Boiler stable combustion energy-saving control method under deep peak regulation of unit

Abstract

The invention relates to the technical field of stable combustion control, in particular to a boiler stable combustion energy-saving control method under deep peak regulation of a unit, which comprises the following steps of: acquiring hearth flame radiation and flue gas temperature signals to generate a heat release table set, calculating radial gradient difference of a same-section center transition adherence area to form a heat release gradient sequence, and calculating the heat release gradient sequence; and the multi-layer perceptron performs screening to generate a target adjustment horizon, generates a stable combustion energy-saving control instruction in combination with the annular gradient and the angle of the air door opening rotational flow blade, and outputs an optimized stable combustion energy-saving control instruction through state pre-estimation optimization. According to the method, a combustion abnormal area and a priority adjustment layer are identified, the opening degree of an air door and the angle of a swirl vane are dynamically optimized, fine regulation and control and flame stabilization in the combustion process are achieved, stable combustion and energy saving requirements under different load and fuel conditions are considered, the operation safety of a unit is improved, and over-combustion and flameout risks are reduced.

Description

[0001] This invention relates to the field of stable combustion control technology, and in particular to a boiler stable combustion and energy-saving control method under deep peak shaving of the unit. Background Technology

[0002] The field of stable combustion control technology mainly involves the adjustment and management of fuel and air supply status of combustion devices under different operating conditions to maintain the continuous stability of the combustion process. This technology is widely used in the operation of power plant boilers, industrial furnaces and kilns, and coal, gas and oil combustion equipment. Its core aspects include fuel supply regulation, primary air and secondary air ratio regulation, furnace negative pressure maintenance, burner operation mode switching, and combustion flame stability regulation. By coordinating and controlling operating parameters such as coal feeder rate, forced draft fan and induced draft fan volume, damper opening, and burner start-up and shutdown sequence, the combustion process can maintain stable operation under different load conditions and adapt to the operating requirements of unit load changes, fuel quality changes, and furnace thermal state changes.

[0003] One traditional method for boiler stable combustion and energy saving control under deep peak shaving of power units refers to a type of operation mode that controls the boiler combustion process to prevent unstable furnace flame or flameout during the low load or deep peak shaving operation of thermal power units. This is usually achieved by reducing the coal feeder and shutting down some burners in a predetermined sequence, while adjusting the air volume ratio of primary air fans and secondary air fans, changing the number of coal mills in operation, adjusting the opening of secondary air dampers and maintaining the negative pressure range of the furnace, and, when necessary, activating oil guns or plasma ignition devices to maintain the furnace flame. This method adapts to the combustion operation requirements of the unit under low load or deep peak shaving conditions by combining and adjusting the operating status of coal feeding equipment, coal milling equipment, air supply equipment and burners in the boiler combustion system.

[0004] In deep peak shaving or low-load operation, existing technologies mainly rely on the combined regulation of equipment such as coal feeders, blowers, burners, and secondary air dampers for combustion process control. The response of these control methods to the combustion state is mainly based on empirical rules and fixed ratio regulation, lacking real-time analysis and quantitative evaluation of the heat release distribution characteristics of different cross sections and orientations in the furnace. This results in a lag in the perception of local flame instability or combustion abnormalities, making it impossible to achieve fine control of the combustion process and difficult to balance stable combustion and energy saving. At the same time, when faced with changes in fuel properties or fluctuations in furnace thermal state, the regulation response is delayed and unbalanced, causing the risk of local overburning or flameout, affecting the unit's operational safety and combustion efficiency. Summary of the Invention

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a boiler stable combustion and energy-saving control method under deep peak shaving of a power unit, comprising the following steps: S1: Collect the output voltage signals of the boiler furnace flame radiation and flue gas temperature sensors, perform analog-to-digital conversion and discrete sampling, extract the radiation intensity and flue gas temperature under the same spatial coordinates, and perform weighted coupling calculations in combination with preset dimension conversion coefficients to generate a heat release characterization set; S2: Extract the heat release characterization values ​​of the center, wall-attached and transition regions of the same cross section from the heat release characterization set, calculate the radial outer gradient difference and the radial inner gradient difference and merge them to construct a heat release gradient sequence; S3: Based on the heat release gradient sequence, extract the radial inner gradient difference and radial outer gradient difference of multiple cross sections, input them into the multilayer perceptron model for feature layer selection, extract the first cross section coordinates of the feature layer selection output, and generate the target adjustment layer; S4: Based on the heat release characterization values ​​of multiple adjacent directions corresponding to the target adjustment layer, calculate multiple directional circumferential gradient difference sequences, extract the current opening degree of the damper actuator and the current angle of the swirl blade adjusting motor in the corresponding direction, reduce the opening degree value of the damper actuator and increase the angle value of the swirl blade adjusting motor, and independently encode and combine the adjusted values ​​to generate a stable combustion and energy-saving control command; S5: Based on the stable combustion and energy-saving control command, extract the target opening value of the damper actuator and the target angle value of the swirl blade adjusting motor and perform state prediction calculation, replace the target opening value of the damper actuator and the target angle value of the swirl blade adjusting motor based on the state prediction value, and construct an optimized stable combustion and energy-saving control command.

[0006] As a further aspect of the present invention, the heat release characterization set includes radiation intensity product, temperature coupled heat intensity, and heat release value corresponding to spatial coordinates; the heat release gradient sequence includes radial inner gradient difference, radial outer gradient difference, and cross-sectional gradient combination; the target adjustment layer includes priority adjustment cross-sectional coordinates, combustion characteristic layer number, and adjustment layer spatial position; the stable combustion and energy-saving control command includes damper actuator adjustment opening degree and swirl vane adjustment motor adjustment angle; and the optimized stable combustion and energy-saving control command includes damper actuator predicted opening degree, swirl vane adjustment motor predicted angle, and control state prediction correction amount.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: By acquiring the output voltage signal of the furnace flame radiation sensor and the output voltage signal of the flue gas temperature sensor and converting them from analog to digital, the continuous voltage amplitude is mapped into a digital amplitude sequence and sampled discretely. The radiation and temperature sensor amplitudes at multiple sampling times are time-aligned and merged to obtain the radiation temperature sampling sequence. S102: Based on the radiation temperature sampling sequence, call the spatial coordinate identifier of the multi-sampling point, convert the radiation sensor amplitude and temperature sensor amplitude into flue gas temperature and radiation intensity, and pair and combine the radiation intensity value and flue gas temperature value under the same spatial coordinate index to obtain the spatial coordinate dual parameter group. S103: Based on the spatial coordinate dual parameter set, extract the multi-coordinate index-related radiation intensity and flue gas temperature, introduce a dimensionless coefficient to normalize the radiation intensity and flue gas temperature, combine the multiplication terms, integrate all product outputs and spatial coordinate index labels, and generate a heat release characterization set.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Obtain the heat release characterization values ​​of the center and transition region of the same cross section in the heat release characterization set, perform difference calculation on the heat release values ​​of the center position and the transition position according to the radial coordinate order of the cross section, and arrange them according to the radial coordinate order to generate a radial inner gradient difference sequence. S202: Based on the heat release characterization set, obtain the heat release characterization value of the wall-attached region, calculate the difference between the heat release values ​​of the transition region and the wall-attached region according to the radial coordinate order, and call the radial inner gradient difference sequence for order verification to obtain the radial outer gradient difference sequence. S203: Call the radial inner gradient difference sequence and the radial outer gradient difference sequence, perform sequence splicing operation according to the unified radial position index, and arrange the two sets of gradient difference data continuously to generate a heat release gradient sequence.

