Fine combustion adjustment method based on full-hearth expansion monitoring

Abstract

The invention relates to the technical field of mechanical stress and thermal deformation monitoring, in particular to a combustion fine adjustment method based on full-hearth expansion monitoring, which is characterized in that expansion sensor arrays are arranged on each layer of a main burner area of a boiler hearth, and vibration monitoring units are arranged in a layered manner from a main burner to an SOFA air area synchronously; normal deformation, three-dimensional expansion displacement data and vibration spectrum data of the four furnace walls are collected; generating a standardized monitoring data set through wavelet noise reduction and temperature compensation processing, fusing boiler operation parameters, and inputting the boiler operation parameters into the reinforcement learning model to generate a combustor adjustment instruction; water-cooled wall expansion rate change data is collected in real time to optimize model weight and update a control strategy, and a closed-loop feedback mechanism is formed. According to the method, through multi-source sensing data fusion and dynamic modeling, sub-meter analysis of thermal load distribution and quick response of the combustion state are achieved, the boiler operation efficiency is improved, and the service life of equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical stress and thermal deformation monitoring, and particularly to a method for fine combustion adjustment based on full-furnace expansion monitoring. Background Art

[0002] Existing boiler combustion control technologies generally adopt means such as monitoring the outlet temperature of the water wall or analyzing the two-dimensional temperature field in the combustion zone, which have the defects of single monitoring dimension and insufficient spatial coverage, and it is difficult to capture the dynamic thermal load distribution in the furnace and the local flame deviation state in real time. Especially in spiral water wall boilers, the coupling effect of the medium flow rate and the thermal load easily causes the temperature monitoring data to be interfered by non-uniform heat transfer, and it is impossible to accurately characterize the actual heat flux density of the heating surface. In addition, under low-load peak shaving conditions, the phenomena of flame skewing and unbalanced thermal stress distribution in the furnace are aggravated. Due to the lack of direct correlation analysis of the deformation characteristics of the water wall, the existing methods are difficult to dynamically identify local overloading areas and combustion instability risks, resulting in lagging combustion adjustment, thermal efficiency loss and reduced equipment life. Based on this, there is an urgent need for a technical solution that integrates thermal deformation monitoring and dynamic combustion control to overcome the deficiencies of existing means in terms of spatial resolution, real-time performance and operating condition adaptability. Summary of the Invention

[0003] Aiming at the deficiencies of the existing technology, the present invention provides a method for fine combustion adjustment based on full-furnace expansion monitoring, which solves the problem of lagging combustion adjustment caused by insufficient spatial resolution of existing temperature monitoring means.

[0004] To solve the above technical problems, the specific technical solution of the present invention is as follows: The method for fine combustion adjustment based on full-furnace expansion monitoring provided by the present invention includes: Arranging an expansion sensor array in each layer of the main burner area of the boiler furnace to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and setting vibration monitoring units in layers from the main burner area to the SOFA air area to obtain vibration spectrum data; Performing wavelet noise reduction filtering and temperature compensation processing on the normal deformation amount, three-dimensional expansion displacement data and the vibration spectrum data respectively to generate a standardized monitoring data set including displacement characteristics and vibration characteristics; Based on the three-dimensional expansion displacement amounts of the four furnace walls at the same height layer in the standardized monitoring data set, calculating the expansion deviation rate of symmetric monitoring points, and constructing a flame center offset vector model through a random forest algorithm according to the expansion deviation rate, and outputting a heat load distribution cloud map reflecting the spatial distribution of the heat load; Extracting the coordinates of the monitoring points where the normal deformation amount exceeds the elastic limit of the material from the standardized monitoring data set, and associating the energy ratio characteristics in the low-frequency band in the vibration spectrum data corresponding to the positions, and identifying local heat flux density abnormal areas through a support vector machine classifier; Input the heat load distribution nephogram output by the flame center offset vector model, the abnormal area coordinate information identified by the support vector machine classifier, and the data of oxygen content, NOx concentration, furnace outlet temperature, desuperheating water volume, and steam temperature in the boiler operation parameters into the reinforcement learning model to generate the burner output adjustment instruction and the damper opening combination parameter; Send the burner output adjustment instruction and the damper opening combination parameter to the burner actuator through the distributed control system; Collect the data of the change in the expansion rate of the water-cooled wall within a preset time period after adjustment in real time, optimize the weight parameters of the reinforcement learning model based on the change data of the expansion rate, and update the control strategy in the combustion adjustment rule base.

[0005] Further, for the combustion refined adjustment method based on full furnace expansion monitoring, arrange an expansion sensor array at each layer in the main burner area of the boiler furnace to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and set vibration monitoring units in layers from the main burner area to the SOFA air area to obtain vibration spectrum data, including: set 4 groups of sensor nodes at each height layer in the main burner area on the four sides of the furnace, and each group of nodes includes a normal displacement sensor and a three-dimensional dilatometer, which are used to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls; The vibration monitoring units are arranged in layers from the main burner area to the SOFA air area, and each layer is equipped with a vibration acceleration sensor and a spectrum analysis module to obtain the vibration spectrum data of the corresponding area.

[0006] Further, for the combustion refined adjustment method based on full furnace expansion monitoring, perform wavelet noise reduction filtering and temperature compensation processing on the normal deformation amount, three-dimensional expansion displacement data, and the vibration spectrum data respectively to generate a standardized monitoring data set including displacement features and vibration features, including: use the wavelet transform algorithm to filter out high-frequency noise from the original vibration spectrum data collected by the vibration monitoring unit; The temperature compensation processing includes: linearly correcting the expansion displacement measurement value measured by the expansion sensor array according to the furnace ambient temperature data to eliminate the deviation of the thermal expansion coefficient.

[0007] Further, for the combustion refined adjustment method based on full furnace expansion monitoring, calculate the expansion deviation rate of symmetric monitoring points based on the three-dimensional expansion displacement amounts of the four furnace walls at the same height layer in the standardized monitoring data set, construct a flame center offset vector model through the random forest algorithm according to the expansion deviation rate, and output a heat load distribution nephogram reflecting the spatial distribution of the heat load. Calculate the expansion deviation rate of symmetric monitoring points at the same height layer based on the expansion displacement amounts of the four furnace walls in the standardized monitoring data set; Establish the mapping relationship between the expansion deviation rate and the flame center coordinates through the random forest algorithm, and generate a spatial distribution matrix of heat load with weight coefficients.

[0008] Further, for the combustion refined adjustment method based on full-furnace expansion monitoring, extracting the coordinates of the monitoring points whose normal deformation exceeds the elastic limit of the material from the standardized monitoring dataset, and correlating the energy ratio characteristics in the low-frequency band of the vibration spectrum data at the corresponding positions, identifying the local heat flux density abnormal area through the support vector machine classifier includes: Extract the coordinates of the monitoring points whose normal deformation exceeds the elastic limit of the water wall tube material from the standardized monitoring dataset, and the elastic limit is dynamically set according to the yield strength threshold of the water wall tube material and combined with the furnace temperature monitored in real time; Combine the time-domain energy distribution characteristics in the vibration spectrum data at the corresponding positions, and determine the risk level of combustion instability through the support vector machine classifier; Generate a set of abnormal area coordinates including priority marks as the input parameters of the reinforcement learning model.

[0009] Further, for the combustion refined adjustment method based on full-furnace expansion monitoring, the working process of the reinforcement learning model includes: Perform spatial encoding on the generated spatial distribution matrix of heat load and the set of abnormal area coordinates; Fuse the data of NOx concentration, oxygen content, furnace outlet temperature, desuperheating water volume and steam temperature in the boiler operation parameters; Solve the adjustment parameter combination under the multi-objective constraints of heat load balance, minimum NOx emission and stable steam temperature through the proximal policy optimization algorithm.

[0010] Further, for the combustion refined adjustment method based on full-furnace expansion monitoring, collecting the data of the change in the expansion rate of the water wall within a preset time period after the adjustment, and optimizing the weight parameters of the reinforcement learning model based on the change data of the expansion rate through the gradient descent method, and updating the control strategy in the combustion adjustment rule library includes: Collect the data of the change in the expansion rate of the water wall within a preset time range after the burner actuator completes the adjustment action; Calculate the expansion suppression efficiency coefficient of the current adjustment strategy according to the change data of the expansion rate; Based on the expansion suppression efficiency coefficient, optimize the weight parameters of the reinforcement learning model through the gradient descent method and update the reward function value.

[0011] Further, the combustion refined adjustment method based on full-furnace expansion monitoring further includes: Input the expansion displacement mode, vibration spectrum characteristics and combustion parameter combination in the historical operation data into the convolutional neural network; Extract spatio-temporal correlation feature vectors, and generate a combustion health index after dimensionality reduction through principal component analysis; Divide the combustion state into three categories: stable state, fluctuating state, and unstable state according to the combustion health index, and feed the classification result back to the reinforcement learning model.

[0012] Furthermore, the combustion refined adjustment method based on full-furnace expansion monitoring further includes: Call the transfer learning algorithm to adapt the parameters of the reinforcement learning model according to the coal quality change data; Dynamically adjust the kernel function parameters of the support vector machine classifier based on the classification result of the combustion health index; Update the combustion health index determination threshold of the convolutional neural network according to the coal quality calorific value change range.

