Boiler heating surface temperature monitoring method and system based on dynamic heat flux density monitoring

Abstract

The invention belongs to the technical field of boiler heating surface high-temperature corrosion monitoring, and relates to a boiler heating surface temperature monitoring method and system based on dynamic heat flux density monitoring. The method comprises the steps of obtaining temperature field data of a plurality of monitoring areas of a boiler heating surface; constructing an improved Tikhonov regularization algorithm based on the temperature field data, and carrying out heat flow field reverse solution by adopting the improved Tikhonov regularization algorithm to obtain reconstructed heat flow field distribution and a heat conduction nuclear matrix after regularization optimization; and generating a temperature evolution trend by adopting a GRU prediction model based on an attention mechanism based on the reconstructed heat flow field distribution and the regularized and optimized heat conduction nuclear matrix. The method can accurately predict the temperature evolution trend through obtaining the comprehensive temperature field data, facilitates the real-time monitoring of the temperature of the heating surface of the boiler, and improves the safety and reliability of the operation of the boiler. Operation personnel can adjust operation parameters in time according to a prediction result, and the operation state of the boiler is optimized.

Description

Technical Field

[0001] This invention belongs to the field of high-temperature corrosion monitoring technology for boiler heating surfaces, and relates to a method and system for monitoring the temperature of boiler heating surfaces based on dynamic heat flux density monitoring. Background Technology

[0002] The outer wall temperature of a boiler's heating surface refers to the surface temperature of the outer side of the tube wall (the side in contact with high-temperature flue gas) of the boiler's internal heating surfaces (such as water-cooled walls, superheaters, reheaters, etc.). This temperature directly reflects the heat transfer efficiency of the heating surface and the heat load state borne by the material, and is a key parameter for evaluating the boiler's operational safety and efficiency. Monitoring the outer wall temperature of boiler heating surfaces is mainly used to prevent material overheating failure, prevent coking, ash accumulation and corrosion, and avoid accidents such as tube rupture caused by overheating.

[0003] In existing technologies, conventional monitoring methods mainly rely on direct temperature monitoring through thermocouples, infrared thermometers, and thermal imagers. These methods are traditional, single-point, static, and passive monitoring methods that cannot achieve prediction and have poor monitoring results. To address this technical problem, there are solutions that use intelligent monitoring systems, which deploy multiple sensors and combine them with analysis software to achieve monitoring and prediction. However, these prediction methods ignore the heat transfer process of the boiler's heating surface, resulting in insufficient prediction accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for monitoring the temperature of boiler heating surfaces based on dynamic heat flux density monitoring, so as to solve the technical problem of insufficient accuracy in predicting the temperature of boiler heating surfaces.

[0005] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention discloses a method for monitoring the temperature of a boiler heating surface based on dynamic heat flux density monitoring, comprising the following steps: Acquire temperature field data for several monitoring areas of the boiler's heating surface; Based on temperature field data, an improved Tikhonov regularization algorithm is used to solve the heat flow field inversely, and the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix are obtained. Based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, a GRU prediction model based on the attention mechanism is used to generate the temperature evolution trend.

[0006] Furthermore, the acquisition of temperature field data for several monitoring areas of the boiler heating surface is as follows: The boiler heating surface is divided into several monitoring areas, and temperature data collected by the sensor array in each monitoring area is obtained. The sensor array is arranged as follows: An a×a miniature thermocouple array is arranged in the near-wall layer of the boiler heating surface; A b×b infrared thermal imager sensor array is configured on the outer layer of the boiler heating surface.

[0007] Furthermore, the improved Tikhonov regularization algorithm constructed based on temperature field data is used to perform inverse solution of the heat flow field, obtaining the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, including: A three-dimensional mathematical model of the unsteady heat conduction inverse problem is established based on temperature field data. An improved Tikhonov regularized generalized function is constructed based on a mathematical model of the three-dimensional unsteady heat conduction inverse problem, and a regularized optimization equation is obtained. The regularized optimization equations were solved using the preprocessed conjugate gradient method to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix.

