Thermal power plant boiler combustion visual monitoring and early warning method and system, storage medium and electronic equipment

Abstract

The invention provides a thermal power plant boiler combustion visual monitoring and early warning method and system, a storage medium and electronic equipment. According to the method, obtained multi-measuring-point temperature data are input into a pre-constructed temperature field reconstruction model, the model comprehensively considers the geometric structure and combustion thermodynamic characteristic factors of a hearth, model parameters are dynamically adjusted through a particle swarm optimization algorithm, the temperature field reconstruction precision is continuously improved, and a three-dimensional real-time temperature distribution diagram of the hearth is generated; fusing and presenting the diagnosed combustion abnormal information and the visual image of the hearth temperature field, generating a visual in-furnace combustion state monitoring interface, and meanwhile, pushing the abnormal information to a mobile terminal to realize remote monitoring and early warning; the combustion state data in the long-period operation process of the boiler are mined and analyzed, typical combustion modes are extracted, a combustion mode library is continuously enriched and optimized, an online learning algorithm is used for adaptively updating a combustion abnormity diagnosis model, and the intelligent level of combustion state sensing and abnormity early warning is continuously improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method, system, storage medium and electronic equipment for visual monitoring and early warning of boiler combustion in a thermal power plant. Background Art

[0002] The perception of boiler combustion status is a complex technical problem. The temperature field distribution inside the furnace directly reflects the combustion status, but due to the large size of the boiler, the harsh working environment, and the high combustion temperature, the traditional contact temperature measurement method is difficult to apply. Non-contact temperature measurement methods such as laser temperature measurement and acoustic temperature measurement also have problems such as complex equipment and severe signal attenuation, making it difficult to stably obtain the temperature field inside the furnace.

[0003] The furnace has a large spatial scale and the temperature field reconstruction problem is complex. It is necessary to design a reasonable measurement point layout and optimize the reconstruction algorithm to accurately invert the entire temperature field. How to achieve dynamic temperature field reconstruction through intelligent algorithms with limited measurement points is a major technical challenge.

[0004] In addition, due to the lack of effective state perception means, boiler combustion anomalies such as partial burning are difficult to detect and warn in time. Partial burning can cause accidents such as coking, wall brushing, and explosion in the furnace, which seriously threatens the safe operation of the boiler. Traditional manual monitoring methods are difficult to accurately judge the combustion state, and intelligent abnormal warning technology is urgently needed. How to achieve real-time diagnosis and warning of partial burning through intelligent algorithms based on the acquired furnace temperature field data to replace manual monitoring is a technical problem that needs to be solved.

[0005] In summary, boiler combustion status perception involves a series of complex technologies such as temperature measurement method selection, measurement point layout optimization, temperature field reconstruction algorithm, and intelligent diagnosis of combustion anomalies. How to break through the limitations of traditional methods and use advanced sensing and artificial intelligence technologies to realize visualization and intelligentization of furnace combustion is a technical problem that needs to be overcome urgently. It is of great significance to improve the safety and efficiency of boiler operation. Summary of the invention

[0006] The present invention provides a method, system, storage medium and electronic equipment for visual monitoring and early warning of boiler combustion in a thermal power plant, aiming to solve one of the above-mentioned technical problems.

[0007] The thermal power plant boiler combustion visualization monitoring and early warning method includes

[0008] A non-contact temperature measurement method is used to obtain real-time temperature data of multiple measuring points inside the furnace. According to the relationship between the boiler combustion state and the temperature field inside the furnace, the measurement point layout is optimized. The genetic algorithm is used to optimize the measurement point position, while meeting the temperature field reconstruction accuracy requirements and minimizing the number of measurement points, reducing system complexity and cost.

[0009] The acquired multi-point temperature data are input into the pre-built temperature field reconstruction model, which comprehensively considers factors such as furnace geometry and combustion thermodynamic characteristics, dynamically adjusts model parameters through particle swarm optimization algorithm, continuously improves the temperature field reconstruction accuracy, and generates a three-dimensional real-time temperature distribution map of the furnace;

[0010] According to the reconstructed furnace temperature field distribution characteristics, support vector machines are used to perform intelligent diagnosis of the combustion state, matching the current temperature field distribution with the pre-established typical combustion mode library to determine whether there is an abnormality in the combustion state. When abnormal modes such as partial burning and coking are identified, an early warning signal is triggered;

[0011] After determining the combustion anomaly, the optimized combustion control strategy is automatically generated. The fuel ratio, primary air volume, secondary air volume and other parameters are adjusted through the fuzzy control algorithm to optimize the combustion process in real time, suppress and eliminate the combustion anomaly, restore the optimal combustion state, and ensure the safe and efficient operation of the boiler.

[0012] The diagnosed combustion anomaly information is integrated with the furnace temperature field visualization image to generate an intuitive furnace combustion status monitoring interface, providing decision-making basis for operators. At the same time, the abnormal information is pushed to the mobile terminal to achieve remote monitoring and early warning.

[0013] Mining and analyzing the combustion state data during the long-term operation of the boiler, refining typical combustion modes, continuously enriching and optimizing the combustion mode library, using online learning algorithms to adaptively update the combustion anomaly diagnosis model, and continuously improving the intelligent level of combustion state perception and anomaly warning;

[0014] By using the updated combustion abnormality diagnosis model and combining it with the current boiler operating status, an optimized combustion control strategy is generated, and a reinforcement learning algorithm is used for intelligent self-optimization control. While meeting the boiler load demand, the furnace temperature field is distributed most evenly, local high temperature and thermal segregation are suppressed, the risk of furnace coking and explosion is reduced, and the safety and efficiency of boiler operation are improved.

[0015] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0016] The present invention discloses a method for intelligent diagnosis and optimization control of boiler combustion state based on reconstruction of furnace temperature field. The method obtains multi-point temperature data of the furnace through non-contact temperature measurement, optimizes the measurement point arrangement by genetic algorithm, and generates a three-dimensional real-time temperature distribution diagram of the furnace in combination with the temperature field reconstruction model optimized by particle swarm optimization. Then, a support vector machine is used to compare the temperature field characteristics with the typical combustion mode library to realize intelligent diagnosis of the combustion state. When an abnormality is detected, an early warning is automatically triggered and a fuzzy control algorithm is used to adjust the combustion parameters. At the same time, the present invention also integrates technologies such as visual presentation, remote monitoring, data mining and online learning to continuously optimize the combustion mode library and the diagnosis model. Finally, intelligent self-optimization control is realized through the reinforcement learning algorithm, which makes the furnace temperature field uniform while ensuring the boiler load, thereby improving the operation safety and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The present invention is a flow chart of a method for visually monitoring and early warning of boiler combustion in a thermal power plant. DETAILED DESCRIPTION

[0018] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0019] like Figure 1 In this embodiment, a method for visual monitoring and early warning of boiler combustion in a thermal power plant may specifically include:

[0020] Step S101, using a non-contact temperature measurement method to obtain real-time temperature data of multiple measuring points inside the furnace. According to the relationship between the boiler combustion state and the temperature field inside the furnace, the measurement point layout is optimized. The genetic algorithm is used to optimize the measurement point position, while meeting the temperature field reconstruction accuracy requirements and minimizing the number of measurement points, reducing system complexity and cost.

