Boiler furnace temperature field measuring system based on multi-sensor fusion

Abstract

The invention relates to the technical field of boiler temperature measurement, and discloses a boiler furnace temperature field measurement system based on multi-sensor fusion. According to a temperature acquisition module of the system, a plurality of groups of infrared temperature measurement sensors and thermocouple sensors are arranged in preset space grids of a boiler hearth, and real-time temperature data of each space grid are synchronously acquired; the data fusion module performs time alignment and space registration on the acquired data to generate a space-time synchronous original temperature data set; the noise suppression module performs multi-scale decomposition on an original data set based on wavelet transform, extracts a high-frequency noise component and performs adaptive filtering, and outputs a temperature data sequence after noise reduction; the temperature field reconstruction module reconstructs a hearth three-dimensional temperature field distribution model through a radial basis function interpolation algorithm according to the denoised data sequence; and the dynamic compensation module calculates the heat conduction lag compensation amount of the temperature field distribution model according to the flowing speed of the flue gas in the hearth and the working condition parameters of the combustor, and generates a compensated dynamic temperature field model.

Description

Technical Field

[0001] This invention relates to the field of boiler temperature measurement technology, specifically to a boiler furnace temperature field measurement system based on multi-sensor fusion. Background Technology

[0002] In modern industrial production systems, power generation boilers are the core thermal equipment in thermal power plants. Their core function is to convert the chemical energy of fuel into the thermal energy of steam, which then drives the steam turbine to complete the power generation process, directly determining the continuity and economy of electricity production. For power plant boilers burning high-alkali coal, the high-alkali content of coal leads to special problems such as slagging and fouling during combustion, posing a more severe challenge to the stability and efficiency of boiler operation.

[0003] The distribution of the furnace temperature field is a key factor affecting the operating quality of power generation boilers burning high-alkali coal, and is directly related to the boiler's operational stability, energy utilization efficiency, and equipment safety.

[0004] Taking thermal power generation as an example, the uniformity and stability of the temperature field within the boiler furnace are key factors in ensuring the continuous and stable operation of the production process. Uneven temperature fields can lead to localized overheating or undercooling. Localized overheating can cause excessive thermal stress on the heating surface pipes, accelerating pipe aging and damage, increasing the risk of accidents such as pipe ruptures, thus affecting the normal operation of the generator unit and even causing shutdowns, resulting in significant losses to power production. Conversely, localized undercooling can lead to incomplete combustion, reducing boiler thermal efficiency, affecting steam output and quality, and consequently impacting the stability of the entire power generation system.

[0005] In terms of energy efficiency, accurately understanding the temperature field distribution in the furnace helps optimize the combustion process and improve energy efficiency. By monitoring and analyzing the temperature field, burner operating parameters, such as the fuel-air mixing ratio, burner nozzle angle, and air velocity, can be adjusted to ensure that the fuel burns fully and evenly in the furnace, reducing incomplete combustion losses and lowering energy consumption.

[0006] The uniformity and stability of the furnace temperature field are directly related to the equipment safety and operation and maintenance costs of power generation boilers burning high-alkali coal. If the temperature field is uneven for a long period of time, in addition to causing slagging and fouling problems, it will also lead to uneven heating of the heating surfaces, aggravate equipment fatigue wear, shorten the boiler's service life, and increase the frequency of maintenance and operation costs. Summary of the Invention

[0007] The purpose of this invention is to provide a boiler furnace temperature field measurement system based on multi-sensor fusion to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides a boiler furnace temperature field measurement system based on multi-sensor fusion, the system comprising: The temperature acquisition module is used to arrange multiple sets of infrared temperature sensors and thermocouple sensors in a preset spatial grid within the boiler furnace to synchronously acquire real-time temperature data of each spatial grid within the furnace. The data fusion module is used to perform time alignment and spatial registration of real-time temperature data collected by infrared temperature sensors and thermocouple sensors to generate a spatiotemporally synchronized raw temperature dataset. The noise suppression module is used to perform multi-scale decomposition of the original temperature dataset based on wavelet transform, extract high-frequency noise components and perform adaptive filtering, and output the noise-reduced temperature data sequence. The temperature field reconstruction module is used to reconstruct the three-dimensional temperature field distribution model of the furnace based on the denoised temperature data sequence using a radial basis function interpolation algorithm. The dynamic compensation module is used to calculate the heat conduction hysteresis compensation amount of the temperature field distribution model based on the flue gas flow velocity in the furnace and the burner operating parameters, and to generate the compensated dynamic temperature field model.

[0009] Preferably, the specific method for arranging multiple sets of infrared temperature sensors and thermocouple sensors in the temperature acquisition module is as follows: Based on the geometry of the boiler furnace, it is divided into several layers of annular monitoring areas, and at least three infrared temperature sensors are evenly deployed in each annular monitoring area along the circumference. A thermocouple sensor chain is vertically installed at the central axis position of each layer of the annular monitoring area. The thermocouple sensor chain consists of multiple thermocouple probes that are evenly distributed. The sampling frequency of the infrared temperature sensor and the thermocouple sensor is dynamically adjusted according to the boiler combustion load. When the load is higher than the threshold, the first sampling frequency is used, and when the load is lower than the threshold, the second sampling frequency is used.

[0010] Preferably, the specific process of generating the spatiotemporally synchronized original temperature dataset in the data fusion module is as follows: The temperature data collected by the infrared temperature sensor is time-stamped and mapped to the unified coordinate system of the furnace through spatial coordinate transformation. Linear interpolation is performed on the temperature data collected by the thermocouple sensor chain to generate a continuous temperature profile that matches the spatial resolution of the infrared temperature sensor. A sliding time window algorithm is used to align the data acquisition time points of the two types of sensors, and the window length is adaptively adjusted according to the boiler combustion fluctuation cycle.

[0011] Preferably, the specific steps for performing adaptive filtering in the noise suppression module include: Discrete wavelet decomposition was performed on the original temperature dataset to obtain approximation coefficients and detail coefficients at different scales; Based on the energy entropy thresholding method, noise-dominant frequency bands in detail coefficients are identified, and soft thresholding is applied to the coefficients of this frequency band. A noise template is constructed using the statistical characteristics of historical furnace temperature data, and periodic interference components are further suppressed through template matching.

