Deep peak shaving and combustion stabilizing system and method for power generating unit coupled with fixed bed oxygen-rich gasification

Abstract

This invention discloses a deep peak-shaving and combustion stabilization system and method for thermal power units coupled with fixed-bed oxygen-enriched gasification, relating to the field of intelligent control technology for thermal power units. The method includes: S1, pre-constructing an AI model library containing a flame state recognition model and a unit load-gasification intensity linkage control model; S2, real-time acquisition of automatic power generation control load commands, multispectral flame images, unit operating parameters, and gasifier operating parameter data, inputting these into the flame state recognition model to output flame state and stability indices; S3, calling the linkage control model and inputting real-time operating data to calculate the syngas calorific value compensation; S4, using the flame stability index as feedback, correcting the feedforward control quantity through a model predictive controller to dynamically adjust the gasifier operating parameters; S5, generating an exit trajectory when exit conditions are met. This invention achieves intelligent control of thermal power units, reduces operating costs, improves flame monitoring accuracy and control response speed, and ensures safe and economical unit operation.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for thermal power units, specifically to a deep peak shaving and combustion stabilization system and method for thermal power units coupled with fixed-bed oxygen enrichment. Background Technology

[0002] With the large-scale grid connection of new energy sources such as wind power and photovoltaics, the power system has placed higher demands on the peak-shaving flexibility of thermal power units. Thermal power units have transformed from traditional "baseload power sources" to "regulating power sources," requiring frequent participation in deep peak shaving (i.e., ultra-low load operation). However, existing thermal power units face many technical challenges in the process of deep peak shaving:

[0003] First, under ultra-low load (<30% of rated load), the furnace temperature drops significantly, making it more difficult for pulverized coal to ignite, resulting in extremely poor combustion stability and a high risk of boiler flameout, seriously threatening the unit's operational safety. Second, traditional combustion stabilization methods mainly rely on oil injection for combustion, which is not only costly but also prone to secondary problems such as air preheater blockage. Plasma ignition technology is poorly adapted to low-quality coal with high ash and low volatile matter content, making it difficult to meet the combustion stabilization requirements under complex coal conditions. Third, there is a lack of low-cost combustion stabilization heat sources that dynamically match the boiler load. Existing heat source supply methods are slow to respond and cannot quickly adapt to frequent load fluctuations. Fourth, traditional flame monitoring technologies mostly rely on single-spectrum or radiation energy signals, resulting in insufficient monitoring accuracy and weak anti-interference capabilities. They are unable to accurately capture early signs of flame instability, leading to delayed control strategy adjustments and further exacerbating combustion instability.

[0004] The aforementioned problems severely restrict the deep peak-shaving capacity and renewable energy consumption efficiency of thermal power units. There is an urgent need to develop a deep peak-shaving and stable combustion technology that combines safety, economy and flexibility to solve the problem of stable combustion under ultra-low load and promote the large-scale implementation of flexible transformation of thermal power plants. Summary of the Invention

[0005] In view of the technical defects of existing thermal power units in deep peak shaving, such as poor combustion stability at ultra-low load, high cost and weak adaptability of traditional combustion-supporting methods, lack of low-cost stable combustion heat sources with dynamic matching, insufficient flame monitoring accuracy and control lag, the core objective of this invention is to provide a deep peak shaving and stable combustion system and control method for thermal power units coupled with fixed-bed oxygen-enriched gasification.

[0006] Specific objectives include: 1) Converting low-quality coal or coal-based solid waste into high-temperature syngas through fixed-bed oxygen-enrichment technology, providing an "on-demand" stable combustion heat source for ultra-low load furnaces, replacing traditional oil-assisted combustion, and reducing stable combustion costs; 2) Employing industrial-grade multispectral flame imaging technology and AI algorithms to achieve high-precision, real-time identification of flame status and accurately capture signs of combustion instability; 3) Constructing an MPC-based "unit load-gasification intensity" linkage control model, achieving multi-objective optimization through objective functions, realizing dynamic matching between gasifier operating parameters and boiler load, and improving the timeliness and robustness of stable combustion control by combining feedforward control and feedback correction; 4) Reducing the difficulty of retrofitting existing thermal power units through modular design, achieving safe, stable, and economical operation under 20%~30% ultra-low rated load, and providing technical support for the consumption of new energy.

[0007] This invention provides a deep peak-shaving and stable combustion control method for thermal power units with coupled fixed-bed oxygen-enriched combustion, comprising the following steps:

[0008] Step S1: Pre-construction and offline training of AI model library. Pre-construct an AI model library containing a flame state recognition model and a unit load-gasification intensity linkage control model based on model predictive control. After offline training and verification, it can be called and updated online.

[0009] Step S2: Real-time monitoring and stable combustion triggering: Real-time acquisition of automatic power generation control load commands, multispectral flame images, unit operating parameters and gasifier operating parameter data; input flame state recognition model to output flame state and stability indicators; start micro fixed bed oxygen enrichment subsystem when triggering conditions are met.

[0010] Step S3: Calculate the feedforward control quantity, call the linkage control model, input real-time operating data to calculate the calorific value compensation of syngas, and issue instructions to control the micro fixed bed oxygen enrichment subsystem to produce high-temperature syngas at the preset temperature.

[0011] Step S4: AI-based feedback correction and rolling optimization, using flame stability index as feedback, the model predicts the controller to correct the feedforward control quantity and dynamically adjust the gasifier operating parameters.

[0012] Step S5: Intelligent exit mechanism. When the exit conditions are met, the exit trajectory is generated by the unit load-gasification intensity linkage control model based on model predictive control, and the micro fixed bed oxygen enrichment subsystem is controlled to standby smoothly.

[0013] Preferably, in step S1, the flame state recognition model adopts a three-layer backpropagation neural network structure. The input is an 11-dimensional fusion feature filtered by partial least squares, and the output is the probability of four types of flame states. The training uses the cross-entropy loss function and the scale conjugate gradient algorithm, with a learning rate of 0.001 and 2000 iterations or a loss of less than 10.-7 Stop when the time comes.

[0014] Preferably, the multi-objective optimization function of the unit load-gasification intensity linkage control model based on model predictive control in step S1 is:

[0015] ;

[0016] in, To optimize the objective function value; This refers to the combustion stability index predicted by the model. The target value for combustion stability index; This is for auxiliary energy consumption of the gasification system; Set the load value for automatic generation control; This represents the actual active power of the generating unit. , , These are the weighting coefficients for each optimization objective, which are dynamically adjusted according to the power plant's operational needs.

[0017] Preferably, step S1 further includes a multispectral flame image preprocessing process, specifically including: region of interest cropping, adaptive median filtering for noise reduction, multispectral channel fusion, adaptive threshold segmentation, and flame contour extraction.

[0018] Preferably, step S1 further includes feature extraction of the flame state recognition model, specifically including 12-dimensional basic features, 7-dimensional Hu invariant moment features, and 20-dimensional Zernike moment features, for a total of 39 original features.

