Deep peak regulation intelligent combustion monitoring and rapid intervention control method and system for coal-fired boiler

The method for monitoring and rapid intervention control of coal-fired boiler combustion through multi-source heterogeneous sensor fusion and intelligent diagnosis solves the combustion instability and safety risks of traditional coal-fired boilers under deep peak shaving conditions, realizes multi-dimensional real-time monitoring and second-level rapid intervention, and improves the stability and safety of combustion.

CN121452559APending Publication Date: 2026-02-03CHINA DATANG CORPORATION SCIENCE AND TECHNOLOGY GENERAL RESEARCH INSTITUTE +1

Patent Information

Application Number
CN202511764498.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-27
Publication Date
2026-02-03

AI Technical Summary

Technical Problem

Traditional coal-fired boiler combustion monitoring technology lacks global perception capabilities, has lagging dynamic identification and early warning capabilities, and slow intervention mechanisms. It cannot meet the needs of multi-dimensional, real-time perception of combustion status and second-level rapid intervention under deep peak shaving conditions, resulting in increased combustion instability and safety risks in low-quality coal boilers.

Method used

An innovative technology system of multi-source heterogeneous sensor fusion, intelligent diagnosis and early warning is adopted. Through spatial redundancy temperature field reconstruction, coal quality dynamic identification and combustion state probability fusion, combined with intelligent self-feedback control of fast-injection oil gun, it realizes real-time monitoring of multi-dimensional combustion state and second-level rapid intervention, and builds a closed-loop control system based on multi-source heterogeneous sensor data.

Benefits of technology

It significantly improves the stability and safety of low-quality coal combustion, shortens the intervention response time, enhances the flexibility and safety of combustion regulation, adapts to the high volatility demand of new energy power generation systems, and strengthens the peak-shaving capacity of thermal power units.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121452559A_ABST
    Figure CN121452559A_ABST
Patent Text Reader

Abstract

The invention relates to an intelligent combustion monitoring and rapid intervention control method and system for deep peak regulation of a coal-fired boiler. The method comprises the following steps: reconstructing a two-dimensional temperature field of a hearth through regularization optimization modeling by utilizing hearth infrared measuring point data; constructing a coal quality dynamic identification model to realize dynamic monitoring and trend prediction of coal quality fluctuation; a combustion state nonlinear probability fusion model is constructed, on the basis of a probability graph model, information such as flame detection signals, hearth negative pressure, flame images, oxygen amount, steam drum water level, coke falling and coal interruption is fused, joint probability distribution is constructed, time sequence incremental learning is conducted in combination with an XGBoost model, and dynamic recognition and early warning of the combustion state are achieved. Key parameters such as a flame image, temperature field distribution and negative pressure fluctuation are collected in real time by means of a hearth multi-dimensional combustion state recognition model, a decision module is triggered when a combustion state score reaches an early warning threshold value, a quick oil gun is triggered to be put into use, and early warning in advance is changed into quick intervention.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a combustion monitoring and control technology for coal-fired boilers, and more particularly to a method and system for intelligent combustion monitoring and rapid intervention control for deep peak shaving in coal-fired boilers. Background Technology

[0002] As the proportion of new energy power generation in the power structure continues to increase, the volatility and uncontrollability of the power system have significantly increased. As the ballast stone of power supply, thermal power units, especially coal-fired boilers, need to undertake deeper peak-shaving tasks and often operate under low-load conditions. However, during deep peak-shaving, especially when burning low-quality coal, the combustion stability of boilers decreases significantly, making them prone to major safety risks such as flame flickering, flameout, furnace coking, and even unplanned shutdowns.

[0003] In terms of stable combustion intervention, traditional fuel nozzle activation control systems (such as the fuel nozzle activation / deactivation control device based on ignition detection signals disclosed in application number CN201811234567.X) have significant response delay issues. These systems typically only initiate intervention after the ignition detection signal is completely lost, with a response time of more than 10 seconds, which cannot meet the demand for second-level rapid stable combustion under deep peak shaving conditions.

[0004] Currently, traditional coal-fired boiler combustion monitoring technologies have the following main shortcomings:

[0005] 1. Limited Monitoring Dimensions and Lack of Global Perception: Traditional coal-fired boiler combustion monitoring technologies primarily rely on single-point parameter measurements, such as furnace negative pressure, flue gas oxygen content, and flame detection signals. While these parameters can reflect the boiler's operating status to some extent, they are insufficient to provide a comprehensive, multi-dimensional perception of the combustion status. Existing flame detection technologies are mostly based on energy signals such as light, heat, and radiation, which can only detect the presence or absence of a flame, but cannot provide a refined assessment of combustion quality and stability. Under deep peak-shaving conditions, the boiler combustion status becomes more complex and sensitive, and this simplified monitoring method can no longer meet the increasingly urgent demand for refined peak-shaving in thermal power generation.

[0006] 2. Lagging dynamic identification and early warning capabilities: The existing system lacks the ability to dynamically identify changes in the quality of coal entering the furnace online, and it is also unable to systematically model multi-source interference factors. When coal quality fluctuates or disturbances such as coke loss or coal shortage occur, the system often only alarms after the disturbance has already had a significant impact on combustion, resulting in insufficient advance warning.