[0021] According to the boiler structure and combustion mode, several infrared probes are arranged around the furnace to form a non-contact temperature measurement array to obtain the temperature data of multiple measuring points inside the furnace in real time. The genetic algorithm is used to optimize the position of the measuring points, and the number of measuring points is minimized while meeting the accuracy requirements of the temperature field reconstruction, so as to obtain the optimal measurement point arrangement scheme. Through the optimized measurement point arrangement scheme, the temperature data of multiple measuring points inside the furnace are collected in real time as the basic data for temperature field reconstruction. According to the acquired multi-point temperature data, the temperature field inside the furnace is reconstructed by an intelligent algorithm to obtain the three-dimensional dynamic temperature distribution inside the furnace. The reconstructed furnace temperature field is visualized and displayed in the form of curves, bar graphs, slices, two-dimensional graphs, three-dimensional graphs, etc., to intuitively reflect the combustion state of the furnace. The image recognition algorithm is used to analyze the visualized temperature field image to determine whether there are abnormal combustion states such as uneven combustion and flame deflection. If an abnormal combustion state is identified, an early warning message is generated and fed back to the operator to prompt him to make combustion adjustments. According to the early warning information and the boiler operation parameters, the operator takes corresponding measures to optimize the combustion state and improve the combustion uniformity of the furnace. The optimized combustion state data is re-input into the temperature field reconstruction and abnormal state identification module to form a closed-loop control, continuously monitor and improve the boiler combustion state, and ensure the safe, efficient and environmentally friendly operation of the boiler.

[0022] Specifically, taking a 600 MW boiler with a four-corner tangential combustion method as an example, the cross-sectional size of its furnace is about 16 meters × 16 meters, and the height is about 60 meters. First, the furnace temperature of the boiler needs to be measured by infrared temperature measurement. Six layers of infrared probes are selected on the water-cooled walls around the furnace, with 16 infrared probes arranged on each layer, and the effective detection distance of each infrared probe is 15 meters. The height of each layer is 10 meters apart, forming a non-contact temperature measurement array with a total of 96 measuring points. Then the data is sent to the intelligent algorithm to calculate the optimal solution for the measuring point. We apply a genetic algorithm, set the initial population size to 50, each individual is the spatial coordinate information of the measuring point, and the encoding scheme is set to the measuring point number. For example, a coordinate point 11 meters, 12 meters and 15 meters is marked as "111215", and the invalidity of a point is set to 10000, such as "00000". The chromosome length is set to 6. In order to meet the accuracy requirement of the relative error of temperature field reconstruction less than 5% and minimize the number of measuring points, the number of iterations is set to 100. After 100 iterations, an optimal measurement point arrangement scheme is obtained. For example, on a plane layer with a height of 10 meters, the infrared probes are installed at four points on the four walls of the furnace: (2,8), (8,2), (8,14), and (14,8) (unit:

[0023] m), this solution can reduce the number of measuring points to 48. 48 infrared probes are arranged according to the optimized scheme, and the sampling frequency is set to 10 times per second, so 480 temperature data are collected per second. The temperature field reconstruction process based on the spatial interpolation method includes the inverse distance weighted method and the minimum curvature method. By setting the algorithm rules and setting the weight to 78 and the minimum curvature to 22 for calculation, a three-dimensional temperature field model is obtained. Through the test set data verification, the relative error is 5%, which meets the accuracy requirements. Using software programming, the obtained data is converted into a slice of the three-dimensional temperature field inside the furnace, and a three-dimensional temperature map is generated at the center. Then, the historical combustion image library of the furnace section is imported, and the support vector machine algorithm is used for recognition. 1,000 normal combustion images and 100 flame deflection images are pre-marked, and the judgment threshold is set to 175. The judgment value of the current temperature image is calculated by algorithm recognition, and the current burning image is marked as an abnormal image. And an alarm is sent to the operation terminal that "the flame on the left side of the furnace is deflected, please verify the fuel quality." The terminal sends combustion optimization information to the operator. For example, the oxygen inlet concentration can be increased by 5% to make the combustion more complete. The combustion adjustment module measures the flue gas temperature and oxygen content through sensors installed on the pipeline, controls the powder feeding amount and other parameters for calculation optimization, and transmits the optimized boiler combustion image to the monitoring center. The temperature field reconstruction program on the central server repeats the previous steps to reconstruct and identify the temperature field, forming a monitoring closed loop. By using temperature sensors and detection programs, efficient, environmentally friendly and safe control of the boiler is achieved.

[0024] Step S102, input the acquired multi-point temperature data into a pre-built temperature field reconstruction model. The model comprehensively considers factors such as furnace geometry and combustion thermodynamic characteristics, dynamically adjusts model parameters through a particle swarm optimization algorithm, continuously improves the temperature field reconstruction accuracy, and generates a three-dimensional real-time temperature distribution map of the furnace.

[0025] According to the boiler structure and combustion mode, an infrared temperature measurement array is designed, and several infrared probes are arranged around the furnace to obtain temperature data at multiple key positions in the furnace. The high-temperature measurement device based on the infrared temperature measurement principle is used to measure the temperature field in the furnace in real time. The probes are arranged between the position above the burnout air nozzle and the position below the screen superheater. The temperature measurement devices are arranged crosswise to obtain a grid-type real-time online measurement system. According to the data obtained by the grid-type real-time online measurement system, the non-contact temperature measurement array layout mode is optimized and the temperature field reconstruction algorithm is developed to determine the real-time construction scheme of the furnace temperature field. If the temperature field is to be reconstructed, it is necessary to study the intelligent algorithm based on multi-point measurement, and obtain the reconstruction results of the furnace temperature field distribution based on the multi-point infrared measurement results through simulation experiments. If there is a reconstruction result of the furnace temperature field distribution of multi-point infrared measurement results, then a three-dimensional numerical model is established for a typical boiler to simulate the combustion in the furnace, and the basic data for verifying the reconstruction effect of the furnace temperature field is obtained. The furnace combustion data obtained by the existing three-dimensional numerical model simulation is obtained, and the effectiveness of the algorithm is judged by comparing it with the reconstruction results of the furnace temperature field. The image recognition algorithm is used to process the acquired temperature field to obtain the partial burning identification and warning information. Based on the partial burning and warning information identified by the image recognition algorithm, the computer vision technology and image recognition algorithm are integrated to determine the partial burning warning model. Through the particle swarm optimization algorithm, the geometric structure of the furnace, the thermodynamic characteristics of combustion and other factors are integrated to adjust the parameters of the pre-built temperature field reconstruction model to obtain the three-dimensional real-time temperature distribution map of the furnace.