[0008] Furthermore, the mathematical model for the three-dimensional unsteady heat conduction inverse problem is as follows:

[0009] in, For heat conduction kernel matrix, Let be the heat flux density distribution vector to be determined. This is the temperature field distribution vector; The improved Tikhonov regularization generalized function is as follows:

[0010]

[0011] in, For regularization terms, For a modified operator based on the characteristics of thermal conduction gradient, This is a priori estimate of the heat flux distribution. For regularization parameters; The specific steps for solving the regularized optimization equation using the preprocessed conjugate gradient method are as follows:

[0012] in, The modification term introduced for regularization. This is the transpose of the heat conduction kernel matrix. For Laplace matrix, This is the transpose of the Laplace matrix; The formula for obtaining the reconstructed heat flow field distribution is as follows:

[0013] in, To reconstruct the heat flow field distribution, Indicates the first discrete points The heat flux density at that location, =1, 2... , The number of discrete points; The regularized optimized heat conduction kernel matrix is ​​as follows:

[0014] in, This is the heat conduction kernel matrix after regularization optimization.

[0015] Furthermore, a time decay factor is introduced into the regularization parameter to obtain a modified regularization parameter. The formula for obtaining the modified regularization parameter is as follows:

[0016] in, These are the corrected regularization parameters; This refers to the decay time; It is an exponential function; This is the time decay factor; This is a set value, set according to the sensor noise level; The formula for obtaining the correction operator based on the thermal conduction gradient feature is as follows:

[0017] in, For a modified operator based on the characteristics of thermal conduction gradient, These are the gradient weighting coefficients. To measure the gradient magnitude matrix of the temperature field, This is a function used to create a diagonal matrix; The formula for constructing the heat conduction kernel matrix is ​​as follows:

[0018] in, For heat conduction core matrix; The quality matrix has dimensions n×n; For time step; Let be the stiffness matrix, with dimensions n×n.

[0019] Furthermore, based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, a GRU prediction model based on an attention mechanism is used to generate the temperature evolution trend, including: The reconstructed heat flow field distribution data and the regularized optimized heat conduction kernel matrix are normalized and then fused with spatiotemporal features. A multidimensional time-series input matrix is ​​generated by combining historical temperature field data. A GRU prediction model based on an attention mechanism is constructed. The encoder of the GRU prediction model based on an attention mechanism consists of two layers of bidirectional GRU units, and the decoder of the GRU prediction model based on an attention mechanism includes a unidirectional GRU layer and an attention computation layer. The multidimensional temporal input matrix is ​​input into the encoder of the attention-based GRU prediction model, and the decoder of the attention-based GRU prediction model generates a context vector containing prior knowledge of heat conduction. The context vector is concatenated with the regularized heat conduction kernel matrix to generate the initial hidden state of the decoder, which is the current hidden state at the initial time step. At each time step of the decoder, the context vector is gated and fused with the current hidden state to generate the physics-guided prediction output for the current moment. The current predicted output value is fed back to the input of the attention-based GRU prediction model to perform iterative prediction of the future temperature field, generate a continuous evolution curve, and obtain the temperature evolution trend.

[0020] Furthermore, the formula for obtaining the prediction output at the current moment is as follows:

[0021] in, This is the predicted output value for time step t; This is the weight matrix of the output layer; Represents the gating weight vector; This is the gate weight matrix; This represents the hidden state of the decoder at time step t; This is element-wise multiplication; This is the optimized heat conduction kernel matrix.

[0022] Furthermore, it also includes the following steps: Real-time gradients are obtained based on temperature evolution trends; Graded early warnings are issued based on real-time gradient, gradient threshold range, predicted temperature, and temperature threshold.

[0023] Furthermore, tiered early warnings are implemented based on real-time gradients, gradient threshold ranges, predicted temperatures, and temperature thresholds, as detailed below: When the real-time gradient is less than or equal to 5℃ / min, a Level 1 warning is triggered; When the real-time gradient is greater than 5℃ / min and less than or equal to 10℃ / min, a level 2 warning is triggered; When the real-time gradient is greater than 10℃ / min and less than or equal to 20℃ / min, a level 3 warning is triggered. A Level 4 warning is triggered when the real-time gradient is greater than 20℃ / min and the predicted temperature is greater than the material's safety limit temperature.

