Intelligent visual system for furnace temperature field based on acoustic perception technology

Abstract

The present application relates to the technical field of acoustic measurement, in particular to a furnace temperature field intelligent visualization system based on acoustic sensing technology, comprising: an array acoustic sensing data acquisition unit; a working condition adaptive inversion unit; a three-dimensional temperature field intelligent visualization unit; a working condition early warning and interaction unit. The present application receives acoustic wave data, working condition parameters and corrected temperature field, extracts acoustic wave frequency-amplitude characteristics and working condition mean value, fluctuation characteristics in branches, dynamically allocates attention weight according to feature credibility to obtain fusion features; the fusion features are input into the inversion model to obtain the current temperature field, the error is calculated based on the measured temperature of the thermocouple, the regularization coefficient negatively related to the error and the number of iterations is used for optimization, and the related parameters are adjusted by disassembling the error, so that the system can adapt to the working condition change without manual intervention, and the system automation and long-term operation reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of acoustic measurement technology, and more specifically, to an intelligent visualization system for furnace temperature field based on acoustic sensing technology. Background Technology

[0002] Accurate monitoring of the furnace temperature field is a core requirement for ensuring the safe operation and efficient temperature control of thermal equipment such as power plant boilers and industrial kilns. Traditional contact temperature measurement methods (such as thermocouples) are susceptible to high-temperature corrosion and dust abrasion, resulting in short lifespans and limitations in acquiring data only at single points. Non-contact infrared and optical temperature measurement technologies, on the other hand, are difficult to accurately capture the temperature across the entire range due to interference from flue gas radiation within the furnace and obstruction of the equipment's viewing angle. Acoustic temperature measurement technology, based on the core principle that "sound wave propagation speed is strongly correlated with medium temperature," possesses advantages such as non-invasiveness, fast response speed, and wide measurement range. It has become a key technological direction in the field of furnace temperature field monitoring. By measuring parameters such as sound wave flight time and combining them with inversion algorithms, the temperature field can be reconstructed and monitored.

[0003] In existing technologies, research has been conducted on acoustic temperature measurement signal optimization and the construction of power plant boiler temperature monitoring systems. For example, the research on acoustic temperature measurement signals based on multi-dimensional optimization strategies for boiler furnace temperature fields focuses on the filtering and noise reduction optimization of acoustic signals, improving the signal-to-noise ratio and reducing noise interference on temperature measurement data through multi-parameter collaborative processing. Another example is the research on power plant boiler temperature monitoring systems based on acoustic temperature measurement principles, which constructs a basic acoustic temperature measurement hardware framework, realizing real-time acquisition and output of temperature data during power plant boiler operation, providing data support for subsequent temperature analysis.

[0004] Despite the design advantages of the above technical solutions, they also have the following technical defects: First, the temperature measurement accuracy is insufficient under complex working conditions. Although the "Research on Acoustic Temperature Measurement Signal of Boiler Furnace Temperature Field Based on Multi-Dimensional Optimization Strategy" can improve the signal quality under a single noise scenario, it does not solve the problem of sound refraction caused by the coupling of high temperature gradient and axial airflow in the furnace, and its noise reduction effect on multi-source superimposed noise such as combustion noise and sootblower working noise is limited. The model construction of "Research on Temperature Monitoring System of Power Plant Boiler Based on Acoustic Temperature Measurement Principle" adopts the assumption of "sound rays propagating in a straight line", which further ignores the influence of airflow disturbance on the sound ray propagation path, resulting in poor temperature reconstruction accuracy in high-temperature areas, which cannot meet the stringent requirements of temperature measurement accuracy in industrial scenarios. Secondly, the reconstruction models are rigid and lack adaptability to operating conditions: the technical solutions in the two aforementioned documents both employ fixed, general reconstruction algorithms (such as the Landweber iterative method). Key parameters in these algorithms, such as regularization parameters and iteration counts, rely on manual experience for pre-setting. When the boiler faces changes in operating conditions, such as load fluctuations, coal type changes, or environmental changes like coking and ash accumulation, the fixed model parameters cannot match the real-time operating conditions. Operators must frequently manually adjust these parameters, increasing maintenance costs and significantly limiting the system's automation, intelligence, and long-term reliability. Thirdly, there is a lack of temperature field visualization capabilities: none of the aforementioned technical solutions construct a temperature field visualization model; they can only output discrete temperature values ​​and cannot achieve a two-dimensional or three-dimensional graphical display of the overall temperature distribution. Maintenance personnel cannot intuitively perceive key information such as furnace temperature gradients and localized overheating, which is out of sync with the "intelligent visualization" requirements of industrial maintenance. Therefore, we propose an intelligent visualization system for the furnace temperature field based on acoustic sensing technology. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent visualization system for furnace temperature field based on acoustic sensing technology, so as to solve the problems mentioned in the background art, such as insufficient temperature measurement accuracy under complex working conditions, fixed reconstruction model and lack of working condition adaptability and lack of temperature field visualization capability.

[0006] To address the aforementioned technical problems, the present invention aims to provide an intelligent visualization system for furnace temperature fields based on acoustic sensing technology, comprising:

[0007] An array-type acoustic sensing data acquisition unit is provided, in which multiple sets of acoustic sensor arrays are arranged along the circumference of the furnace to collect acoustic wave signals inside the furnace; the acoustic wave signals are filtered to suppress characteristic noise, and combined with the airflow data from the differential pressure flow velocity sensor, the basic data for sound ray propagation correction is output.

[0008] The adaptive inversion unit receives real-time operating parameters from the boiler DCS system and basic acoustic signal data output from the array-type acoustic sensing data acquisition unit. Based on the current iterative temperature field distribution and input airflow velocity field, it uses the fast travel method to solve the Eikonal equation to obtain the minimum propagation time of each acoustic path. The time residual is then propagated back to the temperature field update process through an automatic differentiation mechanism. Simultaneously, an improved algorithm with dual-branch attention fusion and residual feedback dynamic regularization is used to extract the time series features of the acoustic data and the steady-state and dynamic features of the operating parameters, respectively. The nonlinear fusion of heterogeneous data is achieved through dynamic allocation of attention weights. During the iteration process, the temperature distribution error is propagated back to dynamically adjust the algorithm parameters by combining the sound ray propagation correction results, and the three-dimensional temperature distribution data of the entire furnace is output.

[0009] The three-dimensional temperature field intelligent visualization unit receives the three-dimensional temperature distribution data of the entire furnace area output by the working condition adaptive inversion unit. After format conversion, it constructs the furnace geometric model and superimposes the temperature information through a general rendering tool to generate a real-time temperature field image. It has the functions of temperature value annotation, regional feature display and view switching.

[0010] The working condition early warning and interaction unit compares the abnormal temperature information identified by the three-dimensional temperature field intelligent visualization unit with the temperature safety range of the furnace equipment. When an early warning is triggered, it outputs an audible and visual alarm signal and pushes the coordinates of the abnormal area. It is equipped with a human-machine interface that can display real-time temperature field images, historical data curves and early warning records, and supports the adjustment of interface parameters.

[0011] As a further improvement to this technical solution, the array-type acoustic sensing data acquisition unit includes an array deployment module, a signal acquisition module, a filtering processing module, and a data fusion module, wherein:

[0012] The array deployment module is used to coordinate the deployment of multiple acoustic sensor arrays and differential pressure flow sensors along the circumference of the furnace. The installation height of the acoustic sensor array is adapted to the center area of ​​the furnace flame, and the differential pressure flow sensors correspond one-to-one with the acoustic sensor arrays in the furnace area. The distance between adjacent sampling points is ≤1.5m to ensure the spatial matching of sound wave and airflow data.

[0013] The signal acquisition module is integrated into the acoustic sensor array of the array deployment module and is used to synchronously acquire acoustic wave signals from different areas inside the furnace. The acquisition frequency is matched with the propagation period of the acoustic wave inside the furnace.

[0014] The filtering module is connected to the signal acquisition module and is used to receive the original acoustic signal. It uses bandpass filtering to retain the relevant characteristic frequency bands of furnace combustion and filter out interference frequency band signals.

[0015] The data fusion module is connected to the filtering module and the differential pressure flow sensor signal, respectively, and is used to synchronize the filtered sound wave signal with the airflow data in time and integrate it into the sound ray propagation correction basic data output to the working condition adaptive inversion unit.

[0016] As a further improvement to this technical solution, the working condition adaptive inversion unit includes a data receiving module, a sound propagation correction module, a feature fusion module, an iterative optimization module, and a result output module, wherein:

[0017] The input terminal of the data receiving module is connected to the boiler DCS system and the array-type acoustic sensing data acquisition unit respectively, and is used to receive real-time operating parameters and basic data of acoustic signals, and synchronously distribute them to the sound propagation correction module and the feature fusion module.