[0009] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the heat release gradient sequence, extract the radial inner gradient difference and radial outer gradient difference of multiple cross sections, call the radial outer gradient difference at the same coordinate position to compare the difference magnitude, record the gradient difference magnitude values ​​of multiple cross sections and serialize them according to the cross section coordinate order to generate the cross section gradient difference magnitude sequence. S302: Based on the cross-sectional gradient difference magnitude sequence, extract the gradient difference magnitude of multiple cross sections, and use a multilayer perceptron model to filter feature layers and extract a set of candidate layers. Select cross-sectional coordinate indices and magnitudes of the candidate layers whose magnitudes exceed a preset gradient magnitude threshold and arrange them in descending order to generate a candidate layer ranking sequence. S303: Call the candidate layer ranking sequence to perform ranking index retrieval, extract the cross-sectional coordinate index that is first in the ranking position, retrieve the corresponding spatial layer parameters and perform layer identification encoding, convert the encoded layer parameters into a coordinate layer command format that can be called by the control terminal, and obtain the target adjustment layer.

[0010] As a further embodiment of the present invention, the gradient amplitude threshold is determined by extracting the gradient amplitude sample values ​​from multiple cross-sections in the thermal release gradient dataset, performing statistical distribution operations on the sample value sequence, obtaining the gradient amplitude mean and gradient amplitude variance, and calculating the gradient fluctuation reference component by weighted summation, and then performing a product operation in combination with a preset sensitivity coefficient.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Obtain the heat release characterization values ​​of four adjacent directions corresponding to the same cross section of the target adjustment layer, perform clockwise subtraction to calculate multiple directional circumferential gradient difference sequences, call the numerical relationship between the heat release characterization values ​​of the four adjacent directions to perform circumferential gradient attribution calculation, and establish the directional circumferential gradient difference sequence. S402: Based on the azimuth circumferential gradient difference sequence, extract the current opening degree of the corresponding azimuth damper actuator and the current angle of the swirl blade adjusting motor, reduce the damper opening value, increase the swirl blade adjusting motor angle value, and generate a damper opening and blade angle adjustment value set. S403: Perform numerical integration and merging on the set of damper opening and blade angle adjustment values, call the accumulated information of multi-directional opening and angle adjustment values ​​for summary calculation, and establish a stable combustion and energy-saving control command.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the stable combustion and energy-saving control command, extract the target opening value of the damper actuator and the target angle value of the swirl blade regulating motor and jointly map them. Perform differential calculations on the opening value and angle value for adjacent angle changes and adjacent opening changes, and perform same-frequency data alignment to generate a damper swirl coordination parameter set. S502: Based on the damper swirl coordination parameter set, perform state prediction calculation, sum the independent sliding intervals of the continuous change trend sequence of opening degree and the continuous change trend sequence of angle respectively, calculate the state amplitude of opening degree and angle interval respectively, and compare it with the preset fluctuation threshold to obtain the damper swirl prediction state group. S503: Based on the damper swirl prediction state group and the stable combustion and energy-saving control command, extract the target opening value of the corresponding damper actuator and the target angle value of the swirl blade adjustment motor inside the command and perform numerical replacement calculation. Reorganize the replacement parameters into a sequence to generate an optimized stable combustion and energy-saving control command.

[0013] As a further aspect of the present invention, the fluctuation threshold is determined by obtaining the amplitude fluctuation statistics of the heat release gradient sequence under stable combustion conditions, and by introducing a dimensionless processing module to convert the distribution standard deviation of the mechanical response error extreme value and the heat release characterization value into dimensionless coefficients, and then performing weighted summation.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by performing real-time quantification and gradient analysis of the heat release distribution at various cross sections and orientations within the furnace, abnormal combustion areas and priority adjustment layers can be accurately identified. The opening of the damper and the angle of the swirl blades can be dynamically optimized and adjusted to achieve refined control of the combustion process, ensuring flame stability and thermal energy balance. At the same time, it can adaptively adjust under different loads and fuel conditions, achieving both stable combustion and energy-saving effects, improving the unit's operational safety and combustion efficiency, reducing the risk of local overburning or flameout, and improving combustion response speed and control accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a boiler stable combustion and energy-saving control method under deep peak shaving of a power unit, including the following steps: S1: Collect the output voltage signals of the boiler furnace flame radiation and flue gas temperature sensors, perform analog-to-digital conversion and discrete sampling, extract the radiation intensity and flue gas temperature under the same spatial coordinates, and perform weighted coupling calculations in combination with preset dimension conversion coefficients to generate a heat release characterization set; S2: Extract the heat release characterization values ​​of the center, wall-attached and transition regions of the same cross section from the heat release characterization set, calculate the radial outer gradient difference and the radial inner gradient difference and merge them to construct the heat release gradient sequence; S3: Based on the thermal release gradient sequence, extract the radial inner gradient difference and radial outer gradient difference of multiple cross sections, input them into the multilayer perceptron model for feature layer selection, extract the first cross section coordinates of the feature layer selection output, and generate the target adjustment layer; S4: Based on the heat release characterization values ​​of multiple adjacent directions corresponding to the target adjustment layer, calculate multiple circumferential gradient difference sequences, extract the current opening degree of the damper actuator and the current angle of the swirl blade adjustment motor in the corresponding direction, reduce the opening value of the damper actuator and increase the angle value of the swirl blade adjustment motor, and independently encode and combine the adjusted values ​​to generate stable combustion and energy-saving control instructions. S5: Based on the stable combustion and energy-saving control command, extract the target opening value of the damper actuator and the target angle value of the swirl blade regulating motor and perform state prediction calculation. Replace the target opening value of the damper actuator and the target angle value of the swirl blade regulating motor based on the state prediction value to construct an optimized stable combustion and energy-saving control command.

[0020] The heat release characterization set includes radiation intensity product, temperature coupled heat intensity, and heat release value corresponding to spatial coordinates. The heat release gradient sequence includes radial inner gradient difference, radial outer gradient difference, and cross-sectional gradient combination. The target adjustment layer includes priority adjustment cross-sectional coordinates, combustion characteristic layer number, and adjustment layer spatial location. The stable combustion and energy-saving control commands include damper actuator adjustment opening and swirl vane adjustment motor adjustment angle. The optimized stable combustion and energy-saving control commands include damper actuator predicted opening, swirl vane adjustment motor predicted angle, and control state prediction correction amount.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: By acquiring the output voltage signal of the furnace flame radiation sensor and the output voltage signal of the flue gas temperature sensor and converting them from analog to digital, the continuous voltage amplitude is mapped into a digital amplitude sequence and sampled discretely. The radiation and temperature sensor amplitudes at multiple sampling times are time-aligned and merged to obtain the radiation temperature sampling sequence. Simulated voltages of flame radiation and flue gas temperature in the furnace physical environment are collected. These simulated voltages are input into an analog-to-digital converter (ADC). The continuous-time simulated voltages are periodically truncated using a set sampling frequency to obtain discrete-time instantaneous voltage values. For example, if the sampling frequency is set to 1000 Hz, the sampling period is 0.001 seconds. The obtained discrete instantaneous voltage values ​​are compared with a reference voltage. Based on the quantization bit width, the instantaneous voltage values ​​are assigned to the corresponding quantization interval, thereby generating digital voltage amplitudes composed of binary code, forming radiation digital amplitude sequences and temperature digital amplitude sequences corresponding to the time axis. The reference voltage is set to 5 volts, the quantization bit width is 12 bits, and the quantization step size is 5 / 4096 = 0.00122 volts. The first sampling timestamp of the radiation digital amplitude sequence and the second sampling timestamp of the temperature digital amplitude sequence are obtained, and the absolute value of the time difference between the first and second sampling timestamps is calculated. A time alignment tolerance threshold is set, which is based on 0.1 times the time scale of the combustion flow field characteristics in the furnace. Through actual flow field measurements in a 500 MW coal-fired power unit furnace, the typical airflow characteristic time scale was found to be 0.1 seconds; therefore, the time alignment tolerance threshold is strictly set to 0.01 seconds. The absolute value of the time difference is then checked to see if it is less than or equal to 0.01 seconds. If the absolute value is less than or equal to 0.01 seconds, the corresponding digital amplitude of radiation and temperature is directly extracted to construct parallel data. If the absolute value is greater than 0.01 seconds, linear interpolation logic is used to calculate the compensation amplitude for the virtual moment using the time deviation ratio. Specifically, the difference in amplitude between two adjacent actual sampling points is multiplied by the ratio of the time difference between the virtual moment and the preceding real sampling point to the adjacent sampling period, and then summed with the amplitude of the preceding real sampling point to obtain the interpolation result. For example, if the digital temperature amplitude is 2.5 volts at time 2 seconds and 2.7 volts at time 2.02 seconds, and we need to align it to the virtual time 2.01 seconds, we calculate the difference between 2.7 and 2.5 (0.2), multiply it by the ratio of 0.01 to 0.02 (0.5), and obtain a difference compensation of 0.1. This is then summed with the previous true amplitude of 2.5 to obtain the interpolated digital temperature amplitude of 2.6 volts. The radiation digital amplitudes and temperature digital amplitudes at the same alignment time are paired and combined to form a radiation temperature sampling sequence.