[0026] Specifically, for a certain type of four-corner tangential combustion boiler, four groups of infrared probes are installed on the walls of the four corners of its furnace, each group contains five probes, forming a 5x4 temperature measurement array, and the probe spacing is set to 5 meters, covering the key area of ​​the furnace height from 15 meters to 25 meters, and equipped with a data acquisition module with a sampling period of 500 milliseconds to obtain real-time temperature data. These probes use the infrared temperature measurement principle and are optimized specifically for the high temperature range of 800 degrees Celsius to 1800 degrees Celsius. They are arranged between 2 meters above the burnout air nozzle and 1 meter below the screen superheater, and are cross-installed diagonally to form a 1 meter x 1 meter grid to ensure full coverage of the temperature field. The temperature measurement device is equipped with dedicated data analysis software. The adaptive weighted average method is applied to optimize the arrangement of these probes, and the inverse distance weighted interpolation method (the weight exponent of the distance is 2) and the least square method are combined to fit the temperature data, and then the furnace temperature field is reconstructed based on the Kriging interpolation algorithm. In order to quickly build a solution, the fast Fourier transform algorithm is also used to accelerate the data processing to achieve the goal of refreshing the temperature field distribution map once a second. For example: the data collected by a certain infrared sensor can be simulated as time 1, furnace temperature 14255 degrees Celsius, and then processed by the multi-point temperature intelligent interpolation algorithm based on the Bayesian optimization algorithm, combined with the temperature values ​​of the three infrared sensors in the adjacent areas at that moment, to generate the furnace temperature distribution prediction of the middle area, and then combined with the temperature field simulation model based on finite element analysis, 50 groups of different combustion conditions are simulated to generate the test set and verification set of the temperature field distribution (for example, using a 7:3 ratio split), and the accuracy of the reconstruction is evaluated by calculating the root mean square error (for example, less than 50 degrees Celsius is the effective range), and then the multi-point temperature intelligent interpolation algorithm and finite element analysis model are optimized in turn according to the accuracy. According to the simulation results, a specific temperature value of a certain position is constructed. At time 1, the furnace temperature at a certain coordinate point is 14423 degrees Celsius. This is the furnace temperature result of a specific position in the furnace at a certain time generated by the temperature field simulation model based on the Bayesian optimization algorithm, the multi-point temperature intelligent interpolation algorithm and the finite element analysis. This data is compared with the results obtained by the established three-dimensional numerical model, and the temperature difference between the two is calculated. For example, the standard deviation is set to not exceed 3%, and the reliability of the algorithm is verified by indicators such as the mean absolute error. Then, the reconstructed temperature field image is processed by the fast edge detection algorithm to detect whether the temperature difference exceeds 200 degrees Celsius to identify the biased burning condition. Through image recognition, for example, at time 1, the highest furnace temperature in the center of a certain area of ​​the furnace is 18251 degrees Celsius and the outermost circle temperature is 16832 degrees Celsius under the biased burning state. After computer vision recognition, it is then constructed based on this, for example, an early warning model based on a support vector machine or a decision tree algorithm is constructed, and the artificial experience knowledge rules are integrated for adjustment. For example, the temperature gradient of three consecutive points is greater than 80 degrees Celsius / meter to trigger an early warning.Furthermore, a particle swarm optimization algorithm is used to correct the temperature field model. In the constructed algorithm model, for example, 1,500 particles are set, the maximum number of iterations is 200 times, the goal is to converge the maximum fitness value to 05, and the height of the furnace is set to 25 meters, the width is 20 meters, and the length is 20 meters. By combining thermodynamic parameters such as thermal enthalpy and flame radiation intensity, it is continuously iterated until the global fitness function is better than 05 degrees Celsius, thereby accurately reconstructing the three-dimensional temperature map of a certain height of the furnace at time 1.

[0027] Step S103, based on the reconstructed furnace temperature field distribution characteristics, a support vector machine is used to perform intelligent diagnosis of the combustion state. The current temperature field distribution is matched with a pre-established typical combustion mode library to determine whether there is an abnormality in the combustion state. When abnormal modes such as partial burning and coking are identified, an early warning signal is triggered.

[0028] Step 1: Obtain infrared temperature measurement array data, and obtain the matrix information of each frame of temperature data according to the temperature measurement range of 200℃-1700℃ and the measurement accuracy controlled within the measurement reading ±1%. Step 2: According to the matrix information of each frame of temperature data, use the intelligent algorithm to reconstruct the three-dimensional temperature field of the furnace to obtain a visualized temperature field distribution image. Step 3: Through the visualized temperature field distribution image, the temperature data in the furnace is compressed and converted to obtain the data information after dimensionality reduction processing. Step 4: Using the temperature data information, a three-dimensional numerical model is established for a typical boiler to simulate the combustion in the furnace and obtain the characteristic parameters of the typical combustion mode. Step 5: If the similarity between the three-dimensional temperature field image features and the characteristic parameters of the typical combustion mode is less than the threshold, it is marked as an abnormal combustion mode and the characteristic data of the abnormal combustion mode is obtained. Step 6: If the abnormal mode characteristic data such as partial burning exceeds the pre-set warning threshold, the warning signal is activated to determine the current combustion state. Step 7: According to the judgment result of the current combustion state, multiple groups of historical data under different combustion states are collected to build a boiler combustion state database. Step 8: Using the boiler combustion state database as a training set, construct and optimize the support vector machine diagnosis model to obtain an image-based boiler combustion anomaly diagnosis system. Step 9: According to the constructed support vector machine diagnosis model, input the real-time temperature field data into the model for classification and recognition to obtain the combustion state diagnosis result.

[0029] Specifically, an infrared temperature measurement array sensor with 64x64 pixels is used, with a total of 4096 temperature measurement points. Each temperature measurement point is controlled within the temperature measurement range of 200℃-1700℃ and the measurement accuracy is controlled within ±1%. 10 frames of data are collected per second to form a 64x64 temperature data matrix. The error of each frame of data is between 2℃-17℃. These temperature matrices are processed by the bicubic interpolation algorithm to smooth the data and enhance the continuity of the data. In addition, spatial filtering is added in the algorithm to reduce the influence of noise. The smooth data obtained in this way constitutes the basic data stream of the furnace temperature change. The temperature matrix of each frame is used as input, and a neural network is constructed and optimized by combining the particle swarm optimization algorithm and the radial basis function neural network algorithm. The two-dimensional temperature data of multiple sensors are used as the input layer, and then calculated by the optimization algorithm, a mathematical model is established, and the spatial geometric model is constructed. After interpolation is performed through spatial analysis, the output layer is finally set to multiple discrete temperature points in the three-dimensional space, so as to reconstruct the three-dimensional temperature field inside the furnace and generate a pseudo-color three-dimensional temperature field distribution image, which can clearly show the key information such as the position, size and shape of the flame and the temperature gradient change. The principal component analysis method is used to compress the data of the three-dimensional temperature field distribution image. The principal components with a contribution rate of more than 95% are selected to reduce the original temperature field data of thousands of dimensions to hundreds of dimensions, and the data is converted into temperature field feature vectors, which greatly reduces the dimensionality of high-dimensional data. The reduced-dimensional data is input into a pre-trained computational fluid dynamics model, which is built based on the three-dimensional geometric structure and operating conditions of a typical 200-megawatt coal-fired boiler. The model outputs key typical combustion mode characteristic parameters such as flame center position coordinates, average temperature, and heat release rate. These key combustion mode characteristic parameters constitute the data of the typical combustion mode. The calculated real-time temperature field image features are then compared with the stored characteristic parameters of the standard combustion mode, and the similarity is calculated using the Euclidean distance. If the similarity score is less than 80 points, it is judged as a deviation from the normal combustion mode and marked as abnormal. Through this process, temperature deviation, temperature variance, and temperature distribution morphology are extracted as abnormal combustion mode feature data. If the temperature deviation detected in real time exceeds the set threshold of 30°C, and the temperature variance is greater than the set value of 15°C2 for 5 consecutive minutes, the warning condition is met, and the program will trigger an alarm, marking that the current combustion state may have abnormal conditions such as partial burning or coking. Then, the abnormal data is collected and saved in the database together with the normal data. The normal condition is marked as 0, the partial burning condition is marked as 1, the coking condition is marked as 2, and the incomplete combustion condition is marked as 3. One month of data was collected, with 24 groups of different states every day, and a total of about 2880 groups of diversified combustion state data samples were obtained. 70% of the data (about 2000 groups) in the boiler combustion state database were used as training samples, and the remaining 30% of the data were used as test samples.These samples are used to train a support vector machine model, and the Gaussian kernel function is used to calculate the input temperature data features. The cross-validation method and grid search method are used to optimize the parameters during the training process. When the model classification accuracy is stable above 95%, the training is stopped, and a diagnostic system that accurately identifies the combustion state of the boiler is constructed. After the training is completed, the model can receive real-time temperature field data as input, map it to the feature space, and perform classification judgment to identify which defined category the current combustion state belongs to, such as normal combustion, or in a state of partial burning, coking or incomplete combustion, thereby completing abnormal diagnosis and classification recognition.