[0013] Furthermore, the combustion refined adjustment method based on full-furnace expansion monitoring further includes: after each combustion adjustment instruction is executed, collect the expansion displacement correlation data of two adjacent monitoring points arranged at preset interval degrees along the furnace circumference in the same height layer, and four monitoring points in the upper and lower adjacent layers; Construct a transfer function of the vibration spectrum feature and the heat flux density change based on the correlation data; Correct the heat load weight coefficient of the flame center offset vector model through the transfer function.

[0014] Furthermore, for the combustion refined adjustment method based on full-furnace expansion monitoring of the present invention, the arrangement method of the expansion sensor array includes a wire-pulling distance measuring hook, a wire-pulling distance measuring sensor, and a wire-pulling distance measuring fixer, wherein the wire-pulling distance measuring hook is fixed on the surface of the furnace outer wall heat preservation iron sheet at a preset interval, and the wire-pulling distance measuring sensor is connected to the hook through a wire to form a distance measuring grid (such as Figure 4 ) for real-time collection of the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls; The vibration monitoring unit includes a vibration measurement module and a vibration measurement probe, and the vibration measurement probe is installed on the surface of the furnace water wall through the fixed bracket of the full-furnace expansion monitoring device for obtaining the vibration spectrum data of the corresponding area; The operation control host module receives the real-time data of the wire-pulling distance measuring sensor, the vibration measurement probe, and the wall temperature measurement probe through the communication antenna, and generates a combustion optimization instruction according to the load condition, coal quality parameters, and air volume data.

[0015] Advantages of the present invention; The beneficial effects of the present invention are as follows: through the multi-source data fusion of the full-furnace expansion sensor array and the vibration monitoring unit, combined with wavelet noise reduction and temperature compensation processing, a high-precision standardized monitoring data set is generated. The random forest algorithm is used to construct a flame center offset vector model to analyze the sub-meter-level heat load distribution. At the same time, the support vector machine classifier is used to associate the deformation exceeding the limit with the vibration spectrum characteristics to realize the dynamic identification of local abnormal areas. Combining with the reinforcement learning model, combustion adjustment instructions are generated under multi-objective constraints and a closed-loop feedback optimization mechanism is formed, effectively improving the spatial resolution of the heat load and the response speed of the combustion state, reducing the risk of misjudgment of flame offset caused by single monitoring dimension, and enhancing the boiler thermal efficiency and equipment operation safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the drawings.

[0017] Figure 1 It is a flowchart of the combustion fine adjustment method based on full-furnace expansion monitoring provided by an embodiment of the present invention.

[0018] Figure 2 It is a schematic cross-sectional view of the main perspective of the full furnace provided by an embodiment of the present invention.

[0019] Figure 3 It is a schematic top view of the full furnace provided by an embodiment of the present invention.

[0020] Figure 4 It is a schematic diagram of the connection mode of the wire-pulling ranging grid layout of the first measuring device provided by an embodiment of the present invention.

[0021] Figure 5 It is a schematic diagram of the connection mode of the wire-pulling ranging grid layout of the second measuring device provided by an embodiment of the present invention Figure 6 It is a schematic diagram of the vibration monitoring unit arranged in layers from the main burner area to the SOFA air area provided by an embodiment of the present invention.

[0022] Description of the reference numerals in the drawings: 1. Pull-wire distance measurement hook; 2. Pull-wire distance measurement sensor; 3. Pull-wire distance measurement fixator; 4. Signal lamp of the full-furnace expansion monitoring device; 5. Power lamp of the full-furnace expansion monitoring device; 6. Communication antenna of the full-furnace expansion monitoring device; 7. Operation control host module of the full-furnace expansion monitoring device; 8. Vibration measurement module; 9. Wall temperature measurement module; 10. Wall temperature measurement probe; 11. Vibration measurement probe; 12. Water-cooled wall of the boiler furnace; 13. Fixing bracket of the full-furnace expansion monitoring device; 14. Shell of the full-furnace expansion monitoring device; 15. Fixing hole of the full-furnace expansion monitoring device; 16. Thermal insulation iron sheet of the furnace outer wall; 17. Thermal insulation layer of the furnace outer wall. Detailed implementation manners

[0023] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below with reference to the drawings. To better understand the objectives of the present invention, the present invention is further described in detail below.

[0024] Please refer to Figures 1 to 6 , the combustion refinement adjustment method based on full-furnace expansion monitoring provided by the present invention includes: Step S101: Arrange an expansion sensor array on each layer in the main burner area of the boiler furnace to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and layer by layer set vibration monitoring units in the area from the main burner area to the SOFA air area to obtain vibration spectrum data; Arrange an expansion sensor array on each layer in the main burner area of the boiler furnace. The array includes a pull-wire distance measurement hook 1, a pull-wire distance measurement sensor 2 and a pull-wire distance measurement fixator 3. The pull-wire distance measurement hook 1 is fixed on the surface of the thermal insulation iron sheet 16 of the furnace outer wall at a preset interval. The pull-wire distance measurement sensor 2 is connected to the hook 1 through a pull wire to form a distance measurement grid, as Figure 4 , for collecting the normal deformation amount and three-dimensional expansion displacement data of the furnace wall surface in real time; Synchronously set vibration monitoring units layer by layer in the area from the main burner to the SOFA air area. The unit includes a vibration measurement module 8 and a vibration measurement probe 11. The vibration measurement probe 11 is installed on the surface of the water-cooled wall 12 of the furnace through a fixing bracket 13 of the full-furnace expansion monitoring device for obtaining the vibration spectrum data of the corresponding area; Further arrange a wall temperature measurement module 9 and a wall temperature measurement probe 10 on the surface of the water-cooled wall 12 of the furnace to collect the water-cooled wall temperature data in real time; The full-furnace expansion monitoring operation control host module 7 receives the real-time data of the wire-drawing ranging sensor 2, the vibration measurement probe 11 and the wall temperature measurement probe 10 through the communication antenna 6, and classifies and summarizes the expansion deformation data, vibration data and wall temperature data to the full-furnace expansion monitoring operation control system host; The operation control system host classifies and processes the data according to the current load condition, coal quality parameters, primary and secondary air volume and temperature, identifies outliers through real-time calculation and generates combustion optimization operation instructions.

[0025] Step S102, perform wavelet noise reduction filtering and temperature compensation processing on the normal deformation amount, three-dimensional expansion displacement data and the vibration spectrum data respectively to generate a standardized monitoring data set including displacement characteristics and vibration characteristics; Step S103, based on the three-dimensional expansion displacement amounts of the four furnace walls at the same height layer in the standardized monitoring data set, calculate the expansion deviation rate of the symmetric monitoring points, and construct a flame center offset vector model through the random forest algorithm according to the expansion deviation rate, and output a heat load distribution cloud map reflecting the spatial distribution of the heat load; Step S104, extract the coordinates of the monitoring points where the normal deformation amount exceeds the material elastic limit from the standardized monitoring data set, and associate the energy ratio characteristics in the low-frequency band of the vibration spectrum data at the corresponding positions, and identify the local heat flux density abnormal area through a support vector machine classifier; Step S105, input the heat load distribution cloud map output by the flame center offset vector model, the coordinate information of the abnormal area identified by the support vector machine classifier, and the data of oxygen content, NOx concentration, furnace outlet temperature, desuperheating water volume and steam temperature in the boiler operation parameters into the reinforcement learning model to generate a burner output adjustment instruction and a damper opening combination parameter; Step S106, issue the burner output adjustment instruction and the damper opening combination parameter to the burner actuator through the distributed control system; Step S107, collect the data of the change in the expansion rate of the water-cooled wall within a preset time period after adjustment in real time, optimize the weight parameters of the reinforcement learning model through the gradient descent method based on the expansion rate change data, and update the control strategy in the combustion adjustment rule base.

[0026] An expansion sensor array is set at each layer in the main burner area of the boiler furnace. The array is distributed in each height layer of the four furnace walls. Each group of sensor nodes includes a normal displacement sensor and a three-dimensional dilatometer, which are used to collect the normal deformation amount of the furnace wall and the three-dimensional expansion displacement data in real time. The vibration monitoring unit is arranged in layers from the main burner area to the SOFA air area, and each layer is equipped with a vibration acceleration sensor and a spectrum analysis module to synchronously obtain the vibration spectrum data of the corresponding area. Through the collaborative layout of multi-sensors, a mechanical deformation and vibration state monitoring network for the entire furnace area is constructed.

[0027] The collected raw data needs to be subjected to signal preprocessing. For the vibration spectrum data, the wavelet transform algorithm is used to separate the high-frequency noise and the effective signal, and filter out the irrelevant noise components introduced by equipment vibration or electromagnetic interference. For the expansion sensor data, based on the real-time monitored value of the furnace environment temperature, a linear correction is performed on the expansion displacement measurement value to eliminate the measurement error caused by the difference in the material thermal expansion coefficient and the environment temperature, and generate a standardized displacement feature and vibration feature data set.