[0007] 3. Slow response of intervention mechanism, unable to meet the second-level stable combustion requirement: In terms of stable combustion intervention, the traditional fuel nozzle activation control system has obvious response delay problems, which cannot meet the second-level response requirement for "early intervention and rapid suppression" of combustion instability under deep peak shaving conditions. The root cause lies in the passive nature of the control logic and the non-optimal design of the actuator, which makes it impossible to implement rapid and precise intervention in the early stage of combustion instability.

[0008] Furthermore, low-quality coal boilers are prone to safety risks such as unstable combustion, coking and blockage, and unplanned outages during deep peak shaving. Traditional combustion monitoring technologies mainly rely on single-point parameters, making it difficult to achieve multi-dimensional, real-time combustion status perception, thus limiting the sensitivity and safety of combustion control. Meanwhile, rapid intervention measures (such as oil gun deployment) require more intelligent control mechanisms to shorten response time. Therefore, to adapt to the increased volatility and uncontrollability of the power system brought about by the continuous increase in the proportion of renewable energy generation, it is necessary to propose an innovative technological system to effectively solve the above-mentioned technical challenges and enhance the more efficient peak shaving and backup capabilities of thermal power units in the power system. Summary of the Invention

[0009] This invention addresses the shortcomings of existing technologies by proposing a method and system for intelligent combustion monitoring and rapid intervention control of coal-fired boilers for deep peak shaving. It proposes an innovative technical system based on multi-source heterogeneous sensor fusion, intelligent diagnosis, early warning and closed-loop control. Based on spatial redundant temperature field reconstruction, dynamic coal quality identification, combustion state probabilistic fusion and multi-source heterogeneous information integration, it enables intelligent rapid oil gun control of coal-fired boilers, improving the stability, safety and peak shaving capacity of low-quality coal combustion.

[0010] The technical solution adopted in this invention is as follows:

[0011] A method for intelligent combustion monitoring and rapid intervention control of deep peak shaving in coal-fired boilers includes step S1: intelligent diagnosis and early warning of multi-dimensional combustion status and step S2: intelligent self-feedback control of boiler stable combustion and rapid oil gun injection.

[0012] Step S1, the multi-dimensional intelligent diagnosis and early warning process of combustion status, includes:

[0013] Step S1.1: Reconstruction of spatial redundancy temperature field. Using infrared measurement data from the furnace, the two-dimensional temperature field of the furnace is reconstructed through regularized optimization modeling to achieve early identification of local overheating, uneven burning, and coking, thereby enhancing the ability to diagnose combustion anomalies.

[0014] Step S1.2: Construct a dynamic coal quality identification model to realize dynamic monitoring and trend prediction of coal quality fluctuations;

[0015] Step S1.3: Construct a nonlinear probabilistic fusion model of combustion state (CPIF). Based on the probabilistic graphical model, it integrates nine types of heterogeneous multi-source interference information, including flame detection signals, furnace negative pressure, flame images, oxygen content, steam drum water level, coke drop, and coal shortage. It constructs a joint probability distribution using conditional dependencies and combines it with the XGBoost model for time-series incremental learning to achieve dynamic identification and early warning of combustion state.

[0016] The aforementioned method for intelligent combustion monitoring and rapid intervention control of coal-fired boilers with deep peak shaving, in step S2, the intelligent self-feedback control process of the boiler's stable combustion and rapid oil gun activation, relies on a multi-dimensional combustion state recognition model of the furnace to collect key parameters such as flame images, temperature field distribution, and negative pressure fluctuations in real time. When the combustion state score reaches the warning threshold, the decision module is triggered, and the system triggers the rapid oil gun activation within seconds, realizing closed-loop control from early warning to rapid intervention. When unstable combustion in the furnace is detected, the system immediately initiates rapid intervention logic, completing the opening of the oil gun solenoid valve, atomized air adjustment, and fuel supply matching within 0.5 seconds, achieving a second-level response to combustion conditions.

[0017] In the aforementioned method for deep peak shaving intelligent combustion monitoring and rapid intervention control of coal-fired boilers, step S1.2 involves constructing a dynamic coal quality identification model. This model integrates the operating parameters of the boiler's flue gas, steam-water, and pulverizing systems. Based on heat balance and the law of conservation of mass, a low-order coal quality component identification algorithm is established. An online calculation model for the coal quality entering the power plant boiler is also established to calculate parameters such as calorific value, moisture content, ash content, and sulfur content of the coal entering the boiler in real time, thereby achieving dynamic monitoring and trend prediction of coal quality fluctuations. The dynamic coal quality identification model calculates the sulfur content of the coal entering the boiler in real time based on the principle of conservation of mass; and calculates the coal inventory in the raw coal silo in real time based on the raw coal silo level calculation model, assessing the time of change in the coal quality entering the boiler.

[0018] A deep peak-shaving intelligent combustion monitoring and rapid intervention control system for coal-fired boilers based on the control method described above consists of a boiler stable combustion state monitoring and early warning module, a coal / fuel state dynamic identification and monitoring module, a two-dimensional temperature field combustion state probability fusion module, and a boiler stable combustion rapid injection oil gun intelligent self-feedback control module. The boiler stable combustion state monitoring and early warning module dynamically constructs a two-dimensional temperature field inside the furnace based on an infrared temperature measuring device, displays the furnace combustion state in real time, and provides real-time early warning for abnormal combustion states.

[0019] The dynamic identification and monitoring module for coal / fuel status establishes an online calculation model for the coal quality entering the power plant boiler based on the operating parameters of the boiler's flue gas system, steam-water system, and pulverizing system. It displays the calorific value, moisture content, and ash content in real time, and calculates the sulfur content of the coal entering the boiler in real time based on the principle of mass conservation. Based on the raw coal bunker level calculation model, it calculates the coal inventory in the raw coal bunker in real time and assesses the time of change in the coal quality entering the boiler.