[0024] Secondly, this invention discloses a boiler heating surface temperature monitoring system based on dynamic heat flux density monitoring, comprising: The data acquisition module is used to acquire temperature field data of several monitoring areas on the boiler heating surface; The data optimization module is used to construct an improved Tikhonov regularization algorithm based on temperature field data, and to use the improved Tikhonov regularization algorithm to solve the heat flow field in reverse, so as to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix. The prediction output module is used to generate temperature evolution trends based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, using an attention-based GRU prediction model.

[0025] Compared with the prior art, the present invention has the following beneficial effects: This invention acquires temperature field data from several monitoring areas of the boiler's heating surface, enabling more detailed and comprehensive monitoring and providing a foundation for subsequent heat flow field analysis and temperature prediction. An improved Tikhonov regularization algorithm is constructed based on the temperature field data, further enhancing the accuracy and stability of the solution. The improved Tikhonov regularization algorithm is used for inverse heat flow field solving, obtaining a reconstructed heat flow field distribution and a regularized optimized heat conduction kernel matrix, which helps to gain a deeper understanding of the heat transfer process on the boiler's heating surface. Based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, a GRU prediction model based on an attention mechanism is used to generate temperature evolution trends. The GRU prediction model based on an attention mechanism can fully utilize historical data and time series characteristics to accurately predict temperature evolution trends, providing a scientific basis for boiler operation and management. This invention, by acquiring comprehensive temperature field data, can accurately predict temperature evolution trends, facilitating real-time monitoring of boiler heating surface temperatures and improving the safety and reliability of boiler operation. In addition, operators can adjust operating parameters in a timely manner based on forecast results, optimize the boiler's operating status, extend the service life of the equipment, and reduce maintenance costs.

[0026] The system of this invention includes: a data acquisition module, a data optimization module, and a prediction output module. The data acquisition module acquires temperature field data of the boiler heating surface; the data optimization module constructs an improved Tikhonov regularization algorithm based on the temperature field data, and uses the improved Tikhonov regularization algorithm to perform inverse solving of the heat flow field, obtaining the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix; the prediction output module generates the temperature evolution trend based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, using a GRU prediction model based on an attention mechanism. The various modules work together to accurately predict the temperature evolution trend, which is beneficial for real-time monitoring of the boiler heating surface temperature. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a system block diagram of an embodiment of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1: See Figure 1 This embodiment provides a method for monitoring the temperature of a boiler heating surface based on dynamic heat flux density monitoring, including the following steps: S1 acquires temperature field data for several monitoring areas of the boiler's heating surface, enabling more detailed and comprehensive monitoring of the heating surface and collecting comprehensive temperature field data, thus providing a foundation for subsequent heat flow field analysis and temperature prediction.

[0031] In this embodiment of the invention, the acquisition of temperature field data for several monitoring areas of the boiler heating surface is specifically as follows: The boiler heating surface is divided into N monitoring areas, and temperature data collected by the sensor array in each monitoring area is obtained. The sensor array is arranged as follows: An a×a miniature thermocouple array is arranged in the near-wall layer of the boiler heating surface; A b×b infrared thermal imager sensor array is configured on the outer layer of the boiler heating surface.

[0032] S2. An improved Tikhonov regularization algorithm is constructed based on temperature field data. This improved algorithm further enhances the accuracy and stability of the solution. Using the improved Tikhonov regularization algorithm for inverse thermal flow field solving, the reconstructed thermal flow field distribution and the regularized optimized heat conduction kernel matrix are obtained, which helps to gain a deeper understanding of the heat transfer process on the boiler's heating surfaces.