[0012] Preferably, the specific method for reconstructing the three-dimensional temperature field distribution model of the furnace using the radial basis function interpolation algorithm in the temperature field reconstruction module is as follows: The denoised temperature data sequence is used as the interpolation node of the radial basis function, and the node weights are solved by the least squares method. A Dirichlet boundary condition constraint interpolation process is introduced at the boundary of the furnace wall, and the boundary temperature value is taken from the boiler design parameters. A multi-layer grid method is used to accelerate the solution of large-scale radial basis function equations, and the number of grid layers is automatically optimized according to the furnace volume.

[0013] Preferably, the dynamic compensation module calculates the heat conduction hysteresis compensation amount as follows: The velocity field distribution of flue gas inside the furnace is obtained through computational fluid dynamics simulation, and the local velocity vector of each spatial grid is extracted. By combining the burner's operating status and fuel characteristic parameters, a mapping relationship between temperature propagation delay time and spatial location is established; Phase correction is performed on the dynamic temperature field model based on the time delay to compensate for the temperature measurement lag effect caused by thermal inertia.

[0014] Preferably, the data fusion module is further configured with an online sensor calibration unit. A dynamic error correction model is established based on the reading deviation between infrared temperature sensors and thermocouple sensors within the same spatial grid under stable operating conditions. When the reading of any sensor suddenly exceeds the preset confidence interval, the cross-validation and smooth transition algorithm based on the data of neighboring sensors is activated.

[0015] Preferably, the threshold determination method of the energy entropy threshold method is as follows: Calculate the energy entropy of detail coefficients at each scale and compare it with an adaptive threshold trained based on a historical noise database; For suspected effective signal frequency bands with energy entropy below the threshold, a local signal-to-noise ratio assessment is introduced for secondary discrimination to avoid the false filtering of effective temperature fluctuations.

[0016] Preferably, the temperature field reconstruction module further includes a model accuracy verification unit. Used to compare the consistency between the reconstructed three-dimensional temperature field distribution model and the adiabatic temperature calculated from the composition of the flue gas at the furnace outlet; If the deviation exceeds the allowable range, the shape parameters of the radial basis function will be automatically adjusted, and the local sensor data will be reacquired.

[0017] Preferably, the calculation process for the heat conduction hysteresis compensation of the dynamic compensation module is periodically updated, and its update strategy is as follows: Real-time monitoring of burner load change rate; when the change rate exceeds the critical value, a new round of computational fluid dynamics simulation is triggered to update the flue gas velocity field distribution. Under steady-state conditions, the mapping relationship is updated periodically according to a preset time interval.

[0018] Compared with the prior art, the beneficial effects of the present invention are: The temperature acquisition module of this invention, by carefully arranging multiple sets of infrared temperature sensors and thermocouple sensors within a preset spatial grid in the boiler furnace, can simultaneously acquire real-time temperature data from each spatial grid within the furnace. Compared to traditional temperature measurement methods, which often have limited measurement points and can only acquire temperature information from a few points within the furnace, making it difficult to comprehensively reflect the distribution of the entire furnace temperature field, the multi-sensor arrangement of this invention acts like weaving a dense temperature monitoring network within the furnace. This significantly increases the number of temperature data acquisition points, enabling a more comprehensive and detailed capture of temperature changes at different locations within the furnace. This effectively solves the problem of traditional methods failing to comprehensively reflect the temperature field, providing a rich data foundation for subsequent analysis and research of the furnace temperature field.

[0019] The data fusion and noise suppression modules ensure data reliability through their processing. The data fusion module performs time alignment and spatial registration on real-time temperature data collected by infrared and thermocouple sensors, eliminating temporal and spatial differences between the sensors and ensuring spatiotemporal consistency of the collected data, generating a spatiotemporally synchronized raw temperature dataset. This is akin to precisely piecing together jigsaw puzzle pieces from different directions to create a complete and orderly picture. The noise suppression module performs multi-scale decomposition of the raw temperature dataset based on wavelet transform, deeply mining detailed information within the data, accurately extracting high-frequency noise components, and performing adaptive filtering. This processing method effectively removes noise interference from the data, like a meticulous cleaning, making the data purer and more accurate, providing reliable data support for subsequent temperature field reconstruction and analysis.

[0020] The temperature field reconstruction module utilizes a radial basis function interpolation algorithm to reconstruct a three-dimensional temperature field distribution model of the furnace based on the denoised temperature data sequence, which has significant practical application value. Through this model, operators can intuitively see the three-dimensional temperature distribution within the furnace, clearly understanding the temperature levels, gradients, and trends in different regions. This is akin to providing operators with a three-dimensional perspective of the furnace's internal temperature, enabling them to gain a more comprehensive and in-depth understanding of the combustion state. Compared to traditional two-dimensional temperature representation methods, the three-dimensional temperature field distribution model provides richer information, helping operators to promptly identify potential problems such as localized overheating, undercooling, or uneven combustion within the furnace, thereby allowing for appropriate adjustments and optimizations to ensure the safe and stable operation of the boiler. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the working principle of the boiler furnace temperature field measurement system based on multi-sensor fusion as described in this invention. Figure 2 A flowchart illustrating the working principle of the sensor arrangement in the temperature acquisition module; Figure 3 A flowchart illustrating the working principle of the data fusion module in generating a spatiotemporally synchronized raw temperature dataset. Detailed Implementation