[0028] Based on the standardized data set, calculate the three-dimensional expansion displacement deviation rate of the symmetric monitoring points on the four furnace walls at the same height layer. Through the random forest algorithm, feature extraction and pattern matching are performed on multiple groups of expansion deviation rate data, establish the mapping relationship between the expansion deviation rate and the spatial position of the flame center, and generate a heat load distribution cloud map reflecting the spatial distribution of the heat load. This model analyzes the sub-meter-level distribution characteristics of the internal heat load of the furnace through multi-dimensional data fusion.

[0029] Based on the displacement feature data, identify the coordinates of the monitoring points where the normal deformation amount exceeds the elastic limit of the water wall tube material. Combining with the energy distribution characteristics in the low-frequency band of the vibration spectrum at the corresponding position, use the support vector machine classifier to analyze the correlation between the time-domain characteristics of the vibration signal and the abnormal heat flux density, determine the risk level of local combustion instability, and generate a set of abnormal area coordinates with priority marks.

[0030] Input the heat load distribution cloud map, the set of abnormal area coordinates, and the data of oxygen content, NOx concentration, furnace outlet temperature, desuperheating water volume, and steam temperature in the boiler operation parameters into the reinforcement learning model. Through the proximal policy optimization algorithm, solve the optimal parameter combination of the burner output adjustment and the damper opening under the multi-objective constraint conditions. This model integrates the requirements of thermodynamic balance, pollutant emission control, and equipment safe operation, and dynamically generates combustion adjustment instructions.

[0031] The adjustment instructions are sent to the burner actuator through the distributed control system to drive the coordinated adjustment of the burner output and the damper opening. During the execution process, real-time collect the data of the change in the expansion rate of the water wall, and quantify the suppression effect of the combustion adjustment on the mechanical deformation of the furnace. Based on the change trend of the expansion rate, use the gradient descent method to optimize the weight parameters of the reinforcement learning model, update the control strategy in the rule base, and form a closed-loop feedback mechanism.

[0032] The above steps realize the dynamic coupling analysis of the combustion state and the mechanical deformation through the closed-loop process of data acquisition, feature analysis, intelligent decision-making, and feedback optimization. The sensor layout and signal preprocessing provide standardized inputs for model construction; the random forest algorithm and the support vector machine classifier respectively solve the problems of heat load distribution analysis and abnormal area identification; the reinforcement learning model generates control instructions under multi-objective constraints and continuously optimizes the decision-making logic through execution feedback.

[0033] Specifically, for the combustion refinement adjustment method based on full-furnace expansion monitoring, an expansion sensor array is arranged in each layer of the main burner area of the boiler furnace to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and vibration monitoring units are arranged in layers from the main burner area to the SOFA air area to obtain vibration spectrum data, including: setting 4 groups of sensor nodes in each height layer of the four main burner areas on the furnace walls, and each group of nodes includes a normal displacement sensor and a three-dimensional dilatometer for collecting the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls; The vibration monitoring units are arranged in layers from the main burner area to the SOFA air area, and each layer is equipped with a vibration acceleration sensor and a spectrum analysis module for obtaining the vibration spectrum data of the corresponding area.

[0034] Four groups of sensor nodes are arranged in each height layer of the four main burner areas on the furnace walls. Each group of nodes consists of a normal displacement sensor and a three-dimensional dilatometer, and they are respectively installed on the preset installation reference surfaces of the four furnace walls. The normal displacement sensor is arranged along the normal direction of the furnace wall and is coupled with the furnace wall surface through a rigid fixing bracket for real-time measurement of the deformation amount perpendicular to the furnace wall plane; the three-dimensional dilatometer adopts a multi-axis displacement sensing structure, and its three measurement axes respectively correspond to the axial, radial, and tangential directions in the furnace coordinate system to synchronously collect the three-dimensional expansion displacement components of the furnace wall. The spatial layout of the four groups of nodes forms a symmetric monitoring network of the four furnace walls at the same height layer, covering the full circumferential deformation characteristics of the burner area.

[0035] Each group of sensor nodes needs to be calibrated for the initial direction during installation. According to the furnace geometric structure parameters, the measurement axis of the normal displacement sensor is adjusted to coincide with the normal direction of the furnace wall, and a mapping relationship is established based on the coordinate system of the three-dimensional dilatometer and the global furnace coordinate system. The sensor nodes achieve synchronous sampling through a distributed data acquisition module to eliminate the phase error caused by signal transmission delay. After redundant verification of the monitoring data of each height layer, a deformation data set with spatio-temporal consistency is generated.

[0036] The vibration monitoring units are arranged in layers from the main burner area to the SOFA air area. An array of vibration acceleration sensors is set in each layer, and its installation position forms a preset angle with the axis of the burner nozzle to optimize the sensitivity of the vibration signal in the combustion disturbance direction. The vibration acceleration sensor adopts a broadband response design, covering the characteristic frequency bands of combustion pulsation and mechanical vibration. The spectrum analysis module is integrated into the data processing unit of each monitoring layer, and performs time-domain segmentation and frequency-domain windowing processing on the original vibration signal to generate vibration spectrum data including energy spectrum, kurtosis, and envelope characteristics.

[0037] The hierarchical layout strategy of the vibration monitoring unit matches the air flow path from the burner area to the SOFA air area. Among them, the monitoring layer in the main burner area focuses on capturing the low-frequency vibration components caused by the pulsation of the combustion flame, while the monitoring layer in the SOFA air area targets the medium-high frequency vibration characteristics generated by the disturbance of the secondary air. Each layer of the spectrum analysis module extracts the vibration modal parameters related to the combustion state through the feature frequency band segmentation algorithm, forming a vibration feature set that is spatio-temporally aligned with the expansion monitoring data.

[0038] The sensor nodes and the vibration monitoring unit achieve data fusion through redundant communication links. Among them, the three-dimensional displacement data of the expansion sensor and the spectrum feature data of the vibration monitoring are aligned by timestamp and then input into the multi-source data verification module. Based on the preset deformation-vibration correlation model, this verification module verifies the physical logic consistency between the data, eliminates abnormal data segments caused by sensor failures or environmental interference, and improves the reliability of the monitoring data.

[0039] The above layout method constructs a high-precision coordinated monitoring system for furnace deformation and vibration through spatial symmetric layout, direction calibration optimization, and multi-source data verification. The composite configuration of the normal displacement sensor and the three-dimensional dilatometer realizes the multi-dimensional analysis of the deformation vector; the hierarchical frequency domain feature extraction of the vibration monitoring unit enhances the identification ability of the combustion dynamic characteristics.

[0040] Specifically, for the combustion refinement adjustment method based on the full-furnace expansion monitoring, the wavelet noise reduction filtering and temperature compensation processing are respectively performed on the normal deformation amount, the three-dimensional expansion displacement data, and the vibration spectrum data to generate a standardized monitoring data set including displacement features and vibration features, including: using the wavelet transform algorithm to filter out the high-frequency noise in the original vibration spectrum data collected by the vibration monitoring unit; The temperature compensation processing includes: linearly correcting the expansion displacement measurement values measured by the expansion sensor array according to the furnace environment temperature data to eliminate the deviation of the thermal expansion coefficient.

[0041] When performing wavelet noise reduction processing on the original vibration spectrum data collected by the vibration monitoring unit, first select an appropriate wavelet basis function according to the frequency domain characteristics of the combustion system vibration signal, and decompose the signal into high-frequency detail components and low-frequency approximation components through multi-scale decomposition. For the noise frequency bands generated by mechanical transmission or electromagnetic interference in the high-frequency components, an adaptive threshold algorithm is used to perform non-linear shrinkage processing on the detail coefficients, and the effective vibration characteristics related to the combustion pulsation are retained. The processed wavelet coefficients of each layer are reconstructed to generate the vibration spectrum data after noise reduction, suppressing the interference of high-frequency noise on the combustion state analysis.

[0042] The selection of wavelet basis functions is based on the typical frequency distribution characteristics of the burner vibration signal. Wavelet basis functions with excellent time-frequency localization characteristics are preferentially selected to match the transient impact characteristics of the combustion vibration signal. The decomposition level is set based on the resonance frequency range of the combustion chamber structure, and the optimal decomposition level that can effectively separate the combustion-related vibration modes is determined through preliminary experiments. In the threshold processing, modulus maxima detection is introduced to distinguish the energy mutation points between noise components and real vibration events, avoiding excessive attenuation of effective signal components.

[0043] For the measurement data of the expansion sensor array, a correlation model between the furnace ambient temperature and the thermal expansion coefficient of the sensor needs to be established for temperature compensation processing. Temperature sensing nodes are synchronously arranged at the sensor installation positions to collect the temperature distribution data on the furnace wall surface in real time. Based on the material thermal expansion characteristic curve, a linear regression model between the measured expansion displacement and the temperature change is constructed, and the temperature compensation coefficient matrix is fitted by the least squares method. In the data preprocessing stage, the real-time temperature data collected by the temperature sensing nodes is input into the compensation model to dynamically correct the original displacement measurement values of the expansion sensors, eliminating the measurement reference drift caused by the furnace temperature fluctuation.