[0030] Step S104, after determining the combustion anomaly, automatically generate an optimized combustion control strategy. Use the fuzzy control algorithm to adjust the fuel ratio, primary air volume, secondary air volume and other parameters, perform real-time optimization control on the combustion process, suppress and eliminate the combustion anomaly, restore the optimal combustion state, and ensure safe and efficient operation of the boiler.

[0031] The information of the online measurement technology of the boiler furnace temperature field and the biased burning warning model based on image recognition is obtained to determine the distribution state and degree of biased burning of the furnace temperature field. According to the distribution state information of the furnace temperature field, the area and degree of uneven distribution of the temperature field are calculated to obtain the deviation value of the temperature field. If the deviation value of the temperature field exceeds the preset threshold, the combustion abnormality diagnosis module is activated and it is judged as an abnormal combustion state. If the abnormal combustion state is activated, the combustion adjustment record corresponding to the same abnormal state in the historical database is called to obtain the historical adjustment parameter data. The fuzzy control algorithm is used to combine the current furnace temperature field information, the degree of biased burning and the historical adjustment parameter data to generate the initial combustion control strategy. Through the combustion simulation module, the initial combustion control strategy is used as the input condition for simulation operation to obtain the prediction result of the combustion state in the furnace. According to the comparison between the prediction result of the combustion state and the optimal combustion state index, the effectiveness of the initial combustion control strategy is evaluated to obtain the strategy evaluation result. If the strategy evaluation result does not meet the preset index, the parameter weight and membership function of the fuzzy control algorithm are adjusted to generate the optimized combustion control strategy. The optimized combustion control strategy is adopted to adjust the fuel ratio, primary air volume, secondary air volume and other actuators, determine the final parameter adjustment instructions and output them to the boiler control system for execution.

[0032] Specifically, the boiler uses an infrared thermal imager to measure the furnace temperature field distribution in real time, and triggers a biased burning warning when the temperature deviation exceeds 30°C. The system calculates the area ratio of the temperature deviation area, and determines it as a combustion abnormality when it exceeds 5%. Retrieving historical biased burning data of the same type, when the temperature deviation is 35°C and the area ratio is 6%, the strategy of adjusting the fuel ratio to 7:3 and increasing the primary air volume by 10% is optimal. The fuzzy control algorithm generates the initial strategy: fuel ratio 8:2, primary air volume increase 8%. The simulation predicts a reduction of NOx emissions by 11%, which is a large gap from the optimal indicator. The optimized strategy is a fuel ratio of 5:5 and a 9% increase in primary air volume. The final NOx emissions are reduced by 18%, meeting the emission standards, and the strategy is issued for execution.

[0033] Step S105: The diagnosed combustion anomaly information is integrated with the furnace temperature field visualization image to generate an intuitive furnace combustion status monitoring interface to provide decision-making basis for operators. At the same time, the abnormal information is pushed to the mobile terminal to achieve remote monitoring and early warning.

[0034] An infrared temperature measurement array is used to obtain real-time temperature data at multiple key locations inside the furnace. According to the layout of the infrared temperature measurement array and the real-time temperature data obtained, the three-dimensional dynamic temperature field distribution inside the furnace is reconstructed through an intelligent algorithm. If the accuracy of the reconstructed temperature field distribution and the numerical simulation results reaches more than 95%, the temperature field reconstruction results are valid.

[0035] As the input of the abnormal combustion state warning model, the image recognition algorithm is used to analyze the temperature field distribution image to determine whether there are abnormal combustion states such as biased burning and wall brushing. If the abnormal combustion state is identified, the warning information is generated, including the type, location, severity of the abnormal state, etc. The diagnosed abnormal combustion warning information is fused with the visualization image of the furnace temperature field to generate an intuitive furnace combustion state monitoring interface. The combustion state monitoring interface is pushed to the large screen of the control room and the mobile terminal of the operator through network communication technology to achieve remote real-time monitoring. The operator can adjust the combustion parameters in time according to the abnormal combustion warning information and temperature field distribution, improve the combustion state in the furnace, and improve the safety and efficiency of boiler operation.

[0036] Specifically, an infrared temperature measurement array is used to obtain real-time temperature data at 100 key positions inside the furnace. The temperature field distribution inside the furnace is reconstructed through a three-dimensional dynamic temperature field reconstruction algorithm combined with a finite element simulation model. The reconstruction is considered valid if the average error between the reconstruction result and the simulation result is less than 5% and the maximum error does not exceed 10%. The reconstructed temperature field data is converted into a 256-color image, and the convolutional neural network algorithm is used for image recognition to determine whether there are six typical abnormal combustion states such as partial burning and wall brushing. If an abnormality is detected, the output is the type of abnormality, location coordinates, and area ratio of the abnormal area. The warning information is superimposed on the color image of the temperature field to generate an intuitive furnace combustion state monitoring interface. The monitoring interface is transmitted to the 60-inch large screen in the control room and the 10-inch pad terminal of the operator in real time through the industrial Ethernet at a resolution of 1080P and a speed of 25 frames per second. According to the warning information and the process parameters, the operator adjusts the burner swing angle, air volume, coal volume, etc. in time to make the furnace temperature field uniform and improve the boiler efficiency by more than 1%.

[0037] Step S106: mining and analyzing the combustion state data during the long-term operation of the boiler, refining typical combustion modes, and continuously enriching and optimizing the combustion mode library. Using online learning algorithms to adaptively update the combustion anomaly diagnosis model, continuously improving the intelligent level of combustion state perception and abnormal warning.