[0022] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1This invention provides a boiler furnace temperature field measurement system based on multi-sensor fusion. The system includes a temperature acquisition module that arranges multiple sets of infrared temperature sensors and thermocouple sensors within a preset spatial grid in the boiler furnace. The infrared temperature sensors and thermocouple sensors synchronously acquire real-time temperature data from each spatial grid within the furnace. The data acquisition process dynamically adjusts the sampling frequency based on the boiler combustion load to ensure data integrity. The data fusion module receives the real-time temperature data acquired by the infrared temperature sensors and thermocouple sensors, performs time alignment and spatial registration processing on the data, and generates a spatiotemporally synchronized original temperature dataset. A sliding time window algorithm is used during the processing to adapt to boiler combustion fluctuations. The noise suppression module performs wavelet transform multi-scale decomposition on the original temperature dataset, extracts high-frequency noise components, and performs adaptive filtering, outputting a denoised temperature data sequence. The filtering process identifies noise frequency bands based on the energy entropy threshold method. The temperature field reconstruction module reconstructs a three-dimensional temperature field distribution model of the furnace using a radial basis function interpolation algorithm based on the denoised temperature data sequence. Boundary condition constraints are introduced during the interpolation process, and a multi-layer grid method is used to accelerate the solution. The dynamic compensation module calculates the heat conduction lag compensation amount of the temperature field distribution model based on the flue gas flow velocity in the furnace and the burner operating parameters, and generates a compensated dynamic temperature field model. The compensation amount is obtained by computational fluid dynamics simulation to obtain the flue gas velocity field distribution and establish a delay mapping relationship.

[0024] Example 1: See Figure 2 The boiler furnace geometry is divided into several annular monitoring zones, based on the characteristics of combustion flow and general thermal distribution patterns within the furnace. At least three infrared temperature sensors are uniformly deployed circumferentially in each annular monitoring zone. The installation locations of these sensors are optimized using computational fluid dynamics simulations to comprehensively capture temperature non-uniformity across the furnace cross-section. The optical lenses of the infrared temperature sensors face the center of the furnace, with precisely calculated viewing angles to avoid mutual obstruction and cover the entire annular area. The sensor housings are protected by water-cooling kits to withstand high-temperature environments. A thermocouple sensor chain is vertically installed along the central axis of each annular monitoring zone. This chain consists of multiple equally spaced thermocouple probes, the spacing of which is determined by the expected axial temperature gradient. The thermocouple sensor chain uses a protective sleeve made of high-temperature resistant alloy material. The thermocouple probes are fixed to the support structure at the top or bottom of the furnace via flanges, forming a vertical measurement line running through the multiple annular monitoring zones. The thermocouple sensor chain is installed to ensure that it coincides with the central axis of the furnace to minimize measurement errors caused by deviation. The leads of the thermocouple sensors are connected to the data acquisition box through shielded cables.

[0025] The sampling frequencies of the infrared temperature sensor and thermocouple sensor are dynamically adjusted according to the boiler combustion load, with real-time data provided directly by the boiler control system. When the load exceeds a threshold, a first sampling frequency is used, with a higher value to accommodate rapidly changing temperature fields during periods of intense combustion fluctuations. When the load falls below the threshold, a second sampling frequency is used, with a lower value to meet basic monitoring needs under stable operating conditions and reduce data storage pressure. The sampling frequency switching logic has a certain hysteresis range to prevent frequent switching of sampling modes when the load fluctuates near the threshold. The frequency adjustment command is executed by the main controller in the data acquisition module. The data acquisition processes of the infrared temperature sensor and thermocouple sensor are synchronized, with the synchronization signal generated by a unified clock source, and each data point is appended with a high-precision timestamp. The spatial grid division is based not only on geometric structure but also on the layout of the burner nozzles and flame morphology. The grid size is denser in the high-temperature flame zone and relatively sparser in the outlet region. The infrared temperature sensor uses a model with a specific infrared wavelength to reduce absorption interference from water vapor and carbon dioxide in the flue gas. Regular calibration of the infrared temperature sensor is performed on-site using a standard blackbody furnace. The thermocouple probes in the thermocouple sensor chain use K-type or S-type thermocouples. Cold junction compensation of the thermocouples is completed within the data acquisition box, and the compensation accuracy meets industrial measurement standards. Verification of the sensor layout scheme is conducted during boiler shutdown and maintenance. Three-dimensional laser scanning confirms the consistency between the actual sensor installation positions and the design coordinates. Any deviations are recorded and corrected by software during data fusion. The preset spatial grid of the boiler furnace is a three-dimensional grid, and the grid node coordinates are stored in the system database as the reference for spatial registration of all temperature data.

[0026] The infrared temperature sensor and thermocouple sensor are powered by separate power supplies, and the signal transmission lines are electromagnetically shielded. The data acquisition module is interference-resistant. The operating status of the temperature acquisition module is monitored in real time, and the module has a built-in self-diagnostic function that can detect faults such as sensor open circuits, short circuits, or abnormal signal attenuation. When a failure of an infrared temperature sensor or thermocouple sensor is detected, the system can temporarily compensate using data from nearby sensors based on redundancy design principles, and maintenance alarm information will be sent to the control room. When dividing the boiler furnace geometry into annular monitoring areas, the number of layers is proportional to the furnace height, and the height of each layer is usually matched with the height of the burner assembly. The division of the annular monitoring area ensures that the difference between the temperature core area and the edge area can be distinguished radially. The infrared temperature sensors are uniformly deployed circumferentially in each layer to form a minimized temperature field reconstruction unit. The axial temperature profile data provided by the thermocouple sensor chain and the radial temperature distribution data provided by the infrared temperature sensor complement each other, forming the basic framework for spatial temperature information acquisition. The synchronization of data acquisition is not only reflected in the timestamp, but also in the hardware triggering mechanism of the acquisition module, which ensures that all channels are sampled and held at the same time, eliminating spatial registration errors caused by time asynchrony.