[0018] The sound ray propagation correction module is signal-connected to the data receiving module and is used to receive the initial temperature field. airflow velocity field and measured sound wave propagation time A corrected temperature field is generated by correcting the sound propagation path error. And output it to the feature fusion module;

[0019] The feature fusion module is connected to the sound propagation correction module and the data receiving module respectively, and is used to extract sound wave features. Operating conditions A fused feature vector is generated through dual-branch attention fusion. The output is sent to the iterative optimization module;

[0020] The iterative optimization module and the feature fusion module are signal-connected, based on Invert the current temperature field Combined with regularization coefficient Perform residual feedback dynamic regularization optimization and output and the corresponding mean absolute error The results are output to the output module, and the non-converged results are also included. Feedback is sent to the sound propagation correction module;

[0021] The result output module is signal-connected to the iterative optimization module and is used to determine the convergence of the iteration. When the temperature field error and gradient stability conditions are met for three consecutive iterations, the output is successful. To a three-dimensional temperature field intelligent visualization unit.

[0022] As a further improvement to this technical solution, the process of sound propagation correction by the sound propagation correction module includes the following steps:

[0023] S22.1 Receive the initial temperature field transmitted by the data receiving module. Real-time airflow velocity field The measured sound wave propagation time output by the array-type acoustic sensing data acquisition unit. ;in The reference temperature field for furnace design or the temperature field results from the previous round fed back by the iterative optimization module;

[0024] S22.2, For any three-dimensional spatial coordinates within the furnace Calculate the sound wave propagation speed after temperature and airflow coupling according to general acoustic relationships. The angle between the airflow and the direction of sound propagation The data was obtained by calculating the airflow direction data from the DCS system and the coordinates of the acoustic sensor deployment.

[0025] S22.3. Taking the sound sensor transmitter as the starting point and the receiver as the ending point, the Eikonal equation is solved using the fast travel method to obtain the shortest path of sound wave propagation and the corresponding theoretical propagation time. ;

[0026] S22.4, Calculation and residual Through automatic differentiation mechanism Reverse conduction to Generate the corrected temperature field And transmit it to the feature fusion module.

[0027] As a further improvement to this technical solution, the sound wave propagation speed in S22.2 The calculation supports airflow humidity adaptation, and the specific adaptation process includes the following steps:

[0028] S22.21 Receive the relative humidity of the airflow inside the furnace output by the DCS system. ;

[0029] S22.22, Introduce a humidity correction factor , The value of varies Linear change, The value range is [0,1];

[0030] S22.23, based on and Adjustment The computational logic enables It is adapted to the sound speed propagation characteristics under corresponding humidity conditions.

[0031] As a further improvement to this technical solution, the feature fusion module performs the dual-branch attention fusion process, which includes the following steps:

[0032] S23.1 Receive the raw acoustic wave data and operating parameters transmitted by the data receiving module, and synchronously receive the corrected temperature field output by the acoustic ray propagation correction module. ;

[0033] S23.2 Perform dual-branch feature extraction. The first branch uses the sliding window method to extract the frequency-amplitude features of the acoustic wave data. The second branch extracts the mean and fluctuation characteristics of the operating condition parameters. ;

[0034] S23.3 Calculating the reliability of acoustic wave features based on the statistical characteristics of feature data. The reliability of the operating condition characteristics is negatively correlated with the standard deviation and positively correlated with the mean. It is negatively correlated with the standard deviation of the operating condition characteristics and positively correlated with the mean;

[0035] S23.4, according to and Assign attention weights The fused feature vector is obtained through weighted calculation. The weights are updated in real time with new feature data and output to the iterative optimization module.

[0036] As a further improvement to this technical solution, the process of residual feedback dynamic regularization optimization performed by the iterative optimization module includes the following steps:

[0037] S24.1, The output of the feature fusion module Input the temperature field inversion model to calculate the global temperature field of the furnace in the current cycle. ;

[0038] S24.2, The actual temperature measured by thermocouples on the furnace wall. Based on the baseline, calculate Mean absolute error , coordinates The measured temperature value at the location;

[0039] S24.3, The value of and It is negatively correlated with the number of iterations. They are negatively correlated; the larger the error, the better in the early stages of iteration. The smaller the value, the smaller the error, and the better in the later stages of iteration. The larger;

[0040] S24.4 Disassembly The source of the sound is determined by adjusting the angle between the airflow and the direction of sound propagation in the sound propagation correction module. If the vocal timbre is imperfect and Synchronous change, i.e., adjustment Calculate the weights; adjust the feature reliability of the feature fusion module. , Computational logic, if feature stability and Synchronization of changes, i.e., correction , The association weights between the median standard deviation and the mean are then output. and To the results output module.

[0041] As a further improvement to this technical solution, the result output module performs the following steps in the process of result determination and output:

[0042] S25.1 Receive the temperature field of the current round transmitted by the iterative optimization module. and corresponding mean absolute error ;

[0043] S25.2 When the preset stability conditions of temperature field error and global temperature gradient are met for three consecutive iterations, the iteration is determined to be converged.

[0044] S25.3 If the iteration converges, output This data serves as the final three-dimensional temperature distribution data for the entire furnace, transmitted to the three-dimensional temperature field intelligent visualization unit. If convergence is not achieved, Feedback is sent to the sound propagation correction module as the initial temperature field for the next iteration. .

[0045] As a further improvement to this technical solution, the three-dimensional temperature field intelligent visualization unit includes a data adaptation module, a model construction module, and a visualization interaction module, wherein:

[0046] The data adaptation module is signal-connected to the working condition adaptive inversion unit and is used to receive three-dimensional temperature distribution data of the entire furnace, perform data format conversion, and match the spatial coordinate system and data accuracy standard of the furnace geometric model in the model construction module.

[0047] The model building module and the data adaptation module are connected by a signal. The model building module is used to call a general rendering tool to build a geometric model that matches the actual furnace, and to overlay the converted temperature data onto the corresponding spatial region of the geometric model to generate a real-time temperature field image of the furnace.

[0048] The visualization interaction module is signal-connected to the model building module and is used to perform temperature value annotation and furnace combustion area feature display operations on the real-time temperature field image, while also supporting multi-view switching function.

[0049] As a further improvement to this technical solution, the working condition early warning and interaction unit includes a working condition comparison module, an early warning triggering module, and a human-machine interaction module, wherein:

[0050] The operating condition comparison module is connected to the three-dimensional temperature field intelligent visualization unit and is used to retrieve the preset safe temperature range of the furnace equipment, compare the temperature field data output by the three-dimensional temperature field intelligent visualization unit, and identify abnormal temperature information that exceeds the safe range.

[0051] The early warning triggering module is signal-connected to the operating condition comparison module. When abnormal temperature information is detected, it outputs an audible and visual alarm signal and simultaneously pushes the coordinates of the furnace area corresponding to the abnormal temperature.

[0052] The human-computer interaction module is connected to the early warning triggering module and the three-dimensional temperature field intelligent visualization unit, respectively, and is used to display real-time temperature field images, historical temperature data curves and early warning records. It also supports the adjustment of interface display parameters.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] 1. This invention deploys acoustic sensors and differential pressure flow velocity sensors around the furnace circumference to simultaneously collect sound wave and airflow data. After bandpass filtering to remove interference, the data is integrated into corrected basic data. Then, the equations are solved using the rapid travel method to obtain the sound wave propagation path and theoretical time. The time residual is calculated and transmitted in reverse to correct the temperature field. In addition, a correction coefficient is introduced in combination with the relative humidity of the airflow to adjust the sound velocity calculation, effectively correcting the sound ray propagation error and improving the accuracy of furnace temperature field reconstruction under complex working conditions.

[0055] 2. This invention receives acoustic wave data, operating parameters, and a corrected temperature field. It extracts acoustic wave frequency-amplitude features, operating mean, and fluctuation features by branching out the data. Based on the feature reliability (negatively correlated with standard deviation and positively correlated with mean), it dynamically allocates attention weights to obtain fused features. The fused features are then input into an inversion model to obtain the current temperature field. The error is calculated based on the thermocouple measured temperature. Regularization coefficients that are negatively correlated with error and iteration count are used for optimization. Furthermore, the error adjustment parameters are decomposed. This allows the system to adapt to changes in operating conditions without manual intervention, improving system automation and long-term operational reliability.