[0020] The two-dimensional temperature field combustion state probabilistic fusion module is based on the state identification of the Combustion Probabilistic Inference Framework (CPIF) nonlinear fusion model of combustion factors. Starting from the complexity of the furnace combustion process and the multi-disturbance coupling mechanism, it constructs a conditional dependency network between disturbance factors based on probabilistic graphical models to achieve systematic modeling of nonlinear, multivariable, and multi-source heterogeneous features.

[0021] The intelligent self-feedback control module for the boiler's stable combustion and quick-start oil gun relies on a multi-dimensional combustion state recognition model in the furnace to collect key parameters such as flame images, temperature field distribution, and negative pressure fluctuations in real time. When the combustion state score reaches the warning threshold, the decision module is triggered, and the system triggers the quick-start oil gun to be put into use within seconds, realizing closed-loop control from early warning to rapid intervention.

[0022] The aforementioned intelligent combustion monitoring and rapid intervention control system for deep peak shaving in coal-fired boilers includes a combustion state probabilistic fusion module. Based on nine core interference factors, including flame detection signals, furnace negative pressure, flame image brightness and morphological characteristics, flue gas oxygen content, steam drum water level fluctuations, coke shedding events, coal shortage status, furnace temperature field changes, and coal quality deterioration index, a Bayesian network structure is constructed among these variables through a joint modeling mechanism. A CPIF model is then established to capture the causal relationships and information transmission paths of the nine core interference factors during the evolution of the combustion state.

[0023] The beneficial effects of this invention are:

[0024] 1. This invention proposes a method and system for deep peak-shaving intelligent combustion monitoring and rapid intervention control of coal-fired boilers based on multi-source heterogeneous sensor fusion. It constructs an innovative technical system based on multi-source heterogeneous sensor fusion, intelligent diagnosis, early warning, and closed-loop control. This system integrates deep-tuning adaptive intelligent diagnosis, early warning, and rapid intervention using multi-source heterogeneous sensor data, aiming to improve the stability, safety, and peak-shaving capacity of low-quality coal combustion. It is suitable for the coordinated dispatch of highly volatile energy sources in new power systems. This invention can be widely applied to the dispatch of highly volatile energy sources in new power systems, enhancing the peak-shaving and backup power capabilities of thermal power units.

[0025] 2. The present invention relates to a method and system for deep peak shaving intelligent combustion monitoring and rapid intervention control of coal-fired boilers. Based on the furnace temperature distribution, the method incorporates multi-dimensional interference factor coupling and multi-data source coupling to intelligently perceive the furnace combustion state. It develops a furnace temperature field measurement system based on infrared thermometry, dynamically constructs a two-dimensional temperature field inside the furnace, and displays the furnace combustion state in real time based on multiple data sources and intelligent algorithms. This achieves real-time perception of the boiler combustion state through multi-source fusion and provides real-time early warning of abnormal combustion states, ensuring the healthy, reliable and safe operation of the boiler.

[0026] 3. The present invention relates to a method and system for deep peak-shaving intelligent combustion monitoring and rapid intervention control of coal-fired boilers. Through deep fusion and intelligent analysis of multi-source heterogeneous sensor data, it significantly improves the shortcomings of traditional combustion monitoring, achieving multi-dimensional, dynamic, and accurate identification of combustion status. Combined with a rapid intervention mechanism, it effectively reduces the risk of combustion instability caused by poor coal quality, improving the boiler's operational safety, economy, and peak-shaving flexibility, and providing strong technical support for the safe and stable operation of new power systems. It has the following advantages:

[0027] 1) By combining state perception and decision-making with multi-source sensor data, the combustion state is assessed in real time and an early warning threshold is triggered; multi-source information fusion is achieved to improve the accuracy and real-time performance of combustion state identification;

[0028] 2) Employ spatial redundancy temperature reconstruction to enhance local anomaly identification and multi-dimensional monitoring capabilities;

[0029] 3) Automatically identifies coal quality change trends, effectively providing early warnings of combustion risks;

[0030] 4) The intelligent control of the fast-acting fuel nozzle significantly shortens the intervention response time and improves the flexibility and safety of combustion adjustment. Fast-acting fuel nozzle control: After the warning reaches the threshold, the system starts the fast-acting fuel nozzle in seconds. It improves the fuel injection efficiency through multi-stage expansion shear nozzles and shortens the fuel injection response time (from more than 10 seconds to less than 3 seconds) by utilizing high-speed gas-liquid crushing and multi-directional mixing combustion technology.

[0031] 5) Rapid intervention closed loop: Enables rapid intervention and closed-loop optimization of combustion instability, improving combustion safety and adjustment sensitivity. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the spatial redundancy temperature field reconstruction of the present invention;

[0033] Figure 2 The combustion state probabilistic fusion model framework of this invention;

[0034] Figure 3 The present invention illustrates the control response process of a fast-dispensing oil gun. Detailed Implementation

[0035] To make the technical concept and advantages of the invention clearer, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the following embodiments are merely preferred embodiments for explaining and illustrating the present invention, and should not be considered as, nor constitute a limitation on, the scope of patent protection claimed by the present invention.