[0027] The threshold setting for dynamic adjustment of the sampling frequency is referenced to the boiler's design load and optimized based on historical operating data. The threshold can be set to multiple levels to adapt to more complex load change patterns. The specific values ​​of the first and second sampling frequencies are determined jointly based on the Nyquist sampling theorem and the characteristic frequencies of the boiler's main thermodynamic processes to avoid frequency aliasing. The arrangement of infrared temperature sensors and thermocouple sensors fully considers the harsh working environment inside the boiler furnace; high temperature, dust, vibration, and other factors are all taken into account in sensor selection and installation structure design. Preliminary validity checks of sensor data are performed within the acquisition module, checking items including signal amplitude range and the rationality of the rate of change. Invalid data is marked and excluded from subsequent processing. Communication between the temperature acquisition module and the boiler's existing distributed control system uses a standard industrial protocol to achieve real-time acquisition of operating parameters such as combustion load. The module's own operating status is also uploaded to the control system. The preset spatial grid information of the boiler furnace is configurable, allowing adjustments based on different types and sizes of boilers. The grid information configuration file is stored in non-volatile memory. The arrangement of infrared temperature sensors and thermocouple sensors forms a three-dimensional sensing network. The topology of this network determines the upper limit of the spatial resolution of the subsequent temperature field reconstruction model. The entire temperature acquisition module is designed to meet the reliability and availability requirements of industrial environments, possessing dustproof, waterproof, and high-temperature resistance characteristics, enabling stable operation during long-term boiler operation. The continuity and integrity of the data stream are ensured through caching technology and data packet retransmission mechanisms, ensuring that even in the event of a brief communication interruption, the temperature data sequence will not suffer significant loss.

[0028] Example 2: See Figure 3The infrared temperature sensor data is timestamped with millisecond-level precision, and the timestamp information originates from a highly stable crystal clock source within the system. The timestamps and temperature values ​​are packaged together to form a data frame, which is transmitted via industrial Ethernet to the buffer memory of the data fusion module. Spatial coordinate transformation maps the temperature values ​​measured by the infrared temperature sensor to a unified coordinate system within the furnace. This unified coordinate system has its origin at the geometric center of the boiler furnace, with three axes pointing to the width, depth, and height of the boiler, respectively. Spatial coordinate transformation requires precise installation position parameters of the infrared temperature sensor within the furnace. These parameters are obtained through total station measurements during boiler installation and commissioning and stored in the system configuration file. Linear interpolation is performed on the temperature data collected by the thermocouple sensor chain. The purpose of linear interpolation is to generate a continuous axial temperature distribution curve between the various thermocouple probes in the thermocouple sensor chain. The measurement points of the thermocouple sensor chain are located on the central axis of the furnace, and their spatial resolution is lower than that of circumferentially arranged infrared temperature sensors. The linear interpolation algorithm inserts multiple virtual measurement points between adjacent thermocouple probes. The interpolated thermocouple data spatially matches the measurement grid of the infrared temperature sensors. Each spatial grid position corresponding to an infrared temperature sensor can obtain interpolated temperature data from the thermocouple sensor chain. The linear interpolation algorithm uses a piecewise linear function, and the interpolation nodes are the actual measurement positions of the thermocouple sensor chain. The interpolation calculation is performed in real time during each data acquisition cycle.

[0029] A sliding time window algorithm is employed to align the data acquisition time points of the infrared temperature sensor and the thermocouple sensor. This algorithm groups data samples from both types of sensors that are temporally close into the same processing window. The window length is set in relation to the fluctuation period of the boiler combustion process, which is obtained by analyzing the frequency characteristics of temperature changes in historical operating data. The window length is adaptively adjusted according to the boiler combustion fluctuation period; it is shortened when combustion is intense to capture rapid dynamic changes, and lengthened when combustion is stable to improve data smoothness. Infrared temperature sensor data and thermocouple sensor data within the sliding time window are unified to the same timestamp sequence using an interpolation algorithm, generating strictly synchronized data pairs. The data fusion module is equipped with an online sensor calibration unit, which utilizes the difference in readings between the infrared temperature sensor and the thermocouple sensor under stable operating conditions within the same spatial grid. Stable operating conditions refer to an operating state where the main operating parameters such as boiler load, fuel quantity, and air volume remain unchanged, resulting in a relatively stable furnace temperature distribution. The online sensor calibration unit compares the readings of an infrared temperature sensor and a thermocouple sensor at the same location, calculating the systematic deviation between them. This systematic deviation may originate from sensor characteristic drift or slight changes in installation position. A dynamic error correction model is established based on long-term statistical data of this systematic deviation. This model is a polynomial function, and its coefficients are obtained by fitting historical deviation data using the least squares method.

[0030] When the reading of any sensor abruptly exceeds a preset confidence interval, a cross-validation and smooth transition algorithm based on neighboring sensor data is activated. The preset confidence interval is set according to the long-term statistical characteristics of the sensors, with the upper and lower limits typically taken as plus or minus three standard deviations of the mean measured value. A sudden reading change may indicate a momentary sensor malfunction or severe interference. The cross-validation algorithm checks data from other sensors within a certain spatial range around the sensor with the abrupt change. If neighboring sensor data does not show a similar abrupt change, the abrupt data is considered an outlier, and the smooth transition algorithm replaces the outlier with a weighted average of the neighboring sensor data. If neighboring sensor data also show coordinated changes, it may reflect a dramatic shift in the actual temperature field; this data is retained but marked for close monitoring. The spatiotemporally synchronized raw temperature dataset is stored in a circular buffer of the data fusion module. The data format is a floating-point array, with each data point containing three-dimensional spatial coordinates, temperature value, timestamp, and data type identifier. The processing logic of the data fusion module is implemented by a programmable logic device, and the algorithm's computation cycle is strictly synchronized with the data acquisition cycle to ensure real-time data processing. During spatial registration, it is necessary to correct for structural deformation errors caused by the thermal expansion of the furnace. The thermal expansion of the furnace is measured by displacement sensors installed on the furnace shell and used for real-time coordinate compensation. The calibration coefficients of the sensor online calibration unit are periodically written to non-volatile memory to prevent loss in case of power failure. The calibration records are also uploaded to the boiler plant's information management system for equipment status analysis.