[0056] 3. This invention receives full-domain three-dimensional temperature data, converts the format, and uses tools to construct a geometric model that matches the actual furnace. It then overlays the temperature data to generate a real-time temperature field image, supporting temperature labeling, combustion zone display, and multi-view switching. Simultaneously, it compares the data with a preset temperature safety range to identify anomalies. When an anomaly is triggered, it outputs an audible and visual alarm and pushes the coordinates of the abnormal area. It can also display real-time images, historical curves, and warning records, and supports parameter adjustment. This allows maintenance personnel to intuitively perceive the temperature gradient and local overheating within the furnace, matching the needs of intelligent operation and maintenance. Attached Figure Description

[0057] Figure 1 This is a system framework diagram of the present invention;

[0058] The meanings of the labels in the diagram are as follows:

[0059] 1. Array-type acoustic sensing data acquisition unit; 11. Array deployment module; 12. Signal acquisition module; 13. Filtering module; 14. Data fusion module;

[0060] 2. Adaptive Inversion Unit for Operating Conditions; 21. Data Receiving Module; 22. Sound Ray Propagation Correction Module; 23. Feature Fusion Module; 24. Iterative Optimization Module; 25. Result Output Module;

[0061] 3. Three-dimensional temperature field intelligent visualization unit; 31. Data adaptation module; 32. Model building module; 33. Visualization interaction module;

[0062] 4. Working condition early warning and interaction unit; 41. Working condition comparison module; 42. Early warning triggering module; 43. Human-computer interaction module. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, this embodiment provides an intelligent visualization system for furnace temperature field based on acoustic sensing technology, including:

[0065] The array-type acoustic sensing data acquisition unit 1 has multiple sets of acoustic sensor arrays arranged around the furnace circumference to collect acoustic wave signals inside the furnace; the acoustic wave signals are filtered to suppress characteristic noise, and combined with the airflow data of the differential pressure flow velocity sensor, the basic data for sound ray propagation correction is output.

[0066] In this embodiment, the array-type acoustic sensing data acquisition unit 1 includes an array deployment module 11, a signal acquisition module 12, a filtering processing module 13, and a data fusion module 14, wherein:

[0067] The array deployment module 11 is used to coordinate the deployment of multiple acoustic sensor arrays and differential pressure flow sensors along the circumference of the furnace. The installation height of the acoustic sensor array is adapted to the center area of ​​the furnace flame, and the differential pressure flow sensors correspond one-to-one with the acoustic sensor arrays in the furnace area. The distance between adjacent sampling points is ≤1.5m to ensure the spatial matching of acoustic wave and airflow data.

[0068] Specifically, the array deployment module 11 divides the furnace circumference into multiple monitoring sub-regions. Each sub-region corresponds to a set of high-temperature resistant and dust-resistant piezoelectric acoustic sensor arrays and a Pitot tube differential pressure sensor, which are deployed together at the height section of the furnace flame center area. The distance between adjacent sampling points is adjusted by a laser rangefinder to ≤1.5m, so as to realize the spatial matching acquisition of sound wave and airflow data.

[0069] The signal acquisition module 12 is integrated into the acoustic sensor array of the array deployment module 11, and is used to synchronously acquire acoustic wave signals in different areas of the furnace. The acquisition frequency is matched with the propagation period of the acoustic wave in the furnace.

[0070] Specifically, the signal acquisition module 12 is integrated into the acoustic sensor array and adopts a multi-channel synchronous acquisition architecture. It synchronously acquires acoustic wave signals from each region at a frequency of 20kHz to 50kHz. The acquisition timing is ensured by the FPGA synchronous trigger circuit, and the analog signal is converted into a digital signal by a 16-bit analog-to-digital converter.

[0071] The filtering module 13 is connected to the signal acquisition module 12 and is used to receive the original acoustic signal. It uses bandpass filtering to retain the relevant characteristic frequency bands of furnace combustion and filter out interference frequency band signals.

[0072] Specifically, the filtering module 13 uses a digital signal processor to implement bandpass filtering. The passband is set to 500Hz~5000Hz to retain the characteristics of combustion-related acoustic waves. A fifth-order Butterworth filtering algorithm is used to suppress power frequency, mechanical vibration and high-frequency electromagnetic noise interference.

[0073] The data fusion module 14 is connected to the filtering module 13 and the differential pressure flow velocity sensor signal, respectively, and is used to synchronize the filtered sound wave signal with the airflow data in time and integrate it into the sound ray propagation correction basic data output to the working condition adaptive inversion unit 2.

[0074] Specifically, the data fusion module 14 uses GPS timing or PTP protocol to add high-precision timestamps to the acoustic signal and airflow data to achieve time synchronization. The synchronized data is then integrated into a structured data packet containing sub-region number, acoustic segment, airflow velocity, and timestamp, and output to the working condition adaptive inversion unit 2 via the Profinet industrial Ethernet interface.

[0075] It is understandable that differential pressure flow sensors employ cooling sleeve protection measures: the sensor probe is placed inside the cooling sleeve, and a cooling medium (such as compressed air) is introduced into the sleeve to isolate the high temperature of the furnace, ensuring that the sensor works stably in an environment of ≤850℃, and avoiding signal distortion or equipment damage caused by high temperature.

[0076] The adaptive inversion unit 2 receives real-time operating parameters from the boiler DCS system and basic acoustic signal data output from the array-type acoustic sensing data acquisition unit 1. Based on the current iterative temperature field distribution and input airflow velocity field, it uses the fast travel method to solve the Eikonal equation to obtain the minimum propagation time of each acoustic path. The time residual is then propagated back to the temperature field update process via an automatic differentiation mechanism. Simultaneously, an improved algorithm using dual-branch attention fusion and residual feedback dynamic regularization is employed to extract the time series features of the acoustic data and the steady-state and dynamic features of the operating parameters. Nonlinear fusion of heterogeneous data is achieved through dynamic allocation of attention weights. During the iteration process, the temperature distribution error is propagated back to dynamically adjust the algorithm parameters, combining the acoustic propagation correction results, and outputting three-dimensional temperature distribution data of the entire furnace area. The adaptive inversion unit 2 includes a data receiving module 21, an acoustic propagation correction module 22, a feature fusion module 23, an iterative optimization module 24, and a result output module 25, wherein:

[0077] In this embodiment, the input terminal of the data receiving module 21 is connected to the boiler DCS system and the array-type acoustic sensing data acquisition unit 1 respectively, and is used to receive real-time operating parameters and basic data of acoustic signals, and synchronously distribute them to the sound propagation correction module 22 and the feature fusion module 23.

[0078] Specifically, the data receiving module 21 adopts a dual-interface architecture to achieve synchronous data reception and distribution, ensuring the real-time performance and consistency of multi-source data. It establishes a communication link with the boiler DCS system via the Modbus TCP protocol, with a transmission rate set to ≥100Mbps. The received real-time operating parameters include furnace load and airflow direction vector. relative humidity of airflow ;

[0079] Meanwhile, the data receiving module 21 interfaces with the array-type acoustic sensing data acquisition unit 1 via the Profinet protocol, receiving the basic data of the acoustic wave signal output by the array-type acoustic sensing data acquisition unit 1, specifically including the measured sound wave propagation time. Airflow velocity field collected by differential pressure flow sensor .

[0080] Furthermore, to avoid timing discrepancies in multi-source data, the data receiving module 21 has a built-in timestamp synchronization engine that adds timestamps with an accuracy of ≤1μs to the two types of received data. After the data timing is aligned, the data receiving module 21 synchronously distributes it to the sound propagation correction module 22 and the feature fusion module 23, providing a unified time reference data source for subsequent processing.

[0081] In this embodiment, the sound propagation correction module 22 is signal-connected to the data receiving module 21 and is used to receive the initial temperature field. airflow velocity field and measured sound wave propagation time A corrected temperature field is generated by correcting the sound propagation path error. The output is then sent to the feature fusion module 23; the sound propagation correction module 22 performs the sound propagation correction process, which includes the following steps:

[0082] S22.1 Receive the initial temperature field transmitted by the data receiving module 21 Real-time airflow velocity field The measured sound wave propagation time output by the array-type acoustic sensing data acquisition unit 1. ;in The reference temperature field for furnace design or the temperature field results from the previous round fed back by the iterative optimization module 24;

[0083] S22.2, For any three-dimensional spatial coordinates within the furnace Calculate the sound wave propagation speed after temperature and airflow coupling according to general acoustic relationships. The angle between the airflow and the direction of sound propagation The sound wave propagation speed in S22.2 is obtained by calculating the airflow direction data from the DCS system and the coordinates of the acoustic sensor deployment. The calculation supports airflow humidity adaptation, and the specific adaptation process includes the following steps:

[0084] S22.21 Receive the relative humidity of the airflow inside the furnace output by the DCS system. ;

[0085] S22.22, Introduce a humidity correction factor , The value of varies Linear change, The value range is [0,1];

[0086] S22.23, based on and Adjustment The computational logic enables It is adapted to the sound speed propagation characteristics under corresponding humidity conditions.

[0087] S22.3. Taking the sound sensor transmitter as the starting point and the receiver as the ending point, the Eikonal equation is solved using the fast travel method to obtain the shortest path of sound wave propagation and the corresponding theoretical propagation time. ;

[0088] S22.4, Calculation and residual Through automatic differentiation mechanism Reverse conduction to Generate the corrected temperature field And transmit it to the feature fusion module 23.