[0036] Example 1: Intelligent Combustion Monitoring and Rapid Intervention Control Method for Deep Peak Shaving in Coal-fired Boilers

[0037] Includes step S1: intelligent diagnosis and early warning of multi-dimensional combustion status; step S2: intelligent self-feedback control of boiler stable combustion and fast oil gun.

[0038] Step S1, the multi-dimensional intelligent diagnosis and early warning process for combustion status, includes:

[0039] Step S1.1: Reconstruction of spatial redundancy temperature field. Using infrared measurement data from the furnace, the two-dimensional temperature field of the furnace is reconstructed through regularized optimization modeling to achieve early identification of local overheating, uneven burning, and coking, thereby enhancing the ability to diagnose combustion anomalies.

[0040] Step S1.2: Construct a dynamic coal quality identification model to realize dynamic monitoring and trend prediction of coal quality fluctuations;

[0041] Step S1.3: Construct a nonlinear probabilistic fusion model for combustion state (CPIF). Based on a probabilistic graphical model, it integrates nine types of heterogeneous multi-source interference information, including flame detector signals, furnace negative pressure, flame images, oxygen content, steam drum water level, coke shedding, and coal shortage. A joint probability distribution is constructed using conditional dependencies, and time-series incremental learning is performed using an XGBoost model to achieve dynamic identification and early warning of combustion state. The foregoing description can be combined with... Figures 1-3 To analyze and explain.

[0042] Example 2: Intelligent Combustion Monitoring and Rapid Intervention Control Method for Deep Peak Shaving in Coal-fired Boilers

[0043] The difference from Example 1 lies in step S2, the intelligent self-feedback control process of the boiler stable combustion quick-start oil gun. This process relies on a multi-dimensional combustion state recognition model of the furnace to collect key parameters such as flame images, temperature field distribution, and negative pressure fluctuations in real time. When the combustion state score reaches the warning threshold, the decision module is triggered, and the system triggers the quick-start oil gun within seconds, achieving closed-loop control from early warning to rapid intervention. When unstable combustion in the furnace is detected, the system immediately initiates rapid intervention logic, completing the opening of the oil gun solenoid valve, atomized air adjustment, and fuel supply matching within 0.5 seconds, achieving a second-level response to combustion conditions.

[0044] Example 3: Intelligent Combustion Monitoring and Rapid Intervention Control Method for Deep Peak Shaving in Coal-fired Boilers

[0045] The difference from Examples 1 and 2 lies in the following: In step S1.2, a dynamic coal quality identification model is constructed. This model integrates the operating parameters of the boiler's flue gas, steam-water, and pulverizing systems. Based on heat balance and the law of conservation of mass, a low-order coal quality component identification algorithm is established. An online calculation model for the coal quality entering the power plant boiler is also established. This model calculates parameters such as calorific value, moisture content, ash content, and sulfur content of the coal entering the boiler in real time, enabling dynamic monitoring and trend prediction of coal quality fluctuations. The dynamic coal quality identification model calculates the sulfur content of the coal entering the boiler in real time based on the principle of conservation of mass; and calculates the coal inventory in the raw coal silo in real time based on the raw coal silo level calculation model, assessing the time of change in the coal quality entering the boiler.

[0046] Example 4: Intelligent Combustion Monitoring and Rapid Intervention Control System for Deep Peak Shaving in Coal-fired Boilers. This system comprises a boiler stable combustion state monitoring and early warning module, a coal / fuel state dynamic identification and monitoring module, a two-dimensional temperature field combustion state probability fusion module, and a boiler stable combustion rapid injection oil gun intelligent self-feedback control module. Based on the methods of the aforementioned examples, it realizes intelligent combustion monitoring and rapid intervention control for deep peak shaving in coal-fired boilers, thereby reducing the risk of combustion instability caused by poor coal quality, improving the boiler's operational safety, economy, and peak shaving flexibility, and providing strong technical support for the safe and stable operation of new power systems.

[0047] Among them, the boiler stable combustion status monitoring and early warning module, based on an infrared temperature measuring device, dynamically constructs a two-dimensional temperature field inside the furnace, displays the combustion status of the furnace in real time, and provides real-time early warning of abnormal combustion status.

[0048] The dynamic identification and monitoring module for coal / fuel status establishes an online calculation model for the coal quality entering the power plant boiler based on the operating parameters of the boiler's flue gas system, steam-water system, and pulverizing system. It displays the calorific value, moisture content, and ash content in real time, and calculates the sulfur content of the coal entering the boiler in real time based on the principle of mass conservation. Based on the raw coal bunker level calculation model, it calculates the coal inventory in the raw coal bunker in real time and assesses the time of change in the coal quality entering the boiler.

[0049] The two-dimensional temperature field combustion state probabilistic fusion module is based on the state identification of the Combustion Probabilistic Inference Framework (CPIF) nonlinear fusion model of combustion factors. Starting from the complexity of the furnace combustion process and the multi-disturbance coupling mechanism, it constructs a conditional dependency network between disturbance factors based on probabilistic graphical models to achieve systematic modeling of nonlinear, multivariable, and multi-source heterogeneous features.

[0050] The intelligent self-feedback control module for the boiler's stable combustion and quick-start oil gun relies on a multi-dimensional combustion state recognition model in the furnace to collect key parameters such as flame images, temperature field distribution, and negative pressure fluctuations in real time. When the combustion state score reaches the warning threshold, the decision module is triggered, and the system triggers the quick-start oil gun to be put into use within seconds, realizing closed-loop control from early warning to rapid intervention.