[0022] S102: Based on the radiation temperature sampling sequence, call the spatial coordinate identifier of multiple sampling points, convert the amplitude of the radiation sensor and the amplitude of the temperature sensor into flue gas temperature and radiation intensity, and pair and combine the radiation intensity value and flue gas temperature value under the same spatial coordinate index to obtain the spatial coordinate dual parameter group. The system retrieves 3D coordinate data from multiple pre-calibrated measuring points within the furnace's 3D space and extracts the hardware channel identification code associated with each data node in the radiation temperature sampling sequence. The hardware channel identification code is matched against the channel coordinates in the mapping dictionary to extract the spatial horizontal, vertical, and depth coordinates uniquely corresponding to each radiation and temperature digital amplitude. The conversion coefficients of the radiation and temperature sensors are then obtained. The radiation sensor conversion coefficient is set based on the linear fitting slope of the blackbody radiation calibration curve. In the calibration of a standard blackbody furnace at 1000°C to 1500°C, the measured voltage output is directly proportional to the radiation intensity, and the fitted conversion coefficient is 45,000 watts per square meter per volt. The temperature sensor conversion coefficient is set based on the average sensitivity of the thermocouple calibration table in the target temperature range. Experimental calculations show that the conversion coefficient is set to 500 degrees Celsius per volt in this high-temperature range. The actual radiation intensity value at the corresponding measuring point is derived by multiplying the radiation digital amplitude with its matched spatial coordinates by the aforementioned radiation sensor conversion coefficient. Simultaneously, the temperature digital amplitude is multiplied by the temperature sensor conversion coefficient to derive the actual flue gas temperature value at the corresponding measuring point. For example, using the data aligned at the virtual time of 2.01 seconds generated in the previous steps, when the radiation digital amplitude input to a certain measuring point channel is 3.2 volts and the temperature digital amplitude is 2.6 volts, 3.2 * 45000 = 144000, yielding an actual radiation intensity value of 144000 watts per square meter, and 2.6 * 500 = 1300, yielding an actual flue gas temperature value of 1300 degrees Celsius. The converted sets of actual radiation intensity values ​​and actual flue gas temperature values ​​are traversed to extract radiation and temperature value nodes with the same spatial horizontal coordinate, spatial vertical coordinate, and spatial depth coordinate. The actual radiation intensity values ​​and actual flue gas temperature values ​​with the same coordinate labels are concatenated into arrays, creating data tuples containing radiation and temperature attributes under the same coordinate index key. All data tuples under the same coordinate index are then summarized to create a spatial coordinate dual-parameter group. To clearly illustrate the data structure of the spatial coordinate dual-parameter group, Table 1 is introduced for explanation.

[0023] Table 1 Mapping Table of Spatial Coordinate Two-Parameter Sets Spatial horizontal coordinates Spatial longitudinal coordinates Spatial depth coordinates Actual radiation intensity value Actual flue gas temperature value 1.5 2.0 5.5 144000 1300 1.5 4.0 5.5 152000 1360 3.0 2.0 5.5 138000 1240 Table 1 lists the coordinates of some spatial measurement points and the corresponding physical quantity values ​​under their indexes.

[0024] S103: Based on the spatial coordinate dual parameter group, extract the multi-coordinate index related radiation intensity and flue gas temperature, introduce a dimensionless coefficient to normalize the radiation intensity and flue gas temperature, combine the multiplication terms and multiply them, integrate all product outputs and spatial coordinate index labels to generate a heat release characterization set. The hierarchical structure stored within the spatial coordinate dual-parameter group is analyzed. The actual radiation intensity and actual flue gas temperature values ​​at each coordinate node are read sequentially according to the multidimensional coordinate array index pointer. The radiation-temperature coupling weight coefficient is obtained. This coefficient is set based on the empirical correlation between heat release rate, temperature, and radiation in combustion dynamics. Multiple linear regression analysis is performed using standard heat release rate data and synchronously measured temperature and radiation data collected over a historical 30-day period under different boiler load conditions to determine the relative relationship between the contribution of radiation to heat release and the contribution of temperature under full load conditions. The calculated radiation-temperature coupling weight coefficient is 0.08. For each extracted coordinate node, the actual radiation intensity value, actual flue gas temperature value, and the radiation-temperature coupling weight coefficient at the current coordinate are multiplied together. For example, substituting the actual radiation intensity of 144,000 watts per square meter and the actual flue gas temperature of 1,300 degrees Celsius at the spatial horizontal coordinate 1.5, spatial vertical coordinate 2.0, and spatial depth coordinate 5.5 calculated in the aforementioned steps, and multiplying them together with the coefficient 0.08, i.e., 144,000 * 1300 * 0.08 = 14,976,000, the heat release product output at this local spatial location is derived to be 14,976,000. A high heat release baseline limit is set. The determination range and setting basis of this high heat release baseline limit value are the top 25 percentile values ​​of the statistical distribution of the heat release product output of the entire furnace under normal operating conditions. Combined with the actual 500 MW unit operating procedures and 50 full furnace cross-section calibration tests, it is determined that the range with values ​​greater than 12,000,000 belongs to the high-activity combustion zone. Therefore, the high heat release baseline limit value is precisely set as 12,000,000. The calculated heat release product output at each location is compared one by one with the high heat release baseline value. In the previous example, the calculated heat release product output of 14,976,000 is significantly greater than the baseline value of 12,000,000, therefore, the region at that coordinate point is determined to be in a high-activity heat release state. All the calculated heat release product outputs, along with their corresponding spatial horizontal coordinates, spatial vertical coordinates, and spatial depth coordinate labels, are matrix-concatenated. Following the coordinate system arrangement rules, all heat release product outputs with coordinate labels are encapsulated into a structured data set, ultimately generating a heat release characterization set.