[0019] Preferably, the triggering condition in step S2 is any one of the following: the automatic power generation control load command is lower than 35% of the rated load; the flame state is moderate or severe instability; the stability index exceeds the preset range and lasts for ≥2s.

[0020] Preferably, in step S3, the input of the linkage control model includes automatic power generation control load commands, real-time coal quality data and furnace combustion status, and the output includes coal feed rate of the gasifier, oxygen enrichment flow rate adjustment, and syngas flow rate and calorific value matching the furnace stable combustion requirements.

[0021] Preferably, the operation of the micro fixed-bed oxygen enrichment subsystem in step S3 needs to meet the following requirements: gasifier temperature ≥ 800℃, oxygen-to-coal ratio ∈ [0.3, 0.8], and furnace main combustion zone temperature ≥ 850℃.

[0022] Preferably, the exit conditions in step S5 must be met simultaneously: the automatic power generation control load recovers to more than 35% of the rated load and lasts for ≥5s; the flame state is stable and the indicators meet the standards for ≥10s; and the temperature of the main combustion zone of the furnace is ≥1100℃.

[0023] This invention also provides a deep peak-shaving and combustion stabilization system for thermal power units coupled with fixed-bed oxygen-enriched gasification, which has a modular architecture and includes:

[0024] The furnace combustion subsystem is the main body of the pulverized coal boiler, equipped with a pulverized coal burner, temperature / pressure sensor, furnace radiation energy detection device, and pulverized coal feeding adjustment unit. It collects combustion status parameters and connects them to the AI ​​intelligent control subsystem.

[0025] The micro fixed-bed oxygen enrichment gasification subsystem is bypassed to the boiler side and integrated with the flue gas system. It includes a coal feeding unit, a gasifier body, an oxygen enrichment supply unit, a waste heat preheating unit, and a syngas conveying unit. The waste heat preheating unit is connected to the secondary air duct, the oxygen enrichment supply unit is connected to the power plant oxygen production station, and the syngas conveying unit is connected to the multi-stage stable combustion nozzle. The execution unit of this subsystem is connected to the AI ​​intelligent control subsystem.

[0026] The multi-source data acquisition subsystem includes a multispectral flame imaging unit, a unit operating parameter acquisition unit, and a gasifier operating parameter acquisition unit. It is adapted to the industrial environment of thermal power plants and is connected to the AI ​​intelligent control subsystem.

[0027] The AI ​​intelligent control subsystem is equipped with an industrial-grade controller, a data interaction module, and a pre-trained AI model library. It seamlessly connects with the multi-source data acquisition subsystem, the micro fixed-bed oxygen enrichment subsystem, the multi-stage stable combustion nozzle assembly, and the power plant distributed control system to achieve data interaction and command issuance.

[0028] The multi-stage stable combustion nozzle assembly is integrated into the pulverized coal burner and adopts a coaxial nested structure. From the inside out, it consists of a high-temperature syngas center nozzle, a dense-phase pulverized coal ring channel, and a secondary air ring channel, which are respectively connected to the syngas delivery unit, the pulverized coal supply pipeline, and the secondary air pipeline.

[0029] The deep peak-shaving and stable combustion system and control method for thermal power units with coupled fixed-bed oxygen-enriched combustion of the present invention have the following advantages compared with the prior art:

[0030] (1) Significantly reduced stable combustion costs and greatly improved raw material adaptability: This invention utilizes waste heat from power plants and by-product oxygen from oxygen production stations to convert low-quality coal, coal slime, and other coal-based solid wastes into ≥ The high-temperature syngas produced eliminates the need for additional high-quality fuel, completely replacing the traditional oil-fueled combustion method and significantly reducing the operating costs of deep peak shaving. At the same time, the gasification system is highly adaptable to low-quality coal with high ash content and low volatile matter, as well as coal-based solid waste, solving the coal quality compatibility problem of traditional plasma ignition technology and realizing the resource utilization of power plant solid waste.

[0031] (2) Strong industrial applicability and high operational reliability: The multispectral flame imaging unit adopted in this invention is adapted to the extreme field environment of thermal power plants with high temperature, high dust and strong vibration. The spectrum covers 400~1700nm. Through anti-dust blowing and sealed heat insulation design, lens contamination and equipment damage are effectively avoided, solving the problem that precision measurement technology is difficult to implement in industry. Combined with preprocessing algorithms such as adaptive median filtering and multispectral channel fusion, the anti-interference ability of flame images is further improved, providing high-quality data support for subsequent multi-dimensional feature extraction (39 dimensions) and state recognition.

[0032] (3) AI intelligent control closed loop, with both stable combustion accuracy and response speed: The flame state recognition model constructed in this invention achieves accurate recognition of four types of flame states through multi-dimensional extraction of basic features, Hu invariant moment features, Zernike moment features and PLS feature fusion, combined with BPNN algorithm (11-32-4 network structure), with an average recognition accuracy of 97.4%, which can accurately capture early signs of combustion instability; The MPC-based “unit load-gasification intensity” linkage control model adopts a rolling time domain optimization strategy, updates the control sequence every 500 ms, and combines feedforward control and feedback correction. It not only ensures the accuracy of load matching through multi-objective optimization of objective function, but also responds quickly to changes in operating conditions, ensuring that furnace combustion remains stable under ultra-low load of 20%~30%.

[0033] (4) Modular design, low modification difficulty and strong promotion: The micro fixed bed oxygen enrichment subsystem of the present invention adopts a bypass arrangement and is seamlessly integrated with the existing boiler flue gas system, with minimal modification to the boiler body; the AI ​​intelligent control subsystem can be directly connected to the existing DCS system of the power plant without large-scale reconstruction of the control system, with low modification cost and short construction period, and is suitable for deep peak-shaving flexibility modification of various pulverized coal boilers, with broad industrial promotion value.

[0034] (5) Multi-objective optimization balance, taking into account both safety and economy: The MPC linkage control model of this invention takes "combustion stability - gasification economy - load tracking" as multiple optimization objectives, and uses weighting coefficients to achieve the balance. , , The priorities of each objective are dynamically adjusted to minimize the auxiliary energy consumption of the gasification system while ensuring stable combustion in the furnace and preventing flameout accidents. At the same time, it ensures that the unit load tracking meets the grid AGC control requirements, thus achieving an organic unity of safety, flexibility and economy in the deep peak shaving process of thermal power units. Attached Figure Description

[0035] Figure 1 This is a system structure diagram of the deep peak shaving and combustion stabilization system for thermal power units with coupled fixed-bed oxygen enrichment according to the present invention;

[0036] Figure 2 This is a flowchart of the deep peak shaving and stable combustion control method for thermal power units with coupled fixed-bed oxygen enrichment according to the present invention.

[0037] Figure 3 This is a flowchart of step S1 of the deep peak shaving and stable combustion control method for thermal power units with coupled fixed-bed oxygen enrichment according to the present invention.