[0031] Data loss or communication interruption is managed by the anomaly handling mechanism of the data fusion module. Short-term data loss is compensated by retaining data from the previous cycle or linear extrapolation, while long-term interruptions trigger system alarms. Data interaction between the data fusion module and the boiler main control system adopts a publish-subscribe model, with temperature data serving as public information subscribed to by other system modules requiring the data. The execution frequency of the online sensor calibration unit is configurable, typically increasing during boiler startup and after significant load fluctuations, and decreasing during stable operation to conserve computing resources. The data fusion module's output interface provides two data streams: the raw temperature dataset and a temperature dataset after preliminary quality control, allowing selection based on different accuracy requirements. Spatial coordinate transformation of infrared temperature sensor data involves coordinate transformation matrix operations. The transformation matrix parameters are fixed after sensor installation and positioning, and coordinate transformation calculations are performed on the graphics processor for accelerated processing. Linear interpolation of the thermocouple sensor chain considers not only axial position but also, due to the thermocouple sensor chain being centrally located in the furnace radial direction, its data represents the temperature along the axis, complementing the offline data measured by the infrared temperature sensor. The sliding window algorithm moves its window step size smaller than the window length, allowing data overlap between adjacent windows and preventing data discontinuities at window boundaries. The dynamic error correction model of the sensor's online calibration unit incorporates a forgetting factor, giving higher weight to recent data than earlier data, enabling the model to track the sensor's slow time-varying characteristics.

[0032] In the cross-validation and smooth transition algorithm, the selection of neighboring sensors is based on spatial distance weighting; the closer the sensor is to the sensor experiencing a sudden change, the higher its data weight. The weighting coefficient is determined based on spatial correlation analysis of historical data. The operational status of the data fusion module is monitored by a hardware watchdog circuit to prevent program crashes or infinite loops. In case of module failure, it can switch to a backup module or send a fault signal to the main control system. The storage format of the raw temperature dataset supports time-series retrieval, facilitating historical data backtracking analysis. The data compression algorithm reduces storage space usage while maintaining accuracy. The operation of the sensor online calibration unit must comply with boiler safety operating procedures. The calibration process must not disturb the boiler combustion stability. Calibration commands are issued by authorized personnel or automatically executed by the system under deemed safe conditions. The software code of the data fusion module adopts a modular design, with coordinate transformation, interpolation calculation, time alignment, and online calibration functions as independent blocks, facilitating individual debugging and upgrade maintenance. The internal data flow of the module uses parallel pipeline processing to improve data throughput, and the real-time operating system ensures that the scheduling of each task thread meets the time limit requirements. The online sensor calibration unit generates a calibration report based on the calibration results. The report includes information such as calibration time, sensor identification, pre-calibration deviation, post-calibration deviation, and changes in calibration model parameters. The data fusion module's hardware platform utilizes an industrial-grade server with redundant power supplies and heat dissipation design, meeting the reliability requirements for long-term continuous operation in power plant environments.

[0033] Example 3: The noise suppression module receives a spatiotemporally synchronized raw temperature dataset from the data fusion module. This raw temperature dataset contains temperature values ​​with spatial coordinates and timestamps provided by infrared temperature sensors and thermocouple sensors. Discrete wavelet decomposition is performed on the raw temperature dataset. The mother wavelet function is used to decompose the temperature signal into approximation coefficients and detail coefficients at different scales. The number of decomposition levels is determined based on the relationship between the main frequency components of the signal and the sampling frequency, ensuring effective separation of noise and true temperature fluctuations in the signal. The discrete wavelet decomposition calculation is performed on a digital signal processor, employing a fast algorithm to improve processing efficiency and meet the system's real-time requirements.

[0034] The energy entropy thresholding method is used to identify noise-dominant frequency bands in detail coefficients. This method quantifies the irregularity of detail coefficients by calculating their energy distribution characteristics at each scale. The formula for calculating energy entropy is: , in: Indicates the first The energy entropy value of detail coefficients at each scale It is the first The detail coefficient at each scale It is the first The total number of detail coefficients at each scale It is the first The sum of the energy of all detail coefficients at each scale, i.e. The level of energy entropy reflects the degree of concentration of coefficient energy distribution; scales with higher energy entropy values ​​usually correspond to noise-dominated frequency bands.

[0035] The calculated energy entropy values ​​at each scale are compared with adaptive thresholds trained on a historical noise database, which stores background noise features collected from the boiler under various typical operating conditions. The adaptive thresholds are classification boundaries obtained through machine learning training on a large number of samples in the historical noise database, and are dynamically adjusted as the database is updated. For scales with energy entropy higher than the adaptive threshold, they are identified as noise-dominant frequency bands, and soft thresholding is applied to the detail coefficients of these bands. Soft thresholding compares the absolute value of the coefficients with a threshold parameter; coefficients less than the threshold are set to zero, and coefficients greater than the threshold are shrunk towards zero. For suspected effective signal frequency bands with energy entropy lower than the adaptive threshold, a local signal-to-noise ratio (SNR) assessment is introduced for secondary discrimination. The SNR assessment calculates the ratio of signal energy to noise energy within a sliding time window. The local SNR assessment can effectively distinguish weak real temperature fluctuations from residual broadband noise, avoiding the misclassification of effective temperature dynamics as noise and its filtering out. The calculated local SNR is compared with an empirical threshold value; frequency bands below the threshold value are still considered noise components that need to be suppressed.

[0036] A noise template is constructed using the statistical characteristics of historical furnace temperature data. Principal component analysis (PCA) is employed to extract common periodic interference patterns from the historical data. The noise template encompasses regular noise characteristics caused by combustion oscillations, fan vibrations, and other factors during boiler operation. Periodic interference components are further suppressed through template matching. Correlation analysis is performed between the detail coefficients of the current signal and the noise template, selectively attenuating periodic components with high matching degrees. The adaptive filtering process is iterative; the residual signal after each filtering step is re-analyzed to check for any remaining significant noise components. The number of iterations is controlled by preset convergence conditions, which can be either the energy change rate of the residual signal falling below a certain minimum or reaching the maximum allowable number of iterations. The filtering parameters are updated dynamically in each iteration based on the statistical characteristics of the current signal, achieving dynamic adaptation.