[0025] Please see Figure 3 The specific steps of S2 are as follows: S201: Obtain the heat release characterization values ​​of the center and transition region of the same cross section in the heat release characterization set, perform difference calculation on the heat release values ​​of the center position and the transition position according to the radial coordinate order of the cross section, and arrange them according to the radial coordinate order to generate a radial inner gradient difference sequence. All data nodes in the heat release characterization set with spatial coordinate labels were extracted, and a subset of horizontal cross-section data with a spatial depth coordinate of 5.5 meters was selected. The spatial lateral and longitudinal coordinates of each measuring point within this subset were analyzed, and the geometric Euclidean distance from each measuring point to the geometric center of the cross-section was calculated. This geometric Euclidean distance was defined as the radial coordinate value. Cross-section region division benchmarks were set, including a central region boundary threshold and a transition region boundary threshold. The central region boundary threshold was set based on the radius of the turbulent core region of the burner's primary and secondary air jet mixing. Fluid dynamics simulations showed the core region radius to be 3 meters under 500 MW conditions; therefore, the central region boundary threshold was set to 3 meters. The transition region boundary threshold was set based on the critical distance at which significant attenuation of radiative heat transfer in the furnace water-cooled wall occurred. Combining this with historical data from 100 cross-section temperature distribution inflection points, this critical distance was determined to be 7 meters; therefore, the transition region boundary threshold was set to 7 meters. Traversing all data nodes of the cross-section, nodes with radial coordinate values ​​less than or equal to 3 meters are classified as central region nodes, and their corresponding central location heat release characterization values ​​are extracted. Nodes with radial coordinate values ​​greater than 3 meters and less than or equal to 7 meters are classified as transition region nodes, and their corresponding transition location heat release characterization values ​​are extracted. Substituting the heat release product output of 14,976,000 calculated in the previous steps, since its corresponding spatial horizontal coordinate of 1.5 and spatial vertical coordinate of 2.0 results in a radial coordinate value of 2.5 meters, which is less than 3 meters, 14,976,000 is extracted as the first central location heat release characterization value. Simultaneously, the transition region node with a radial coordinate value of 4.5 meters on the same radial ray is obtained, and its transition location heat release characterization value of 11,200,000 is extracted. The difference between the heat release characterization value at the center position and the heat release characterization value at the transition position on the corresponding radial coordinate is calculated. Specifically, the difference between the heat release characterization value at the first center position (14976000) and the corresponding heat release characterization value at the transition position (11200000) is calculated, resulting in an inner difference value of 3776000 for this radial position. The same difference calculation is performed on all nodes on all radial rays within the cross-section to obtain multiple inner difference values. These extracted inner difference values ​​are then arranged in ascending order of their radial coordinate values. The sorted inner difference values ​​and their corresponding radial coordinate labels are encapsulated into a one-dimensional data array to generate a radial inner gradient difference sequence. The calculated inner difference value of 3776000 is compared with the inner abnormal fluctuation baseline limit of 4000000. Since 3776000 is less than 4000000, it indicates that the current inner combustion heat transfer gradient is in a stable transition state.

[0026] S202: Obtain the heat release characterization value of the wall-attached region based on the heat release characterization set, calculate the difference between the heat release values ​​of the transition region and the wall-attached region according to the radial coordinate order, and call the radial inner gradient difference sequence for order verification to obtain the radial outer gradient difference sequence. The extracted horizontal cross-sectional data subset is retrieved, and the remaining data nodes within the subset that are not classified as central or transitional regions are read. A boundary threshold for the wall-adhering region is set, based on the maximum thermal boundary thickness extending inward from the physical isolation layer of slag on the furnace inner wall. Through statistical measurements of the slag thickness on the water-cooled wall during previous unit overhauls, combined with the abrupt change in near-wall heat flux density, this boundary thickness is determined to be within 0.5 meters. Given that the furnace half-width is 10 meters, the interval from 9.5 meters to 10 meters belongs to the wall-adhering influence zone. Therefore, the radial coordinate judgment interval for the wall-adhering region is set to be greater than 9.5 meters and less than or equal to 10 meters. Data nodes with radial coordinate values ​​within this judgment interval are selected, classified as wall-adhering region nodes, and their corresponding wall-adhering region heat release characterization values ​​are extracted. A wall-adhering region node with a radial coordinate value of 9.8 meters on the same radial ray is obtained, and its corresponding wall-adhering region heat release characterization value of 6,500,000 is extracted. The heat release characterization value of 11200000 at the transition position with a radial coordinate of 4.5 meters, obtained in the previous steps, is retrieved. The heat release characterization value at the transition position in the same radial direction is extracted and subtracted from the heat release characterization value of the wall-attached region. Specifically, the difference between the transition position heat release characterization value 11200000 and the wall-attached region heat release characterization value 6500000 is calculated, resulting in an outer difference value of 4700000 for this radial position. This subtraction process is repeated for all defined radial rays within the cross-section to obtain multiple outer difference values. These outer difference values ​​are then arranged in ascending order of their corresponding radial coordinate values ​​to generate an initial radial outer difference value data sequence. The radial inner gradient difference value sequence generated in the previous steps is retrieved, and its radial ray index label set is read. The ray index labels of the initial radial outer difference value data sequence are compared and verified one-to-one with the ray index labels of the radial inner gradient difference value sequence. If a ray index exists in the inner sequence but is missing in the outer sequence, the corresponding inner difference node is removed, ensuring an absolute correspondence between the inner and outer sequences in space. After sequential verification and synchronous reduction, the final radial outer gradient difference sequence is output. The calculated outer difference value of 4700000 is compared with the outer slag formation warning baseline limit of 5000000. Since 4700000 is less than 5000000, it indicates that a slag layer that severely hinders heat transfer has not yet appeared in the current wall-attached region.

[0027] S203: Call the radial inner gradient difference sequence and the radial outer gradient difference sequence, perform sequence splicing operation according to the unified radial position index, and continuously arrange the two sets of gradient difference data to generate the heat release gradient sequence; The generated radial inner gradient difference sequence and radial outer gradient difference sequence are analyzed, and data tuples are extracted from the two sequences respectively. Each data tuple contains a unique radial azimuth index and radial distance index. A data splicing synchronization tolerance range is set, which is based on the error range of the angle deflection effect caused by airflow swirl within the same horizontal cross section. By measuring the airflow rotation angle deviation generated by tangential combustion at the four corners of the furnace, the maximum reasonable deflection angle is determined to be 2 degrees, so the splicing synchronization tolerance range of the angle index is set to 2 degrees. The first inner difference value 3776000 and its corresponding radial azimuth index of 15 degrees are extracted from the radial inner gradient difference sequence. In the radial outer gradient difference sequence, outer difference data with azimuth indexes between 13 and 17 degrees are searched. The outer difference value within the corresponding angle range is found to be 4700000, which was calculated above. The radial azimuth index corresponding to the found outer difference value is forcibly corrected to 15 degrees, thus completing the unified operation of the radial position index. Under a unified radial azimuth index of 15 degrees, and extending along the physical distance from the center to the wall, the inner difference values ​​from the inner radial gradient difference sequence are placed at the beginning of the array, and the outer difference values ​​from the outer radial gradient difference sequence are placed at the end of the array. Sequence concatenation is then performed. Specifically, the inner difference value 3776000 and the outer difference value 4700000 are sequentially placed into the same data structure in ascending order of radial coordinates, constructing a continuously arranged single-ray gradient data chain. All unified radial position indices are traversed, and the single-ray gradient data chains generated under different azimuth angles are combined and arranged in a counter-clockwise order from 0 degrees to 360 degrees. All the concatenated continuous data chains are encapsulated in a multi-dimensional array, and a label corresponding to a cross-sectional height of 5.5 meters is added to the data header, ultimately generating a complete heat release gradient sequence characterizing the heat transfer trend of the entire horizontal cross-section. The total length of the concatenated data chain is compared and verified with the baseline value of 72 rays in the cross-sectional design. Assuming that the actual number of data chains completed is 72, which is exactly the same as the baseline value of 72, it indicates that the gradient sequence of the entire cross section is completely spliced ​​without any missing data.