[0051] The combustion state probability fusion module is based on nine core interference factors, including flame detection signals, furnace negative pressure, flame image brightness and morphological characteristics, flue gas oxygen content, steam drum water level fluctuations, coke shedding events, coal shortage status, furnace temperature field changes and coal quality deterioration index. Through a joint modeling mechanism, a Bayesian network structure is constructed among these variables, and a CPIF model is established to capture the causal relationships and information transmission paths of the nine core interference factors in the combustion state evolution process.

[0052] Example 5: A deep-adaptive intelligent combustion monitoring and intervention control method and system based on spatial redundancy temperature field reconstruction, dynamic coal quality identification, probabilistic fusion of combustion states, integration of multi-source heterogeneous information, and intelligent fast-acting oil gun control. This method can significantly improve the combustion stability, safety, and peak-shaving capability of boilers, including:

[0053] I. Multi-dimensional intelligent diagnosis and early warning of combustion status

[0054] 1. Spatial Redundancy Temperature Field Reconstruction (SRC-TI): Utilizing infrared measurement data from the furnace, the two-dimensional temperature field of the furnace is reconstructed through regularized optimization modeling, enabling early identification of local overheating, uneven burning, and coking, enhancing the ability to diagnose combustion anomalies, and reducing reliance on dense sensor arrangements.

[0055] 2. Coal quality dynamic identification model: Integrating parameters of boiler flue gas, steam and water, and pulverizing system, a low-order coal quality component identification algorithm is established based on heat balance and material conservation. The calorific value, moisture, ash content, sulfur content and other parameters of the coal fed into the furnace are calculated in real time to realize dynamic monitoring and trend prediction of coal quality fluctuations.

[0056] 3. Combustion State Nonlinear Probabilistic Fusion Model (CPIF): Based on the probabilistic graphical model, it integrates nine types of heterogeneous multi-source interference information, including flame detection signals, furnace negative pressure, flame images, oxygen content, steam drum water level, coke drop, and coal shortage. It constructs a joint probability distribution using conditional dependencies and combines it with the XGBoost model for time-series incremental learning to achieve dynamic identification and early warning of combustion state.

[0057] See the framework of the combustion state probabilistic fusion model. Figure 2 ,include:

[0058] Multi-source heterogeneous sensor data input;

[0059] Data preprocessing and feature engineering layers;

[0060] The probabilistic graphical model layer includes node definition, establishing conditional dependencies, and joint probability distribution calculation;

[0061] Temporal incremental learning models, including XGBoost models, online learning, and dynamic weight adjustment;

[0062] A comprehensive assessment of the combustion status is used for decision-making and early warning: if the risk probability is greater than the threshold, a rapid intervention mechanism is triggered; if the risk probability is normal, continuous monitoring continues.

[0063] Figure 2 The combustion state probability fusion model framework is as follows:

[0064] First layer: Data preprocessing and feature engineering layer

[0065] Input: Original 9 types of heterogeneous data.

[0066] 1. Flame detection signal (intensity, frequency) 2. Furnace negative pressure (mean, fluctuation amplitude) 3. Flame image (features extracted through image processing: centroid, area, flicker frequency) 4. Oxygen signal 5. Steam drum water level 6. Coke shedding signal (vibration / sound characteristics) 7. Coal shortage signal 8. From the SRC-TI module: probability of local high temperature zone, temperature distribution uniformity index 9. From the coal quality identification module: current coal calorific value deviation, moisture / ash content.

[0067] Processing procedure:

[0068] Data alignment and cleaning: Time synchronization of data with different frequencies and delays, and handling of missing and outlier values;

[0069] Feature extraction: Extracting meaningful statistical features (such as mean, variance, gradient, and spectral features) from the original signal.

[0070] Standardization: Normalize all features to the same dimension, making them easier for the model to process.

[0071] Second layer: Probabilistic graphical model layer

[0072] This is the framework's static knowledge core, encapsulating domain experts' understanding of the inherent causal relationships within combustion systems.

[0073] Node definition:

[0074] Root node (input node): Corresponds to the key features extracted from the above 9 types of data. For example: "low fire detection intensity", "large negative pressure fluctuation", "image centroid shift", "high probability of defocusing", etc.

[0075] Intermediate nodes (hidden states): These represent crucial intermediate states that cannot be directly measured. Examples include: "local flame instability," "abnormal fuel supply," and "deterioration of the aerodynamic field."

[0076] Leaf nodes (output nodes): These are the "overall combustion instability risk" that we are ultimately concerned with.

[0077] Conditional dependency:

[0078] Construct a Bayesian network by connecting nodes with directed edges. For example:

[0079] "Low calorific value of coal" → "Abnormal fuel supply"

[0080] "Abnormal fuel supply" + "Inappropriate airflow" → "Localized flame instability"

[0081] "Localized flame instability" + "High probability of charring" → "Risk of overall combustion instability"

[0082] These relationships are based on thermodynamics, combustion science, and field operating experience.

[0083] Function:

[0084] Given a set of real-time input data, the network can calculate the posterior probability of the state of all nodes;

[0085] The final output is P (overall combustion instability risk | all observation data), which is a probability value between 0 and 1. This probability value incorporates the uncertainty of all evidence.

[0086] Third layer: Temporal incremental learning engine

[0087] This is the core of the framework's dynamic adaptation, enabling the model to learn from historical data and continuously evolve.