[0033] In this embodiment of the invention, the step of constructing an improved Tikhonov regularization algorithm based on temperature field data, and using the improved Tikhonov regularization algorithm to perform inverse solution of the heat flow field to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix includes: A three-dimensional mathematical model of the unsteady heat conduction inverse problem is established based on temperature field data. An improved Tikhonov regularized generalized function is constructed based on a mathematical model of the three-dimensional unsteady heat conduction inverse problem, and a regularized optimization equation is obtained. The regularized optimization equations were solved using the preprocessed conjugate gradient method to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix.

[0034] In this embodiment of the invention, the mathematical model for the three-dimensional unsteady-state inverse heat conduction problem is as follows:

[0035] in, For heat conduction kernel matrix, Let be the heat flux density distribution vector to be determined. This is the temperature field distribution vector; The improved Tikhonov regularization generalized function is as follows:

[0036]

[0037] in, For regularization terms, For a modified operator based on the characteristics of thermal conduction gradient, This is a priori estimate of the heat flux distribution. For regularization parameters; The specific steps for solving the regularized optimization equation using the preprocessed conjugate gradient method are as follows:

[0038] in, The modification term introduced for regularization. This is the transpose of the heat conduction kernel matrix. For Laplace matrix, This is the transpose of the Laplace matrix; The formula for obtaining the reconstructed heat flow field distribution is as follows:

[0039] in, To reconstruct the heat flow field distribution, Indicates the first discrete points The heat flux density at that location, =1, 2... , The number of discrete points; The regularized optimized heat conduction kernel matrix is ​​as follows:

[0040] in, This is the heat conduction kernel matrix after regularization optimization.

[0041] In this embodiment of the invention, a time decay factor is introduced into the regularization parameter to obtain a modified regularization parameter. The formula for obtaining the modified regularization parameter is as follows:

[0042] in, These are the corrected regularization parameters; This refers to the decay time; It is an exponential function; This is the time decay factor; This is a set value, set according to the sensor noise level; The formula for obtaining the correction operator based on the thermal conduction gradient feature is as follows:

[0043] in, For a modified operator based on the characteristics of thermal conduction gradient, These are the gradient weighting coefficients. To measure the gradient magnitude matrix of the temperature field, This is a function used to create a diagonal matrix; The formula for constructing the heat conduction kernel matrix is ​​as follows:

[0044] in, For heat conduction core matrix; The quality matrix has dimensions n×n; For time step; Let be the stiffness matrix, with dimensions n×n.

[0045] S3, based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, employs an attention-based GRU prediction model to generate temperature evolution trends. This attention-based GRU prediction model can fully utilize historical data and time series characteristics to accurately predict temperature evolution trends, providing a scientific basis for boiler operation and management.

[0046] In this embodiment of the invention, the step of generating a temperature evolution trend based on a GRU prediction model using an attention mechanism, based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, includes: The reconstructed heat flow field distribution data and the regularized optimized heat conduction kernel matrix are normalized and then fused with spatiotemporal features. A multidimensional time-series input matrix is ​​generated by combining historical temperature field data. A GRU prediction model based on an attention mechanism is constructed. The encoder of the GRU prediction model based on an attention mechanism consists of two layers of bidirectional GRU units, and the decoder of the GRU prediction model based on an attention mechanism includes a unidirectional GRU layer and an attention computation layer. The multidimensional temporal input matrix is ​​input into the encoder of the attention-based GRU prediction model, and the decoder of the attention-based GRU prediction model generates a context vector containing prior knowledge of heat conduction. The context vector is concatenated with the regularized heat conduction kernel matrix to generate the initial hidden state of the decoder, which is the current hidden state at the initial time step. At each time step of the decoder, the context vector is gated and fused with the current hidden state to generate the physics-guided prediction output for the current moment. The current predicted output value is fed back to the input of the attention-based GRU prediction model to perform iterative prediction of the future temperature field, generate a continuous evolution curve, and obtain the temperature evolution trend.

[0047] In this embodiment of the invention, the formula for obtaining the prediction output at the current moment is as follows:

[0048] in, This is the predicted output value for time step t; This is the weight matrix of the output layer; Represents the gating weight vector; This is the gate weight matrix; This represents the hidden state of the decoder at time step t; This is element-wise multiplication; This is the optimized heat conduction kernel matrix.