[0028] Please see Figure 4 The specific steps of S3 are as follows: S301: Extract the radial inner gradient difference and radial outer gradient difference of multiple cross sections based on the thermal release gradient sequence, call the radial outer gradient difference at the same coordinate position to compare the difference magnitude, record the gradient difference magnitude values ​​of multiple cross sections and serialize them according to the cross section coordinate order to generate the cross section gradient difference magnitude sequence. The generated heat release gradient sequence, including the section height marker of 5.5 meters, is extracted. Simultaneously, heat release gradient sequences from multiple additional sections with heights of 10.5 meters and 15.5 meters are read in ascending order of spatial depth coordinates. For each section's heat release gradient sequence, the encapsulated single-ray gradient data chain is parsed, and the radial inner gradient difference and radial outer gradient difference at each radial azimuth angle are extracted. Substituting these values ​​into the previously calculated radial inner gradient difference of 3776000 and radial outer gradient difference of 4700000 at a section height of 5.5 meters and a radial azimuth angle of 15 degrees, the absolute difference calculation is performed on the radial outer gradient difference and radial inner gradient difference at the same section height and radial azimuth angle to derive the gradient difference amplitude at that spatial coordinate position. Substituting the radial outer gradient difference of 4700000 and the radial inner gradient difference of 3776000 into the calculation, the initial gradient difference amplitude at the current coordinate position is derived to be 924000. An amplitude correction factor was set, which was obtained by integrating the heat flux density decay curve in the vertical direction of the furnace. Considering that the 5.5-meter height of the furnace is located in the main combustion zone of the burner, where radiative heat transfer is dominant, after nearly 30 days of wall heat flux meter calibration tests, the amplitude correction factor for this height was determined to be 1.2. The obtained initial gradient difference amplitude value was multiplied by the amplitude correction factor to derive the corrected gradient difference amplitude value. The initial gradient difference amplitude value of 924000 was multiplied by the amplitude correction factor of 1.2 to derive the corrected gradient difference amplitude value of 1108800. The above absolute difference calculation and correction product calculation were repeated for all radial azimuth angles at all cross-sectional heights to obtain multiple sets of corrected gradient difference amplitude values ​​corresponding to cross-sectional positions. For the multiple sets of corrected gradient difference amplitude values ​​obtained, a one-dimensional array was reorganized and serialized according to the spatial depth coordinates from small to large and the radial azimuth angle within the same cross section from 0 degrees to 360 degrees. Each corrected gradient difference amplitude value was tightly bound to its corresponding spatial coordinate label, ultimately generating a cross-sectional gradient difference amplitude sequence. The calculated corrected gradient difference amplitude value of 1108800 was compared with the set amplitude fluctuation benchmark limit value of 1500000. It was determined that 1108800 is less than 1500000, indicating that the difference in heat transfer gradient between the inner and outer sides of the current cross section is within a safe and controllable reasonable range.

[0029] S302: Based on the cross-sectional gradient difference magnitude sequence, extract the gradient difference magnitude of multiple cross sections, and use a multilayer perceptron model to filter feature layers and extract a set of candidate layers. Select cross-sectional coordinate indices and magnitudes of the candidate layers whose magnitudes exceed a preset gradient magnitude threshold and sort them in descending order to generate a candidate layer ranking sequence. The generated cross-sectional gradient difference amplitude sequence is retrieved, and the multi-dimensional array structure in the sequence is parsed to extract a set of multi-sectional gradient difference amplitude values ​​containing spatial depth coordinate labels. The extracted multi-sectional gradient difference amplitude values ​​are preprocessed with extreme value normalization. The maximum value limit of 5,000,000 and the minimum value limit of 0 recorded in the historical database are read. The current extracted amplitude value is subtracted from the minimum value limit, and then divided by the difference between the maximum and minimum values, transforming it into standard normalized input data distributed between 0 and 1. Substituting the corrected gradient difference amplitude value of 1108800 obtained in the previous steps, subtracting 0, and dividing by 5,000,000, the normalized input data is derived as 0.22176. A multilayer perceptron network structure is constructed, setting the number of input layer nodes to 72, corresponding to the 72 radial azimuth angle input data of a single cross-section, and a fully connected network topology is constructed. Three hidden layers were configured: the first hidden layer had 128 neurons, the second had 64 neurons, and the third had 32 neurons. All hidden layers were fully connected, and a linear rectified activation function was used for non-linear feature mapping to suppress the vanishing gradient phenomenon. The output layer had one neuron and used a logistic sigmoid activation function to output the magnitude risk assessment index corresponding to the given cross-section. The training process of this multilayer perceptron network used the mean squared error loss function. The error level was assessed by calculating the sum of squared deviations between the magnitude risk assessment index predicted by the network and the true risk labels labeled by experts. An adaptive moment estimation optimization algorithm was used to iteratively update the network parameters. The learning rate was set to 0.001, the batch size to 32, and the maximum number of iterations to 500. Table 2 is provided to clearly illustrate the selection process for the multi-cross-section data.

[0030] Table 2 Multi-section gradient amplitude evaluation table Spatial depth coordinates Correcting the gradient difference magnitude value Magnitude Risk Assessment Index 5.5 1108800 0.35 10.5 2850000 0.82 15.5 1620000 0.51 Table 2 lists some cross-sectional assessment data after model processing. The normalized data is input into the trained multilayer perceptron network, and the network outputs the amplitude risk assessment index for each cross-section during forward propagation. The coordinate indices and amplitude values ​​of cross-sections exceeding a preset gradient amplitude threshold are extracted. The preset gradient amplitude threshold is set based on the statistical upper limit of the risk index under safe operating conditions. Combined with historical cases of superheater tube wall overheating, it is determined that an index exceeding 0.7 indicates a critical tube burst risk; therefore, the threshold is set to 0.7. In the above processing results, the amplitude risk assessment index of the 10.5-meter cross-section (0.82) is greater than 0.7, so its coordinate index and amplitude value are extracted. All extracted cross-sectional data exceeding the threshold are sorted in descending order of amplitude risk assessment index. The sorted coordinate indices and values ​​are encapsulated into a one-dimensional column vector to generate a candidate layer ranking sequence.

[0031] S303: Call the candidate layer ranking sequence to perform ranking index retrieval, extract the cross-sectional coordinate index that is first in the ranking position, retrieve the corresponding spatial layer parameters and encode the layer identifier, convert the encoded layer parameters into a coordinate layer command format that can be called by the control terminal, and obtain the target adjustment layer; The generated candidate layer ranking sequence is invoked, and its memory starting address is read to perform a ranking index retrieval operation. Based on the descending order data structure characteristics, the cross-sectional coordinate index at position 1 is directly located and extracted. Substituting the ranking results obtained from the previous process, since the amplitude risk assessment index of 0.82 corresponding to the spatial depth coordinate of 10.5 meters is the largest among all data exceeding the preset gradient amplitude threshold, the spatial depth coordinate of 10.5 meters is extracted as the first cross-sectional coordinate index. The underlying configuration database of the unit's distributed control is accessed, and multiple spatial layer parameters strictly corresponding to this spatial layer are retrieved by searching for the spatial depth coordinate of 10.5 meters. The retrieved spatial layer parameters include the burner layer number of layer 3 to which this height belongs, and the physical equipment address code and real-time opening feedback benchmark value of the secondary damper damper arranged at this layer. The layer identification feature code is obtained. The setting of this layer identification feature code is based on the security encryption verification header file required for control identification of different physical layer heights. By consulting the communication protocol specification of the programmable logic controller, the identification feature code of layer 3 is determined to be a fixed value of 103. Extract the burner level number 3 from the spatial layer parameters and perform a combined multiplication operation with the layer identifier feature code 103. Set the basic layer control protocol offset, and determine that the offset is set to 2000 based on the offset addressing rules of the communication bus. Multiply the burner level number 3 by 103 and add 2000 to calculate the target instruction base address 2309. According to the specific message structure supported by distributed control, merge the derived target instruction base address 2309 with the corresponding spatial depth coordinate value 10.5 and the damper device address code into a byte stream, perform layer identifier encoding, and then convert the encoded layer parameters into a coordinate layer instruction format that can be called by the control terminal. Parse the payload segment of this instruction format and perform cyclic redundancy check. If the check result is consistent with the preset standard check, the data packet that has undergone strict encoding and format conversion is finally output, and it is established as the target regulating layer. The established target adjustment level command is sent to the damper actuator. The target command base address 2309 is compared with the upper and lower limits of the executable address space, which is 2000 to 3000. It is determined that 2309 is within the reasonable range, indicating that the generated control command conforms to the device addressing specification and can be accurately identified by the actuator.