[0088] Model platform: XGBoost (Extreme Gradient Boosting Tree). It was chosen because of its efficiency, accuracy, and support for online learning.

[0089] Input: intermediate node probabilities (such as the probability of "local flame instability") and output node probabilities calculated by the probabilistic graphical model layer; temporal features in the original features (such as trends and fluctuations over the past N seconds).

[0090] Output: A final "combustion status score" calibrated and enhanced by a machine learning model. This score is more accurate than the result from a purely probabilistic graphical model.

[0091] Incremental learning mechanism:

[0092] The system will continuously collect new data samples (including intervention results, such as whether the situation is stable after oil injection).

[0093] The parameters of the XGBoost model are updated periodically or when triggering conditions are met, using new data and an online learning approach.

[0094] Effects: Enables the model to adapt to the slow aging of boiler equipment, changes in coal type, etc.; continuously corrects potentially inaccurate prior assumptions in the probabilistic graphical model, achieving self-optimization.

[0095] Fourth layer: Decision-making and early warning layer

[0096] Input: The "burning status score" (a value of 0-100 or 0-1) output by the time-series incremental learning engine.

[0097] Decision-making logic:

[0098] Set one or more dynamic early warning thresholds (e.g., warning threshold, danger threshold);

[0099] When the score continues to exceed the threshold for a preset duration (to avoid false alarms), the system determines that the combustion has entered an unstable state.

[0100] Output:

[0101] Warning signal: sent to operations personnel;

[0102] Intervention command: Directly and instantly triggers the "fast-dispensing oil gun intelligent self-feedback control technology" to form a closed-loop control.

[0103] II. Intelligent Self-Feedback Control of Boiler Stable Combustion and Rapid Oil Gun Dosing

[0104] The intelligent self-feedback control technology for boiler stable combustion and rapid oil gun injection is an advanced combustion optimization technology that integrates real-time status perception, intelligent judgment, and rapid execution. For example... Figure 3 As shown, this invention relies on a multi-dimensional combustion state recognition model in the furnace to collect key parameters such as flame images, temperature field distribution, and negative pressure fluctuations in real time. When the combustion state score reaches the warning threshold, the decision module is triggered, and the system triggers the rapid-acting oil gun within seconds, realizing closed-loop control from early warning to rapid intervention, significantly improving the ability to prevent combustion instability. When furnace combustion instability is detected, the system immediately initiates rapid intervention logic, completing the opening of the oil gun solenoid valve, atomization air adjustment, and fuel supply matching within 0.5 seconds, achieving a second-level response to combustion conditions.

[0105] The optimized design of the quick-acting fuel injector features a three-stage swirling atomizing nozzle, refining fuel particles to below 50μm. Combined with a streamlined flame-stabilizing chamber, this creates a low-speed recirculation zone, enhancing fuel mist residence time and ensuring a 40% increase in ignition energy density. The fuel injector rod is equipped with a 1200℃ heat-resistant ceramic bushing and a reverse cooling channel, effectively solving high-temperature erosion problems and extending service life by three times. This technology intelligently predicts combustion deterioration trends and proactively injects fuel to stabilize combustion before the flame detector signal is completely lost, avoiding the risk of boiler deflagration caused by delayed manual intervention and reducing the combustion interruption accident rate by 90%.

[0106] 1. Status perception and decision-making: Combine multi-source sensor data to assess the combustion status in real time and trigger early warning thresholds.

[0107] 2. Fast-acting fuel gun control: After the warning reaches the threshold, the system starts the fast-acting fuel gun in seconds. It improves the fuel injection efficiency through multi-stage expansion shear nozzles and shortens the fuel injection response time (from more than 10 seconds to less than 3 seconds) by using high-speed gas-liquid crushing and multi-directional mixing combustion technology.

[0108] 3. Rapid intervention closed loop: Enables rapid intervention and closed-loop optimization of combustion instability, improving combustion safety and adjustment sensitivity.

[0109] Example 6: A Deep Peak Shaving Intelligent Combustion Monitoring and Rapid Intervention Control Method and System for Coal-fired Boilers Based on Multi-Source Heterogeneous Sensor Fusion. This method, based on furnace temperature distribution, incorporates multi-dimensional interference factor coupling, and multi-data source coupling to intelligently perceive the furnace combustion state. An infrared thermometry-based furnace temperature field measurement system was developed, dynamically constructing a two-dimensional temperature field within the furnace. Using multiple data sources and intelligent algorithms as the core, the system displays the furnace combustion state in real time, achieving real-time perception of the boiler combustion state through multi-source fusion and providing real-time early warnings for abnormal combustion states, ensuring the healthy, reliable, and safe operation of the boiler. The system mainly consists of boiler stable combustion state monitoring and early warning, fuel state monitoring, a two-dimensional temperature field, and a novel rapid-injection oil gun. Furthermore, the system also features historical data query and temperature field inversion functions.

[0110] 1. Spatial Redundancy Temperature Field Reconstruction

[0111] Furnace temperature is an important parameter of boiler combustion status and can promptly characterize combustion instability. By arranging multiple infrared furnace temperature measuring points on the same cross section in the burnout air area and calculating the two-dimensional temperature field based on these points, the furnace temperature distribution level can be effectively monitored and the flame deviation can be displayed in real time.

