating parameters of each mechanism do not exceed the preset safety threshold, and the interlocking protection is not triggered. All adjustment actions are executed smoothly.

[0030] The feedback module collects the actual combustion parameters and operating conditions of the boiler after adjustment in real time according to the sampling frequency and time stamp synchronized with the acquisition module. It calculates the deviation between the predicted value and the actual value of the combustion state, and the deviation between the target value and the actual value of the adjustment command. Since neither type of deviation data exceeds the preset deviation threshold, the deviation data is fed back to the modeling module according to the system's preset cycle for periodic iterative optimization of the model parameters.

[0031] After the system was applied, the combustion thermal efficiency of the coal-fired boiler increased by 1.9%, the emission concentration of flue gas pollutants such as nitrogen oxides and sulfur dioxide decreased by about 13%, the real-time fluctuation of the furnace combustion conditions decreased significantly, and the energy consumption of auxiliary equipment such as forced and induced draft fans was also reasonably controlled, thus fully achieving the preset goal of intelligent optimization control of boiler combustion.

[0032] In summary, the system in the above embodiments can accurately collect multi-dimensional parameters of the entire boiler combustion process and complete structured traceable storage. At the same time, it retains effective parameter features by combining adaptive filtering based on operating conditions. The extracted core features are more in line with the actual combustion state. The standardized processing takes into account the characteristics of the operating condition range and the parameter change trend, making the feature input more accurate. In addition, it predicts the future combustion state based on the time-series coupling model. The comprehensive evaluation index calculated by combining multi-dimensional coefficients can comprehensively evaluate the combustion condition. Moreover, the model continuously iterates and optimizes the parameters through self-supervised learning. The dual-objective optimization adjustment scheme not only optimizes the combustion state but also makes the adjustment action more gentle. Overall, it greatly improves the boiler combustion thermal efficiency and operational stability, effectively reduces flue gas pollutant emissions, and reduces auxiliary machine energy consumption.

[0033] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine learning-based intelligent optimization control system for boiler combustion, characterized in that, include: The data acquisition module is used to collect data on fuel characteristics, furnace temperature, flue gas composition, air supply, and auxiliary equipment operating parameters during the boiler combustion process, and to record the collected parameters. The extraction module is used to synchronously receive the raw acquisition parameters, preprocess the parameters to extract the core characteristic parameters that characterize the combustion state, and standardize the core characteristic parameters. The modeling module receives the standardized core feature parameters, builds a boiler combustion state prediction model based on the core feature parameters, updates the model parameters through autonomous learning, and outputs real-time combustion state assessment results. The decision-making module is used to combine the combustion state assessment results and generate optimal adjustment instructions for fuel supply, air ratio and auxiliary machine operating parameters based on preset combustion control targets. The adjustment module is used to receive optimal adjustment commands and drive the boiler fuel supply mechanism, air supply mechanism and related auxiliary machines to perform corresponding actions. The feedback module is used to monitor the actual combustion parameters and combustion conditions of the boiler after combustion adjustment in real time, capture adjustment deviation data, and feed it back to the modeling module.

2. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The acquisition module has a built-in multi-source sensing unit, a spatiotemporal synchronization unit, and a data archiving unit. The multi-source sensing unit includes in-situ sensing components distributed in each combustion zone of the furnace, fuel supply link sensing components, air supply system sensing components, and auxiliary machine operation sensing components. Each sensing component collects fuel characteristics, furnace temperature, flue gas composition, air supply and auxiliary machine operation parameters at the corresponding location according to a preset sampling frequency. The spatiotemporal synchronization unit adds a unified furnace combustion time stamp and spatial location code to all collected parameters. The time stamp is synchronized with the boiler's combustion cycle, and the spatial location code corresponds one-to-one with the furnace partition and equipment link location of the parameter collection. The data archiving unit stores the collected parameters in a structured manner and marks them with traceable indexes according to the order of time stamps and the hierarchical relationship of spatial location codes. Among them, fuel characteristics include the calorific value, volatile matter content, ash content, and particle size distribution parameters of the fuel fed into the furnace.

3. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The extraction module preprocesses the original collected parameters, including time alignment, outlier removal, missing value completion, and noise filtering. After processing, the core feature parameters characterizing the combustion state are extracted. The timing alignment operation is based on the timing stamp of the acquired parameters, aligning all parameter sequences to a unified sampling time node; The outlier removal operation is based on the baseline fluctuation range of the operating condition range to which the parameter belongs, and parameter values ​​that exceed the preset fluctuation threshold are identified as outliers and removed. The missing value completion operation targets the missing values ​​generated after removing outliers and the missing values ​​in the original collected data. First, it determines the operating condition range and combustion cycle node to which the missing value belongs. Then, it retrieves the baseline change trend curve of the corresponding parameter within the operating condition range. Combining the effective values ​​of the parameters at two adjacent effective sampling times before and after the missing value, it calculates the missing value according to the slope of the baseline change trend curve within the time range and the missing duration, thus completing the missing value completion of the entire parameter sequence. The noise filtering operation adopts an operating condition adaptive variable order filtering method, that is, according to the reference noise characteristics of the operating condition range to which the parameter belongs, the corresponding filtering order and cutoff frequency are matched. The core feature parameters extracted by the extraction module include the average temperature and temperature gradient of each combustion zone in the furnace, the oxygen content of the flue gas, the exhaust gas temperature, the fuel feed rate, the ratio of primary air to secondary air, the furnace negative pressure, the auxiliary machine operating power, the combustion thermal efficiency, and the excess air coefficient.