[0038] Figure 4 This is a schematic diagram of the flame state identification confusion matrix of the deep peak shaving and stable combustion control method for thermal power units with coupled fixed-bed oxygen enrichment according to the present invention. Detailed Implementation

[0039] The following detailed implementation of the deep peak shaving and stable combustion system and control method for thermal power units with coupled fixed-bed oxygen enrichment is described in detail with reference to specific embodiments. The embodiments of the present invention are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0040] To address the shortcomings of existing deep peak-shaving systems for thermal power units, such as poor furnace combustion stability under ultra-low loads, high costs of traditional oil-assisted combustion, poor adaptability of plasma ignition to low-quality coal, lack of low-cost combustion stabilization methods for dynamic matching with boiler load, and insufficient accuracy and strong control lag of traditional flame monitoring technologies, this invention provides a deep peak-shaving and combustion stabilization system and control method for thermal power units coupled with fixed-bed oxygen-enriched gasification. This invention achieves stable furnace combustion under ultra-low loads using high-temperature syngas generated by fixed-bed oxygen-enriched gasification. Simultaneously, it achieves high-precision real-time identification of flame status based on industrial-grade multispectral flame imaging technology and neural network algorithms. Combined with model predictive control (MPC), it realizes dynamic linkage optimization between unit load and gasification intensity, ensuring safe, stable, and economical operation of thermal power units at 20%~30% of rated load, and solving the core pain point of flexible retrofitting of thermal power plants under the background of new energy consumption.

[0041] Example 1: Implementation of a deep peak shaving and combustion stabilization system for thermal power units coupled with fixed-bed oxygen-enriched gasification.

[0042] Combined with appendix Figure 1 As shown, the present invention provides a deep peak shaving and stable combustion system for thermal power units with coupled fixed-bed oxygen-enriched gasification.

[0043] A deep peak-shaving and combustion stabilization system for thermal power units coupled with fixed-bed oxygen-enriched gasification, featuring a modular architecture, includes:

[0044] The furnace combustion subsystem is the main body of the pulverized coal boiler, equipped with a pulverized coal burner, temperature / pressure sensor, furnace radiation energy detection device, and pulverized coal feeding adjustment unit. It collects combustion status parameters and connects them to the AI ​​intelligent control subsystem.

[0045] The micro fixed-bed oxygen enrichment gasification subsystem is bypassed to the boiler side and integrated with the flue gas system. It includes a coal feeding unit, a gasifier body, an oxygen enrichment supply unit, a waste heat preheating unit, and a syngas conveying unit. The waste heat preheating unit is connected to the secondary air duct, the oxygen enrichment supply unit is connected to the power plant oxygen production station, and the syngas conveying unit is connected to the multi-stage stable combustion nozzle. The execution unit of this subsystem is connected to the AI ​​intelligent control subsystem.

[0046] The multi-source data acquisition subsystem includes a multispectral flame imaging unit, a unit operating parameter acquisition unit, and a gasifier operating parameter acquisition unit. It is adapted to the industrial environment of thermal power plants and is connected to the AI ​​intelligent control subsystem.

[0047] The AI ​​intelligent control subsystem is equipped with an industrial-grade controller, a data interaction module, and a pre-trained AI model library. It seamlessly connects with the multi-source data acquisition subsystem, the micro fixed-bed oxygen enrichment subsystem, the multi-stage stable combustion nozzle assembly, and the power plant distributed control system to achieve data interaction and command issuance.

[0048] The multi-stage combustion-stabilizing nozzle assembly, integrated within the pulverized coal burner, employs a coaxial nested structure. From the inside out, it consists of a high-temperature syngas central nozzle, a dense-phase pulverized coal annular channel, and a secondary air annular channel, respectively connected to the syngas delivery unit, pulverized coal supply pipeline, and secondary air pipeline. Specific structural details are as follows:

[0049] This system adopts a modular architecture, with interconnected signals between modules. It is adaptable to the retrofitting and integration of existing equipment in thermal power plants. Specifically, it includes a furnace combustion subsystem, a micro fixed-bed oxygen enrichment subsystem, a multi-source data acquisition subsystem, an artificial intelligence (AI) intelligent control subsystem, and a multi-stage combustion stabilization nozzle assembly. The structure and connection relationships of each component are as follows:

[0050] Furnace Combustion Subsystem: This is the main body of the pulverized coal boiler in the thermal power unit. It is equipped with a pulverized coal burner, furnace temperature / pressure sensor, furnace radiation energy detection device, and pulverized coal feeding adjustment unit. This subsystem is the core working unit of the thermal power unit, realizing the combustion of pulverized coal and the conversion of electrical energy. At the same time, it collects basic combustion state parameters of the furnace, providing underlying data support for stable combustion control. All its detection devices are connected to the AI ​​intelligent control subsystem.

[0051] Miniature fixed-bed oxygen-enriched gasification subsystem: The bypass is located on the side of the pulverized coal boiler and is seamlessly integrated with the boiler flue gas system. It serves as the heat source supply unit for the stable combustion system. It includes a coal feeding unit, a fixed-bed gasifier body, an oxygen-enriched supply unit, a waste heat preheating unit, and a syngas conveying unit connected in sequence. The waste heat preheating unit is connected to the boiler's secondary air duct to obtain waste heat, the oxygen-enriched supply unit is connected to the by-product oxygen from the power plant's oxygen production station, and the output end of the syngas conveying unit is connected to the multi-stage stable combustion nozzle assembly. The execution units of the entire subsystem are all connected to the AI ​​intelligent control subsystem to receive its control commands and provide feedback on operating parameters.

[0052] Multi-stage stable combustion nozzle assembly: Integrated into the pulverized coal burner of the furnace combustion subsystem, it is the core component for the mixed combustion of high-temperature syngas and pulverized coal. It adopts a coaxial nested gas / pulverized coal composite structure, consisting of a high-temperature syngas central nozzle, a dense-phase pulverized coal ring channel, and a secondary air ring channel from the inside out. The high-temperature syngas central nozzle is sealed to the syngas delivery unit, and the dense-phase pulverized coal ring channel and the secondary air ring channel are connected to the boiler pulverized coal supply pipeline and the secondary air pipeline, respectively.

[0053] Multi-source data acquisition subsystem: This is the sensing end of the stable combustion system, which is connected to the AI ​​intelligent control subsystem. It includes a multispectral flame imaging unit, a unit operating parameter acquisition unit, and a gasifier operating parameter acquisition unit. All acquisition units are adapted to the high-temperature, high-dust, and strong-vibration industrial environment of thermal power plants, and have the characteristics of anti-interference and high reliability, realizing the real-time acquisition and transmission of various operating data and flame images.