[0112] This invention presents a spatial redundancy compensation-based inverse temperature field reconstruction method (SRC-TI) and establishes a two-dimensional furnace temperature field reconstruction model. Without relying on densely distributed temperature measurement points, this method utilizes infrared online furnace temperature measurement data and constructs a regularized optimization solution process for a spatially distributed temperature field. This enables continuous reconstruction and dynamic visualization of temperature information on the furnace cross-section, effectively identifying the risks of localized overheating, uneven combustion, and coking caused by coal quality deterioration, uneven air-coal distribution, or combustion deviation.

[0113] 2. Dynamic coal quality identification model

[0114] By utilizing the operating parameters of the boiler's flue gas system, steam-water system, and pulverizing system, an online calculation model for the coal quality fed into the power plant boiler is established. This model displays the calorific value, moisture content, and ash content in real time. Based on the principle of mass conservation, the sulfur content of the coal fed into the boiler is calculated in real time. Using the raw coal bunker level calculation model, the coal inventory in the bunker is calculated in real time, and the time frame for changes in the coal quality fed into the boiler is assessed.

[0115] 3. Combustion State Probabilistic Fusion Model

[0116] At the level of combustion state identification, this invention proposes a state identification method based on the Combustion Probabilistic Inference Framework (CPIF). Starting from the complexity of the furnace combustion process and the coupling mechanism of multiple disturbances, this method constructs a conditional dependency network among disturbance factors based on probabilistic graphical models, achieving systematic modeling of nonlinear, multivariable, and multi-source heterogeneous characteristics.

[0117] The combustion state of coal-fired boilers is influenced by multiple factors, including coal quality fluctuations, operating condition disturbances, and control system lags, typically exhibiting nonlinear, non-stationary, strongly coupled, and highly uncertain characteristics. To address these challenges, the CPIF model designs a joint modeling mechanism for nine core disturbance factors, including flame detector signals, furnace negative pressure, flame image brightness and morphological characteristics, flue gas oxygen content, steam drum water level fluctuations, coke shedding events, coal shortage conditions, furnace temperature field changes, and coal quality deterioration index. The model constructs a Bayesian network structure among these variables to capture their causal relationships and information transmission paths during the evolution of the combustion state.

[0118] 4. Intelligent self-feedback control module for boiler stable combustion and rapid oil injection gun

[0119] The intelligent self-feedback control technology for boiler stable combustion and rapid oil gun activation is an integrated technology combining real-time status perception, intelligent judgment, and rapid execution. This technology relies on a multi-dimensional combustion state recognition model of the furnace to collect key parameters such as flame images, temperature field distribution, and negative pressure fluctuations in real time. When the combustion state score reaches the warning threshold, the decision module is triggered, and the system activates the rapid oil gun within seconds. This achieves closed-loop control from early warning to rapid intervention, significantly improving the ability to prevent combustion instability.

[0120] This invention relates to a deep peak-shaving intelligent combustion monitoring and rapid intervention technology and system for coal-fired boilers based on multi-source heterogeneous sensor fusion. It utilizes a furnace temperature distribution, introduces multi-dimensional interference factor coupling, and incorporates multi-data source coupling to intelligently perceive the furnace combustion state. A furnace temperature field measurement system based on infrared thermometry is developed, dynamically constructing a two-dimensional temperature field within the furnace. Using multiple data sources and intelligent algorithms as the core, the system displays the furnace combustion state in real time, achieving real-time perception of the boiler combustion state through multi-source fusion and providing real-time early warnings for abnormal combustion conditions, ensuring the healthy, reliable, and safe operation of the boiler.

Claims

1. A coal-fired boiler deep peak shaving intelligent combustion monitoring and rapid intervention control method, comprising steps S1: multi-dimensional combustion state intelligent diagnosis and early warning and step S2: boiler stable combustion fast oil gun intelligent self-feedback control; characterized in that: Step S1, the multi-dimensional combustion state intelligent diagnosis and early warning process comprises: Step S1.1: spatial redundant temperature field reconstruction, using the furnace infrared measuring point data, reconstructing the furnace two-dimensional temperature field through regularized optimization modeling, realizing early identification of local overheating, partial burning and coking, and enhancing the combustion abnormality diagnosis capability; Step S1.2: constructing a coal quality dynamic identification model to realize dynamic monitoring and trend prediction of coal quality fluctuation; Step S1.3: constructing a combustion state nonlinear probability fusion model CPIF, based on a probability graph model, fusing nine types of heterogeneous multi-source interference information such as flame detection signal, furnace negative pressure, flame image, oxygen content, steam drum water level, coking and coal cutting, constructing a joint probability distribution using conditional dependence, combining an XGBoost model for time series incremental learning, and realizing dynamic identification and early warning of the combustion state.

2. The intelligent combustion monitoring and fast intervention control method for deep load following of coal-fired boilers according to claim 1, characterized in that: The step S2, the boiler stable combustion fast oil gun intelligent self-feedback control process, relies on the furnace multi-dimensional combustion state identification model, collects key parameters such as flame image, temperature field distribution and negative pressure fluctuation in real time, triggers the decision module when the combustion state score reaches the early warning threshold, and the system triggers the fast oil gun use in seconds, realizing closed-loop control from early warning to rapid intervention.

3. The intelligent combustion monitoring and fast intervention control method for deep load following of coal-fired boilers according to claim 2, characterized in that: When the furnace combustion is unstable, the system immediately starts the rapid intervention logic, and completes the opening of the oil gun electromagnetic valve, the adjustment of the atomizing air and the matching of the fuel supply amount within 0.5 seconds, realizing the second-level response of the combustion condition.