[0032] Please see Figure 5 The specific steps of S4 are as follows: S401: Obtain the heat release characterization values ​​of four adjacent directions corresponding to the same cross section of the target adjustment layer, perform clockwise subtraction to calculate multiple azimuth circumferential gradient difference sequences, call the numerical relationship between the heat release characterization values ​​of the four adjacent directions to perform circumferential gradient attribution calculation, and establish the azimuth circumferential gradient difference sequence. The command for the target adjustment layer with a spatial depth coordinate of 10.5 meters was extracted. The corresponding cross-sectional spatial location was analyzed, and four adjacent physical sensor nodes located on the same horizontal plane of this cross-section were retrieved. These four adjacent orientations were defined as 0°, 90°, 180°, and 270°. The sensor data cache from the past 10 minutes was read, and the corresponding thermal emission characterization values ​​for these four orientations were extracted. The obtained thermal emission characterization values ​​were 15,000,000 for 0°, 14,500,000 for 90°, 13,800,000 for 180°, and 14,800,000 for 270°. Table 3 is provided to clearly illustrate the multi-directional data structure.

[0033] Table 3. Test table for heat release characterization in four adjacent directions. Directional names Sensor node number Heat release characterization value 0 degrees Node 11 15000000 90-degree orientation Node 12 14500000 180-degree orientation Node 13 13800000 270-degree orientation Node 14 14800000 Table 3 lists the four adjacent azimuths within the cross-section and their corresponding measurement values. The first circumferential difference is derived by subtracting the thermal release values ​​at the 0° and 90° azimuths clockwise. Substituting 15,000,000 and 14,500,000 into the subtraction operation yields a first circumferential difference of 500,000. Similarly, subtracting the thermal release values ​​at the 90° and 180° azimuths (14,500,000 minus 13,800,000) yields a second circumferential difference of 700,000. Subtracting the thermal release values ​​at the 180° and 270° azimuths (13,800,000 minus 14,800,000) results in a third circumferential difference of -1,000,000. Subtracting the 270-degree azimuth from the 0-degree azimuth (i.e., subtracting 15,000,000 from 14,800,000) yields the fourth circumferential difference of -200,000. The numerical relationship between the heat release characterization values ​​of the four adjacent azimuths is then used to perform circumferential gradient assignment calculations. A gradient assignment benchmark parameter is set, calculated by linearly fitting the streamline deflection curvature of a cold-state aerodynamic field test. Combined with calibration data from 50 historical ignition tests of this specific boiler model, this benchmark parameter is precisely set to 300,000. The absolute values ​​of each calculated circumferential difference are extracted and compared with this gradient assignment benchmark parameter. For the first circumferential difference of 500,000, its absolute value is greater than the benchmark parameter 300,000, indicating a significant thermal bias in this azimuth interval, and it is assigned the strong bias gradient label. For the fourth circumferential difference of 200,000, its absolute value is less than the benchmark parameter 300,000, and it is assigned the weak bias gradient label. The multiple azimuth circumferential differences assigned to the attribution labels are sorted in clockwise spatial angles of 0 degrees, 90 degrees, 180 degrees, and 270 degrees. The difference data, along with the corresponding azimuth labels and attribution labels, are stored sequentially in a one-dimensional memory structure to establish an azimuth circumferential gradient difference sequence.

[0034] S402: Based on the azimuth circumferential gradient difference sequence, extract the current opening degree of the corresponding azimuth damper actuator and the current angle of the swirl blade adjusting motor, reduce the damper opening value, increase the swirl blade adjusting motor angle value, and generate a set of damper opening and blade angle adjustment values; Retrieve the generated azimuth circumferential gradient difference sequence. Analyze the data nodes with strong skew gradient labels in the sequence and extract the first circumferential difference value of 500000 from the corresponding 0-degree azimuth node. Based on this 0-degree azimuth label, access the lower-level control register of the programmable logic controller (PLC) to read the current opening value of the damper actuator and the current angle value of the swirl vane adjusting motor for the corresponding azimuth. The current opening value of the damper actuator at the 0-degree azimuth is found to be 65 degrees, and the current angle value of the swirl vane adjusting motor is 45 degrees. Perform a reduction operation on the damper opening value. Set a damper opening reduction conversion coefficient, which is based on the negative correlation fitting function between the excess air coefficient and the heat release deviation. This coefficient was calculated by collecting 100 sets of secondary damper disturbance test data under full-load conditions. The reduction conversion coefficient is strictly set to a damper closing action of 1 unit per 100,000 units of heat release difference. Dividing the first circumferential difference of 500,000 by the base unit of 100,000, and then multiplying it by the reduction conversion coefficient of 1, yields a damper opening reduction of 5. Subtracting the current opening value of 65 from the damper opening reduction of 5, the adjusted target damper opening value is 60. The swirl blade adjustment motor angle value is extracted and increased. A blade angle increase conversion coefficient is set, which is derived from the equation of conservation of rotational momentum in the air-powder mixing flow field. After on-site cold velocity measurement and calibration, it is set to correspond to a blade outward expansion angle of 1.5 degrees for every 100,000 units of heat release difference. Dividing the first circumferential difference of 500,000 by the base unit of 100,000, and then multiplying it by the increase conversion coefficient of 1.5, yields a blade angle increase of 7.5 degrees. Adding the current angle value of 45 to the blade angle increase of 7.5, the adjusted target swirl blade angle value is 52.5 degrees. The calculated target damper opening value of 60 degrees and the target swirl blade angle value of 52.5 degrees are bundled into a data structure, and a physical location label of 0 degrees is attached before being stored in the memory cache stack. The circumferential gradient difference values ​​for the remaining three azimuths are traversed, and the multiplication and addition / subtraction operations for opening reduction and angle increase are repeatedly performed to obtain the target opening and angle adjustment data for all azimuths. The data nodes for all azimuths are combined and arranged into a two-dimensional data array to generate a set of damper opening and blade angle adjustment values.

[0035] S403: Performs numerical integration and merging of damper opening and blade angle adjustment value sets, calls up the accumulated information of multi-directional opening and angle adjustment values ​​for summary calculation, and establishes stable combustion and energy-saving control commands; Extract the generated set of damper opening and blade angle adjustment values, which includes multiple target damper opening values ​​and target swirl blade angle values. Perform numerical integration and merging operations on the discrete data within this set of damper opening and blade angle adjustment values. Following a clockwise spatial sorting rule, sequentially extract the target damper opening values ​​corresponding to the 0-degree, 90-degree, 180-degree, and 270-degree azimuths and store them in independent one-dimensional arrays of damper opening values. Store the previously calculated target damper opening value of 60 (0-degree azimuth) at the beginning of the array, and simultaneously read the extracted target damper opening values ​​of 58 (90-degree azimuth), 62 (180-degree azimuth), and 65 (270-degree azimuth), and complete the array sequence accordingly. Perform a summary calculation using the accumulated information of the multi-azimuth opening and angle adjustment values. A summation operation is performed on the four opening values ​​in the one-dimensional array of damper openings: 60, 58, 62, and 65 are summed to derive a total cumulative damper opening value of 245. This total cumulative damper opening value of 245 is divided by 4 to calculate the average opening baseline value for the entire cross-section as 61.25. A tolerance control range for the total air volume of the cross-section is set. This range is determined based on the air distribution design specifications required for stable operation of the boiler's overall aerodynamic field, combined with historical data from 300 unit performance tests, and is determined to be a safe and economical reasonable range of 50 to 75. The calculated average opening baseline value of 61.25 for the entire cross-section is compared with this tolerance control range for the total air volume of the cross-section. It is determined that 61.25 strictly falls within the value range of 50 to 75, indicating that the current opening reduction and distribution scheme in each direction has not caused a severe shortage or excess of the overall air supply. The target instruction base address 2309 generated in the previous steps is retrieved. Using the target instruction base address 2309 as the address header feature code of the communication message, a bitstream data splicing operation is performed on the data in the integrated one-dimensional array of damper opening and the set of swirl blade angle adjustment values ​​generated by merging using the same logic. A cyclic redundancy parity check bit is appended to the end of the spliced ​​instruction data packet to establish a continuous byte stream structure containing the address path, the execution equipment type identification code, and specific multi-directional opening and angle action parameters, ultimately establishing the stable combustion and energy-saving control instruction.