[0054] AI Intelligent Control Subsystem: This subsystem serves as the core for decision-making and control of the stable combustion system. It is equipped with an industrial-grade controller, a data interaction module, and a pre-trained AI model library. Through the data interaction module, this subsystem seamlessly connects with the multi-source data acquisition subsystem, the micro fixed-bed oxygen enrichment subsystem, the multi-stage stable combustion nozzle assembly, and the power plant's distributed control system (DCS) to achieve real-time data interaction and precise issuance of control commands. It completes flame status identification, stable combustion trigger judgment, control quantity calculation, closed-loop optimization, and smooth exit control of the gasification subsystem.

[0055] Example 2: Implementation of a deep peak shaving and stable combustion control method for thermal power units with coupled fixed-bed oxygen-enriched gasification.

[0056] Combined with appendix Figures 2-4 As shown, this invention provides a deep peak shaving and stable combustion control method for thermal power units with coupled fixed-bed oxygen-enriched combustion. Based on the aforementioned deep peak shaving and stable combustion system for thermal power units with coupled fixed-bed oxygen-enriched combustion, it follows a control logic of "real-time monitoring - intelligent judgment - dynamic adjustment - closed-loop optimization," constructing a complete control closed loop of "perception - decision-making - execution." Specifically, it includes steps S1 to S5, with smooth transitions between steps, adapting to the dynamic operating conditions of deep peak shaving in thermal power units. The method includes the following steps:

[0057] Step S1: Pre-construction and offline training of the AI ​​model library.

[0058] In step S1, the flame state recognition model and the "unit load-gasification intensity" linkage control model based on model predictive control (MPC) are constructed, trained offline, and verified in advance, forming an AI model library that can be called online and supports online updates. The model training data are all from the actual operation and test data of the target thermal power unit, ensuring the industrial applicability and robustness of the model.

[0059] In step S1, an AI model library is pre-built, which includes a flame state recognition model and a unit load-gasification intensity linkage control model based on model predictive control. After offline training and verification, it can be called and updated online.

[0060] In step S1, the flame state recognition model employs a three-layer backpropagation neural network structure. The input is an 11-dimensional fusion feature filtered by partial least squares, and the output is the probability of four flame states. Training uses a cross-entropy loss function and a scale conjugate gradient algorithm with a learning rate of 0.001 and 2000 iterations or a loss of less than 10. -7 Stop when the time comes.

[0061] Step S1 also includes a multispectral flame image preprocessing process, specifically including: region of interest cropping, adaptive median filtering for noise reduction, multispectral channel fusion, adaptive threshold segmentation, and flame contour extraction.

[0062] Step S1 also includes feature extraction from the flame state recognition model, specifically including 12-dimensional basic features, 7-dimensional Huinstein moment features, and 20-dimensional Zernike moment features, for a total of 39 original features. The specific process is as follows:

[0063] Step S11: Construction and offline training of the flame state recognition model.

[0064] This model is built based on multispectral flame image fusion features and backpropagation neural network (BPNN). Its core function is to realize real-time quantitative identification of the combustion state of the furnace flame. It outputs four types of flame states: stable combustion, mild instability, moderate instability, and severe instability, as well as the corresponding quantitative indicators of combustion stability, providing a reliable decision basis for stable combustion control. The specific construction and training process is from step S111 to step S115.

[0065] Step S111: Construction of the training dataset.

[0066] (1) Conduct deep peak shaving steady-state / dynamic tests at 20%~100% rated load for the target thermal power unit, and simultaneously collect multispectral flame images of the main combustion zone of the furnace, furnace operating parameters and combustion stability calibration results;

[0067] (2) The collected multispectral flame images are labeled into four categories according to the flame combustion state: stable combustion, mild instability, moderate instability, and severe instability, and the original flame image dataset is constructed.

[0068] (3) The original flame image dataset is divided into training set and test set in a ratio of 8:2 using stratified sampling method to ensure that the distribution of samples of each type of flame state is consistent in the training set and test set, and to avoid model overfitting.

[0069] Step S112: Multispectral flame image preprocessing.

[0070] A unified preprocessing operation is performed on all multispectral flame images in both the training and test sets to adapt to the interference characteristics of industrial field images, such as dust, noise, and haze, thereby improving the effectiveness of feature extraction. Specific operations include:

[0071] (1) Region of Interest (ROI) cropping: The flame image area of ​​the main combustion zone of the furnace is cropped, and irrelevant background areas such as boiler wall and observation hole are removed to obtain the ROI for flame feature analysis;

[0072] (2) Industrial noise suppression: The ROI image is filtered by an adaptive median filtering algorithm to remove bright spots and salt and pepper noise caused by dust, electromagnetic interference, and equipment vibration, while preserving the edge and shape features of the flame.

[0073] (3) Multispectral channel fusion: Weighted fusion of images from different band channels of a multispectral camera is performed to enhance the effective features of the flame, such as brightness, temperature distribution, and outline, thereby improving the image's recognizability.

[0074] (4) Adaptive threshold segmentation: The fused image is binarized and segmented using an adaptive threshold method based on the local gray-level mean, effectively separating the flame region from the background region, where the pixel value of the flame region is 1 and the pixel value of the background region is 0.

[0075] (5) Flame contour extraction: The Canny edge detection operator is used to extract the flame boundary of the binarized image to obtain the flame contour features, which lays the foundation for subsequent geometric feature extraction.

[0076] Step S113: Multi-dimensional flame feature extraction.

[0077] For the preprocessed multispectral flame image, three categories of feature parameters totaling 39 dimensions are extracted: basic features, Hu invariant moment features, and Zernike moment features. This achieves dimensionality reduction and effective information mining of high-dimensional image data. All features have clear physical meaning or morphological representation capabilities, adapting to the identification needs of industrial flame states. The specific extraction rules for the three types of features are as follows:

[0078] (1) Basic features: There are 12 dimensions in total, which characterize the combustion intensity, geometric shape and spatial distribution of the flame. They are the core basic features for flame state identification. The definitions and calculation formulas of each feature are as follows:

[0079] ① Vertical range of flame The vertical distance from the upper wall of the furnace cavity to the bottom of the flame represents the longitudinal distribution range of the flame.

[0080] ② Flame peak position The horizontal and vertical coordinates corresponding to the maximum grayscale values ​​of pixels in a multispectral fusion image represent the spatial location of the point with the highest flame intensity.

[0081] ③ Flame centroid coordinates The spatial distribution center of the flame as a whole is represented by the following formula:

[0082] ;

[0083] In the formula, , These are the x and y coordinates of the pixels within the image; For pixels grayscale value; The x-coordinate of the centroid of the flame image; The ordinate of the centroid of the flame image;

[0084] ④ Flame area The total number of pixels in the flame region of the binarized image, representing the scale of the flame combustion;

[0085] ⑤ Flame circumference The total pixel length of the flame outline boundary, after being converted into actual physical length, represents the boundary scale of the flame.