4. The intelligent combustion monitoring and fast intervention control method for deep load following of coal-fired boilers according to claim 1, 2 or 3, characterized in that: In step S1.2, the coal quality dynamic identification model is constructed: the operation parameters of the boiler air and smoke, steam and water, and the pulverizing system are fused, a coal quality low-grade component identification algorithm is established based on heat balance and mass conservation, an online calculation model of the coal quality of the power station boiler is established, the calorific value, moisture content, ash content and sulfur content of the incoming coal are calculated in real time, and dynamic monitoring and trend prediction of coal quality fluctuation are realized.

5. The intelligent combustion monitoring and fast intervention control method for deep load following of coal-fired boilers according to claim 4, characterized in that: According to the principle of mass conservation, the coal quality dynamic identification model calculates the sulfur content of the incoming coal in real time; according to the raw coal bunker level calculation model, the raw coal bunker coal storage is calculated in real time, and the incoming coal quality change time is evaluated.

6. A coal-fired boiler deep peak shaving intelligent combustion monitoring and rapid intervention control system based on the control method of claim 1, comprising a boiler stable combustion state monitoring and early warning module, a coal quality / fuel state dynamic identification and monitoring module, a two-dimensional temperature field combustion state probability fusion module, and a boiler stable combustion fast oil gun intelligent self-feedback control module, characterized in that: The boiler stable combustion state monitoring and early warning module dynamically constructs the two-dimensional temperature field in the furnace based on the infrared temperature measuring device, realizes real-time display of the furnace combustion state, and realizes real-time early warning of the abnormal combustion state; The coal quality / fuel state dynamic identification monitoring module establishes an online calculation model of the coal quality of the power station boiler through the operation parameters of the boiler air and smoke system, steam and water system, and pulverizing system, displays the calorific value, moisture content and ash content in real time, calculates the sulfur content of the incoming coal according to the principle of mass conservation, calculates the raw coal bunker coal storage in real time according to the raw coal bunker level calculation model, and evaluates the incoming coal quality change time; The two-dimensional temperature field combustion state probability fusion module is based on a nonlinear probability fusion model of combustion factors, and is used for state recognition from the complexity and multi-disturbance coupling mechanism of the furnace combustion process. The conditional dependence network among interference factors is constructed based on a probability graph model, so as to realize systematic modeling of nonlinear, multi-variable and multi-source heterogeneous characteristics. The boiler stable combustion and fast oil gun intelligent self-feedback control module is based on the multi-dimensional combustion state recognition model of the furnace. Key parameters such as flame image, temperature field distribution and negative pressure fluctuation are collected in real time. When the combustion state score reaches the early warning threshold, the decision module is triggered. The system triggers the fast oil gun to use in seconds, realizing the closed-loop control from early warning to rapid intervention.

7. The coal-fired boiler deep peak shaving intelligent combustion monitoring and rapid intervention control system according to claim 6, characterized in that: The combustion state probability fusion module is based on nine types of core interference factors including flame detection signals, furnace negative pressure, flame image brightness and morphological characteristics, flue gas oxygen content, drum water level fluctuation, coking event, coal cutting state, furnace temperature field change and coal quality deterioration index. The Bayesian network structure among these variables is constructed through a joint modeling mechanism. The CPIF model is established to capture the causal relationship and information transmission path of the nine types of core interference factors in the combustion state evolution process.

8. The coal-fired boiler deep peak shaving intelligent combustion monitoring and rapid intervention control system according to claim 7, characterized in that: The combustion state probability fusion model framework is as follows: Multi-source heterogeneous sensor data input; Data preprocessing and feature engineering layer; Probability graph model layer, including node definition, conditional dependence relationship establishment and joint probability distribution calculation; Time series incremental learning model, including XGBoost model, online learning and dynamic weight adjustment; Combustion state comprehensive score, decision and early warning: if the risk probability is greater than the threshold, the rapid intervention mechanism is triggered; If the risk probability is normal, continuous monitoring is performed.

9. The coal-fired boiler deep peak shaving intelligent combustion monitoring and rapid intervention control system according to claim 6, 7 or 8, characterized in that: The boiler stable combustion and fast oil gun intelligent self-feedback control module is based on the multi-dimensional combustion state recognition model of the furnace. Key parameters such as flame image, temperature field distribution and negative pressure fluctuation are collected in real time. When the combustion state score reaches the early warning threshold, the decision module is triggered. The system triggers the fast oil gun to use in seconds, realizing the closed-loop control from early warning to rapid intervention; When the furnace combustion is unstable, the system immediately starts the rapid intervention logic, and completes the opening of the oil gun electromagnetic valve, the adjustment of the atomizing air and the matching of the fuel supply amount within 0.5 seconds, realizing the second-level response of the combustion condition.

10. The coal-fired boiler deep peak shaving intelligent combustion monitoring and rapid intervention control system according to claim 9, characterized in that: The fast oil gun supporting structure optimization design adopts a three-stage cyclone atomizing nozzle, which refines the fuel particle size to below 50 μm. Combined with the design of a streamline stable flame chamber, a low-speed backflow area is formed to strengthen the oil mist residence time. The oil gun rod body is additionally provided with a heat-resistant ceramic liner and a reverse cooling channel.

Citation Information

Patent Citations

  • Atomizer and atomization method thereof

    CN109303958B

Cited By

  • Thermal power generating unit deep peak regulation and combustion stabilization system and method coupled with fixed bed oxygen-enriched gasification

    CN122063910A