[0038] The combustion state data of the boiler during long-term operation is collected, including the temperature data of the infrared temperature measurement array, the operating parameter data, and the equipment operation performance data obtained through the intelligent analysis of the equipment state perception data, so as to obtain a comprehensive boiler combustion state data set. The temperature data of the infrared temperature measurement array obtained by the infrared array measurement technology of the furnace temperature field is used to obtain the temperature information of different positions in the furnace, and the key factors affecting the combustion state are determined by combining the boiler structural parameters and coal quality data. The furnace temperature field data obtained by the digitization and visualization of the furnace temperature field are integrated to clean, convert and normalize the combustion state data set during the long-term operation of the boiler, and analyze it through the intelligent algorithm to obtain the data feature information reflecting the combustion state. The temperature field distribution is reconstructed according to the intelligent algorithm, and the combustion state data is classified by the cluster analysis algorithm through the constructed data feature information reflecting the combustion state. If the system commissioning rate under the normal operation of the boiler reaches the predetermined threshold, different combustion modes are extracted to determine the typical combustion mode set. According to the boiler combustion abnormal state early warning technology, the image recognition algorithm is used to identify the temperature field and obtain the information of the early warning, obtain the temperature field image and operating parameter data corresponding to the abnormal state, and use the classification algorithm to distinguish the abnormal type. If the accuracy of a certain type of combustion abnormality recognition result exceeds the set threshold, then the initial combustion abnormality diagnosis model is obtained by combining the typical combustion mode extracted under normal operation. The initial combustion abnormality diagnosis model is used to identify the infrared temperature measurement data, operating parameters and image recognition data obtained under the historical operation state, obtain the operation accuracy of the model, and determine its effectiveness. The combustion state data that has been obtained is marked, the existing combustion mode data is marked, and the initial combustion abnormality diagnosis model is also marked. Then a historical information set is constructed, and a useful data model and feature set are obtained through data analysis and preprocessing. Through online learning of the historical information set, the supervised learning algorithm and the historical combustion state data, combustion mode data and the initial combustion abnormality diagnosis model marking are used, and the gradient descent method is used to obtain the weights required for model update. Based on the update weights learned by the online learning algorithm and combined with the new combustion state data, if the model update parameters meet the iteration requirements, the model parameters are adaptively updated to continuously improve the intelligence level of combustion state perception and abnormal warning.

[0039] Specifically, 30 infrared temperature probes deployed around the boiler furnace collect temperature data every 5 seconds, and record the corresponding 10 operating parameters, such as coal feeder speed, primary wind speed, etc., and record equipment performance data such as grate speed fluctuation rate. These data constitute the original data set, with a capacity of about 100,000 data per day. During the operation time of up to 1 year, the total number of data collected is as high as 36.5 million data records. Then, using the temperature data of these 30 infrared temperature probes, and considering that the length, width and height of the boiler furnace are 10 meters, 8 meters and 20 meters respectively, combined with historical coal quality data including 65% fixed carbon and 30% volatile matter, the key factors affecting combustion are determined to include the primary and secondary air ratios, induced draft fan frequency and other 6 key factors. For 100,000 data items per day, about 1,000 erroneous data items are deleted first, and then the data conversion format is unified into 32-bit floating point numbers. Finally, the numerical range of all parameters is normalized to between 0 and 1. Each data item after processing is a vector containing 15 features, which can display the temperature field in the form of a heat map. Combined with the coal feeding amount and 6 key factors, a comprehensive analysis is performed through the logistic regression algorithm, and finally 5 characteristic values ​​that can reflect the current combustion state are extracted, including the flame center temperature, flame coverage area, etc. Then, the K-means clustering algorithm is applied, and the K value is set to 5, that is, the data is automatically divided into 5 categories, and the 3 million data items obtained in the last month are clustered. The 5 categories obtained by clustering are counted, and the number of these 5 categories and the proportion of normal operating mode data are counted, and the 3 categories with a proportion of more than 85% are determined as typical combustion modes. And the mean variance of the coal feeder speed of the three typical modes, the mean variance of the primary and secondary air ratios and other operating parameter statistics are recorded to construct a collection of typical combustion modes. Get about 100 temperature field images of each abnormal state, each image size is 512x512 pixels, and combine with the operating parameter data, use the convolutional neural network algorithm to classify these data, set the number of network layers to 5, and configure the cross entropy loss function. And set if the abnormal recognition accuracy can reach 90%, then combine with the three typical combustion modes extracted previously to form an initial combustion abnormality diagnosis model containing 8 modes. This model contains a total of 3+100x5x5=2503 parameters. Use a total of 3 million combustion data obtained in the past month, combined with the aforementioned abnormal diagnosis model, to verify the model. Statistic the recognition accuracy and response time of each abnormal mode. All these 3 million data are manually annotated, and the annotation content includes 3 specific types of normal or abnormal. At the same time, the existing combustion mode data are also annotated to form a data set. And the initial combustion abnormality diagnosis model is marked and weighted. A historical information set containing about 100 million data is constructed from these annotated data.First, 1 million data are randomly extracted for analysis and preprocessing, and the continuous temperature change curve is discretized. The key node information is obtained through data analysis, and 15 parameters are retained for each node. 500,000 data nodes are obtained, and about 20,000 different combustion state feature sets are formed. On the basis of determining the label content, combustion mode and initial weights of model parameters, supervised learning algorithm is used for online learning. For example, a fully connected neural network with 3 hidden layers and 50 nodes in each layer is used, and then online learning is performed. 500,000 data are sampled from historical information for training to obtain the initial parameters of the model. Based on the current model parameters, the past 100 data points are used as input to predict the combustion state at the next moment, and the predicted results are compared with the actual results to calculate the error. The gradient descent method is iterated with a learning rate of 0.01 to obtain the weights required for model update. Combined with the combustion state data newly obtained every 5 seconds, according to the current parameters of the model and the update weights, the model update condition is set to the value of the loss function when the change in the value of the loss function is less than 0.1 for 10 consecutive iterations. The model parameters are adaptively updated. This completes a complete intelligent analysis closed loop and improves the combustion status perception and abnormal warning capabilities.

[0040] Step S107, using the updated combustion anomaly diagnosis model, combined with the current boiler operating status, generates an optimized combustion control strategy. Reinforcement learning algorithm is used for intelligent self-optimization control, which makes the furnace temperature field distribution most uniform while meeting the boiler load demand, suppresses local high temperature and thermal segregation, reduces the risk of furnace coking and explosion, and improves the safety and efficiency of boiler operation.

[0041] According to the boiler structure and combustion mode, several infrared probes are arranged around the furnace to form an infrared temperature measurement array to obtain temperature data of some key positions in the furnace in real time. The intelligent algorithm is used to fuse the multi-point temperature data obtained by the infrared temperature measurement array to reconstruct the three-dimensional distribution of the temperature field inside the furnace. The accuracy of the temperature field reconstruction algorithm is verified by simulation experiments. If the deviation between the reconstruction result and the simulation result is greater than 5%, the algorithm is returned to step 2 to optimize until the accuracy requirement is met. The reconstructed furnace temperature field distribution data is obtained and converted into a temperature distribution image as the input of the combustion anomaly diagnosis model. The temperature distribution image is analyzed by an image recognition algorithm to determine whether there are combustion anomalies such as partial burning and wall brushing. If so, an early warning signal is output. According to the combustion anomaly early warning signal, combined with the current operating parameters of the boiler, an optimized combustion control strategy is generated through a reinforcement learning algorithm. The optimized combustion control strategy is sent to the DCS system to adjust the burner secondary air ratio, coal powder injection parameters, etc. to suppress combustion anomalies. The boiler operating parameters and furnace temperature field distribution data are continuously collected and input into the combustion anomaly diagnosis model to dynamically evaluate the combustion control effect. If the abnormal combustion phenomenon is eliminated, the optimization of the combustion control strategy is completed; otherwise, return to step 6 to continue iterative optimization until the furnace temperature field is evenly distributed and local high temperature and thermal segregation are eliminated.