[0086] ⑥ Circumference ratio The ratio of flame circumference to area represents the complexity of the flame's shape. The formula for calculation is:

[0087] ;

[0088] ⑦ Equivalent circle radius The radius corresponding to the circle with the same circumference as the flame is calculated using the following formula:

[0089] ;

[0090] ⑧ Aspect Ratio The ratio of the length of the flame region along the airflow direction to its length in the vertical direction characterizes the flame's extension shape.

[0091] 9. Roundness : Characterizes the similarity between the flame area and a circle, calculated using the following formula:

[0092] ;

[0093] ⑩ Total signal strength The sum of the grayscale values ​​of all pixels in the image represents the overall heat release and combustion intensity of the flame. The calculation formula is:

[0094] ;

[0095] (2) Hu invariant moment features: 7 dimensions in total, translation, rotation, and scale invariant features constructed based on image moments, effectively characterizing the geometric morphology of the flame, unaffected by slight flame displacement or morphological scaling. The calculation process is as follows:

[0096] ①For Digital images, defined First-order origin moment With central moment The calculation formula is:

[0097] ;

[0098] , , ;

[0099] In the formula, The number of pixel rows in a digital image; The number of pixel columns in a digital image; x-axis powers of; : Vertical axis powers of; : First-order origin moment; : The first central moment; Pixels in a digital image grayscale value; : The x-coordinate of a pixel within a digital image; : The vertical coordinate of a pixel within a digital image; The x-coordinate of the gray-level centroid of a digital image, calculated as follows: ; The ordinate of the gray-level centroid of a digital image, calculated as follows: ; Image grayscale quality; First-order moment at the origin, used to calculate the abscissa of the centroid of the image gray level; : First-order origin moment, used to calculate the ordinate of the centroid of the image grayscale.

[0100] ② To eliminate the effects of scale variations, a normalized central moment is defined. and normalization coefficient :

[0101] ;

[0102] , ;

[0103] In the formula, Normalized central moments; : Normalization coefficient, calculated as follows ; zeroth order origin moment of Power of 1 It represents the total grayscale quality that characterizes the scale and spatial dimensions of flame combustion. It is used as a whole to normalize the central moment to eliminate the interference of scale changes in the flame image.

[0104] ③ Based on the derivation of the second-order and third-order normalized central moments, seven Hu invariant moments are obtained, realizing a robust characterization of the flame morphology.

[0105] (3) Zernike moment features: 20 dimensions in total, features constructed based on orthogonal polynomials within the unit circle, possessing rotational invariance and strong noise resistance, accurately characterizing the fine morphology and texture features of flames, and adaptable to flame feature extraction under industrial noise interference; the first 7 Zernike moments are extracted, totaling 20 effective feature parameters, calculated as follows:

[0106] ;

[0107] In the formula, : Step The Zernike moment is the Zernike moment feature parameter of the flame image; : The radial order of the Zernike moment; Angular frequency of the Zernike moment; Pi (π) : The x-coordinate of a pixel within the image; : The ordinate of a pixel within the image; Pixels in an image grayscale value; : The conjugate of the Zernike polynomial; : The distance from the origin of the coordinate system to the pixel in the image The distance; : With images The angle between axes.

[0108] Step S114: Feature selection and fusion based on partial least squares (PLS).

[0109] To reduce the redundancy of the 39-dimensional original features and improve the model's computational efficiency and recognition accuracy, partial least squares (PLS) is used to filter and fuse the original features, fully considering the correlation between the features and the flame state category. The specific steps are as follows:

[0110] (1) Data standardization: Z-score standardization is performed on all 39-dimensional feature data to make the feature data conform to Normal distribution eliminates dimensional differences between different characteristics;

[0111] (2) PLS iterative calculation: To reduce the influence of random factors, the PLS calculation is repeated 100 times; before each calculation, random samples are taken. A subset of samples is constructed, and PLS regression is performed based on this subset to extract the latent variables with the largest covariance with the flame state category, and the top 10 features with the highest projective importance (VIP) coefficients are retained.

[0112] (3) Feature importance statistics: Count the frequency of each feature in 100 PLS calculations, calculate the feature importance and sort them in descending order;

[0113] (4) Optimal fusion feature selection: retain features with importance greater than 50 to obtain 11-dimensional fusion features as input to the BPNN recognition model. This fusion feature retains the core discriminative information of the flame state and minimizes feature redundancy.

[0114] Step S115: Construction, training, and validation of the BPNN recognition model

[0115] A single-hidden-layer BPNN recognition model is constructed to achieve multi-class classification of flame states. The model structure and training parameters are adapted to the real-time and stability requirements of industrial online control. The specific process is as follows:

[0116] Step S1151 Model Structure Construction: A three-layer feedforward network structure is adopted, including an input layer, a hidden layer, and an output layer. Inter-layer connections are fully connected, while intra-layer connections are not. The node settings and neuron output rules for each layer are as follows:

[0117] (1) Input layer: The number of nodes is 11, corresponding to the 11-dimensional fusion features obtained in step S114, which are responsible for receiving feature data and passing it to the hidden layer;

[0118] (2) Hidden layer: Single hidden layer, with the number of nodes optimized to 32, and the activation function is Sigmoid function, which balances non-linear fitting ability and computational efficiency;

[0119] (3) Output layer: The number of nodes is 4, corresponding to four flame states: stable combustion, mild instability, moderate instability, and severe instability. The activation function is the Softmax function, which outputs the probability values ​​of each flame state. The Softmax function calculation formula is as follows:

[0120] ;

[0121] The output value is The probability between the two categories is such that the sum of the probabilities of all categories is 1.

[0122] In the formula, Softmax The Softmax activation function is used to convert the raw scores of the neural network output layer into probability values ​​corresponding to various flame states. : No. The original output score for the flame-like state; To traverse all When the first category is... The unnormalized raw log score of the neural network output for each category; : Total number of flame status categories, with a value of 4; : Natural exponential function; : The traversal index of the flame state category.

[0123] (4) The formula for calculating the output of a single neuron is:

[0124] ;

[0125] In the formula, Neuron output value; Neuron activation function; Neuron input dimension; : No. The weights corresponding to each input; : No. One input value; Neuron bias; : Traversal index of the neuron's input dimension.

[0126] Step S1152, Model training parameter settings.

[0127] (1) Loss function: The cross-entropy loss function is adopted to adapt to the error calculation of multi-classification tasks;

[0128] (2) Training algorithm: The scale conjugate gradient algorithm is adopted to improve the convergence speed and training stability of the model;

[0129] (3) Learning rate: set to 0.001 to balance training efficiency and model convergence accuracy;

[0130] (4) Training stopping condition: The number of iterations reaches 2000, or the loss function value is lower than 10. -7 Training will stop if any condition is met.

[0131] (5) Validation method: 100 repeated training and testing were conducted to reduce random errors, and the average recognition accuracy and recognition result stability were used as the performance evaluation indicators of the model.

[0132] Step S1153, Model training and validation.