[0089] Specifically, the sound propagation correction module 22 is signal-connected to the data receiving module 21, and performs sound propagation path error correction according to the steps in the claims, ultimately generating a corrected temperature field. The specific process is as follows:

[0090] First, the initial temperature field transmitted by the data receiving module 21 is received. Real-time airflow velocity field The measured sound wave propagation time output by the array-type acoustic sensing data acquisition unit 1. .in, The reference temperature field for furnace design (retrieved from the boiler design database) or the temperature field results from the previous round fed back by the iterative optimization module 24; It is vector data, in the format of , representing coordinates airflow at the location , , The directional velocity component is acquired by a differential pressure flow sensor in the array-type acoustic sensing data acquisition unit 1, with a sensor accuracy class of 0.5. The sound sensor transmitter (denoted as) ) to the receiving end (denoted as The actual propagation time is calculated as follows:

[0091] ;

[0092] In the formula, This indicates the moment when the acoustic sensor emits sound waves. This indicates the moment when the acoustic sensor receives the sound wave; the time measurement accuracy is ≤1ms.

[0093] Subsequently, for any three-dimensional spatial coordinates within the furnace Calculate the sound wave propagation speed after temperature and airflow coupling according to general acoustic relationships. The calculation formula is:

[0094] ;

[0095] In the formula, The speed of sound in dry air, determined solely by temperature, is expressed in m / s and is given by the formula:

[0096] ;

[0097] in, Represents the coordinates in the initial temperature field The temperature value at that location, in °C;

[0098] Representing coordinates The airflow velocity vector at the location (in the format of) ), The magnitude (scalar, in m / s) of the airflow velocity vector is calculated from the magnitude of the airflow velocity field vector, using the following formula:

[0099] ;

[0100] in, These represent the airflow velocity vectors respectively. exist The directional component of velocity, all in m / s;

[0101] The airflow direction vector, expressed in rad (radians), represents the angle between the airflow and the direction of sound propagation, output by the DCS system. With the direction vector of the sound ray ( The formula for calculating using the vector dot product is:

[0102] ;

[0103] in, Represents the airflow direction vector. Indicates that the sound ray originates from the transmitting end. to the receiving end directional vector, Let represent the magnitudes of the two vectors respectively.

[0104] Meanwhile, to accommodate the effect of airflow humidity on sound velocity, a humidity correction coefficient is introduced. The specific adaptation process is as follows:

[0105] Receive relative humidity of airflow inside the furnace from the DCS system (Values ​​range [0,1], unitless, humidity sensor measurement accuracy ≤±3%RH); Humidity correction factor The calculation formula is:

[0106] ;

[0107] The derivation of the above formula is based on the conclusion regarding the relationship between air speed and humidity in the authoritative textbook in the field of acoustics, *Fundamentals of Acoustics* (by Du Gonghuan, Zhu Zhemin, and Gong Xiufen, 3rd edition): When air humidity increases, the average molecular weight of air decreases because the molecular weight of water vapor (18 g / mol) is smaller than that of dry air (approximately 29 g / mol). Since the speed of sound is inversely proportional to the square root of the molecular weight of the medium, at the same temperature, the speed of sound increases by approximately 0.2% for every 100% increase in relative humidity (RH). Based on this physical mechanism, a linear relationship is used to adapt the effect of humidity on the speed of sound.

[0108] based on and Adjusting the sound velocity calculation logic yields a sound wave propagation speed adapted to humidity levels. The formula is:

[0109] ;

[0110] In the formula, This represents the speed of sound propagation after temperature and airflow coupling. This indicates the final sound wave propagation speed after humidity adjustment, and the unit is m / s.

[0111] Next, using the acoustic sensor transmitter... For the starting point and the receiving end Using the fast travel method as the endpoint, the Eikonal equation is solved to obtain the shortest path for sound wave propagation and the corresponding theoretical propagation time. The Eikonal equation is in the form of:

[0112] ;

[0113] In the formula, Indicates theoretical propagation time The spatial gradient reflects the rate of change of time in three-dimensional space; This represents the sound wave propagation speed after humidity adjustment, in m / s. During the solution process, the three-dimensional space of the furnace is divided into a grid with a grid size of ≤0.5m. This value is based on the requirement in the "Power Plant Boiler Performance Test Procedure" (GB / T10184-2025) that "the density of temperature measurement points in the furnace must meet the requirement of a grid size ≤0.5m". The speed of sound is calculated iteratively for each grid node using the fast travel method. Finally, the receiving end is taken place As a voice The theoretical propagation time.

[0114] Finally, the residual between the theoretical propagation time and the measured time is calculated. The formula is:

[0115] ;

[0116] In the formula, Indicates vocal timbre The theoretical propagation time, measured in seconds (s); Indicates vocal timbre The measured propagation time, in seconds. Then, an automatic differentiation mechanism is used to... Reverse conduction to the initial temperature field Generate a corrected temperature field The formula is:

[0117] ;

[0118] In the formula, This represents the correction factor, with a value range of [0.1, 0.5]. Residual For the initial temperature field The partial derivatives are calculated using PyTorch's autograd automatic differentiation module; Represents the coordinates in the corrected temperature field The temperature value at the location, in °C, is finally output to the feature fusion module 23.

[0119] Furthermore, the process of solving the Eikonal equations using the fast travel method includes the following steps:

[0120] Initialization rules: Set the acoustic sensor transmitter end Assuming it's a "determined point", its theoretical propagation time is... ;by Centered on the grid, the surrounding adjacent grid nodes are set as "active points" (included in narrowband management), and the remaining nodes are set as "undetermined points".

[0121] Narrowband management method: Uses a binary heap (priority queue) to store active points, and the queue is ordered by... Sort by size from smallest to largest to ensure that each retrieval is successful. Update the smallest active point.

[0122] The nearest neighbor update formula uses a first-order upwind scheme (adapted to the hyperbolic properties of the Eikonal equation) to calculate the nearest neighbor points. The formula is:

[0123] ;

[0124] In the formula, For the current neighboring points to be updated, For the already determined points, The spacing between grid nodes is ≤0.5m. for The sound velocity is adapted to the humidity level; after the update, if... of If convergence is achieved, the point is converted into a "determined point" and its undetermined neighboring points are added to the active point queue.

[0125] Difference scheme selection: The first-order upwind scheme is adopted because the solution of the Eikonal equation increases monotonically along the direction of sound ray propagation. The upwind scheme can avoid numerical oscillations and ensure computational stability.

[0126] Furthermore, the calculation is performed using PyTorch's autograd automatic differentiation module. The logic is as follows:

[0127] Input parameter format: Initial temperature field It is stored in a three-dimensional mesh data structure, and the temperature value of each mesh node is a differentiable variable;

[0128] Forward propagation path: As input, calculate sequentially. The final output residual ;

[0129] Backpropagation: Using an automatic differentiation algorithm (based on the chain rule), from the residual... Reverse calculation of its pair Partial derivative of temperature value at each grid node The partial derivative directly reflects the degree of influence of temperature change on the residual.

[0130] Traditional sound propagation corrections often ignore the effects of humidity or simplify sound ray as a straight line, and empirical temperature adjustment still relies on manual intervention—inaccurate sound velocity calculations under complex flow fields can affect theoretical propagation time. Large deviations and simplification of the vocal ray path can lead to residuals. If the wrong direction is used, manual temperature adjustment is slow and inaccurate, and the temperature field correction accuracy can never keep up with changes in operating conditions. In this embodiment, the sound ray propagation correction module 22 adopts a method of "coupled sound velocity calculation of three factors: temperature, airflow, and humidity + fast travel method to solve the Eikonal equation (0.5m grid tracking sound ray bending) + automatic differential residual back correction of the temperature field", which is like equipping the sound ray with a "precise navigator". Its core function is to correct the sound ray propagation under complex operating conditions. More realistic It can directly provide feedback to adjust the initial temperature field. This addresses the issues of traditional models neglecting humidity and simplifying sound, allowing the corrected model to... More accurate, laying a solid data foundation for subsequent inversion.

[0131] In this embodiment, the feature fusion module 23 is signal-connected to the sound propagation correction module 22 and the data receiving module 21, respectively, and is used to extract sound wave features. Operating conditions A fused feature vector is generated through dual-branch attention fusion. The output is sent to the iterative optimization module 24; the feature fusion module 23 performs the dual-branch attention fusion process, which includes the following steps:

[0132] S23.1 Receive the raw acoustic wave data and operating parameters transmitted by the data receiving module 21, and synchronously receive the corrected temperature field output by the sound propagation correction module 22. ;

[0133] S23.2 Perform dual-branch feature extraction. The first branch uses the sliding window method to extract the frequency-amplitude features of the acoustic wave data. The second branch extracts the mean and fluctuation characteristics of the operating condition parameters. ;

[0134] S23.3 Calculating the reliability of acoustic wave features based on the statistical characteristics of feature data. The reliability of the operating condition characteristics is negatively correlated with the standard deviation and positively correlated with the mean. It is negatively correlated with the standard deviation of the operating condition characteristics and positively correlated with the mean;

[0135] S23.4, according to and Assign attention weights The fused feature vector is obtained through weighted calculation. The weights are updated in real time with the new feature data and output to the iterative optimization module 24.