[0049] In this embodiment of the invention, the following steps are also included: Real-time gradients are obtained based on temperature evolution trends; A tiered early warning system is implemented based on real-time gradient, gradient threshold range, predicted temperature, and temperature threshold, as detailed below: When the real-time gradient is less than or equal to 5℃ / min, a Level 1 warning is triggered; When the real-time gradient is greater than 5℃ / min and less than or equal to 10℃ / min, a level 2 warning is triggered; When the real-time gradient is greater than 10℃ / min and less than or equal to 20℃ / min, a level 3 warning is triggered. A Level 4 warning is triggered when the real-time gradient is greater than 20℃ / min and the predicted temperature is greater than the material's safety limit temperature.

[0050] This invention overcomes the limitations of traditional monitoring methods, transforming from single-point static passive monitoring to comprehensive dynamic monitoring and prediction. By acquiring comprehensive temperature field data, gaining a deeper understanding of the heat transfer process, and accurately predicting temperature evolution trends, it achieves all-round monitoring and management of boiler heating surface temperatures. This facilitates real-time monitoring of boiler heating surface temperatures, effectively preventing various accidents caused by temperature anomalies and improving the safety and reliability of boiler operation. Furthermore, operators can adjust operating parameters promptly based on the prediction results, optimizing boiler operating conditions, extending equipment lifespan, and reducing maintenance costs.

[0051] Based on the above method, the present invention also discloses a boiler heating surface temperature monitoring system based on dynamic heat flux density monitoring, see [link to relevant documentation]. Figure 2 ,include: The data acquisition module is used to acquire temperature field data of the boiler heating surface; The data optimization module is used to construct an improved Tikhonov regularization algorithm based on temperature field data, and to use the improved Tikhonov regularization algorithm to solve the heat flow field in reverse, so as to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix. The prediction output module is used to generate temperature evolution trends based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, using an attention-based GRU prediction model.

[0052] The various modules of the system in this invention work together to accurately predict temperature evolution trends, which is beneficial for real-time monitoring of boiler heating surface temperature.

[0053] Example 2: See Figure 1 This embodiment provides a method for monitoring the temperature of a boiler heating surface based on dynamic heat flux density monitoring. The specific steps are as follows: Step S1: Divide the boiler heating surface into N monitoring areas, deploy thermocouple and infrared thermal imager sensor arrays, and collect temperature field data; Step S2: Based on the temperature field data, the improved Tikhonov regularization algorithm is used to solve the heat flow field inversely, and the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix are output. Tikhonov regularization is a mathematical method for solving ill-conditioned and overfitting problems.

[0054] Step S3: Construct a GRU (Gated Recurrent Unit) prediction model based on the attention mechanism to generate temperature evolution trends; Step S4: Trigger a graded warning when the real-time gradient or predicted temperature exceeds the threshold.

[0055] In one implementation, step S1 is specifically implemented as follows: the boiler heating surface is divided into 8 to 12 monitoring areas, and a sensor array is arranged in each monitoring area. The sensor array is arranged as follows: near the wall layer: 16×16 miniature thermocouple array is arranged; outer layer: 8×8 infrared thermal imager sensor array is configured. In one implementation, step S2 is specifically implemented as follows: Step S2.1: Establish a three-dimensional mathematical model for the unsteady-state inverse heat conduction problem: ,in For heat conduction kernel matrix, Let be the heat flux density distribution vector to be determined. This is the temperature field distribution vector; Step S2.2: Construct the improved Tikhonov regularization generalized function: ,in, For regularization terms, For a modified operator based on the characteristics of thermal conduction gradient, This is a priori estimate of the heat flux distribution. For regularization parameters; Step S2.3: Solve the regularized optimization equation using the preprocessed conjugate gradient method:

[0056] in, The modification term introduced for regularization. This is the transpose of the heat conduction kernel matrix. For Laplace matrix, This is the transpose of the Laplace matrix; Step S2.4: Based on the solution in step S2.3, obtain the reconstructed heat flow field distribution and the optimized heat conduction kernel matrix; Reconstructed heat flow field distribution: ,in, Indicates the first discrete points The heat flux density at that location, =1, 2... , The number of discrete points; Optimized heat conduction kernel matrix: .