[0133] (1) The BPNN recognition model is trained in a supervised manner using training set data, with 11-dimensional fusion features as input and flame state category labels as output;

[0134] (2) After training, the model is validated and its parameters are optimized using test set data. Ultimately, the model’s recognition accuracy and real-time performance meet the requirements of online monitoring and closed-loop control of thermal power units, and can accurately capture signs of instability in the furnace flame.

[0135] Step S12: Construction and offline training of the MPC-based “unit load-gasification intensity” linkage control model.

[0136] This model is built based on the Model Predictive Control (MPC) algorithm. Its core function is to dynamically match the gasification intensity of the gasifier with the unit load. It uses combustion stability and overall system energy consumption as dual optimization objectives, providing feedforward control for adjusting gasifier operating parameters. The model combines predictive, rolling, and constraining characteristics, adapting to the dynamic operating conditions of deep peak shaving in thermal power units. The specific construction and training process is as follows: steps S121 to S123.

[0137] Step S121, define the model input and output.

[0138] (1) Input variables: Automatic generator control (AGC) load command, real-time coal quality industrial analysis data, current operating parameters of gasifier, real-time combustion stability quantitative index of furnace, and furnace radiation energy signal;

[0139] (2) Output variables: coal feed rate adjustment, oxygen enrichment flow rate adjustment, oxygen-coal ratio setting, and syngas injection flow rate setting.

[0140] Step S122: Optimize the objective and constraints.

[0141] Step S1221, set a multi-objective optimization function: taking into account furnace combustion stability, minimizing auxiliary energy consumption of the gasification system, and unit load tracking accuracy, the optimization function is:

[0142] ;

[0143] In the formula, To optimize the objective function value; This refers to the combustion stability index predicted by the model. The target value for combustion stability index; This is for auxiliary energy consumption of the gasification system; Set the AGC load value; This represents the actual active power of the generating unit. , , These are the weighting coefficients for each optimization objective, which can be dynamically adjusted according to the power plant's operational needs.

[0144] Model-predicted combustion stability index The method for determining the value is as follows: the predicted value output by the model is compared with the set target combustion stability index. Calculate the norm deviation, weighted The weighted average is incorporated into the loss function, and its value is constrained to approach the set target during optimization; auxiliary energy consumption of the gasification system. The method for determining the value is as follows: directly extract the auxiliary energy consumption value of the actual operation of the gasification system, and then weight it. After weighting, it is included as an independent term in the loss function, thereby achieving the constraint of minimizing auxiliary energy consumption during optimization.

[0145] Step S1222: Set constraints: Based on the equipment safety and operating characteristics of the gasifier and boiler, set hard constraints to ensure the safety of system operation. The specific constraints are as follows:

[0146] ① Gasifier furnace temperature ② Oxygen-to-coal ratio in the gasifier ③ Coal feed rate to the gasifier ,in ④ Rated coal feed rate of the gasifier; ⑤ Syngas injection flow rate ,in This is the rated injection flow rate of syngas.

[0147] Step S123, Model training and online update mechanism.

[0148] (1) Offline training: Collect historical operation data of deep peak shaving of target thermal power units, including boiler combustion data, gasifier operation data, coal quality data and load response data under different load conditions, construct training dataset, and complete offline identification and parameter optimization of MPC prediction model;

[0149] (2) Online update: When the model is running online, a rolling time domain optimization strategy is adopted. The prediction time domain is set to 10 and the control time domain is set to 3. The control sequence is updated every 500ms. The model is corrected online based on the real-time feedback of combustion status and operating parameters, which effectively compensates for model errors and dynamic changes in on-site working conditions, and ensures control accuracy.

[0150] Step S2: Real-time monitoring and stable combustion triggering.

[0151] In step S2, the automatic power generation control load command, multispectral flame image, unit operating parameters and gasifier operating parameter data are collected in real time. The flame state recognition model is input and the flame state and stability index are output. When the triggering conditions are met, the micro fixed bed oxygen enrichment subsystem is started.

[0152] The triggering condition in step S2 is any of the following: the automatic power generation control load command is below 35% of the rated load; the flame state is moderate or severe instability; the stability index exceeds the preset range and lasts for ≥2 seconds. The specific process is as follows:

[0153] (1) The AI ​​intelligent control subsystem collects the unit's AGC load command, furnace multispectral flame image, unit operating parameters, and gasifier operating parameters in real time and synchronously through the multi-source data acquisition subsystem;

[0154] (2) Input the real-time acquired multispectral flame image into the flame state recognition model pre-trained in step S11. The model outputs the current furnace flame state category and the corresponding combustion stability quantification index in real time.

[0155] (3) When any of the following triggering conditions are met, the AI ​​intelligent control subsystem automatically triggers the start of the micro fixed bed oxygen enrichment subsystem and enters the deep peak shaving and stable combustion control process:

[0156] The unit's AGC load command is lower than the preset deep peak shaving threshold, which is 35% of the rated load; the flame status identification model outputs a flame status of moderate instability or severe instability; the flame stability quantification index exceeds the preset stability range and lasts for more than 2 seconds.

[0157] Step S3: Calculate the feedforward control quantity.

[0158] In step S3, the linkage control model is invoked, real-time operating data is input to calculate the calorific value compensation of the syngas, and instructions are issued to control the micro fixed bed oxygen enrichment subsystem to produce high-temperature syngas at the preset temperature.

[0159] In step S3, the linkage control model inputs include automatic power generation control load commands, real-time coal quality data, and furnace combustion status, and outputs the coal feed rate of the gasifier, the oxygen enrichment flow rate adjustment, and the syngas flow rate and calorific value matching the furnace combustion stability requirements.

[0160] In step S3, the micro fixed-bed oxygen-enriched gasification subsystem must meet the following requirements for operation: gasifier temperature ≥ 800℃, oxygen-to-coal ratio ∈ [0.3, 0.8], and main combustion zone temperature ≥ 850℃. The specific process is as follows:

[0161] (1) The AI ​​intelligent control subsystem calls the unit load-gasification intensity linkage control model pre-trained in step S12, takes the current unit AGC load command, real-time coal quality data and furnace combustion status as input, and combines the heat load demand of the boiler under ultra-low load to accurately calculate the high temperature syngas calorific value compensation required to maintain stable combustion in the furnace under the current working condition.

[0162] (2) Based on the calorific value compensation, the unit load-gasification intensity linkage control model generates feedforward control commands to determine the target coal feed rate, target oxygen enrichment flow rate, and target oxygen-coal ratio of the gasifier;

[0163] (3) The AI ​​intelligent control subsystem sends the feedforward control command to the coal feeding unit and oxygen supply unit of the micro fixed bed oxygen enrichment subsystem, controls its rapid response, and produces high-temperature syngas with a temperature ≥800℃, ensuring that the flow rate and calorific value of the syngas accurately match the current stable combustion requirements of the furnace.

[0164] Step S4: AI-based feedback correction and scrolling optimization.