[0044] In the temperature compensation model, different compensation parameters are set for different furnace wall structure zones according to the material. For the water-cooled wall tube area, considering the difference in the thermal conductivity of the tube material and the fin, the thermal expansion influence weights of the metal components and the non-metal sealing materials are calculated respectively. The compensation coefficient matrix is dynamically updated according to the furnace operating conditions. When the temperature gradient change in the furnace exceeds the set threshold, the recalibration process of the compensation parameters is triggered to maintain the long-term stability of the displacement measurement. After the compensated displacement data and the vibration spectrum data are aligned in time stamp, they are input into the normalization processing module to generate a monitoring data set with a unified time reference and physical dimension.

[0045] The above processing flow optimizes the vibration feature extraction accuracy through signal decomposition and reconstruction techniques, and improves the environmental adaptability of the expansion measurement by combining the temperature compensation mechanism. Wavelet denoising processing retains the time-frequency characteristics of the combustion-related vibration modes, providing high-quality input for subsequent spectrum analysis; the temperature compensation algorithm eliminates the systematic deviation of the thermal expansion effect on the deformation measurement, enhancing the physical consistency of the displacement data. The collaborative application of these two preprocessing techniques lays a data foundation for the flame center offset modeling and abnormal area identification.

[0046] Specifically, for the combustion fine-tuning method based on the full-furnace expansion monitoring, based on the three-dimensional expansion displacement amounts of the four-sided furnace walls at the same height layer in the standardized monitoring data set, calculate the expansion deviation rate of the symmetric monitoring points, and construct a flame center offset vector model through the random forest algorithm according to the expansion deviation rate, and output a heat load distribution cloud map reflecting the spatial distribution of the heat load. Based on the expansion displacement amounts of the four-sided furnace walls in the standardized monitoring data set, calculate the expansion deviation rate of the symmetric monitoring points at the same height layer; The mapping relationship between the expansion deviation rate and the flame center coordinates is established by the random forest algorithm, and a heat load spatial distribution matrix with weight coefficients is generated.

[0047] Based on the standardized monitoring data set, the three-dimensional expansion displacement vector of the symmetric monitoring points on the four furnace walls at the same height layer is decomposed, and the axial, radial and tangential displacement components of each monitoring point in the furnace coordinate system are extracted. By calculating the three-dimensional displacement vector modulus length difference ratio of the symmetric monitoring point pairs and combining the cosine value of the displacement direction included angle, a set of expansion deviation rate parameters characterizing the asymmetric deformation of the four furnace walls is generated. The selection of the symmetric monitoring point pairs is based on the geometric symmetry of the furnace. Each group contains two sets of sensor nodes arranged at intervals of 180 degrees along the furnace circumference, covering four orthogonal symmetric axes in the burner area.

[0048] The calculation of the expansion deviation rate adopts normalization processing, which converts the difference value of the three-dimensional displacement components into a dimensionless proportional coefficient to eliminate the dimensional influence caused by the absolute position difference of the monitoring points. For the four sets of symmetric monitoring point pairs at each height layer, three types of characteristic parameters, namely the axial expansion difference rate, the radial offset angle and the tangential displacement gradient, are respectively generated to form a set of characteristic vectors characterizing the asymmetry of the heat load distribution at this layer.

[0049] The model training of the random forest algorithm is based on the corresponding relationship between the expansion deviation rate characteristic vectors and the measured coordinates of the flame center in the historical operation data set. The input features include the axial expansion difference rate, the radial offset angle and the tangential displacement gradient at each height layer, and the output label is the three-dimensional coordinates of the flame center calibrated by the infrared thermal imager under the corresponding working conditions. In the feature selection stage, the Gini coefficient is used to evaluate the contribution of each feature to the prediction of the flame center position, and redundant features are removed to optimize the generalization ability of the model. The trained random forest model can output the coordinate prediction value and the confidence index of the flame center in the furnace space according to the real-time expansion deviation rate characteristic vectors.

[0050] The construction of the heat load spatial distribution matrix is based on the flame center coordinate prediction results, and the radial basis function interpolation algorithm is used to generate the heat load density distribution of the furnace cross-section. The weight coefficient allocation of the matrix follows the furnace heat transfer characteristics. Higher weights are assigned to the grid cells in the area close to the flame center, and the weights in the edge area decay exponentially with the distance. The matrix update frequency is synchronized with the sensor data acquisition period to realize the dynamic visualization expression of the heat load distribution. The dynamic adjustment mechanism of the weight coefficient introduces the number of operating burner layers and the opening degree of the secondary air dampers as correction factors to enhance the adaptability of the model to changes in operating conditions.

[0051] During the model verification stage, the accuracy of the flame center offset vector model is evaluated by comparing the predicted heat load distribution with the results of boiler thermal calculations. When the deviation between the real-time monitoring data and the model prediction value exceeds the preset threshold, the online optimization process of model parameters is triggered. The incremental learning algorithm is used to update the splitting nodes of the random forest decision tree to maintain the prediction accuracy of the model under variable operating conditions. The above process establishes a quantitative relationship between the furnace thermal state and structural deformation by integrating mechanical deformation monitoring data and machine learning algorithms.

[0052] Specifically, for the combustion refinement adjustment method based on full-furnace expansion monitoring, extracting the coordinates of the monitoring points where the normal deformation exceeds the elastic limit of the material from the standardized monitoring data set, and associating the energy proportion characteristics of the low-frequency band in the vibration spectrum data at the corresponding position, and identifying the local heat flux density abnormal area through the support vector machine classifier includes: Extracting the coordinates of the monitoring points where the normal deformation exceeds the elastic limit of the water wall pipe material from the standardized monitoring data set, and the elastic limit is dynamically set according to the yield strength threshold of the water wall pipe material and combined with the furnace temperature monitored in real time; Combining the time-domain energy distribution characteristics in the vibration spectrum data at the corresponding position, and determining the risk level of combustion instability through the support vector machine classifier; Generating a set of abnormal area coordinates including priority marks as the input parameters of the reinforcement learning model.

[0053] During the processing of the standardized monitoring data set, when dynamically setting the elastic limit threshold of the water wall pipe material, according to the yield strength characteristic curve of the pipe material and combined with the furnace temperature field distribution data collected in real time, a temperature compensation factor matrix is established. The elastic modulus attenuation coefficient of the pipe material under different temperature gradients is calculated through the material thermodynamics model, and the elastic limit judgment threshold of each monitoring point is dynamically adjusted to make the deformation exceeding limit judgment more in line with the actual working conditions. For the area with uneven temperature distribution, the adjacent monitoring point temperature weighted average algorithm is used to correct the local threshold to avoid the threshold setting deviation caused by sparse temperature measurement points.

[0054] When extracting the coordinates of the monitoring points where the normal deformation exceeds the limit, a multi-dimensional sliding window scan is performed on the standardized data set to identify the monitoring points where the deformation exceeds the dynamic threshold for three consecutive sampling periods. For the identified abnormal points, the time-domain energy distribution data of the vibration spectrum at the corresponding position is synchronously retrieved, and the energy entropy value and the energy proportion characteristics of the low-frequency band of the vibration signal within the preset time window are extracted. The length of the time window is set to match the combustion adjustment period, so that the vibration characteristics can reflect the cumulative effect of combustion state changes.

[0055] The training sample set of the support vector machine classifier includes the vibration spectrum feature vectors corresponding to the deformation overrun events in historical operations, as well as the combustion instability state labels verified by thermal calculations. When constructing the feature vectors, the overrun amplitude of the deformation amount, the vibration energy entropy value, and the low-frequency energy ratio are normalized to form a multi-dimensional input space. The choice of the classifier kernel function is based on the test results of the linear separability of the feature vectors, and the radial basis function is preferentially used to realize the mapping in the high-dimensional space. During the training process, a cross-validation mechanism is introduced to optimize the classification boundary and improve the ability to distinguish between slight combustion disturbances and severe instability states.

[0056] The output results of the classifier include the risk level label and the confidence score. The risk level is divided according to the plastic deformation rate of the pipe that may be caused by combustion instability. For the high-risk level abnormal points with high confidence, they are marked as the first priority in the coordinate set; for the abnormal points with medium and low risks but with continuous multi-cycle trigger records, they are marked as the second priority. The priority marking rule library integrates expert experience knowledge. When there are more than three adjacent abnormal points at the same height layer, the priority level of this area is automatically increased. The marked coordinate set is bound with the timestamp and the risk level code to form a structured input parameter format.

[0057] The above recognition process enhances environmental adaptability through dynamic threshold adjustment, improves the reliability of abnormal determination through multi-source data correlation analysis, and optimizes the control resource allocation through the priority marking mechanism. The dynamic calculation of the elastic limit solves the problem of temperature field interference; the extraction of the vibration time-domain energy characteristics captures the combustion dynamic characteristics; the support vector machine classification realizes the multi-dimensional feature fusion decision.

[0058] Specifically, for the refined combustion adjustment method based on full-furnace expansion monitoring, the working process of the reinforcement learning model includes: Perform spatial encoding on the generated heat load spatial distribution matrix and the abnormal area coordinate set; Fuse the data of NOx concentration, oxygen content, furnace outlet temperature, desuperheating water volume, and steam temperature in the boiler operation parameters; Solve the adjustment parameter combination under the multi-objective constraints of heat load balance, minimum NOx emission, and stable steam temperature through the proximal policy optimization algorithm.