[0036] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the stable combustion and energy-saving control command, extract the target opening value of the damper actuator and the target angle value of the swirl blade regulating motor and jointly map them. Perform differential calculations on the opening value and angle value for adjacent angle changes and adjacent opening changes, and perform same-frequency data alignment to generate a damper swirl coordination parameter set. The stable combustion and energy-saving control command is parsed, and the effective payload data segment in the command byte stream is read. The target opening value of the damper actuator (60) and the target angle value of the swirl vane regulating motor (52.5 degrees) corresponding to the 0-degree azimuth in the current control cycle are extracted. The historical target opening value of the damper (55) and the historical target angle value of the swirl vane (45 degrees) issued 500 milliseconds ago at the corresponding 0-degree azimuth are read. A joint mapping operation is performed between the target opening value of the damper actuator in the current cycle and the historical target opening value. A double-key binding mechanism combining spatial azimuth coordinate labels and control timestamps is used to establish a memory pointer association between the opening value and angle value corresponding to the same 0-degree physical azimuth and two adjacent control cycles in the time series. The current target opening value (60) and the historical opening value (55) are extracted and subjected to discrete subtraction difference operation, deriving that the adjacent opening change of the damper actuator is 5. The current target angle value of 52.5 degrees and the historical angle value of 45 degrees, after synchronous extraction and correlation, are subjected to discrete subtraction difference operation, which derives the adjacent angle change of the swirl vane regulating motor as 7.5 degrees. The original sampling frequency of the opening feedback signal of the bottom damper actuator is 10 times per second, and the original sampling frequency of the angle feedback signal of the swirl vane regulating motor is 5 times per second. In view of the time axis misalignment caused by asynchronous sampling, the above-calculated changes are processed for same-frequency data alignment. The global reference synchronous sampling frequency is set to 5 times per second, and the opening change data buffer queue with the higher sampling frequency is extracted. The arithmetic mean of two consecutive opening changes with a time span of exactly 100 milliseconds is performed, that is, the previously calculated current change of 5 is added to the change of 3 in the next sampling cycle, and then divided by the smoothing constant 2, which derives the adjacent opening change after same-frequency alignment as 4. The aligned adjacent opening change amount 4 and the adjacent angle change amount 7.5 degrees at the reference frequency are horizontally concatenated into an array, assigned a 0-degree azimuth feature label and a unified synchronization timestamp, and stored in a circular buffer to generate a damper swirl coordination parameter set containing action increment information.

[0037] S502: Based on the damper swirl coordination parameter set, perform state prediction calculation, sum the independent sliding intervals of the continuous change trend sequence of opening and the continuous change trend sequence of angle respectively, calculate the state amplitude of opening and angle intervals respectively, and compare it with the preset fluctuation threshold to obtain the damper swirl prediction state group. The generated damper swirl coordination parameter set is retrieved and state prediction calculations are performed. The historical data sequence recorded in the annular buffer is parsed, and the changes in adjacent opening degrees and adjacent angles at 0 degrees for the most recent five consecutive synchronization cycles are extracted to form a continuous trend sequence. The set of adjacent opening degree changes in the continuous trend sequence is read, and the historical value sequences obtained are 4, 3, 5, 4, and 2. The set of adjacent angle changes is read synchronously, and the historical value sequences obtained are 7.5 degrees, 6.0 degrees, 5.5 degrees, 4.0 degrees, and 3.0 degrees. The change data in the above sequence set is used to perform a time-series recursive calculation operation. A sliding time window of a set length of five cycles is extracted, and the adjacent opening degree changes falling within the sliding interval are summed. That is, the sequence values ​​4, 3, 5, 4 and 2 are continuously added together, deriving the interval state amplitude of the corresponding damper opening as 18. Using the same sliding interval summation calculation logic, the adjacent angle changes of 7.5 degrees, 6.0 degrees, 5.5 degrees, 4.0 degrees, and 3.0 degrees falling within the time window are continuously added together, resulting in a corresponding interval state amplitude of 26 degrees for the swirl blade angle. A preset fluctuation threshold is set, based on the critical airflow change rate for anti-surge testing of the burner region aerodynamic field. Through simulation of wind pressure surges caused by different action rates on a cold-state ventilation test bench, the preset fluctuation threshold for the opening is determined to be 15 degrees, and the preset fluctuation threshold for the angle is 20 degrees over five cycles. The interval state amplitude of the damper opening (18 degrees) is compared with the preset fluctuation threshold of 15, and 18 is determined to be greater than 15. Simultaneously, the interval state amplitude of the angle (26 degrees) is compared with the preset fluctuation threshold of 20 degrees, and 26 degrees is determined to be greater than 20 degrees. Based on the above dual limit determination results, the control channel at the current 0-degree azimuth position is assigned a sudden limit-crossing feature label. The feature data with the limit-crossing label and the corresponding interval state amplitude are encapsulated into a multi-dimensional structure to establish the damper swirl prediction state group.

[0038] S503: Based on the damper swirl prediction state group and stable combustion energy-saving control command, extract the target opening value of the corresponding damper actuator and the target angle value of the swirl blade adjustment motor inside the command and perform numerical replacement calculation. Then, reorganize the replacement parameters into a sequence to generate an optimized stable combustion energy-saving control command. The generated damper swirl prediction state group is analyzed, and the 0-degree azimuth data node containing the action change over-limit feature label is identified. Based on the state feedback signal, the stable combustion and energy-saving control instruction byte stream temporarily stored in memory is accessed synchronously, and the target opening value of the damper actuator (60) and the target angle value of the swirl blade regulating motor (52.5 degrees) corresponding to the 0-degree azimuth are extracted. For the mechanical overshoot risk state revealed in the prediction state group, the target opening value and target angle value are extracted and a smooth numerical replacement operation is performed. A dynamic mitigation coefficient is set, which is based on the normalized ratio of the over-limit difference of the interval state amplitude. The over-limit difference value of 3 is obtained by subtracting the preset fluctuation threshold value of 15 from the previously calculated damper opening interval state amplitude value of 18. Dividing this difference value of 3 by the preset fluctuation threshold value of 15, the over-limit ratio is derived to be 0.2. Setting the standard mitigation base constant to 0.8, and subtracting the over-limit ratio of 0.2 from the standard mitigation base constant 0.8, the dynamic mitigation coefficient on the damper side is derived to be 0.6. The historical damper target opening value of 55 at the 0-degree azimuth and the adjacent opening change amount of 5 are obtained from the previous steps. The adjacent opening change amount of 5 is multiplied by the dynamic mitigation coefficient 0.6 to obtain the limit increment 3. The historical damper target opening value of 55 and the limit increment 3 are added together to derive the replacement opening value of 58. Using the same logic, the mitigation ratio on the swirl blade side is calculated. The angle interval state amplitude of 26 degrees is subtracted from the preset angle fluctuation threshold of 20 degrees to obtain the over-limit difference of 6 degrees. This difference of 6 degrees is divided by the threshold of 20 degrees to obtain the ratio 0.3. Subtracting 0.3 from the base constant 0.8 yields 0.5. Multiplying the adjacent angle change of 7.5 by 0.5 yields a limit increment of 3.75 degrees. Adding this limit increment to the historical angle value of 45 degrees, the replacement angle value is derived to be 48.75 degrees. The replacement opening value 58 and the replacement angle value 48.75 degrees are then used to overwrite and replace the corresponding target opening value 60 and target angle value 52.5 degrees within the original control command. A sequence recombination operation is performed on the replaced 0-degree azimuth parameter and other retained parameters that do not exceed the limit. Based on the clockwise increasing physical index order of spatial azimuth, all updated opening and angle values ​​are concatenated bitwise, and a new address header and data checksum are added to generate the optimized stable combustion and energy-saving control command. Comparing the original target opening value 60 with the optimized replacement opening value 58, the decrease in this value indicates that the feedforward correction logic actively intervened and reduced overly aggressive air intake command operations.