[0037] The noise suppression module's processing flow is integrated onto a dedicated real-time signal processor. The algorithm is optimized to reduce computational complexity, ensuring it can handle high-frequency sampled temperature data streams. Memory management employs a block-based processing strategy, dividing the large-scale temperature dataset into smaller blocks for sequential processing, reducing instantaneous memory resource requirements. The historical noise database has an online update mechanism; when the boiler undergoes major modifications or fuel characteristics change significantly, the database can be retrained to maintain the accuracy of the noise model. The adaptive threshold training process of the energy entropy threshold method uses a support vector machine algorithm, training features including multiple statistics such as energy entropy, variance, skewness, and kurtosis for detail coefficients at various scales. The trained model parameters are stored in the noise suppression module's non-volatile memory and loaded into memory for real-time computation upon system startup. Noise template matching calculation involves signal convolution operations; a fast Fourier transform is used to convert the time-domain signal to the frequency domain for matching, improving computational speed. The output of the noise suppression module is a denoised temperature data sequence, which maintains the same data structure and temporal-spatial alignment properties as the original temperature dataset. The noise reduction effect is internally evaluated by comparing the energy spectral density of the signals before and after filtering. The evaluation results are recorded in the system log but do not affect real-time data output. The module has a self-monitoring function, capable of detecting boundary effects or reconstruction errors that may occur during wavelet decomposition, and employing methods such as symmetric extension to mitigate the impact of these effects. The entire noise suppression process aims to preserve the true thermodynamic process information within the boiler furnace while minimizing sensor electronic noise, transmission interference, and environmental background noise, providing a high-quality data foundation for subsequent temperature field reconstruction. The denoised temperature data sequence is sent to the temperature field reconstruction module for further processing; the data transmission interface uses a high-throughput parallel bus protocol.

[0038] Example 4: The temperature field reconstruction module receives the denoised temperature data sequence from the noise suppression module. The denoised temperature data sequence contains spatial coordinate information and the corresponding temperature measurement value. The denoised temperature data sequence is used as interpolation nodes for the radial basis function. Each data point constitutes a node in the interpolation network, and the node position is determined by its three-dimensional coordinates in the unified coordinate system of the boiler furnace. The node weights are solved using the least squares method. The least squares solution process constructs an overdetermined system of equations. The solution of the system minimizes the sum of squares of the differences between the calculated and measured values ​​of the reconstructed temperature field at the nodes. A Tikhonov regularization term is introduced during the solution process to improve the ill-conditioned nature of the problem and prevent excessive numerical fluctuations in the weight coefficients. A Dirichlet boundary condition is introduced at the furnace wall boundary to constrain the interpolation process. The Dirichlet boundary condition specifies the known temperature value of the furnace wall. The boundary temperature value is taken from the boiler design parameters, including the thermal conductivity of the furnace material, the design operating temperature, and the design conditions of the cooling system. The boundary conditions are applied by adding additional constraint equations to the radial basis function interpolation equations to ensure that the reconstructed temperature field at the boundaries is consistent with the physical settings.

[0039] A multi-layer mesh method is employed to accelerate the solution of large-scale radial basis function equations. This method divides the computational grid into a series of meshes with varying densities. The number of mesh layers is automatically optimized based on the furnace volume; boilers with larger furnace volumes use more mesh layers to balance computational accuracy and efficiency. The solution process rapidly captures the macroscopic trend of the temperature field on the coarse mesh and finely corrects local features on the fine mesh. Information transfer and correction between meshes improve convergence speed. The temperature field reconstruction module includes a model accuracy verification unit, which compares the reconstructed 3D temperature field distribution model with the adiabatic temperature calculated based on the composition of the flue gas at the furnace outlet. The composition of the flue gas at the furnace outlet is measured in real time by a gas analyzer installed on the flue, and the adiabatic temperature is calculated based on the heat balance of the combustion reaction. The comparison process calculates the deviation between the average temperature of the entire 3D temperature field distribution model in the furnace outlet region and the calculated adiabatic temperature. If the deviation exceeds the allowable range, the model accuracy verification unit automatically adjusts the shape parameters of the radial basis functions. These shape parameters control the width of the radial basis functions, affecting the smoothness of the interpolation results. The adjustment process uses an optimization algorithm to find shape parameter values ​​that minimize the deviation. The optimization algorithm can employ gradient descent or the simplex method. The model accuracy verification unit triggers the re-acquisition of local sensor data. The re-acquisition command is sent to the temperature acquisition module, prioritizing the infrared temperature sensors and thermocouple sensors corresponding to areas with larger deviations.

[0040] The radial basis function interpolation algorithm of the temperature field reconstruction module uses a Gaussian function as the kernel function, and the width parameter of the Gaussian function is related to the average spacing of the data points. The linear equations generated by the interpolation calculation are solved iteratively using the preprocessed conjugate gradient method. The preprocessing matrix is ​​constructed based on the eigenvalue distribution of the coefficient matrix to improve convergence. The three-dimensional temperature field distribution model is output in the form of a voxel grid. Each voxel contains the coordinates of its center point and the reconstructed temperature value. The grid resolution can be configured to meet different application requirements. The allowable range threshold of the model accuracy verification unit is dynamically set according to the boiler type, fuel type, and operating standards. The threshold is usually expressed as a percentage interval of the back-calculated adiabatic temperature. The historical records of the comparison results are stored in the system database to analyze the long-term accuracy trend of the temperature field reconstruction model. The data records include information such as timestamps, deviation values, and adjusted shape parameters. During the local sensor data re-acquisition process, the temperature acquisition module temporarily increases the sampling frequency of the sensors in the specified area to obtain denser temperature data for local model correction.

[0041] The computational tasks of the temperature field reconstruction module are deployed on high-performance computing nodes equipped with large-capacity memory to store the massive coefficient matrix of the interpolation equations. The module provides an application programming interface (API) for the boiler control system to call, allowing the system to request the temperature distribution of a specific cross-section or the temperature time series of a point in space. The parameter configurations of the radial basis function interpolation algorithm, including basis function type, regularization coefficients, and convergence tolerance, are stored in a configuration file, allowing for offline adjustment. The computation cycle of the model accuracy verification unit is correlated with the boiler operating conditions, increasing the verification frequency during periods of rapid load changes and decreasing it under stable conditions. The re-acquired sensor data, after preprocessing by the data fusion and noise suppression modules, is input again into the temperature field reconstruction module for local reconstruction. The local reconstruction results are integrated with the global temperature field through the data fusion algorithm. The temperature field reconstruction module has a self-checking function, periodically checking memory usage, computational task completion status, and communication connections with other modules. Table 1 shows the typical configuration ranges of key parameters for the radial basis function interpolation algorithm in the temperature field reconstruction module; these parameters affect the accuracy and computational efficiency of the reconstructed model. Table 1: Parameter Configuration for Radial Basis Function Interpolation Algorithm