[0059] During the spatial encoding process of the heat load spatial distribution matrix, the furnace three-dimensional grid division technology is adopted to map the weight coefficients in the matrix to the corresponding physical coordinate grid cells. The heat load value of each grid cell is converted into an element of the state vector through normalization processing, and the abnormal area coordinate set is marked with the one-hot encoding method to mark the affected grids, generating a two-dimensional encoding matrix containing heat load intensity and abnormal position information. During the encoding process, a spatio-temporal dimension expansion mechanism is introduced, and the heat load change gradient of the adjacent three sampling periods is superimposed on the spatial encoding matrix as a time series feature to enhance the model's understanding ability of the combustion dynamic process.

[0060] The fusion processing of boiler operation parameters adopts a multi-source data standardization method, which respectively performs sliding window averaging and variance normalization processing on NOx concentration, oxygen content, furnace outlet temperature, desuperheating water volume, and steam temperature to eliminate the influence of dimension differences on the model input. The standardized parameter data is synchronized with the spatial coding matrix through the timestamp alignment module to construct a three-channel input tensor containing heat load distribution, abnormal area marking, and operation parameters. Before the tensor is input into the reinforcement learning model, the key feature dimensions are screened by the feature importance evaluation module to reduce the interference of data redundancy on policy optimization.

[0061] During the implementation of the proximal policy optimization algorithm, a composite reward function is defined, which includes heat load balance degree, NOx emission intensity, furnace outlet temperature gradient, desuperheating water volume gradient, and steam temperature deviation degree. The heat load balance degree is quantified by calculating the load difference entropy value of adjacent grids in the spatial coding matrix; the NOx emission intensity is evaluated by the cumulative emission gradient within the sliding window; the steam temperature deviation degree is calculated based on the standard deviation from the set value. The policy network explores the action space of the burner output adjustment and damper opening combination in each iteration through the gradient update mechanism under trust region constraints to find the Pareto optimal solution set under multi-objective constraints.

[0062] During the policy optimization process, an action mask mechanism is introduced to transform the physical constraint conditions of the burner actuator into the feasible domain boundary of the action space. For the damper opening adjustment action, the single-step adjustment amplitude limit is set according to the mechanical response characteristics of the actuator; the burner output adjustment is associated with the flow control margin of the fuel supply system to generate discrete action options that conform to the actual control ability. The optimized adjustment parameter combination passes through the feasibility verification module to eliminate the aggressive strategies that may cause transient impacts on the equipment and generate a safe and controllable adjustment instruction sequence.

[0063] The above workflow realizes the digital reconstruction of the combustion state through spatial coding, establishes a global optimization goal through multi-source data fusion, and balances multi-dimensional control requirements through proximal policy optimization. The coding mechanism solves the mathematical representation problem of the heat load spatial distribution and abnormal areas; data standardization and feature screening improve the quality of the model input; the composite reward function and action constraint design ensure the engineering applicability of the control strategy. The technical features of each link serve the in-depth application of mechanical deformation monitoring data and the generation of combustion control decisions.

[0064] Specifically, for the combustion fine adjustment method based on full-furnace expansion monitoring, the change data of the water wall expansion rate within a preset time period after the adjustment is collected in real time, and the weight parameters of the reinforcement learning model are optimized based on the change data of the expansion rate through the gradient descent method, and the control strategy in the combustion adjustment rule library is updated, including: Collect the data of the change in the expansion rate of the water-cooled wall within a preset time range after the burner actuator completes the adjustment action; Calculate the expansion suppression efficiency coefficient of the current adjustment strategy according to the data of the change in the expansion rate; Based on the expansion suppression efficiency coefficient, optimize the weight parameters of the reinforcement learning model by the gradient descent method and update the reward function value.

[0065] After the burner actuator completes the adjustment action, set a dynamic response time window to collect the data of the change in the expansion rate of the water-cooled wall. The starting moment of the time window is synchronized with the moment when the adjustment instruction is issued, and the window length is dynamically adjusted according to the current load rate of the boiler. The collection period is shortened under high load conditions to match the fast thermal response characteristics. During the data collection process, cross-verify the expansion rate of the same monitoring point through redundant sensor nodes, and eliminate abnormal data segments caused by local sensor failures to improve the credibility of the data set.

[0066] The calculation of the expansion suppression efficiency coefficient adopts a multi-dimensional evaluation method, and the expansion rate change curve within the time window is decomposed into a transient response stage and a steady-state convergence stage. In the transient stage, the second derivative extreme points of the rate change curve are calculated to identify the time when the adjustment strategy takes effect on suppressing mechanical deformation; in the steady-state stage, the integral operation is used to obtain the proportion of the rate drop area, and combined with the heat transfer safety weight distribution coefficient of different regions of the water-cooled wall, an efficiency coefficient that comprehensively reflects the adjustment effect is generated. The weight distribution coefficient is dynamically set according to the structural stress distribution map of the water-cooled wall tube screen, and a higher weight is given to the suppression effect in the high heat load area.

[0067] During the optimization process of the gradient descent method, the expansion suppression efficiency coefficient is used as a negative penalty term in the loss function of the policy network, and forms a multi-objective optimization problem with the heat load balance degree and NOx emission index in the original reward function. When optimizing and iterating, an adaptive learning rate adjustment mechanism is adopted, and the parameter update step size is dynamically scaled according to the historical gradient amplitude to avoid falling into local optimal solutions. After the model weights are updated, the coefficients of each objective term in the reward function are normalized and redistributed to improve the priority of the expansion suppression effect in policy decision-making, and at the same time maintain the balance of pollutant emission control, combustion flow field, and steam temperature stability.

[0068] The updated model parameters are imported into the online rule library after offline simulation verification, replacing the original policy entries. During the verification process, a virtual combustion scenario is constructed, and typical disturbance patterns in the historical operation data are injected to test the generalization ability of the new policy under various working conditions. For the policy entries that pass the verification, record the corresponding improvement amplitude of the efficiency coefficient and the applicable working condition range as the basis for the priority ranking of subsequent policy calls. The above process forms a closed-loop control architecture of "execution - monitoring - evaluation - optimization".

[0069] Specifically, the refined combustion adjustment method based on full-furnace expansion monitoring further includes: Input the expansion displacement pattern, vibration spectrum characteristics, and combustion parameter combination in the historical operation data into the convolutional neural network; Extract the spatio-temporal correlation feature vector, and generate the combustion health index after dimensionality reduction by principal component analysis; According to the combustion health index, divide the combustion state into three categories: stable state, fluctuating state, and unstable state, and feedback the classification result to the reinforcement learning model.

[0070] In the preprocessing stage of the historical operation data, construct the expansion displacement pattern, vibration spectrum characteristics, and combustion parameter combination into a three-dimensional spatio-temporal data cube. The time dimension of the cube is sliced according to the combustion adjustment period, the space dimension corresponds to the furnace monitoring grid coordinates, and the channel dimension includes parameters such as displacement vector, vibration energy spectrum, and oxygen content. Before the data cube is input into the convolutional neural network, eliminate the range differences of different sensors through sliding window normalization, and use zero-padding technology to maintain the continuity of the space dimension.

[0071] The architecture design of the convolutional neural network includes a spatio-temporal feature parallel extraction path. Among them, the expansion displacement pattern captures the spatial distribution characteristics through a two-dimensional convolutional layer, and the vibration spectrum time series extracts the dynamic evolution law through a one-dimensional time series convolutional layer. After the two types of features are deeply fused in the network, a three-dimensional max-pooling layer is used to compress the data dimension and generate a feature vector containing the spatio-temporal correlation characteristics of the furnace thermal state. The stride setting of the pooling layer is matched with the combustion adjustment period, so that the feature vector can represent the key state transition nodes of the complete combustion process.

[0072] In the principal component analysis dimensionality reduction process, perform singular value decomposition on the covariance matrix of the feature vector, and select the feature vectors with the cumulative contribution rate exceeding the set threshold to construct the projection space. The low-dimensional vectors after dimensionality reduction are linearly weighted and fused to generate the combustion health index, and the weight distribution is dynamically adjusted according to the correlation analysis result between the feature vector and the historical failure events. The scaling range of the health index is set according to the boiler design parameters, with the lower limit corresponding to the ideal stable combustion state and the upper limit associated with the critical condition of the pipe material plastic deformation.

[0073] The Gaussian mixture model clustering algorithm is used for combustion state classification, and the time series change trajectory of the combustion health index is divided into three categories: stable state, fluctuating state, and unstable state. The classification boundary is iteratively optimized by the expectation maximization algorithm, and the clustering center is manually calibrated in combination with expert experience to eliminate misclassification caused by uneven data distribution. The classification result is fed back to the reinforcement learning model in the form of a state code, serving as an additional conditional variable in the input layer of the policy network to guide the model to preferentially call the historically verified and effective adjustment strategies under specific combustion states.