[0136] Specifically, the feature fusion module 23 is signal-connected to the sound propagation correction module 22 and the data receiving module 21, respectively. It extracts and integrates features through dual-branch attention fusion to generate a fused feature vector. The specific process is as follows:

[0137] First, the system receives the raw acoustic wave data and operating parameters transmitted by the data receiving module 21, and the corrected temperature field output by the sound ray propagation correction module 22. The three types of data are preprocessed to remove invalid values ​​(such as abnormal values ​​of temperature and flow rate that are outside the reasonable range) that occur during transmission. The data are then classified and associated according to the furnace space region (corresponding to the sampling area of ​​the array-type acoustic sensing data acquisition unit 1), and the temperature field is corrected. The calculation cycle is synchronized with the 10-second statistical cycle of the working condition characteristics to ensure the matching of data in the spatiotemporal dimensions and to guarantee the targeting of feature extraction.

[0138] Then, the first branch extracts acoustic features. (Frequency-amplitude characteristics) The raw acoustic data was processed using the sliding window method, with a window size W=2048 and a step size S=1024; the acoustic time-domain signal within each window was processed. ( Perform a Fourier transform on the time (time) to extract the frequency-amplitude features. The formula is:

[0139] ;

[0140] In the formula, This indicates the peak frequency of the sound wave within the window, expressed in Hz. This represents the sound wave amplitude corresponding to the peak frequency, in Pa (amplitude range 0-100 Pa). This represents the average amplitude of all frequency components within the window, in Pa.

[0141] The second branch extracts working condition features. (Mean and fluctuation characteristics), select three key operating parameters (load) output by the boiler DCS system. airflow speed ,humidity The statistical period is set to 10 seconds; the mean and fluctuation characteristics are extracted using the following formula:

[0142] ;

[0143] In the formula, This represents the average furnace load over 10 seconds. This represents the standard deviation of the furnace load over 10 seconds. This represents the average airflow velocity over 10 seconds. This represents the standard deviation of airflow velocity over 10 seconds. This represents the average relative humidity of the airflow over 10 seconds, and has no unit. It represents the standard deviation of relative humidity of airflow within 10 seconds.

[0144] Next, the confidence level is calculated based on the statistical characteristics of the feature data, specifically the confidence level of the acoustic wave features. The formula is:

[0145] ;

[0146] In the formula, This represents the mean value of the average amplitude in the characteristics of a sound wave. The standard deviation of the average amplitude in the characteristics of a sound wave is expressed in Pa. The value range is [0,1], and Negative correlation with They are positively correlated.

[0147] Reliability of operating condition characteristics The formula is:

[0148] ;

[0149] In the formula, These represent the coefficients of variation (dimensionless) for load, airflow velocity, and humidity, respectively; the denominator "3" represents the number of operating parameter types, ensuring... Normalize to the range [0,1]. It is negatively correlated with the coefficient of variation of each parameter, reflecting the stability of the operating condition.

[0150] Finally, according to and The attention weights are dynamically assigned using the following formula:

[0151] ;

[0152] ;

[0153] In the formula, Attention weights representing the characteristics of sound waves The attention weights represent the characteristics of the working conditions, and the sum of the two is 1.

[0154] And on , Normalized features are obtained by using Min-Max standardization. , (Value range [0,1]), weighted calculation of fused feature vector The formula is:

[0155] ;

[0156] In the formula, This represents the normalized sound wave characteristics. This represents the normalized operating condition characteristics. This represents the final fused feature vector (9 dimensions), which is output to the iterative optimization module 24.

[0157] In traditional multi-source feature fusion, fixed weights are often used to concatenate features, regardless of their reliability. The effective information of acoustic features can be masked by fluctuating operating parameters, and their role is not highlighted when the operating conditions are stable. Therefore, feature fusion... The inversion of the temperature field does not accurately reflect the actual operating conditions. In this embodiment, the feature fusion module 23 employs a dual-branch extraction method to extract the acoustic wave's 'frequency-amplitude' and the operating condition's 'mean-fluctuation' features, along with reliability quantification. , Combining statistical characteristics with dynamic attention weight allocation is like equipping features with an "intelligent sorter." Its core function is to assign higher weights to high-confidence features, avoiding fusion bias caused by fixed weights under fluctuating operating conditions. It more accurately matches the current working conditions, solves the problem of insufficient representation of simple splicing of heterogeneous data, and improves the input quality of subsequent temperature field inversion.

[0158] In this embodiment, the iterative optimization module 24 and the feature fusion module 23 are signal-connected, based on Invert the current temperature field Combined with regularization coefficient Perform residual feedback dynamic regularization optimization and output and the corresponding mean absolute error The results are output to module 25, and the non-converged results are also included. Feedback is sent to the sound propagation correction module 22; the process of residual feedback dynamic regularization optimization performed by the iterative optimization module 24 includes the following steps:

[0159] S24.1, The output of the feature fusion module 23 Input the temperature field inversion model to calculate the global temperature field of the furnace in the current cycle. ;

[0160] S24.2, The actual temperature measured by thermocouples on the furnace wall. Based on the baseline, calculate Mean absolute error , coordinates The measured temperature value at the location;

[0161] S24.3, The value of and It is negatively correlated with the number of iterations. They are negatively correlated; the larger the error, the better in the early stages of iteration. The smaller the value, the smaller the error, and the better in the later stages of iteration. The larger;

[0162] S24.4 Disassembly The source of the sound is adjusted by changing the angle between the airflow and the direction of sound propagation in the sound propagation correction module 22. If the vocal timbre is imperfect and Synchronous change, i.e., adjustment Calculate the weights; adjust the feature fusion module 23 feature confidence. , Computational logic, if feature stability and Synchronization of changes, i.e., correction , The association weights between the median standard deviation and the mean are then output. and To the results output module 25.

[0163] Specifically, the iterative optimization module 24 is signal-connected to the feature fusion module 23. Based on the fused features, the temperature field is inverted and residual feedback dynamic regularization optimization is performed. The specific process is as follows:

[0164] First, the feature fusion module 23 outputs... Input the temperature field inversion model, which is a fully connected neural network containing two hidden layers (64 neurons each) and using ReLU as the activation function; invert the global furnace temperature field for the current cycle. The formula is:

[0165] ;

[0166] In the formula, This represents a fully connected neural network. This represents the fused feature vector of the input. This represents the pre-training parameters (weights, biases) of the neural network. Indicates the number of iterations (initial). =1), Indicates the first Coordinates in the temperature field of the next iteration The temperature value at that location is expressed in °C.

[0167] Then, the temperature was measured by thermocouples on the furnace wall. As a benchmark ( For the first The coordinates of the thermocouples, , ≥8), calculate Mean absolute error The formula is:

[0168] ;

[0169] In the formula, Indicates the number of thermocouples on the furnace wall; Indicates the first Coordinates in the temperature field of the next iteration The temperature value at that location, in °C; Representing coordinates The measured temperature value of the thermocouple, in °C.

[0170] Next, the regularization coefficient and Negatively correlated with the number of iterations They are negatively correlated, and the calculation formula is:

[0171] ;

[0172] In the formula, This represents the initial regularization coefficient, with a value range of [0.01, 0.1], determined by model pre-training; This represents the error impact coefficient, with a value range of [1, 5], ensuring... The larger, The smaller; This represents the influence coefficient of the iteration number, with a value range of [0.05, 0.2], ensuring... The larger, The larger; Indicates the first The regularization coefficient of the next iteration, which is dimensionless.

[0173] Finally, disassemble The source and dynamically adjust the correlation parameters: if the acoustic residual and Synchronous changes (such as) When it increases (Also increase), then adjust the angle between the airflow and the direction of sound propagation in the sound propagation correction module 22. The weights are calculated and adjusted using the following formula:

[0174] ;

[0175] In the formula, Indicates the adjusted included angle; Indicates the included angle before adjustment; This represents the angle adjustment coefficient, with a value range of [0.05, 0.1]. Indicates the first The mean absolute error of the iterations. Indicates the first The mean absolute error of the iterations.

[0176] If the characteristic stability is Synchronous changes (e.g.) Decrease If the value is increased, the calculation logic for feature confidence in feature fusion module 23 is modified to adjust the confidence of acoustic features. For example, the corrected formula is:

[0177] ;

[0178] In the formula, This indicates the reliability of the corrected acoustic wave characteristics; This represents the correction factor, with a value range of [0.8, 1.2]. This represents the mean of the average amplitude in the characteristics of a sound wave. It represents the standard deviation of the average amplitude in the characteristics of a sound wave.