[0042] Specifically, according to the boiler structure and combustion mode, 20 infrared temperature probes are arranged around the furnace to form a 4x5 infrared temperature measurement array to obtain temperature data at key positions in the furnace in real time, with a sampling frequency of 1Hz. The convolutional neural network algorithm is used to fuse the multi-point temperature data obtained by the infrared temperature measurement array, and the temperature field distribution of the furnace with a resolution of 1000x1000x1000 is reconstructed by three-dimensional interpolation. The temperature field reconstruction algorithm is verified by 10 sets of simulation test data under different working conditions, and the average reconstruction error is 2%, which meets the accuracy requirements. The reconstructed furnace temperature field distribution data is converted into a 256x256 pixel grayscale image as the input of the CNN image recognition model to determine whether there are six typical combustion abnormalities such as partial burning and wall brushing, with an accuracy rate of more than 95%. The abnormal diagnosis results are input into the reinforcement learning model together with the boiler operation parameters, and the combustion control strategy including the secondary air ratio and coal powder injection parameters of 20 burners is generated through iterative optimization, and sent to the DCS system for real-time adjustment. The boiler operating parameters and furnace temperature field distribution data are continuously collected and input into the combustion anomaly diagnosis model every 30 seconds to evaluate the control effect. If the abnormal phenomenon is not eliminated within 3 minutes, the reinforcement learning model is returned to continue optimization until the deviation between the highest temperature point in the furnace and the average temperature is less than 50°C, the local thermal segregation index is less than 2, and the combustion control strategy optimization is completed.

[0043] The present invention further provides a thermal power plant boiler combustion visualization monitoring and early warning system, comprising:

[0044] The acquisition unit is used to obtain the real-time temperature data of multiple measuring points inside the furnace by non-contact temperature measurement method, optimize the arrangement of measuring points according to the relationship between the boiler combustion state and the temperature field inside the furnace, and use genetic algorithm to optimize the position of measuring points, so as to minimize the number of measuring points while meeting the accuracy requirements of temperature field reconstruction, and reduce system complexity and cost;

[0045] The reconstruction unit is used to input the acquired multi-measurement point temperature data into the pre-built temperature field reconstruction model. The model comprehensively considers the furnace geometry structure and combustion thermodynamic characteristics, dynamically adjusts the model parameters through the particle swarm optimization algorithm, continuously improves the temperature field reconstruction accuracy, and generates a three-dimensional real-time temperature distribution map of the furnace;

[0046] The trigger unit is used to perform intelligent diagnosis of the combustion state using a support vector machine based on the reconstructed furnace temperature field distribution characteristics, match the current temperature field distribution with a pre-established typical combustion mode library, and determine whether the combustion state is abnormal. When the abnormal mode of partial burning or coking is identified, an early warning signal is triggered;

[0047] The control unit is used to automatically generate an optimized combustion control strategy after determining that the combustion is abnormal. It adjusts the fuel ratio, primary air volume, and secondary air volume parameters through the fuzzy control algorithm to optimize the combustion process in real time, suppress and eliminate combustion abnormalities, restore the optimal combustion state, and ensure safe and efficient operation of the boiler;

[0048] The diagnostic unit is used to fuse the diagnosed combustion abnormality information with the visual image of the furnace temperature field to generate an intuitive furnace combustion status monitoring interface, provide decision-making basis for operators, and push abnormal information to mobile terminals to achieve remote monitoring and early warning;

[0049] The update unit is used to mine and analyze the combustion state data during the long-term operation of the boiler, extract typical combustion modes, continuously enrich and optimize the combustion mode library, and use online learning algorithms to adaptively update the combustion anomaly diagnosis model, continuously improving the intelligent level of combustion state perception and anomaly warning;

[0050] The optimization unit is used to utilize the updated combustion anomaly diagnosis model and combine it with the current boiler operating status to generate an optimized combustion control strategy, and adopt a reinforcement learning algorithm for intelligent self-optimization control. While meeting the boiler load requirements, it makes the furnace temperature field distribution most uniform, suppresses local high temperature and thermal segregation, reduces the risk of furnace coking and explosion, and improves the safety and efficiency of boiler operation.

[0051] The present invention further provides a storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute any one of the aforementioned methods.

[0052] The present invention further provides an electronic device, comprising a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, and the program instructions implement the aforementioned method when loaded and executed by the processor.

[0053] The present invention provides a method for visual monitoring and early warning of boiler combustion in a thermal power plant. The method obtains multi-point temperature data of the furnace through non-contact temperature measurement, optimizes the measurement point arrangement by using a genetic algorithm, and generates a three-dimensional real-time temperature distribution diagram of the furnace in combination with a temperature field reconstruction model optimized by a particle swarm. Then, a support vector machine is used to compare the temperature field characteristics with a typical combustion mode library to realize intelligent diagnosis of the combustion state. When an abnormality is detected, an early warning is automatically triggered and a fuzzy control algorithm is used to adjust the combustion parameters. At the same time, the present invention also integrates technologies such as visual presentation, remote monitoring, data mining, and online learning to continuously optimize the combustion mode library and the diagnostic model. Finally, intelligent self-optimization control is realized through a reinforcement learning algorithm, which makes the furnace temperature field uniform while ensuring the boiler load, thereby improving operational safety and efficiency.

[0054] The above only lists some preferred embodiments of the present invention, but the present invention is not limited thereto, and many improvements and changes can be made. As long as the improvements and changes are made on the basis of the basic principles of the present invention, they should be regarded as falling within the protection scope of the present invention.

Claims

1. A method for visual monitoring and early warning of boiler combustion in a thermal power plant, characterized in that: The method comprises: A non-contact temperature measurement method is used to obtain real-time temperature data of multiple measuring points inside the furnace. According to the relationship between the boiler combustion state and the temperature field inside the furnace, the measurement point layout is optimized. The genetic algorithm is used to optimize the measurement point position, while meeting the temperature field reconstruction accuracy requirements and minimizing the number of measurement points, reducing system complexity and cost. The acquired multi-point temperature data are input into the pre-built temperature field reconstruction model. The model comprehensively considers the furnace geometry structure and combustion thermodynamic characteristics, dynamically adjusts the model parameters through the particle swarm optimization algorithm, continuously improves the temperature field reconstruction accuracy, and generates a three-dimensional real-time temperature distribution map of the furnace; According to the reconstructed furnace temperature field distribution characteristics, support vector machines are used to perform intelligent diagnosis of the combustion state, matching the current temperature field distribution with the pre-established typical combustion mode library to determine whether there is an abnormality in the combustion state. When abnormal modes such as partial burning and coking are identified, an early warning signal is triggered. After determining the combustion anomaly, the optimized combustion control strategy is automatically generated. The fuel ratio, primary air volume, and secondary air volume parameters are adjusted through the fuzzy control algorithm to optimize the combustion process in real time, suppress and eliminate the combustion anomaly, restore the optimal combustion state, and ensure the safe and efficient operation of the boiler. The diagnosed combustion anomaly information is integrated with the furnace temperature field visualization image to generate an intuitive furnace combustion status monitoring interface, providing decision-making basis for operators. At the same time, the abnormal information is pushed to the mobile terminal to achieve remote monitoring and early warning. Mining and analyzing the combustion state data during the long-term operation of the boiler, refining typical combustion modes, continuously enriching and optimizing the combustion mode library, using online learning algorithms to adaptively update the combustion anomaly diagnosis model, and continuously improving the intelligent level of combustion state perception and anomaly warning; By using the updated combustion abnormality diagnosis model and combining it with the current boiler operating status, an optimized combustion control strategy is generated, and a reinforcement learning algorithm is used for intelligent self-optimization control. While meeting the boiler load demand, the furnace temperature field is distributed most evenly, local high temperature and thermal segregation are suppressed, the risk of furnace coking and explosion is reduced, and the safety and efficiency of boiler operation are improved.