[0042] The deviation calculation in the model accuracy verification unit employs a weighted average method, assigning different weights to temperature deviations at different locations in the furnace outlet region based on their flow characteristics. The weighting coefficients are determined based on the flue gas residence time distribution obtained from computational fluid dynamics simulations, with regions having longer residence times receiving higher weights. The output data format of the temperature field reconstruction module supports multiple industrial standards, such as VTK format for scientific visualization and HDF5 format for big data storage and analysis. The model accuracy verification unit is integrated with the boiler combustion optimization system; when a persistent large deviation is detected, it not only adjusts the reconstruction model parameters but also sends calibration suggestions to the optimization system. The calculation process of the temperature field reconstruction module is optimized for real-time requirements, reducing processing latency through algorithm parallelization, memory pre-allocation, and computational pipelines. The module's configuration interface allows users to adjust parameters according to specific boiler characteristics, and provides parameter sensitivity analysis to help users understand the impact of parameter changes on the reconstruction results. The entire temperature field reconstruction process aims to generate a physically reasonable and numerically accurate three-dimensional temperature field distribution, providing spatial temperature information for boiler combustion status monitoring and optimization. The reconstruction results are transmitted to the monitoring system via the industrial network and displayed to the operators in real time in various forms such as 3D cloud maps, isothermal surfaces, and temperature profiles.

[0043] Example 5: The dynamic compensation module receives a three-dimensional temperature field distribution model from the temperature field reconstruction module. This model is a static temperature field reconstructed based on sensor data at the current moment. The flue gas velocity field distribution within the furnace is obtained through computational fluid dynamics (CFD) simulation. This simulation is based on a precise three-dimensional geometric model of the boiler furnace, the current burner operating status, fuel characteristic parameters, and boundary condition settings. The CFD simulation runs on a dedicated high-performance computing cluster. The simulation software solves the Navier-Stokes equations and energy equations to simulate the flow and heat transfer processes of the flue gas inside the furnace. The simulation generates physical field data such as flue gas velocity vectors, pressure, and temperature for each spatial grid node within the furnace. Local velocity vectors for each spatial grid are extracted from the CFD simulation results. These local velocity vectors are three-dimensional vectors containing both magnitude and direction information. The spatial grid division is consistent with the grid system used by the temperature field reconstruction module to ensure direct data correspondence. The local velocity vector data is stored in the memory array of the dynamic compensation module, with array indices corresponding one-to-one with grid coordinates. A mapping relationship between temperature propagation delay time and spatial location is established by combining the burner operating status and fuel characteristic parameters. Burner operational status information comes from the boiler control system, including which burners are operating, their respective fuel quantities, and air volumes. Fuel characteristic parameters include the fuel's lower heating value, composition analysis, density, specific heat capacity, and other physicochemical properties. The temperature propagation delay time characterizes the time required for temperature information to travel from the furnace combustion zone to the sensor installation location; this delay time is inversely proportional to the local flue gas velocity. The mapping relationship is established through an empirical model. The model inputs are local flow velocity, fuel characteristics, and burner location; the output is the delay time from each grid point to the upstream reference point.

[0044] Phase correction is performed on the dynamic temperature field model based on time delay. The phase correction algorithm shifts the static three-dimensional temperature field distribution model along the flue gas flow direction in time. For each spatial location in the furnace, the temperature value of the corresponding upstream point at an earlier time is found based on its time delay, and the current measurement value is replaced by the earlier temperature value. The phase correction process compensates for the temperature measurement lag effect caused by thermal inertia, making the dynamic temperature field model closer to the actual instantaneous combustion state. The calculation process of the heat conduction lag compensation in the dynamic compensation module is updated periodically, and the update strategy is dynamically adjusted according to the boiler operating conditions. The burner load change rate is monitored in real time, and the burner load change rate is obtained by calculating the change in fuel quantity per unit time. When the change rate exceeds a critical value, a new round of computational fluid dynamics simulation is triggered to update the flue gas velocity field distribution. The critical value is set according to the dynamic characteristics of the boiler, usually corresponding to the condition of large load fluctuation. Under steady-state conditions, the burner load change rate is lower than the critical value, and the update strategy updates the mapping relationship periodically at a preset time interval. The preset time interval is determined based on the boiler thermal inertia time constant, for example, updating once every 30 minutes.

[0045] The initialization and execution of the computational fluid dynamics simulation are automated, with simulation input parameters automatically obtained from a real-time database. An unstructured mesh is used for the simulation, with localized refinement near the burner nozzle and in corner areas to capture complex flow details. Simulation results are validated by comparing the simulated furnace outlet parameters with measured data to ensure reliability. Fuel characteristic parameters are stored in an online database, which is automatically updated to the current industrial analysis data of the coal being fed into the furnace when different coal types are blended in the coal-fired power plant. The time-delay mapping model is periodically retrained using historical operating data to adapt to changes in boiler equipment status. The phase correction algorithm employs digital filtering technology, with the filter designed as a full-pass filter that only delays the signal phase without altering the amplitude. The output of the dynamic compensation module is a compensated dynamic temperature field model, output as a time series, with each time step corresponding to a hysteresis-compensated three-dimensional temperature field. This dynamic temperature field model is sent to the boiler combustion optimization system via a high-speed data interface for real-time adjustment of control parameters such as the air-coal ratio and burner tilt angle. The module has an internal caching mechanism to store historical temperature and velocity field data for a recent period of time, which is used for backtracking analysis and model validation.