[0165] Step S4 uses the flame stability index as feedback to correct the feedforward control quantity through a model predictive controller, dynamically adjusting the gasifier's operating parameters. The specific process is as follows:

[0166] (1) After the high-temperature syngas is injected into the main combustion zone of the furnace through the multi-stage stable combustion nozzle assembly, the multi-spectral flame imaging unit continuously collects furnace flame images, and the flame state recognition model outputs the combustion stability quantitative index in real time, and inputs it as the core feedback signal to the MPC controller.

[0167] (2) Based on the real-time feedback of combustion status, gasifier dynamic characteristics and model prediction results, the MPC controller performs rolling optimization and closed-loop correction on the feedforward control quantity generated in step S3.

[0168] (3) Based on the optimization and correction results, the AI ​​intelligent control subsystem dynamically adjusts the operating parameters of the gasifier, such as coal feed rate, oxygen enrichment flow rate, and oxygen-coal ratio, to achieve multi-objective optimization of maintaining the furnace combustion stability index within the preset target range, minimizing the auxiliary energy consumption of the gasification system, and meeting the grid AGC control requirements for unit load tracking deviation, ultimately achieving the deep peak shaving control target of combustion stability and gasification economy.

[0169] Step S5, intelligent exit mechanism.

[0170] When step S5 meets the exit conditions, the exit trajectory is generated by the unit load-gasification intensity linkage control model based on model predictive control, and the micro fixed bed oxygen enrichment subsystem is controlled to standby smoothly.

[0171] In step S5, the exit conditions must be met simultaneously: the automatic power generation control load recovers to above 35% of the rated load and remains there for ≥5 seconds; the flame is in stable combustion and the indicators meet the standards for ≥10 seconds; and the temperature in the main combustion zone of the furnace is ≥1100℃. The specific process is as follows:

[0172] (1) The AI ​​intelligent control subsystem determines the furnace combustion status and unit load conditions in real time. When all of the following exit conditions are met simultaneously, it determines that the furnace combustion has self-sustaining capability and initiates the smooth exit process of the micro fixed bed oxygen enrichment subsystem:

[0173] ① The unit's AGC load command rises to above 35% of the rated load and lasts for more than 5 seconds;

[0174] ② The flame state recognition model continuously determines that the flame state is stable combustion, the combustion stability index is maintained within the target range, and the duration exceeds 10 seconds;

[0175] ③ The temperature in the main combustion zone of the furnace is stable above 1100℃, and the pulverized coal airflow can achieve autonomous ignition and stable combustion.

[0176] (2) The AI ​​intelligent control subsystem generates a smooth exit trajectory through MPC model optimization, and gradually reduces the coal feed and oxygen enrichment flow of the gasifier according to the preset gradient, thereby gradually reducing the gasification intensity.

[0177] (3) After the gasifier operating parameters drop to the standby threshold, control the micro fixed bed oxygen enrichment subsystem to return to the standby state smoothly and wait for the next stable combustion trigger to avoid unit load fluctuations and furnace combustion disturbances caused during the start-up and shutdown of the gasification system.

[0178] Comparative examples and experimental comparisons.

[0179] Comparison of Flame State Recognition Accuracy of Different Algorithms. To verify the industrial applicability and recognition performance of the multispectral feature fusion + backpropagation neural network (BPNN) algorithm used in this invention, it was compared with two classic algorithms, Convolutional Neural Network (CNN) and Support Vector Machine (SVM), using the same multispectral flame image dataset from thermal power units. The test set included four types of operating conditions: stable combustion, mild instability, moderate instability, and severe instability. All algorithms were trained on the same training set. The following are the comparison results of recognition accuracy under various operating conditions.

[0180] Table 1. Accuracy Comparison of Different Algorithms

[0181]

[0182] The algorithm of this invention achieves an average recognition accuracy of 97.4% under four flame conditions, outperforming CNN and SVM algorithms overall. Particularly in the most challenging industrial conditions of "mild instability" and "moderate instability," this invention integrates the geometric, rectangular, and textural depth features of the flame through PLS feature filtering, effectively suppressing interference from industrial dust and noise. Its recognition accuracy is 7%–11% higher than SVM and 2%–3% higher than CNN. These results demonstrate that the algorithm of this invention, while maintaining high accuracy, possesses superior robustness and can accurately capture early signs of instability in furnace combustion, meeting the online control requirements for deep peak shaving in thermal power units.

[0183] The confusion matrix of the algorithm's recognition results in this invention is simulated. Using Python's sklearn and matplotlib libraries, the confusion matrix of the algorithm's recognition results on the test set is simulated and visualized to intuitively demonstrate the model's classification performance under various flame conditions. (See attached diagram.) Figure 4 The proposed algorithm for flame state identification confusion matrix shows that the model has an extremely strong ability to distinguish between stable combustion and severe instability, with accuracies of 98% and 99% respectively, exhibiting almost no misclassification and reliably determining extreme combustion states. In the most challenging industrial setting, distinguishing between mild and moderate instability, the model performs exceptionally well, with only a small number of samples (approximately 3%–4%) experiencing confusion, and no misclassification to other categories.