2. The method according to claim 1, characterized in that The non-contact temperature measurement method is used to obtain real-time temperature data of multiple measuring points inside the furnace; the measurement point arrangement scheme is optimized according to the relationship between the boiler combustion state and the temperature field inside the furnace; the measurement point positions are optimized using a genetic algorithm to minimize the number of measurement points while meeting the temperature field reconstruction accuracy requirements, thereby reducing system complexity and cost, including: According to the boiler structure and combustion mode, several infrared probes are arranged around the furnace to form a non-contact temperature measurement array to obtain temperature data of multiple measuring points inside the furnace in real time; Genetic algorithm is used to optimize the measurement point positions, minimize the number of measurement points while meeting the temperature field reconstruction accuracy requirements, and obtain the optimal measurement point layout plan; Through the optimized measurement point arrangement plan, the temperature data of multiple measurement points inside the furnace are collected in real time as the basic data for temperature field reconstruction; Based on the multi-point temperature data obtained, the temperature field inside the furnace is reconstructed using an intelligent algorithm to obtain the three-dimensional dynamic temperature distribution inside the furnace; The reconstructed furnace temperature field is visualized and displayed in the form of curves, bar graphs, slice graphs, two-dimensional graphs and three-dimensional graphs to intuitively reflect the furnace combustion status; Use image recognition algorithms to analyze the visualized temperature field image to determine whether there is an abnormal combustion state such as uneven combustion and flame deflection; If an abnormal combustion state is identified, an early warning message is generated and fed back to the operator to prompt him to make combustion adjustments; Based on the early warning information and combined with the boiler operating parameters, the operating personnel take corresponding measures to optimize the combustion state and improve the furnace combustion uniformity; The optimized combustion state data is re-input into the temperature field reconstruction and abnormal state identification module to form a closed-loop control, continuously monitor and improve the boiler combustion state, and ensure the safe, efficient and environmentally friendly operation of the boiler.

3. The method according to claim 1, characterized in that The obtained multi-point temperature data is input into a pre-built temperature field reconstruction model; the model comprehensively considers the furnace geometry structure and combustion thermodynamic characteristics, dynamically adjusts the model parameters through the particle swarm optimization algorithm, continuously improves the temperature field reconstruction accuracy, and generates a three-dimensional real-time temperature distribution map of the furnace, including: Design an infrared temperature measurement array based on the boiler structure and combustion mode, arrange several infrared probes around the furnace to obtain temperature data at multiple key locations in the furnace; The high temperature measuring device based on infrared temperature measurement principle is used to measure the temperature field in the furnace in real time. The probes are arranged above the burnout air nozzle and below the screen superheater. The temperature measuring devices are arranged crosswise to obtain a grid-type real-time online measurement system. Based on the data obtained by the grid-type real-time online measurement system, the non-contact temperature measurement array layout optimization design and temperature field reconstruction algorithm development are carried out to determine the real-time construction plan of the furnace temperature field; If the temperature field reconstruction is to be performed, it is necessary to study the intelligent algorithm based on multi-point measurement and obtain the furnace temperature field distribution reconstruction results based on multi-point infrared measurement results through simulation experiments; If the furnace temperature field distribution reconstruction results of multi-point infrared measurement results are available, then a three-dimensional numerical model is established for a typical boiler to simulate the combustion in the furnace and obtain basic data for verifying the reconstruction effect of the furnace temperature field; Obtain the furnace combustion data obtained by simulating the existing three-dimensional numerical model, and compare it with the reconstruction results of the furnace temperature field to determine the effectiveness of the algorithm; Use image recognition algorithm to process the acquired temperature field to obtain bias burning identification and early warning information; According to the partial burn and warning information identified by the image recognition algorithm, the partial burn warning model is determined by integrating computer vision technology and image recognition algorithm; Through the particle swarm optimization algorithm, the furnace geometry and combustion thermodynamic characteristics are integrated, the parameters of the pre-built temperature field reconstruction model are adjusted, and the three-dimensional real-time temperature distribution map of the furnace is obtained.

4. The method according to claim 1, characterized in that The method uses a support vector machine to perform intelligent diagnosis on the combustion state based on the reconstructed furnace temperature field distribution characteristics; matches the current temperature field distribution with a pre-established typical combustion mode library to determine whether the combustion state is abnormal; and triggers an early warning signal when an abnormal mode of partial burning or coking is identified, including: Step 1: Obtain infrared temperature measurement array data, and obtain matrix information of each frame of temperature data according to the temperature measurement range of 200℃-1700℃ and the measurement accuracy controlled within ±1% of the measurement reading; Step 2: Based on the temperature data matrix information of each frame, an intelligent algorithm is used to reconstruct the three-dimensional temperature field of the furnace to obtain a visualized temperature field distribution image; Step 3: By visualizing the temperature field distribution image, compress and convert the temperature data in the furnace to obtain the data information after dimensionality reduction processing; Step 4: Using temperature data information, a three-dimensional numerical model is established for a typical boiler to simulate the combustion in the furnace and obtain characteristic parameters of a typical combustion mode; Step 5: If the similarity between the three-dimensional temperature field image features and the characteristic parameters of the typical combustion mode is less than a threshold, then it is marked as an abnormal combustion mode, and the abnormal combustion mode characteristic data is obtained; Step 6: If the characteristic data of the abnormal mode of partial combustion exceeds the preset warning threshold, the warning signal is activated to determine the current combustion state; Step 7: Based on the judgment result of the current combustion state, collect multiple sets of historical data under different combustion states to build a boiler combustion state database; Step 8: Using the boiler combustion state database as a training set, constructing and optimizing the support vector machine diagnosis model, and obtaining an image-based boiler combustion abnormality diagnosis system; Step 9: According to the constructed support vector machine diagnosis model, the real-time temperature field data is input into the model for classification and identification to obtain the combustion state diagnosis result.

5. The method according to claim 1, characterized in that After determining the combustion anomaly, the optimized combustion control strategy is automatically generated; the fuel ratio, primary air volume, and secondary air volume parameters are adjusted through the fuzzy control algorithm to optimize the combustion process in real time, suppress and eliminate the combustion anomaly, restore the optimal combustion state, and ensure the safe and efficient operation of the boiler, including: Obtain information on the boiler furnace temperature field online measurement technology and the biased burning warning model based on image recognition to determine the distribution state of the furnace temperature field and the degree of biased burning; According to the distribution state information of the furnace temperature field, the area and degree of uneven temperature field distribution are calculated to obtain the deviation value of the temperature field; If the deviation value of the temperature field exceeds the preset threshold, the combustion abnormality diagnosis module is activated and it is judged as an abnormal combustion state; If the combustion abnormal state is activated, the combustion adjustment record corresponding to the same abnormal state in the historical database is called to obtain the historical adjustment parameter data; The fuzzy control algorithm is used to generate the initial combustion control strategy by combining the current furnace temperature field information, the degree of biased burning and the historical adjustment parameter data; Through the combustion simulation module, the initial combustion control strategy is used as the input condition for simulation calculation to obtain the prediction result of the combustion state in the furnace; By comparing the predicted results of the combustion state with the optimal combustion state indicators, the effectiveness of the initial combustion control strategy is evaluated and the strategy evaluation results are obtained; If the strategy evaluation result does not meet the preset indicators, the parameter weights and membership functions of the fuzzy control algorithm are adjusted to generate an optimized combustion control strategy; The optimized combustion control strategy is adopted to adjust the fuel ratio, primary air volume and secondary air volume actuators, determine the final parameter adjustment instructions and output them to the boiler control system for execution.