[0074] The above process enhances the state representation ability through spatio-temporal data fusion, improves the efficiency of feature engineering by principal component analysis, and realizes a hierarchical description of the combustion state through clustering and classification. The feature extraction of the convolutional neural network solves the problem of fusion analysis of multi-source heterogeneous data; the construction of the health index quantifies the overall stability of the combustion system; the state classification feedback mechanism enhances the environmental adaptability of the reinforcement learning model.

[0075] Specifically, the combustion refinement adjustment method based on full-furnace expansion monitoring further includes: Invoking a transfer learning algorithm according to the coal quality change data to adapt the parameters of the reinforcement learning model; Dynamically adjusting the kernel function parameters of the support vector machine classifier based on the classification results of the combustion health index; Updating the determination threshold of the combustion health index of the convolutional neural network according to the change range of the coal quality calorific value.

[0076] During the processing of coal quality change data, a mapping relationship library between coal quality feature vectors and combustion monitoring data sets is established. The feature vectors include coal quality industrial analysis parameters and elemental analysis parameters, and effective sample data are screened through an outlier detection algorithm. When it is detected that the current coal quality feature vector deviates from the historical data distribution, a transfer learning process is triggered, the parameter of the policy network layer strongly related to fuel characteristics in the reinforcement learning model is frozen, the general parameters of the thermodynamic equilibrium constraint layer are retained, and incremental training is carried out using the limited sample data under the new coal quality condition to achieve rapid adaptation of the model parameters.

[0077] During the transfer learning process, a feature alignment technique is adopted to reduce the distribution difference between the old and new coal quality data in the combustion feature space through the maximum mean discrepancy algorithm. The trained adapted model retains the original multi-objective optimization ability and at the same time enhances the response sensitivity to the current coal quality combustion characteristics. In the model verification stage, a coal quality gradual change test sequence is injected to observe the smooth transition characteristics of the output adjustment parameters of the policy network and prevent the actuator from oscillating due to parameter mutation.

[0078] The dynamic adjustment mechanism of the kernel function parameters of the support vector machine classifier constructs a feedback loop based on the classification results of the combustion health index. When the health index continuously remains in a stable state, a larger kernel width parameter is adopted to enhance the fault tolerance ability of the classification boundary; when the index enters a fluctuating state or an unstable state, the kernel width is reduced and the penalty factor weight is increased to improve the identification accuracy of abnormal features. The parameter adjustment amplitude is positively correlated with the degree of deviation of the health index from the reference value, and the exponential moving average algorithm is used to smooth the adjustment process to avoid frequent parameter jumps.

[0079] During the process of adjusting the kernel function parameters, the mapping relationship of the feature space is optimized synchronously. For the feature dimensions with relatively low contribution degrees in the combustion health index, their weight ratios in the kernel function calculation are gradually reduced, and the influence of key features on the classification decision is strengthened. This optimization process works in coordination with the feature alignment module of transfer learning to form a classifier architecture with self-adaptive combustion state sensitivity, improving the stability of anomaly detection under different coal quality conditions.

[0080] The update strategy for the determination threshold of the combustion health index of the convolutional neural network divides the dynamic adjustment stage according to the change range of coal quality calorific value. When the coal quality calorific value is within the designed coal type range, the reference threshold is used for state classification; when the calorific value deviates beyond the preset window, the compensation coefficient is calculated based on the calorific value-expansion rate correlation curve, and the determination threshold is scaled proportionally. A hysteresis mechanism is introduced during the threshold update process to prevent threshold oscillation caused by short-term fluctuations in coal quality. At the same time, a maximum adjustment amplitude limit is set to maintain the consistency of the physical meaning of the health index evaluation system.

[0081] The above technical process constructs a combustion monitoring and control system with self-adaptive fuel characteristics through a collaborative mechanism of driving model parameter adaptation by coal quality characteristics, optimizing classification accuracy through health status feedback, and updating the calorific value-related threshold. Transfer learning maintains the generalization ability of the model for new coal qualities, the dynamic adjustment of the kernel function enhances the sensitivity of anomaly detection, and the threshold compensation mechanism ensures the reliability of state classification.

[0082] Specifically, the refined combustion adjustment method based on full-furnace expansion monitoring further includes: after each combustion adjustment instruction is executed, collecting the correlation data of the expansion displacements of two adjacent monitoring points arranged at preset interval degrees along the furnace circumference in the same height layer, and four monitoring points in the upper and lower adjacent layers; Constructing a transfer function of the vibration spectrum characteristics and the change of heat flux density based on the correlation data; Correcting the heat load weight coefficient of the flame center offset vector model through the transfer function.

[0083] When collecting the expansion displacement correlation for the preset monitoring point combination after the combustion adjustment instruction is executed, the spatial layout of the monitoring points adopts a cross-layer coupling design. Two adjacent monitoring points arranged at intervals along the furnace circumference in the same height layer form a horizontal correlation pair, and four monitoring points in the upper and lower adjacent layers form a pyramid-shaped spatial correlation group. The data acquisition module synchronously records the time-domain waveforms of the three-dimensional expansion displacements of each monitoring point, calculates the phase difference and amplitude ratio of the displacement waveforms through the cross-correlation algorithm, and generates a set of correlation parameters characterizing the heat expansion propagation characteristics.

[0084] The processing of the correlation data adopts a system identification algorithm to perform time-delay matching between the dominant frequency components in the vibration spectrum characteristics and the change rate of the heat flux density. When constructing the transfer function, the vibration acceleration signal is used as the input variable, and the change gradient of the heat flux density is used as the output response, and the recursive least squares method is used to solve the coefficient matrix of the transfer function. The transfer function includes two dimensions of amplitude-frequency characteristics and phase-frequency characteristics, reflecting the dynamic coupling relationship between mechanical vibration and heat load fluctuation under a specific combustion state.

[0085] The correction of the flame center offset vector model by the transfer function is achieved through the dynamic compensation of the heat load weight coefficient. During the update period of the heat load spatial distribution matrix, the resonant peak frequency components in the amplitude-frequency characteristics of the transfer function are extracted to identify the heat flux density sensitive frequency band. According to the energy distribution intensity of the sensitive frequency band, a gain compensation is applied to the heat load weight coefficient of the corresponding spatial grid to strengthen the correction effect of the vibration characteristics on the heat load distribution. After the corrected weight coefficient is normalized, it is redistributed to the feature importance evaluation module of the random forest algorithm to form a two-way coupling mechanism of the model parameters.

[0086] The verification of the correction effect adopts a virtual combustion disturbance test method. A standard heat flux step signal is injected into the simulation environment to compare the prediction deviation of the flame center coordinates before and after the transfer function correction. When the improvement amplitude of the prediction accuracy of the corrected model reaches the preset threshold, the current transfer function parameters are solidified into the model library as the reference correction parameters for the same type of combustion conditions. The closed-loop correction process continuously optimizes the analytical accuracy of the heat load distribution through the online identification of the vibration-heat flux correlation characteristics.

[0087] The following is an explanatory description of the main technical feature terms in the technical solution of the present invention: Expansion sensor array: A monitoring device composed of a normal displacement sensor and a three-dimensional dilatometer, which is arranged at multiple height layers on the four furnace walls of the furnace according to a preset geometric law. The normal displacement sensor is installed along the normal direction of the furnace wall to measure the deformation perpendicular to the furnace wall plane; the three-dimensional dilatometer synchronously collects the axial, radial and tangential displacement components through a multi-axis sensing structure to form a three-dimensional deformation monitoring ability. This array realizes the high-density capture of the circumferential deformation characteristics of the furnace through a space symmetric layout and a redundancy verification mechanism.

[0088] Vibration monitoring unit: A monitoring layer composed of a vibration acceleration sensor and a spectrum analysis module, which is arranged in layers along the main burner to the SOFA air region. The vibration acceleration sensor is installed at a specific inclination angle to optimize the signal sensitivity in the combustion disturbance direction; the spectrum analysis module performs time-frequency domain windowing processing on the original vibration signal to extract characteristic parameters such as energy spectrum and kurtosis, reflecting the coupling effect of combustion dynamic characteristics and mechanical vibration.

[0089] Wavelet denoising processing: A signal preprocessing method based on wavelet transform. The vibration spectrum data is separated into high-frequency noise components and low-frequency effective signals through multi-scale decomposition. The adaptive threshold algorithm is used to perform non-linear shrinkage on the high-frequency coefficients, retaining the time-frequency characteristics of the vibration modes related to combustion and suppressing the interference of electromagnetic interference and mechanical noise on subsequent analysis.

[0090] Temperature compensation processing: According to the real-time data of the furnace wall surface temperature, the displacement measurement value of the expansion sensor is dynamically corrected. A linear compensation model is established based on the material thermal expansion coefficient curve to eliminate the measurement reference drift caused by environmental temperature fluctuations and ensure the physical consistency between the displacement data and the true deformation state.

[0091] Flame center offset vector model: A machine learning model constructed using the random forest algorithm. The feature vector of the expansion deviation rate of symmetric monitoring points at the same height layer is input, and the predicted values of the three-dimensional coordinates of the flame center and the spatial distribution matrix of the heat load are output. The model establishes the mapping relationship between the expansion deviation rate and the heat load distribution through historical data training to achieve sub-meter-level heat flux density analysis.