[0179] After the adjustment is completed, the iterative optimization module 24 outputs. and corresponding The results are output to module 25, and the non-converged results are also included. Feedback is sent to the sound propagation correction module 22 as the initial temperature field for the next iteration. .

[0180] Furthermore, the pre-training details of the temperature field inversion model are as follows:

[0181] Network Dimension Matching: Input layer dimension is 9 (and fused feature vector) (Consistent with 9 dimensions), both hidden layers have 64 neurons each, and the output layer dimension is... (In this embodiment, the number of nodes is consistent with the number of nodes in the furnace grid) The mapping relationship between the output layer and the temperature field mesh is "output layer 1". The first neuron corresponds to the furnace. Temperature value of each grid node;

[0182] The pre-training dataset is sourced from historical data of a 300MW power plant boiler over a continuous month, including: acoustic signals, operating parameters (load, airflow velocity, humidity), and measured temperatures of the furnace wall thermocouples (a total of 8 measuring points).

[0183] Training process: The Adam optimizer was used to train for 50 epochs with the loss function being the "mean absolute error between the predicted temperature field and the measured temperature" and the learning rate being set to 0.001.

[0184] In addition, "vocal residual" and The quantitative judgment criterion for "synchronous change" is based on five consecutive iterations. and The data implementation is as follows:

[0185] calculate Rate of change: The formula for rate of change is:

[0186] ;

[0187] In the formula, Indicates the first The iteration relative to the first The rate of change of acoustic residuals in each iteration (dimensionless). For the first The acoustic residual of the nth iteration (i.e., the 1st iteration) (Theoretical and experimental results of the next iteration) For the first The vocal timbre residual of the next iteration;

[0188] calculate Rate of change: The formula for rate of change is:

[0189] ;

[0190] In the formula, Indicates the first The iteration relative to the first The average absolute error change rate (dimensionless) of each iteration. For the first The mean absolute error of the iterations. For the first The mean absolute error of the next iteration;

[0191] Determine the conditions for synchronous change: First calculate and The Pearson correlation coefficient ρ (dimensionless, range of values) The formula is:

[0192] ;

[0193] In the formula, express and The degree of linear correlation; Index of iteration number (take the first iteration number) To the (a total of 5 iterations of data); For the first The vocal timbre residual of the next iteration For these 5 times The average value; For the first The mean absolute error of the iterations. For these 5 times The average value.

[0194] If the following two conditions are met, then Δτ and Ek are considered to change "synchronously," and the calculation weight of the vocal liner angle needs to be adjusted:

[0195] (express and (showing a strong positive linear correlation)

[0196] and The symbols are consistent (indicating) and The changes are in the same direction, such as increasing or decreasing simultaneously.

[0197] In traditional temperature field iterative optimization, the regularization coefficient is fixed, and it's unclear how to adjust it when the error becomes large—a large coefficient in the early stages of iteration limits accuracy improvement, while a small coefficient in the later stages easily leads to overfitting and increased error. Increasing the value only allows for general model adjustments, and the inversion results always drift under high-risk conditions. In this embodiment, the iterative optimization module 24 uses a "dynamic regularization coefficient ( Follow , Adaptive adjustment) + Error decomposition and source tracing (correlation) The method involves "feature stability parameter adjustment + thermocouple measured error anchoring". Its core function is to maintain accuracy in the early stages of iteration, prevent overfitting in later stages, and, when errors are large, to specifically adjust the acoustic ray angle and feature confidence weights. It more closely matches the measured values ​​of thermocouples, solves the problems of slow convergence and overfitting in traditional iteration, and improves the stability of the inversion results.

[0198] In this embodiment, the result output module 25 is signal-connected to the iterative optimization module 24 and is used to determine the convergence of the iteration. When the temperature field error and gradient stability conditions are met for three consecutive iterations, the output is successful. The process of determining and outputting results in the three-dimensional temperature field intelligent visualization unit 3 includes the following steps:

[0199] S25.1 Receive the current round temperature field transmitted by the iterative optimization module 24. and corresponding mean absolute error ;

[0200] S25.2 When the preset stability conditions of temperature field error and global temperature gradient are met for three consecutive iterations, the iteration is determined to be converged.

[0201] S25.3 If the iteration converges, output The final three-dimensional temperature distribution data of the entire furnace is sent to the three-dimensional temperature field intelligent visualization unit 3; if convergence is not achieved, Feedback is sent to the sound propagation correction module 22 as the initial temperature field for the next iteration. .

[0202] Specifically, the result output module 25 is signal-connected to the iterative optimization module 24, and is responsible for determining the convergence of the iteration and outputting the final temperature field. The detailed process is as follows:

[0203] First, the temperature field of the current round is received from the iterative optimization module 24. and corresponding mean absolute error Structure the storage of two types of data (format: "iteration count"). ", and establish an index linking the number of iterations with the results to facilitate tracing and retrieval.

[0204] Then, when the preset conditions of "temperature field error stability" and "global temperature gradient stability" are met for three consecutive iterations, the iteration is considered to have converged. The "error stability" condition is as follows:

[0205] ;

[0206] In the formula, , , They represent the first , , The mean absolute error of the iterations, in °C; This represents the error threshold, with a value range of [1,3]℃, based on the temperature measurement accuracy requirements of industrial boilers (such as "the allowable error for furnace temperature measurement is ≤3℃" in GB / T10184-2025).

[0207] The condition for "global temperature gradient stability" is:

[0208] ;

[0209] In the formula, , They represent the first , The spatial gradient of the temperature field in each iteration is expressed in °C / m. This represents the temperature gradient threshold, with a value range of [5,10]℃ / m.

[0210] Finally, if the iteration is determined to be converged, the result output module 25 outputs the result. The final three-dimensional temperature distribution data of the entire furnace is transmitted to the three-dimensional temperature field intelligent visualization unit 3 via an industrial Ethernet interface; if convergence is not achieved, Feedback is sent to the sound propagation correction module 22 as the initial temperature field for the next iteration. At the same time, the maximum number of iterations is set to 20 (to avoid infinite loops; the value of 20 is based on engineering practice of numerical iterative algorithms). If the number of iterations reaches 20 and the algorithm still has not converged, a device self-test prompt will be triggered.

[0211] The 3D temperature field intelligent visualization unit 3 receives the furnace full-domain 3D temperature distribution data output by the working condition adaptive inversion unit 2. After format conversion, it constructs the furnace geometric model and superimposes the temperature information through a general rendering tool to generate a real-time temperature field image. It has the functions of temperature value annotation, regional feature display and view switching.

[0212] In this embodiment, the three-dimensional temperature field intelligent visualization unit 3 includes a data adaptation module 31, a model building module 32, and a visualization interaction module 33, wherein:

[0213] The data adaptation module 31 is connected to the working condition adaptive inversion unit 2 by signal, and is used to receive the three-dimensional temperature distribution data of the entire furnace, perform data format conversion, and match the spatial coordinate system and data accuracy standard of the furnace geometric model in the model construction module 32.

[0214] Specifically, the data adaptation module 31 receives the three-dimensional temperature distribution data of the furnace full domain output by the working condition adaptive inversion unit 2 through an industrial Ethernet interface (such as using the Profinet protocol with a transmission rate of ≥100Mbps). This data is a structured dataset (such as a custom binary format or JSON format) containing spatial coordinates (x, y, z) and corresponding temperature values.

[0215] Meanwhile, the data adaptation module 31 has a built-in coordinate transformation engine that converts the data's computational coordinate system (defined by the inversion unit) into a Cartesian world coordinate system consistent with the model building module 32. The transformation process is achieved through preset coordinate offsets and rotation matrices (which can be calibrated on-site according to the actual installation position of the furnace). At the same time, the temperature data accuracy is normalized, and two decimal places are uniformly retained to ensure complete matching with the spatial coordinate system and data accuracy standards of the model building module 32. The processed data is temporarily stored in an intermediate format that the rendering tool can recognize (such as CSV or a data format specific to the rendering tool), providing a basis for subsequent temperature superposition.

[0216] The model building module 32 is connected to the data adaptation module 31 by signal. It is used to call a general rendering tool to build a geometric model that matches the actual furnace, and to overlay the converted temperature data onto the corresponding spatial area of ​​the geometric model to generate a real-time temperature field image of the furnace.

[0217] Specifically, the model building module 32 calls general rendering tools (such as using the OpenGL graphics library to implement low-level rendering, or using the Unity engine to improve interactivity) to import the CAD 3D model file of the furnace (such as STEP format, which contains the size parameters of key structures such as the furnace inner wall and burners), or to parametrically build a geometric model based on the actual size of the furnace (diameter, height, combustion area position, etc.) to ensure that the shape and size of the model deviate from the actual furnace by ≤5%.