[0184] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0185] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for deep peak shaving and combustion stabilization of a fossil power plant coupled with a fixed bed oxygen-enriched gasification, characterized in that, Includes the following steps: Step S1, Pre-construction and offline training of AI model library: Pre-construct an AI model library containing a flame state recognition model and a unit load-gasification intensity linkage control model based on model predictive control, and call and update it online after offline training and verification; Step S2, Real-time monitoring and stable combustion triggering: Real-time acquisition of automatic power generation control load commands, multispectral flame images, unit operating parameters and gasifier operating parameter data, input flame state recognition model output flame state and stability indicators, and start micro fixed bed oxygen enrichment subsystem when triggering conditions are met; Step S3, Calculation of feedforward control quantity: Call the linkage control model, input real-time operating data to calculate the calorific value compensation of syngas, and issue instructions to control the micro fixed bed oxygen enrichment subsystem to produce high-temperature syngas at the preset temperature. Step S4, AI-based feedback correction and rolling optimization: Using the flame stability index as feedback, the model predicts the controller to correct the feedforward control quantity and dynamically adjust the gasifier operating parameters. Step S5, Intelligent Exit Mechanism: When the exit conditions are met, the exit trajectory is generated by the unit load-gasification intensity linkage control model based on model predictive control, and the micro fixed bed oxygen enrichment subsystem is controlled to standby smoothly. Step S1 also includes feature extraction of the flame state recognition model, specifically including 12-dimensional basic features, 7-dimensional Hu invariant moment features, and 20-dimensional Zernike moment features, for a total of 39 original features; Hu invariant moment features: 7 dimensions in total, translational, rotational, and scale-invariant features constructed based on image moments, effectively characterizing the geometric morphology of flames, unaffected by slight flame displacement or morphological scaling. The calculation process is as follows: For a digital image of , define order origin moments and central moments with the formula ； ， ， ； In the formula, The number of pixel rows in a digital image; The number of pixel columns in a digital image; x-axis powers of; : Vertical axis powers of; : First-order origin moment; : The central moment of the first order; Pixels in a digital image grayscale value; : The x-coordinate of a pixel within a digital image; : The vertical coordinate of a pixel within a digital image; The x-coordinate of the gray-level centroid of a digital image, calculated as follows: ; The ordinate of the gray-level centroid of a digital image, calculated as follows: ; Image grayscale quality; First-order moment at the origin, used to calculate the abscissa of the centroid of the image gray level; First-order moment at the origin, used to calculate the ordinate of the centroid of the image grayscale; To eliminate the effect of scale variations, the normalized central moments are defined and the normalized coefficients : ； ， ； In the formula, Normalized central moments; : Normalization coefficient, calculated as follows ; zeroth order origin moment of Power of 1 It is the total grayscale quality that characterizes the scale and spatial scale of flame combustion. It is used as a whole to normalize the central moment to eliminate the interference of flame image scale changes. Seven Hu invariant moments are derived based on the second-order and third-order normalized central moments, achieving a robust characterization of flame morphology. Zernike moments feature: 20 dimensions in total, based on features constructed from orthogonal polynomials within the unit circle, possessing rotational invariance and strong noise resistance, accurately representing the fine morphology and texture features of flames, and suitable for flame feature extraction under industrial noise interference; the first 7 Zernike moments, totaling 20 effective feature parameters, are extracted, and the calculation formula is as follows: ； In the formula, : Step The second Zernike moment is the Zernike moment feature parameter of the flame image; : The radial order of the Zernike moment; Angular frequency of the Zernike moment; Pi (π) : The x-coordinate of a pixel within the image; : The ordinate of a pixel within the image; Pixels in an image grayscale value; : The conjugate of the Zernike polynomial; : The distance from the origin of the coordinate system to the pixel in the image The distance; : With images The angle between the axes; The multi-objective optimization function of the unit load-gasification intensity linkage control model based on model predictive control in step S1 is: ； in, To optimize the objective function value; This refers to the combustion stability index predicted by the model. The target value for combustion stability index; This is for auxiliary energy consumption of the gasification system; Set the load value for automatic generation control; This represents the actual active power of the generating unit. , , These are the weighting coefficients for each optimization objective, which are dynamically adjusted according to the power plant's operational needs.

2. The method for deep peak shaving and stable combustion of a thermal power unit with coupled fixed-bed oxygen-enriched gasification according to claim 1, characterized in that, In step S1, the flame state recognition model employs a three-layer backpropagation neural network structure. The input is an 11-dimensional fusion feature filtered by partial least squares, and the output is the probability of four flame states. Training uses a cross-entropy loss function and a scale conjugate gradient algorithm with a learning rate of 0.001 and 2000 iterations or a loss of less than 10. -7 Stop when the time comes.

3. The method for deep peak shaving and stable combustion of a thermal power unit with coupled fixed-bed oxygen-enriched gasification according to claim 1, characterized in that, Step S1 also includes a multispectral flame image preprocessing process, specifically including: region of interest cropping, adaptive median filtering for noise reduction, multispectral channel fusion, adaptive threshold segmentation, and flame contour extraction.

4. The method for deep peak shaving and stable combustion of a thermal power unit with coupled fixed-bed oxygen-enriched gasification according to claim 1, characterized in that, The triggering condition in step S2 is any one of the following: the automatic power generation control load command is lower than 35% of the rated load; the flame state is moderate or severe instability; the stability index exceeds the preset range and lasts for ≥2s.

5. The method for deep peak shaving and stable combustion of a thermal power unit with coupled fixed-bed oxygen-enriched gasification according to claim 1, characterized in that, In step S3, the input of the linkage control model includes automatic power generation control load commands, real-time coal quality data and furnace combustion status, and the output includes coal feed rate of gasifier, oxygen enrichment flow rate adjustment, and syngas flow rate and calorific value matching furnace stable combustion requirements.

6. The method for deep peak shaving and stable combustion of a thermal power unit with coupled fixed-bed oxygen-enriched gasification according to claim 1, characterized in that, In step S3, the operation of the micro fixed-bed oxygen enrichment subsystem must meet the following requirements: gasifier temperature ≥ 800℃, oxygen-to-coal ratio ∈ [0.3, 0.8], and furnace main combustion zone temperature ≥ 850℃.

7. The method for deep peak shaving and stable combustion of a thermal power unit with coupled fixed-bed oxygen-enriched gasification according to claim 1, characterized in that, The exit conditions in step S5 must be met simultaneously: the automatic power generation control load recovers to more than 35% of the rated load and lasts for ≥5s; the flame is in stable combustion and the indicators meet the standards for ≥10s; and the temperature of the main combustion zone in the furnace is ≥1100℃.

8. A deep peak-shaving and combustion stabilization system for thermal power units with coupled fixed-bed oxygen-enriched combustion, used to implement the deep peak-shaving and combustion stabilization method for thermal power units with coupled fixed-bed oxygen-enriched combustion as described in any one of claims 1-7, characterized in that, The system has a modular architecture, including: The furnace combustion subsystem is the main body of the pulverized coal boiler, equipped with a pulverized coal burner, temperature / pressure sensor, furnace radiation energy detection device, and pulverized coal feeding adjustment unit. It collects combustion status parameters and connects them to the AI ​​intelligent control subsystem. The micro fixed-bed oxygen enrichment gasification subsystem is bypassed to the boiler side and integrated with the flue gas system. It includes a coal feeding unit, a gasifier body, an oxygen enrichment supply unit, a waste heat preheating unit, and a syngas conveying unit. The waste heat preheating unit is connected to the secondary air duct, the oxygen enrichment supply unit is connected to the power plant oxygen production station, and the syngas conveying unit is connected to the multi-stage stable combustion nozzle. The execution unit of this subsystem is connected to the AI ​​intelligent control subsystem. The multi-source data acquisition subsystem includes a multispectral flame imaging unit, a unit operating parameter acquisition unit, and a gasifier operating parameter acquisition unit. It is adapted to the industrial environment of thermal power plants and is connected to the AI ​​intelligent control subsystem. The AI ​​intelligent control subsystem is equipped with an industrial-grade controller, a data interaction module, and a pre-trained AI model library. It seamlessly connects with the multi-source data acquisition subsystem, the micro fixed-bed oxygen enrichment subsystem, the multi-stage stable combustion nozzle assembly, and the power plant distributed control system to achieve data interaction and command issuance. The multi-stage stable combustion nozzle assembly is integrated into the pulverized coal burner and adopts a coaxial nested structure. From the inside out, it consists of a high-temperature syngas center nozzle, a dense-phase pulverized coal ring channel, and a secondary air ring channel, which are respectively connected to the syngas delivery unit, the pulverized coal supply pipeline, and the secondary air pipeline.

Images