[0046] The critical values ​​for updating the strategy are not fixed but are adaptively adjusted based on factors such as seasonal changes and equipment aging. The algorithm analyzes the relationship between historical load change data and the boiler's dynamic response characteristics to optimize the setting of critical values. When triggering a new round of computational fluid dynamics simulation, the system assesses the current computing resource usage. If resources are scarce, the simulation grid resolution is appropriately reduced or a simplified model is used to ensure real-time calculation. There is bidirectional communication between the dynamic compensation module and the boiler main control system. The main control system provides real-time operation commands to the dynamic compensation module, and the dynamic compensation module feeds back the dynamic characteristics of the temperature field to the main control system. The module has comprehensive fault diagnosis and handling logic. When the computational fluid dynamics simulation fails to converge or data is abnormal, it automatically switches to the backup simplified hysteresis model and issues a maintenance alarm.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A boiler furnace temperature field measurement system based on multi-sensor fusion, characterized in that, include: The temperature acquisition module is used to arrange multiple sets of infrared temperature sensors and thermocouple sensors in a preset spatial grid within the boiler furnace to synchronously acquire real-time temperature data of each spatial grid within the furnace. The data fusion module is used to perform time alignment and spatial registration of real-time temperature data collected by infrared temperature sensors and thermocouple sensors to generate a spatiotemporally synchronized raw temperature dataset. The noise suppression module is used to perform multi-scale decomposition of the original temperature dataset based on wavelet transform, extract high-frequency noise components and perform adaptive filtering, and output the noise-reduced temperature data sequence. The temperature field reconstruction module is used to reconstruct the three-dimensional temperature field distribution model of the furnace based on the denoised temperature data sequence using a radial basis function interpolation algorithm. The dynamic compensation module is used to calculate the heat conduction hysteresis compensation amount of the temperature field distribution model based on the flue gas flow velocity in the furnace and the burner operating parameters, and to generate the compensated dynamic temperature field model.

2. The boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 1, characterized in that, The specific method for arranging multiple sets of infrared temperature sensors and thermocouple sensors in the temperature acquisition module is as follows: Based on the geometry of the boiler furnace, it is divided into several layers of annular monitoring areas, and at least three infrared temperature sensors are evenly deployed in each annular monitoring area along the circumference. A thermocouple sensor chain is vertically installed at the central axis position of each layer of the annular monitoring area. The thermocouple sensor chain consists of multiple thermocouple probes that are evenly distributed. The sampling frequency of the infrared temperature sensor and the thermocouple sensor is dynamically adjusted according to the boiler combustion load. When the load is higher than the threshold, the first sampling frequency is used, and when the load is lower than the threshold, the second sampling frequency is used.

3. The boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 2, characterized in that, The specific process of generating the spatiotemporally synchronized original temperature dataset in the data fusion module is as follows: The temperature data collected by the infrared temperature sensor is time-stamped and mapped to the unified coordinate system of the furnace through spatial coordinate transformation. Linear interpolation is performed on the temperature data collected by the thermocouple sensor chain to generate a continuous temperature profile that matches the spatial resolution of the infrared temperature sensor. A sliding time window algorithm is used to align the data acquisition time points of the two types of sensors, and the window length is adaptively adjusted according to the boiler combustion fluctuation cycle.

4. The boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 3, characterized in that, The specific steps for performing adaptive filtering in the noise suppression module include: Discrete wavelet decomposition was performed on the original temperature dataset to obtain approximation coefficients and detail coefficients at different scales; Based on the energy entropy thresholding method, noise-dominant frequency bands in detail coefficients are identified, and soft thresholding is applied to the coefficients of this frequency band. A noise template is constructed using the statistical characteristics of historical furnace temperature data, and periodic interference components are further suppressed through template matching.

5. A boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 4, characterized in that, The specific method for reconstructing the three-dimensional temperature field distribution model of the furnace using the radial basis function interpolation algorithm in the temperature field reconstruction module is as follows: The denoised temperature data sequence is used as the interpolation node of the radial basis function, and the node weights are solved by the least squares method. A Dirichlet boundary condition constraint interpolation process is introduced at the boundary of the furnace wall, and the boundary temperature value is taken from the boiler design parameters. A multi-layer grid method is used to accelerate the solution of large-scale radial basis function equations, and the number of grid layers is automatically optimized according to the furnace volume.

6. A boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 5, characterized in that, The dynamic compensation module calculates the heat conduction hysteresis compensation amount as follows: The velocity field distribution of flue gas inside the furnace is obtained through computational fluid dynamics simulation, and the local velocity vector of each spatial grid is extracted. By combining the burner's operating status and fuel characteristic parameters, a mapping relationship between temperature propagation delay time and spatial location is established; Phase correction is performed on the dynamic temperature field model based on the time delay to compensate for the temperature measurement lag effect caused by thermal inertia.

7. A boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 1, characterized in that, The data fusion module is also equipped with an online sensor calibration unit. A dynamic error correction model is established based on the reading deviation between infrared temperature sensors and thermocouple sensors within the same spatial grid under stable operating conditions. When the reading of any sensor suddenly exceeds the preset confidence interval, the cross-validation and smooth transition algorithm based on the data of neighboring sensors is activated.

8. A boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 4, characterized in that, The threshold determination method of the energy entropy threshold method is as follows: Calculate the energy entropy of detail coefficients at each scale and compare it with an adaptive threshold trained based on a historical noise database; For suspected effective signal frequency bands with energy entropy below the threshold, a local signal-to-noise ratio assessment is introduced for secondary discrimination to avoid the false filtering of effective temperature fluctuations.

9. A boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 5, characterized in that, The temperature field reconstruction module also includes a model accuracy verification unit. Used to compare the consistency between the reconstructed three-dimensional temperature field distribution model and the adiabatic temperature calculated from the composition of the flue gas at the furnace outlet; If the deviation exceeds the allowable range, the shape parameters of the radial basis function will be automatically adjusted, and the local sensor data will be reacquired.

10. A boiler furnace temperature field measurement system based on multi-sensor fusion according to claim 6, characterized in that, The calculation process for the heat conduction hysteresis compensation of the dynamic compensation module is periodically updated, and its update strategy is as follows: Real-time monitoring of burner load change rate; when the change rate exceeds the critical value, a new round of computational fluid dynamics simulation is triggered to update the flue gas velocity field distribution. Under steady-state conditions, the mapping relationship is updated periodically according to a preset time interval.

Images