[0218] Meanwhile, the model building module 32 overlays the temperature data converted by the data adaptation module 31 onto the corresponding spatial region of the geometric model through a color mapping algorithm (such as setting a temperature-color mapping table: ≤800℃ is green, 800-1000℃ is yellow, and ≥1000℃ is red). The temperature value of each spatial coordinate point is associated with the model vertex one by one. Real-time color rendering is achieved by using vertex shader technology, and the image refresh rate is maintained at 10-15 frames / second to ensure the real-time visualization of the temperature field.

[0219] The visualization interaction module 33 is connected to the model building module 32 by signal, and is used to perform temperature value annotation and furnace combustion area feature display operations on the real-time temperature field image, while also supporting multi-view switching function.

[0220] Specifically, the visualization interaction module 33 constructs a human-computer interaction interface based on the generated real-time temperature field image. The temperature value annotation function is implemented by embedding UI controls in the rendering interface. When the user clicks or selects a model area, the module triggers a coordinate query event, reads the temperature value of the corresponding spatial point, and displays it in the form of a floating window (the numerical accuracy is consistent with that of the data adaptation module 31). The furnace combustion area feature display function automatically identifies and marks high-temperature concentrated areas by setting a preset high temperature threshold (such as ≥1100℃), highlighting them with a red high-brightness outline or dynamic flashing effect.

[0221] Meanwhile, the multi-view switching function supports top view (view directly above the furnace), side view (view from the side of the furnace), and three-dimensional free rotation view. Users can switch views by using interface buttons, shortcut keys, or mouse dragging (scroll wheel zoom, left-click rotate). The camera parameters (position, angle, field of view) of each preset view can be preset through configuration files to ensure the stability and convenience of view switching.

[0222] The working condition early warning and interaction unit 4 compares the abnormal temperature information identified by the three-dimensional temperature field intelligent visualization unit 3 with the temperature safety range of the furnace equipment. When an early warning is triggered, it outputs an audible and visual alarm signal and pushes the coordinates of the abnormal area. It is equipped with a human-machine interface that can display real-time temperature field images, historical data curves and early warning records, and supports the adjustment of interface parameters.

[0223] In this embodiment, the working condition early warning and interaction unit 4 includes a working condition comparison module 41, an early warning triggering module 42, and a human-computer interaction module 43, wherein:

[0224] The working condition comparison module 41 is connected to the three-dimensional temperature field intelligent visualization unit 3 by signal, and is used to retrieve the preset safe temperature range of the furnace equipment, compare the temperature field data output by the three-dimensional temperature field intelligent visualization unit 3, and identify abnormal temperature information that exceeds the safe range.

[0225] Specifically, the working condition comparison module 41 receives the furnace full-domain temperature field data (including real-time temperature values ​​and corresponding coordinate information of each spatial coordinate point) output by the three-dimensional temperature field intelligent visualization unit 3 through the industrial Ethernet interface (adapted to the output interface of the three-dimensional temperature field intelligent visualization unit 3, using the Ethernet / IP protocol, with a transmission rate ≥100Mbps).

[0226] Meanwhile, the operating condition comparison module 41 has a built-in local configuration database that pre-stores the zone temperature safety ranges based on the material characteristics of the furnace equipment and the operating safety standards (for example, the safe temperature of the furnace water-cooled wall zone is ≤600℃, and the safe temperature of the burner outlet zone is ≤1200℃, which can be modified and updated through the subsequent human-machine interaction module 43).

[0227] Furthermore, during the comparison process, the operating condition comparison module 41 compares the real-time temperature value with the preset safety range of the corresponding area in real time according to the "spatial coordinate point by spatial point" logic. If the temperature of a certain coordinate point exceeds the upper limit of the safety range or falls below the lower limit of the safety range for 10 consecutive seconds (configurable) (if there is a low temperature safety requirement, such as pipeline antifreeze), the coordinate point is marked as an "abnormal temperature point". The coordinates and temperature values ​​of all abnormal temperature points are integrated to form a structured abnormal temperature information package (including the time of abnormal occurrence, abnormal coordinate set, corresponding area safety range, and actual temperature value), which is then transmitted to the early warning trigger module 42.

[0228] The early warning triggering module 42 is connected to the working condition comparison module 41 by signal. When abnormal temperature information is detected, it outputs an audible and visual alarm signal and simultaneously pushes the coordinates of the furnace area corresponding to the abnormal temperature.

[0229] Specifically, the early warning trigger module 42 establishes a signal connection with the operating condition comparison module 41 through the RS485 bus. After receiving the abnormal temperature information packet transmitted by the operating condition comparison module 41, it first analyzes the abnormal level in the packet (which can be pre-divided according to the temperature deviation range: deviation ≤ 50℃ is "Level 1 early warning", deviation > 50℃ is "Level 2 early warning").

[0230] Furthermore, for different warning levels, the warning trigger module 42 drives the external industrial-grade audible and visual alarm (which can be installed in a conspicuous location in the control room and in the duty room near the furnace equipment) to output audible and visual alarm signals of corresponding intensity.

[0231] Meanwhile, the early warning trigger module 42 pushes the coordinates of the furnace area corresponding to the abnormal temperature (which are consistent with the Cartesian coordinate system of the three-dimensional temperature field intelligent visualization unit 3) in two ways: first, it sends the coordinates to the human-machine interaction module 43 in real time and highlights them in the interface pop-up window; second, it sends the coordinates to the boiler DCS system via the Modbus protocol (if linkage control is required, such as triggering burner load adjustment), to ensure that maintenance personnel and the control system can obtain abnormal location information synchronously.

[0232] In addition, the early warning triggering module 42 will automatically record the triggering time, early warning level, abnormal coordinates, and processing status (initially "unprocessed") of each early warning, and synchronize the records to the early warning record database of the human-computer interaction module 43.

[0233] The human-computer interaction module 43 is connected to the early warning trigger module 42 and the three-dimensional temperature field intelligent visualization unit 3 respectively. It is used to display real-time temperature field images, historical temperature data curves and early warning records, and also supports the adjustment of interface display parameters.

[0234] It should be added that, taking the inversion period (1s) of the adaptive inversion unit 2 as the time synchronization benchmark, the acquisition period of the array-type acoustic sensing data acquisition unit 1 is 100ms. Within each inversion period of the adaptive inversion unit 2, the array-type acoustic sensing data acquisition unit 1 continuously acquires 10 sets of data and stores them in a dedicated data cache pool for use by the adaptive inversion unit 2 during inversion. The control period of the early warning and interaction unit 4 is 5s. It needs to perform average fusion based on the 5 inversion results continuously output by the adaptive inversion unit 2 before executing the feedback control operation to achieve synchronization matching between the control period and the inversion period.

[0235] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0236] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A furnace temperature field intelligent visualization system based on acoustic sensing technology, characterized in that, include: An array-type acoustic sensing data acquisition unit (1) is provided with multiple sets of acoustic sensor arrays arranged along the circumference of the furnace to collect acoustic wave signals inside the furnace. The acoustic signal is filtered to suppress characteristic noise, and combined with the airflow data from the differential pressure flow velocity sensor, the basic data for sound ray propagation correction is output. The adaptive inversion unit (2) receives the real-time operating parameters of the boiler DCS system and the basic data of the acoustic signal output by the array-type acoustic sensing data acquisition unit (1). Based on the current iterative temperature field distribution and the input airflow velocity field, the Eikonal equation is solved by the fast travel method to obtain the minimum propagation time of each acoustic path. The time residual is transmitted back to the temperature field update process through the automatic differentiation mechanism. At the same time, the improved algorithm of dual-branch attention fusion and residual feedback dynamic regularization is adopted to extract the time series features of acoustic data and the steady-state and dynamic features of operating parameters respectively. The nonlinear fusion of heterogeneous data is realized by dynamically allocating attention weights. In the iterative process, the temperature distribution error is transmitted back to dynamically adjust the algorithm parameters by combining the sound ray propagation correction results, and the three-dimensional temperature distribution data of the entire furnace is output. The three-dimensional temperature field intelligent visualization unit (3) receives the furnace full-domain three-dimensional temperature distribution data output by the working condition adaptive inversion unit (2), and after format conversion, constructs the furnace geometric model and superimposes the temperature information through a general rendering tool to generate a real-time temperature field image, which has the functions of temperature value labeling, regional feature display and view switching. The working condition early warning and interaction unit (4) compares the abnormal temperature information identified by the three-dimensional temperature field intelligent visualization unit (3) based on the safe temperature range of the furnace equipment; when the early warning is triggered, it outputs an audible and visual alarm signal and pushes the coordinates of the abnormal area; it is equipped with a human-machine interaction interface, which can display real-time temperature field images, historical data curves and early warning records, and supports the adjustment of interface parameters.

2. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 1, characterized in that, The array-type acoustic sensing data acquisition unit (1) includes an array deployment module (11), a signal acquisition module (12), a filtering module (13), and a data fusion module (14), wherein: The array layout module (11) is used to coordinate the layout of multiple acoustic sensor arrays and differential pressure flow sensors along the circumference of the furnace. The installation height of the acoustic sensor array is adapted to the center area of ​​the furnace flame, and the differential pressure flow sensor and the acoustic sensor array correspond one-to-one to the furnace area. The signal acquisition module (12) is integrated into the acoustic sensor array of the array deployment module (11) and is used to synchronously acquire acoustic wave signals in different areas of the furnace. The acquisition frequency is matched with the propagation period of the acoustic wave in the furnace. The filtering module (13) is connected to the signal acquisition module (12) and is used to receive the original acoustic signal. It uses bandpass filtering to retain the relevant characteristic frequency bands of furnace combustion and filter out interference frequency band signals. The data fusion module (14) is connected to the filtering module (13) and the differential pressure flow velocity sensor signal respectively, and is used to synchronize the filtered sound wave signal with the airflow data in time and integrate it into the sound ray propagation correction basic data output to the working condition adaptive inversion unit (2).

3. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 2, characterized in that, The adaptive inversion unit (2) includes a data receiving module (21), a sound propagation correction module (22), a feature fusion module (23), an iterative optimization module (24), and a result output module (25), wherein: The input terminal of the data receiving module (21) is connected to the boiler DCS system and the array-type acoustic sensing data acquisition unit (1) respectively, and is used to receive real-time operating parameters and basic data of acoustic signals, and synchronously distribute them to the sound propagation correction module (22) and the feature fusion module (23). The acoustic propagation correction module (22) is signal-connected to the data receiving module (21) and is used to receive the initial temperature field. airflow velocity field and measured sound wave propagation time A corrected temperature field is generated by correcting the sound propagation path error. , and output to the feature fusion module (23); The feature fusion module (23) is connected to the sound propagation correction module (22) and the data receiving module (21) respectively, and is used to extract sound wave features. Operating conditions A fused feature vector is generated through dual-branch attention fusion. , and output to the iterative optimization module (24); The iterative optimization module (24) and the feature fusion module (23) are connected by a signal, based on Invert the current temperature field Combined with regularization coefficient Perform residual feedback dynamic regularization optimization and output and the corresponding mean absolute error To the result output module (25), and simultaneously the unconverged Feedback is sent to the sound propagation correction module (22); The result output module (25) is signal-connected to the iterative optimization module (24) and is used to determine the convergence of the iteration. When the temperature field error and gradient stability conditions are met for three consecutive iterations, the output is... To the three-dimensional temperature field intelligent visualization unit (3).

4. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 3, characterized in that, The process of sound propagation correction by the sound propagation correction module (22) includes the following steps: S22.1, Receive the initial temperature field transmitted by the data receiving module (21) Real-time airflow velocity field The measured sound wave propagation time output by the array-type acoustic sensing data acquisition unit (1) ;in The previous round temperature field results are fed back by the furnace design reference temperature field or iterative optimization module (24); S22.2, For any three-dimensional spatial coordinates within the furnace Calculate the sound wave propagation speed after temperature and airflow coupling according to general acoustic relationships. The angle between the airflow and the direction of sound propagation The data was obtained by calculating the airflow direction data from the DCS system and the coordinates of the acoustic sensor deployment. S22.

3. Taking the sound sensor transmitter as the starting point and the receiver as the ending point, the Eikonal equation is solved using the fast travel method to obtain the shortest path of sound wave propagation and the corresponding theoretical propagation time. ; S22.4, Calculation and residual Through automatic differentiation mechanism Reverse conduction to Generate the corrected temperature field And transmit it to the feature fusion module (23).

5. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 4, characterized in that, The speed of sound wave propagation in S22.2 The calculation supports airflow humidity adaptation, and the specific adaptation process includes the following steps: S22.21 Receive the relative humidity of the airflow inside the furnace output by the DCS system. ; S22.22, Introduce a humidity correction factor , The value of varies Linear change, The value range is [0,1]; S22.23, based on and Adjustment The computational logic enables It is adapted to the sound speed propagation characteristics under corresponding humidity conditions.

6. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 5, characterized in that, The process of performing dual-branch attention fusion by the feature fusion module (23) includes the following steps: S23.1 Receive the raw acoustic wave data and operating parameters transmitted by the data receiving module (21), and synchronously receive the corrected temperature field output by the acoustic ray propagation correction module (22). ; S23.2 Perform dual-branch feature extraction. The first branch uses the sliding window method to extract the frequency-amplitude features of the acoustic wave data. The second branch extracts the mean and fluctuation characteristics of the operating condition parameters. ; S23.3 Calculating the reliability of acoustic wave features based on the statistical characteristics of feature data. The reliability of the operating condition characteristics is negatively correlated with the standard deviation and positively correlated with the mean. It is negatively correlated with the standard deviation of the operating condition characteristics and positively correlated with the mean; S23.4, according to and Assign attention weights The fused feature vector is obtained through weighted calculation. The weights are updated in real time with the new feature data and output to the iterative optimization module (24).

7. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 6, characterized in that, The process of performing residual feedback dynamic regularization optimization by the iterative optimization module (24) includes the following steps: S24.1, The output of the feature fusion module (23) Input the temperature field inversion model to calculate the global temperature field of the furnace in the current cycle. ; S24.2, The actual temperature measured by thermocouples on the furnace wall. Based on the baseline, calculate Mean absolute error , coordinates The measured temperature value at the location; S24.3, The value of and It is negatively correlated with the number of iterations. They are negatively correlated; the larger the error, the better in the early stages of iteration. The smaller the value, the smaller the error, and the better in the later stages of iteration. The larger; S24.4 Disassembly The source of the sound is adjusted by setting the angle between the airflow and the direction of sound propagation in the sound propagation correction module (22). If the vocal timbre is imperfect and Synchronous change, i.e., adjustment Calculate the weights; adjust the feature fusion module (23) feature credibility. , Computational logic, if feature stability and Synchronization of changes, i.e., correction , The association weights between the median standard deviation and the mean are then output. and To the result output module (25).

8. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 7, characterized in that, The result output module (25) performs the result determination and output process, which includes the following steps: S25.1 Receive the current round temperature field transmitted by the iterative optimization module (24). and corresponding mean absolute error ; S25.2 When the preset stability conditions of temperature field error and global temperature gradient are met for three consecutive iterations, the iteration is determined to be converged. S25.3 If the iteration converges, output As the final three-dimensional temperature distribution data of the entire furnace, it is sent to the three-dimensional temperature field intelligent visualization unit (3); if it does not converge, Feedback is sent to the sound propagation correction module (22) as the initial temperature field for the next iteration. .

9. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 8, characterized in that, The three-dimensional temperature field intelligent visualization unit (3) includes a data adaptation module (31), a model building module (32), and a visualization interaction module (33), wherein: The data adaptation module (31) is connected to the working condition adaptive inversion unit (2) by signal, and is used to receive the three-dimensional temperature distribution data of the entire furnace, perform data format conversion, and match the spatial coordinate system and data accuracy standard of the furnace geometric model in the model construction module (32). The model building module (32) is connected to the data adaptation module (31) by signal, and is used to call the general rendering tool to build a geometric model that matches the actual furnace, and to superimpose the converted temperature data onto the corresponding spatial area of ​​the geometric model to generate a real-time temperature field image of the furnace. The visualization interaction module (33) is connected to the model building module (32) by signal, and is used to perform temperature value annotation and furnace combustion area feature display operations on the real-time temperature field image, while supporting multi-view switching function.

10. The intelligent visualization system for furnace temperature field based on acoustic sensing technology according to claim 9, characterized in that, The working condition early warning and interaction unit (4) includes a working condition comparison module (41), an early warning triggering module (42), and a human-computer interaction module (43), wherein: The working condition comparison module (41) is connected to the three-dimensional temperature field intelligent visualization unit (3) by signal, and is used to retrieve the preset safe temperature range of the furnace equipment, compare the temperature field data output by the three-dimensional temperature field intelligent visualization unit (3), and identify abnormal temperature information that exceeds the safe range. The early warning triggering module (42) is connected to the working condition comparison module (41) by signal, and is used to output an audible and visual alarm signal when abnormal temperature information is detected, and simultaneously push the coordinates of the furnace area corresponding to the abnormal temperature. The human-computer interaction module (43) is connected to the early warning trigger module (42) and the three-dimensional temperature field intelligent visualization unit (3) respectively, and is used to display real-time temperature field images, temperature historical data curves and early warning records, while supporting the adjustment of interface display parameters.

Images