[0092] Support vector machine classifier: A supervised learning model based on kernel function mapping. The low-frequency energy characteristics of the vibration spectrum of the deformation overrun monitoring points are input, and the optimal classification hyperplane is constructed through the high-dimensional feature space to determine the risk level of local combustion instability and generate an abnormal coordinate set with priority marks.

[0093] Reinforcement learning model: A decision-making model using the proximal policy optimization algorithm. It integrates the heat load distribution, abnormal area coordinates, and boiler operation parameters to generate adjustment instructions for the burner output and damper opening under multi-objective constraints. The model quantifies the heat load balance degree, emission control, and deformation suppression effect through the reward function to drive the policy network to explore the optimal action space.

[0094] Transfer function correction: A dynamic correction mechanism constructed based on the correlation data between the vibration spectrum and the heat flux density. The vibration-heat flux transfer function is established through the system identification algorithm, and the energy characteristics of the sensitive frequency band are extracted to perform gain compensation on the heat load weight coefficient to optimize the dynamic response accuracy of the flame center offset model.

[0095] An expansion sensor array is installed on the four-sided furnace walls of each layer in the main burner area of the boiler furnace. Each group of sensor nodes includes a normal displacement sensor and a three-dimensional dilatometer, which are symmetrically distributed along the circumference of the furnace to form a deformation monitoring network that covers the entire height. The normal displacement sensor is coupled with the furnace wall surface through a rigid bracket to collect the deformation amount perpendicular to the furnace wall plane in real time. The three-dimensional dilatometer adopts a multi-axis sensing structure to synchronously record the axial, radial, and tangential displacement components. The vibration monitoring unit is arranged in layers from the main burner to the SOFA air area. The vibration acceleration sensors are installed at a preset angle to capture the characteristic frequency spectrum in the direction of combustion disturbance. After the direction calibration and coordinate system mapping of the sensor nodes, the transmission delay is eliminated through the synchronous sampling module to generate a spatio-temporally consistent deformation and vibration data set.

[0096] In the preprocessing stage of the original data, wavelet basis functions are used for multi-scale decomposition of the vibration spectrum data. High-frequency mechanical noise is filtered out through an adaptive threshold algorithm, and the low-frequency pulsation characteristics related to combustion are retained. The data of the expansion sensors are combined with the real-time monitored values of the furnace wall surface temperature, and the displacement measurement values are dynamically corrected based on the thermal expansion coefficient of the material to eliminate the reference drift caused by the temperature gradient. The preprocessed standardized data set is input into the random forest algorithm to calculate the three-dimensional expansion deviation rate of the symmetric monitoring points at the same height layer, and a mapping model between the expansion deviation rate and the spatial position of the flame center is established. This model is trained with historical operation data and outputs a heat load distribution cloud map with weight coefficients to analyze the sub-meter-level heat flux density distribution of the furnace cross-section.

[0097] When identifying local abnormal areas, the elastic limit threshold of the water-cooled wall tubes is dynamically calculated, and the deformation overrun judgment standard is corrected in combination with the temperature field distribution. For continuously overrun monitoring points, the low-frequency energy entropy value characteristics of the vibration spectrum at the corresponding positions are extracted, and the risk level of combustion instability is evaluated through a support vector machine classifier to generate a set of abnormal coordinates with priority marks. The reinforcement learning model integrates operation parameters such as the heat load cloud map, abnormal area coordinates, and NOx concentration, and uses the proximal policy optimization algorithm to generate adjustment instructions for the burner output and damper opening under multi-objective constraints. After the instructions are executed, the data of the change in the expansion rate of the water-cooled wall is collected, and the model weights are optimized based on the gradient descent method and the control strategy library is updated to form a closed-loop feedback mechanism. This implementation method realizes the collaborative optimization control of the combustion state and mechanical deformation through the dynamic correlation analysis of deformation and heat load and real-time policy iteration.

[0098] Flame center offset vector model: A machine learning model constructed based on the random forest algorithm. The characteristic vector of the three-dimensional expansion displacement deviation rate of symmetric monitoring points at the same height layer is input, and a non-linear mapping relationship between the expansion deviation rate and the three-dimensional coordinates of the flame center is established through the ensemble learning mechanism of multiple decision trees. This model evaluates the importance of features through the Gini coefficient, screens the key deformation features sensitive to flame offset, and outputs a heat load distribution cloud map with spatial weight coefficients to realize the sub-meter-level spatial analysis of the heat flux density of the furnace cross-section.

[0099] Support vector machine classifier: A supervised learning model using a radial basis kernel function. It inputs the low-frequency energy entropy value and energy ratio characteristics of the vibration spectrum of the input method deformation over-limit monitoring points, constructs an optimal classification hyperplane in the high-dimensional feature space, and divides the local area into risk levels of stable state, transition state, and unstable state. The classification boundary is optimized through cross-validation, and an abnormal area coordinate set is generated in combination with the priority marking rule base, providing spatial constraint conditions for the combustion instability risk for the reinforcement learning model.

[0100] Reinforcement learning model: A multi-objective decision-making model based on the proximal policy optimization (PPO) algorithm. It inputs the heat load distribution cloud map, abnormal area coordinates, and boiler operation parameters, and explores the action space of burner output and damper opening through the policy network. The model defines a composite reward function to quantify the heat load balance degree (entropy difference between adjacent grids), NOx emission gradient, furnace outlet temperature gradient, desuperheating water volume gradient, and steam temperature deviation degree, and generates an adjustment instruction sequence that meets the requirements of equipment safety and environmental protection under the trust region constraint to achieve dynamic optimization of combustion parameters.

[0101] Transfer function correction mechanism: A vibration-thermal flow dynamic coupling model constructed based on system identification technology. It inputs the expansion displacement correlation data and vibration spectrum characteristics of a specific monitoring point group, and uses the recursive least squares method to solve the transfer function coefficient matrix of the vibration signal and heat flux density change. By extracting the resonance peak frequency components in the amplitude-frequency characteristics of the transfer function, frequency domain gain compensation is performed on the heat load weight coefficient of the flame center offset model to eliminate the analytical error of the heat load distribution caused by combustion disturbances.

[0102] Convolutional neural network model: A spatio-temporal feature extraction model for generating combustion health indices. It inputs the spatio-temporal data cube of historical expansion displacement patterns, vibration spectra, and combustion parameters, captures the spatial deformation distribution characteristics through a two-dimensional convolutional layer, and extracts the vibration dynamic evolution law through a one-dimensional time series convolutional layer. The fused high-order features are dimensionally reduced by principal component analysis to generate a scalar health index representing the overall stability of the combustion system, providing an input benchmark for state classification.

[0103] Transfer learning adaptation module: A model parameter optimization module for coal quality changes. It reduces the distribution difference of new and old coal quality data in the combustion feature space through a feature alignment algorithm, freezes the general parameter layer related to thermodynamic equilibrium in the reinforcement learning model, and only performs incremental training on the policy network layer sensitive to fuel characteristics. This module retains the generalization ability of historical policies and adapts to the combustion dynamics characteristics of the current coal quality, avoiding inaccurate control instructions caused by fuel changes in the model.

[0104] Combustion health index generation model: A clustering and classification model based on the Gaussian mixture model. It inputs the dimensionality-reduced feature vectors extracted by the convolutional neural network, iteratively optimizes the clustering centers through the expectation-maximization algorithm, and divides the combustion health index into three categories: stable state, fluctuating state, and unstable state. The classification results are fed back to the reinforcement learning model in the form of state codes, restricting the action exploration range of the policy network under specific combustion conditions and enhancing the condition adaptability of control instructions.

[0105] The present invention solves the problem of combustion adjustment lag caused by insufficient spatial resolution of existing temperature monitoring means through the following technical solutions: Full-dimensional deformation monitoring and data fusion: An expansion sensor array is arranged on all four sides of each layer in the main burner area of the boiler furnace, and vibration monitoring units are set layer by layer synchronously from the main burner to the SOFA air area. The expansion sensor array collects the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and the vibration monitoring unit obtains the combustion dynamic vibration spectrum data, constructing a collaborative monitoring network of mechanical deformation and vibration covering the entire furnace area. High-frequency noise interference is filtered out through wavelet denoising, and the expansion measurement values are linearly compensated in combination with the real-time furnace temperature data to generate a standardized displacement and vibration feature dataset, breaking through the limitation of a single temperature monitoring dimension and realizing the spatial multi-dimensional analysis of the heat load distribution.

[0106] Dynamic heat load modeling and anomaly identification: Based on the standardized dataset, calculate the expansion deviation rate of symmetric monitoring points at the same height layer, establish the mapping relationship between the expansion deviation rate and the flame center coordinates through the random forest algorithm, and generate a heat load spatial distribution matrix with sub-meter resolution. At the same time, extract the coordinates of the monitoring points with excessive normal deformation, correlate the low-frequency energy characteristics of the vibration spectrum, and use the support vector machine classifier to identify the local heat flux density anomaly area. Through the deformation-heat load correlation model, the method realizes the second-level dynamic tracking of the flame center offset and the accurate positioning of the anomaly area, eliminating the analysis error caused by the heat transfer interference in temperature monitoring.