[0039] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit, characterized in that, Includes the following steps: S1: Collect the output voltage signals of the boiler furnace flame radiation and flue gas temperature sensors, perform analog-to-digital conversion and discrete sampling, extract the radiation intensity and flue gas temperature under the same spatial coordinates, and perform weighted coupling calculations in combination with preset dimension conversion coefficients to generate a heat release characterization set; S2: Extract the heat release characterization values ​​of the center, wall-attached and transition regions of the same cross section from the heat release characterization set, calculate the radial outer gradient difference and the radial inner gradient difference and merge them to construct a heat release gradient sequence; S3: Based on the heat release gradient sequence, extract the radial inner gradient difference and radial outer gradient difference of multiple cross sections, input them into the multilayer perceptron model for feature layer selection, extract the first cross section coordinates of the feature layer selection output, and generate the target adjustment layer; S4: Based on the heat release characterization values ​​of multiple adjacent directions corresponding to the target adjustment layer, calculate multiple directional circumferential gradient difference sequences, extract the current opening degree of the damper actuator and the current angle of the swirl blade adjustment motor in the corresponding direction, reduce the opening value of the damper actuator and increase the angle value of the swirl blade adjustment motor, and independently encode and combine the adjusted values ​​to generate stable combustion and energy-saving control instructions.

2. The boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 1, characterized in that, The heat release characterization set includes radiation intensity product, temperature coupled heat intensity, and heat release value corresponding to spatial coordinates. The heat release gradient sequence includes radial inner gradient difference, radial outer gradient difference, and cross-sectional gradient combination. The target adjustment layer includes priority adjustment cross-sectional coordinates, combustion characteristic layer number, and adjustment layer spatial position. The stable combustion and energy-saving control command includes adjusting the opening degree of the damper actuator and adjusting the angle of the swirl blade adjustment motor.

3. The boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: By acquiring the output voltage signal of the furnace flame radiation sensor and the output voltage signal of the flue gas temperature sensor and converting them from analog to digital, the continuous voltage amplitude is mapped into a digital amplitude sequence and sampled discretely. The radiation and temperature sensor amplitudes at multiple sampling times are time-aligned and merged to obtain the radiation temperature sampling sequence. S102: Based on the radiation temperature sampling sequence, call the spatial coordinate identifier of the multi-sampling point, convert the radiation sensor amplitude and temperature sensor amplitude into flue gas temperature and radiation intensity, and pair and combine the radiation intensity value and flue gas temperature value under the same spatial coordinate index to obtain the spatial coordinate dual parameter group. S103: Based on the spatial coordinate dual parameter set, extract the multi-coordinate index-related radiation intensity and flue gas temperature, introduce a dimensionless coefficient to normalize the radiation intensity and flue gas temperature, combine the multiplication terms, integrate all product outputs and spatial coordinate index labels, and generate a heat release characterization set.

4. The boiler stable combustion and energy-saving control method under deep peak shaving of a unit according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Obtain the heat release characterization values ​​of the center and transition region of the same cross section in the heat release characterization set, perform difference calculation on the heat release values ​​of the center position and the transition position according to the radial coordinate order of the cross section, and arrange them according to the radial coordinate order to generate a radial inner gradient difference sequence. S202: Based on the heat release characterization set, obtain the heat release characterization value of the wall-attached region, calculate the difference between the heat release values ​​of the transition region and the wall-attached region according to the radial coordinate order, and call the radial inner gradient difference sequence for order verification to obtain the radial outer gradient difference sequence. S203: Call the radial inner gradient difference sequence and the radial outer gradient difference sequence, perform sequence splicing operation according to the unified radial position index, and arrange the two sets of gradient difference data continuously to generate a heat release gradient sequence.

5. The boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the heat release gradient sequence, extract the radial inner gradient difference and radial outer gradient difference of multiple cross sections, call the radial outer gradient difference at the same coordinate position to compare the difference magnitude, record the gradient difference magnitude values ​​of multiple cross sections and serialize them according to the cross section coordinate order to generate the cross section gradient difference magnitude sequence. S302: Based on the cross-sectional gradient difference magnitude sequence, extract the gradient difference magnitude of multiple cross sections, and use a multilayer perceptron model to filter feature layers and extract a set of candidate layers. Select cross-sectional coordinate indices and magnitudes of the candidate layers whose magnitudes exceed a preset gradient magnitude threshold and arrange them in descending order to generate a candidate layer ranking sequence. S303: Call the candidate layer ranking sequence to perform ranking index retrieval, extract the cross-sectional coordinate index that is first in the ranking position, retrieve the corresponding spatial layer parameters and perform layer identification encoding, convert the encoded layer parameters into a coordinate layer command format that can be called by the control terminal, and obtain the target adjustment layer.

6. The boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 5, characterized in that, The gradient amplitude threshold is determined by extracting the gradient amplitude sample values ​​from multiple cross-sections in the thermal release gradient dataset, performing statistical distribution operations on the sample value sequence, obtaining the gradient amplitude mean and gradient amplitude variance, and calculating the gradient fluctuation baseline component by weighted summation, and then performing a product operation in combination with a preset sensitivity coefficient.

7. The boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Obtain the heat release characterization values ​​of four adjacent directions corresponding to the same cross section of the target adjustment layer, perform clockwise subtraction to calculate multiple directional circumferential gradient difference sequences, call the numerical relationship between the heat release characterization values ​​of the four adjacent directions to perform circumferential gradient attribution calculation, and establish the directional circumferential gradient difference sequence. S402: Based on the azimuth circumferential gradient difference sequence, extract the current opening degree of the corresponding azimuth damper actuator and the current angle of the swirl blade adjusting motor, reduce the damper opening value, increase the swirl blade adjusting motor angle value, and generate a damper opening and blade angle adjustment value set. S403: Perform numerical integration and merging on the set of damper opening and blade angle adjustment values, call the accumulated information of multi-directional opening and angle adjustment values ​​for summary calculation, and establish a stable combustion and energy-saving control command.

8. The boiler stable combustion and energy-saving control method under deep peak shaving of a unit according to claim 1, characterized in that, The method further includes: S5: Based on the stable combustion and energy-saving control command, extract the target opening value of the damper actuator and the target angle value of the swirl vane adjusting motor and perform state prediction calculation. Replace the target opening value of the damper actuator and the target angle value of the swirl vane adjusting motor based on the state prediction value to construct an optimized stable combustion and energy-saving control command. The optimized stable combustion and energy-saving control commands include the predicted opening degree of the damper actuator, the predicted angle of the swirl vane adjustment motor, and the predicted correction amount of the control state.

9. The boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 8, characterized in that, The specific steps of S5 are as follows: S501: Based on the stable combustion and energy-saving control command, extract the target opening value of the damper actuator and the target angle value of the swirl blade regulating motor and jointly map them. Perform differential calculations on the opening value and angle value for adjacent angle changes and adjacent opening changes, and perform same-frequency data alignment to generate a damper swirl coordination parameter set. S502: Based on the damper swirl coordination parameter set, perform state prediction calculation, sum the independent sliding intervals of the continuous change trend sequence of opening degree and the continuous change trend sequence of angle respectively, calculate the state amplitude of opening degree and angle interval respectively, and compare it with the preset fluctuation threshold to obtain the damper swirl prediction state group. S503: Based on the damper swirl prediction state group and the stable combustion and energy-saving control command, extract the target opening value of the corresponding damper actuator and the target angle value of the swirl blade adjustment motor inside the command and perform numerical replacement calculation. Reorganize the replacement parameters into a sequence to generate an optimized stable combustion and energy-saving control command.

10. A boiler stable combustion and energy-saving control method under deep peak shaving of a generating unit according to claim 9, characterized in that, The fluctuation threshold is determined by obtaining the amplitude fluctuation statistics of the heat release gradient sequence under stable combustion conditions, and by introducing a dimensionless processing module to convert the distribution standard deviation of the mechanical response error extreme value and the heat release characterization value into dimensionless coefficients, and then performing weighted summation.

Images