4. The intelligent optimization control system for boiler combustion based on machine learning according to claim 3, characterized in that, The standardization processing of the core feature parameters by the extraction module follows the following rules: For any core feature parameter, the preprocessed measured value at the k-th sampling time Its standardized value ; In the formula: m is The operating condition interval number to which it belongs; This is the lower limit of the core feature parameter corresponding to the working condition interval numbered m; This is the upper limit of the core feature parameter corresponding to the working condition interval numbered m; This represents the change of the core feature parameter at time k relative to the previous valid sampling time. This is the baseline allowable fluctuation range of the core characteristic parameter within the operating condition interval numbered m; The extraction module determines the core feature parameters whose standardized values ​​exceed the preset valid range as invalid features and removes them, retaining only the valid core feature parameters and outputting them to the modeling module.

5. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The boiler combustion state prediction model constructed by the modeling module is a time-coupled prediction model. The model takes the time sequence of core feature parameters after standardized processing at multiple consecutive sampling times as input and the boiler combustion state dimension parameters at multiple time steps within a preset future duration as output. The combustion state dimension parameters output by the model include combustion efficiency coefficient, combustion stability coefficient, and pollutant emission control coefficient. The modeling module calculates the real-time combustion state assessment result based on the combustion state dimension parameters, and the calculation formula is as follows: ； Where: I is the comprehensive evaluation index of combustion state; E is the combustion efficiency coefficient; S is the combustion stability coefficient; P is the pollutant emission control coefficient; The modeling module uses the measured combustion state dimension parameters corresponding to the actual boiler combustion conditions collected by the feedback module as the true labels for self-supervised learning, and takes the maximization of the comprehensive evaluation index of combustion state as the optimization objective. The network weights and bias parameters of the model are iteratively updated through self-supervised autonomous learning.

6. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The decision module takes maximizing the comprehensive evaluation index of combustion state and minimizing the adjustment action as dual optimization objectives, constructs an optimization solution space for adjustment commands, and generates the optimal adjustment commands for fuel supply, air ratio and auxiliary machine operating parameters. The decision-making module calculates the optimal adjustment increment for each type of controllable variable among fuel supply, air ratio, and auxiliary machine operating parameters. The formula is: ； In the formula: j is the number of the control quantity to be adjusted; , For control variable j, specify the upper and lower limits of its operation. A comprehensive evaluation index for the preset target combustion state; This is a comprehensive evaluation index of the combustion state at the current moment; Let j be the normalized sensitivity coefficient of the control quantity with respect to the combustion state, and its value range is [0,1]. The change in the comprehensive evaluation index of combustion state when the control quantity j, calculated by the boiler combustion state prediction model, is adjusted by a preset unit step size; The maximum absolute value of the change in the comprehensive evaluation index of combustion state when each of the control variables to be adjusted individually by a preset unit step size; The decision module superimposes the optimal adjustment increment of all control variables to be adjusted onto the current operating value of the corresponding control variable to obtain the optimal adjustment command, and the value of the optimal adjustment command does not exceed the range defined by the upper and lower operating limits of the corresponding control variable.

7. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The adjustment module is integrated with an instruction parsing unit, a partition execution unit, and an action locking unit; The instruction parsing unit decomposes the optimal adjustment instruction into branch execution instructions for the corresponding fuel supply mechanism, air supply mechanism and each associated auxiliary machine, and the branch execution instructions are matched with the control protocol of the corresponding actuator; The partition execution unit executes the corresponding fuel supply and air supply adjustment commands for different combustion zones of the furnace. The action interlocking unit monitors the operating status of the actuator in real time. When the operating parameters of the actuator exceed the preset safety threshold, it immediately triggers the interlocking protection, stops the adjustment action of the corresponding mechanism, and maintains the current operating status.

8. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The feedback module maintains a synchronized sampling frequency and time stamp with the acquisition module, and collects the actual combustion parameters and combustion conditions of the boiler after adjustment in real time. It also calculates the deviation between the predicted value and the actual value of the combustion state and the deviation between the target value and the actual value of the adjustment command in real time, obtaining two types of adjustment deviation data. When the adjustment deviation data exceeds the preset deviation threshold, the adjustment deviation data is immediately fed back to the modeling module to trigger the autonomous learning and updating of the model parameters. When the adjustment deviation data does not exceed the preset deviation threshold, the adjustment deviation data is fed back to the modeling module according to the preset cycle for periodic iterative optimization of the model parameters.

9. The intelligent optimization control system for boiler combustion based on machine learning according to claim 1, characterized in that, The acquisition module is interactively connected to the extraction module via a local area network. The extraction module is interactively connected to the modeling module via a local area network. The modeling module is interactively connected to the decision-making module and the feedback module via a local area network. The decision-making module and the feedback module are interactively connected to the adjustment module via a local area network.

Images