[0107] Closed-loop combustion control and strategy optimization: Input the heat load distribution cloud map, anomaly area coordinates, and boiler operation parameters into the reinforcement learning model, and generate multi-objective optimization adjustment instructions for the burner output and damper opening through the proximal policy optimization algorithm. After the adjustment is executed, collect the data of the change in the expansion rate of the water wall in real time, optimize the model weights based on the gradient descent method, and update the control strategy library to form a closed-loop control architecture of "monitoring - decision-making - execution - feedback". This mechanism realizes the rapid response and self-adaptive adjustment of the heat load imbalance through the dynamic coupling analysis of the combustion state and mechanical deformation, effectively solving the combustion instability risk caused by the lag of traditional temperature monitoring.

Claims

1. A combustion refinement adjustment method based on full-furnace expansion monitoring, characterized in that Including: Arranging an expansion sensor array on each layer in the main burner area of the boiler furnace to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and arranging vibration monitoring units in layers from the main burner area to the SOFA air area to obtain vibration spectrum data; Performing wavelet denoising filtering and temperature compensation processing on the normal deformation amount, three-dimensional expansion displacement data, and the vibration spectrum data respectively to generate a standardized monitoring data set including displacement characteristics and vibration characteristics; Based on the three-dimensional expansion displacement amounts of the four furnace walls at the same height layer in the standardized monitoring data set, calculating the expansion deviation rate of symmetric monitoring points, and constructing a flame center offset vector model through a random forest algorithm according to the expansion deviation rate, and outputting a heat load distribution cloud map reflecting the spatial distribution of heat load; Extracting the coordinates of the monitoring points where the normal deformation amount exceeds the material elastic limit from the standardized monitoring data set, and correlating the energy ratio characteristics in the low-frequency band in the vibration spectrum data at the corresponding positions to identify the local heat flux density abnormal area through a support vector machine classifier; Inputting the heat load distribution cloud map output by the flame center offset vector model, the coordinate information of the abnormal area identified by the support vector machine classifier, and the oxygen content, furnace outlet temperature, desuperheating water volume, NOx concentration, and steam temperature data in the boiler operation parameters into a reinforcement learning model to generate a burner output adjustment instruction and a damper opening combination parameter; Sending the burner output adjustment instruction and the damper opening combination parameter to the burner actuator through a distributed control system; Real-time collecting the change data of the water-cooled wall expansion rate within a preset time period after adjustment, optimizing the weight parameters of the reinforcement learning model through the gradient descent method based on the expansion rate change data, and updating the control strategy in the combustion adjustment rule base.

2. The combustion fine-tuning method based on full-furnace expansion monitoring according to claim 1, wherein: The arranging an expansion sensor array on each layer in the main burner area of the boiler furnace to collect the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls, and arranging vibration monitoring units in layers from the main burner area to the SOFA air area to obtain vibration spectrum data includes: Setting 4 groups of sensor nodes at each height layer in the main burner area on the four sides of the furnace, and each group of nodes includes a normal displacement sensor and a three-dimensional dilatometer for collecting the normal deformation amount and three-dimensional expansion displacement data of the four furnace walls; The vibration monitoring units are arranged in layers from the main burner area to the SOFA air area, and each layer is configured with a vibration acceleration sensor and a spectrum analysis module for obtaining the vibration spectrum data of the corresponding area.

3. The combustion fine adjustment method based on full furnace expansion monitoring according to claim 1, characterized in that: The performing wavelet denoising filtering and temperature compensation processing on the normal deformation amount, three-dimensional expansion displacement data, and the vibration spectrum data respectively to generate a standardized monitoring data set including displacement characteristics and vibration characteristics includes: Using a wavelet transform algorithm to filter out high-frequency noise from the original vibration spectrum data collected by the vibration monitoring unit; The temperature compensation processing includes: linearly correcting the expansion displacement measurement value measured by the expansion sensor array according to the furnace ambient temperature data to eliminate the thermal expansion coefficient deviation.

4. The combustion fine-tuning method based on full-furnace expansion monitoring according to claim 1, characterized in that: Based on the three-dimensional expansion displacement amounts of the four furnace walls at the same height level in the standardized monitoring dataset, calculate the expansion deviation rate of symmetric monitoring points, and construct a flame center offset vector model through the random forest algorithm based on the expansion deviation rate, and output a heat load distribution cloud map reflecting the spatial distribution of heat load; Based on the expansion displacement amounts of the four furnace walls in the standardized monitoring dataset, calculate the expansion deviation rate of symmetric monitoring points at the same height level; Establish the mapping relationship between the expansion deviation rate and the flame center coordinates through the random forest algorithm, and generate a heat load spatial distribution matrix with weight coefficients.

5. The combustion refinement adjustment method based on full-furnace expansion monitoring according to claim 1, characterized in that: The method of extracting the coordinates of the monitoring points whose normal deformation amount exceeds the elastic limit of the material from the standardized monitoring dataset and associating the energy ratio characteristics in the low-frequency band of the vibration spectrum data at the corresponding positions, and identifying the local heat flux density abnormal area through the support vector machine classifier includes: Extract the coordinates of the monitoring points whose normal deformation amount exceeds the elastic limit of the water-cooled wall tube material from the standardized monitoring dataset, and the elastic limit is dynamically set according to the yield strength threshold of the water-cooled wall tube material and in combination with the furnace temperature monitored in real time; Combine the time-domain energy distribution characteristics in the vibration spectrum data at the corresponding positions, and determine the combustion instability risk level through the support vector machine classifier; Generate a set of abnormal area coordinates including priority marks as the input parameters of the reinforcement learning model.

6. The combustion fine adjustment method based on full furnace expansion monitoring according to claim 5, characterized in that: The reinforcement learning model includes: Perform spatial encoding on the generated heat load spatial distribution matrix and the set of abnormal area coordinates; Fuse the NOx concentration, oxygen content, furnace outlet temperature, desuperheating water volume, and steam temperature data in the boiler operation parameters; Solve the adjustment parameter combination under the multi-objective constraints of heat load balance, minimum NOx emission, and stable steam temperature through the proximal policy optimization algorithm.

7. The combustion fine-tuning method based on full-furnace expansion monitoring according to claim 1, characterized in that: The method of collecting the change data of the water-cooled wall expansion rate within a preset time period after adjustment in real time, and optimizing the weight parameters of the reinforcement learning model through the gradient descent method based on the expansion rate change data, and updating the control strategy in the combustion adjustment rule base includes: Collect the change data of the water-cooled wall expansion rate within a preset time range after the burner actuator completes the adjustment action; Calculate the expansion suppression efficiency coefficient of the current adjustment strategy according to the expansion rate change data; Based on the expansion suppression efficiency coefficient, optimize the weight parameters of the reinforcement learning model through the gradient descent method and update the reward function value.

8. The combustion fine-tuning method based on full-furnace expansion monitoring according to claim 7, characterized in that It also includes: Input the expansion displacement pattern, vibration spectrum characteristics, and combustion parameter combination in the historical operation data into the convolutional neural network; Extract the spatio-temporal correlation feature vector, and generate a combustion health index after dimensionality reduction through principal component analysis; Divide the combustion state into three categories: stable state, fluctuating state, and unstable state according to the combustion health index, and feedback the classification result to the reinforcement learning model.

9. The combustion fine-tuning method based on full-furnace expansion monitoring according to claim 8, characterized in that It also includes: Call the transfer learning algorithm to adapt the parameters of the reinforcement learning model according to the coal quality change data; Dynamically adjust the kernel function parameters of the support vector machine classifier based on the classification result of the combustion health index; Update the combustion health index determination threshold of the convolutional neural network according to the coal quality calorific value change range; It further includes: after each combustion adjustment instruction is executed, collecting the correlation data of the expansion displacements of two adjacent monitoring points arranged at preset interval degrees along the circumferential direction of the furnace at the same height level, and four monitoring points in the upper and lower adjacent layers; Constructing a transfer function of the vibration spectrum characteristics and the change of heat flux density based on the correlation data; Correcting the heat load weight coefficient of the flame center offset vector model through the transfer function.

10. The combustion refinement adjustment method based on full-furnace expansion monitoring according to claim 1, characterized in that: The arrangement mode of the expansion sensor array includes a wire-pulling distance measuring hook (1), a wire-pulling distance measuring sensor (2) and a wire-pulling distance measuring fixer (3). The wire-pulling distance measuring hook (1) is fixed on the surface of the furnace wall outer wall heat preservation iron sheet (16) at preset intervals. The wire-pulling distance measuring sensor (2) is connected with the hook (1) through a wire to form a distance measuring grid for real-time collecting the normal deformation amount of the four furnace walls and the three-dimensional expansion displacement data; The vibration monitoring unit includes a vibration measurement module (8) and a vibration measurement probe (11). The vibration measurement probe (11) is installed on the surface of the furnace water wall (12) through a full-furnace expansion monitoring device fixing bracket (13) for obtaining the vibration spectrum data of the corresponding area; The operation control host module (7) receives the real-time data of the wire-pulling distance measuring sensor (2), the vibration measurement probe (11) and the wall temperature measurement probe (10) through the communication antenna (6), and generates a combustion optimization instruction according to the load condition, the coal quality parameter and the air volume data.

Images