6. The method according to claim 1, characterized in that The combustion state data of the boiler during long-term operation is mined and analyzed to extract typical combustion modes, and the combustion mode library is continuously enriched and optimized; the combustion anomaly diagnosis model is adaptively updated using an online learning algorithm to continuously improve the intelligent level of combustion state perception and abnormal warning, including: Collect the combustion status data of the boiler during long-term operation, including the temperature data and operating parameter data of the infrared temperature measurement array and the equipment operation performance data obtained through intelligent analysis of the equipment status perception data, to obtain a comprehensive boiler combustion status data set; Based on the temperature data of the infrared temperature measurement array obtained by the infrared array measurement technology of the furnace temperature field, the temperature information of different positions in the furnace is obtained, and the key factors affecting the combustion state are determined by combining the boiler structural parameters and coal quality data; The furnace temperature field data obtained by integrating the digitization and visualization of the furnace temperature field are cleaned, converted and normalized for the combustion state data set during the long-term operation of the boiler. The data is analyzed through intelligent algorithms to obtain data feature information reflecting the combustion state. Reconstruct the temperature field distribution according to the intelligent algorithm, and classify the combustion state data by constructing the data feature information reflecting the combustion state. If the system operation rate under the normal operation of the boiler reaches the predetermined threshold, different combustion modes are extracted and the typical combustion mode set is determined. According to the boiler combustion abnormal state early warning technology, the image recognition algorithm is used to identify the temperature field and obtain the information of early warning, obtain the temperature field image and operation parameter data corresponding to the abnormal state, and use the classification algorithm to distinguish the abnormal type. If the accuracy of the recognition result of a certain type of combustion abnormality exceeds the set threshold, then the initial combustion abnormality diagnosis model is obtained by combining the typical combustion mode extracted under normal operating conditions; Using the initial combustion anomaly diagnosis model, the infrared temperature measurement data and operating parameters and image recognition data obtained under the historical operating state are identified to obtain the operating accuracy of the model and determine its effectiveness; Annotate the acquired combustion state data, annotate the existing combustion mode data, and also mark the initial combustion abnormality diagnosis model; Then, a historical information set is constructed, and useful data models and feature sets are obtained through data analysis and preprocessing; Through online learning of historical information sets, using supervised learning algorithms and historical combustion state data and combustion mode data as well as initial combustion anomaly diagnosis model labels, using the gradient descent method, the weights required for model updating are obtained; Based on the update weights learned by the online learning algorithm and combined with the new combustion state data, if the model update parameters meet the iteration requirements, the model parameters are adaptively updated to continuously improve the intelligence level of combustion state perception and abnormal warning.

7. The method according to claim 1, characterized in that The updated combustion abnormality diagnosis model is used in combination with the current boiler operation status to generate an optimized combustion control strategy; a reinforcement learning algorithm is used for intelligent self-optimization control to make the furnace temperature field distribution most uniform while meeting the boiler load demand, suppress local high temperature and thermal segregation, reduce the risk of furnace coking and explosion, and improve the safety and efficiency of boiler operation, including: Step 1: According to the boiler structure and combustion mode, several infrared probes are arranged around the furnace to form an infrared temperature measurement array to obtain temperature data of some key positions in the furnace in real time; Step 2: Use intelligent algorithms to fuse the multi-point temperature data obtained by the infrared temperature measurement array to reconstruct the three-dimensional distribution of the temperature field inside the furnace; Step 3: Verify the accuracy of the temperature field reconstruction algorithm through simulation experiments. If the deviation between the reconstruction result and the simulation result is greater than 5%, return to step 2 to optimize the algorithm until the accuracy requirement is met; Step 4: Obtain the reconstructed furnace temperature field distribution data and convert it into a temperature distribution image as the input of the combustion anomaly diagnosis model; Step 5: Use image recognition algorithm to analyze the temperature distribution image to determine whether there is abnormal phenomenon of partial burning and wall brushing combustion. If so, output a warning signal; Step 6: Based on the abnormal combustion warning signal and the current operating parameters of the boiler, an optimized combustion control strategy is generated through a reinforcement learning algorithm; Step 7: Send the optimized combustion control strategy to the DCS system to adjust the burner secondary air ratio and pulverized coal injection parameters to suppress combustion anomalies; Step 8: Continuously collect boiler operating parameters and furnace temperature field distribution data, input them into the combustion anomaly diagnosis model, and dynamically evaluate the combustion control effect; Step 9: If the abnormal combustion phenomenon is eliminated, the combustion control strategy optimization is completed; otherwise, return to step 6 to continue iterative optimization until the furnace temperature field is evenly distributed and local high temperature and thermal segregation are eliminated.

8. A thermal power plant boiler combustion visual monitoring and early warning system, characterized in that: include: The acquisition unit is used to obtain the real-time temperature data of multiple measuring points inside the furnace by non-contact temperature measurement method, optimize the arrangement of measuring points according to the relationship between the boiler combustion state and the temperature field inside the furnace, and use genetic algorithm to optimize the position of measuring points, so as to minimize the number of measuring points while meeting the accuracy requirements of temperature field reconstruction, and reduce system complexity and cost; The reconstruction unit is used to input the acquired multi-measurement point temperature data into the pre-built temperature field reconstruction model. The model comprehensively considers the furnace geometry structure and combustion thermodynamic characteristics, dynamically adjusts the model parameters through the particle swarm optimization algorithm, continuously improves the temperature field reconstruction accuracy, and generates a three-dimensional real-time temperature distribution map of the furnace; The trigger unit is used to perform intelligent diagnosis of the combustion state using a support vector machine based on the reconstructed furnace temperature field distribution characteristics, match the current temperature field distribution with a pre-established typical combustion mode library, and determine whether the combustion state is abnormal. When the abnormal mode of partial burning or coking is identified, an early warning signal is triggered; The control unit is used to automatically generate an optimized combustion control strategy after determining that the combustion is abnormal. It adjusts the fuel ratio, primary air volume, and secondary air volume parameters through the fuzzy control algorithm to optimize the combustion process in real time, suppress and eliminate combustion abnormalities, restore the optimal combustion state, and ensure safe and efficient operation of the boiler; The diagnostic unit is used to fuse the diagnosed combustion abnormality information with the visual image of the furnace temperature field to generate an intuitive furnace combustion status monitoring interface, provide decision-making basis for operators, and push abnormal information to mobile terminals to achieve remote monitoring and early warning; The update unit is used to mine and analyze the combustion state data during the long-term operation of the boiler, extract typical combustion modes, continuously enrich and optimize the combustion mode library, and use online learning algorithms to adaptively update the combustion anomaly diagnosis model, continuously improving the intelligent level of combustion state perception and anomaly warning; The optimization unit is used to utilize the updated combustion anomaly diagnosis model and combine it with the current boiler operating status to generate an optimized combustion control strategy, and adopt a reinforcement learning algorithm for intelligent self-optimization control. While meeting the boiler load requirements, it makes the furnace temperature field distribution most uniform, suppresses local high temperature and thermal segregation, reduces the risk of furnace coking and explosion, and improves the safety and efficiency of boiler operation.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, and the program instructions implement the method according to any one of claims 1 to 7 when loaded and executed by the processor.

Images