[0057] In one implementation, in step S2.2, the regularization parameter The time decay factor is introduced for correction, as expressed below:

[0058] in, These are the corrected regularization parameters; This refers to the decay time; It is an exponential function; This is the time decay factor; This is a set value, set according to the sensor noise level.

[0059] In one implementation, in step S2.2, the correction operator is... The calculation method is as follows:

[0060] in, These are the gradient weighting coefficients. To measure the gradient magnitude matrix of the temperature field, This is a function used to create a diagonal matrix.

[0061] In one embodiment, the heat conduction core matrix The construction specifically includes:

[0062] in, For time step; Here is the stiffness matrix, with dimensions n×n; Let be the quality matrix, with dimensions n×n.

[0063] In one implementation, step S3 is specifically implemented as follows: Step S3.1: Normalize the reconstructed heat flow field distribution data and the regularized optimized heat conduction kernel matrix, and perform spatiotemporal feature fusion to generate a multidimensional time series input matrix by combining historical temperature field data; Step S3.2: Construct a GRU prediction model based on a multi-head attention mechanism, wherein the encoder consists of two layers of bidirectional GRU units, the decoder includes a unidirectional GRU layer and an attention computation layer, the encoder receives the multi-dimensional temporal input matrix, and the decoder generates a context vector containing prior knowledge of heat conduction. Step S3.3: The context vector and the regularized optimized heat conduction kernel matrix are tensor concatenated to generate the decoder, wherein the initial hidden state is the current hidden state at the initial time step; Step S3.4: At each time step t of the decoder, the context vector... With the current hidden state Gated fusion is performed to generate physically guided prediction output:

[0064] in, The predicted output value at time step t (representing the local heat flux density or temperature field characteristics of the outer wall of the boiler heating surface). This is the weight matrix of the output layer; Represents the gating weight vector; This is the gate weight matrix; This represents the hidden state of the decoder at time step t; This is element-wise multiplication; This is the optimized heat conduction kernel matrix.

[0065] In one implementation, the method further includes step S3.5: feeding back the current predicted output value to the input of the attention-based GRU prediction model to perform iterative prediction of the temperature field for the next 6-12 hours, generating a continuous evolution curve, and obtaining the temperature evolution trend.

[0066] In one implementation, step S4 is specifically implemented as follows: Graded early warnings are issued based on real-time gradients and predicted temperatures.

[0067] Among them: Level 1 warning: Real-time gradient ≤ 5℃ / min; Level II warning: 5℃ / min < real-time gradient ≤ 10℃ / min; Level 3 warning: 10℃ / min < real-time gradient ≤ 20℃ / min; Level 4 warning: Real-time gradient > 20℃ / min or predicted temperature exceeds the material's safe temperature limit.

[0068] To achieve the above objectives, the present invention also provides an intelligent monitoring system for the outer wall temperature of a boiler heating surface based on dynamic heat flux density monitoring, comprising: Data acquisition module: Divide the boiler heating surface into N monitoring areas, deploy thermocouple and infrared thermal imager sensor arrays to collect temperature field data; Data optimization module: Based on temperature field data, an improved Tikhonov regularization algorithm is used to solve the heat flow field inversely, and outputs the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix; Prediction output module: Constructs a GRU prediction model based on the attention mechanism to generate temperature evolution trends; Early warning module: Triggers a tiered early warning when the real-time gradient or predicted temperature exceeds a threshold.

[0069] This invention divides the boiler heating surface into multiple monitoring areas, enabling more detailed and comprehensive monitoring of the heating surface and collecting comprehensive temperature field data, thus providing a foundation for subsequent heat flow field analysis and temperature prediction.

[0070] Based on the collected temperature field data, this invention uses an improved Tikhonov regularization algorithm to solve the heat flow field inversely. The Tikhonov regularization algorithm can improve the ill-posedness of the problem by introducing a regularization term; the improved algorithm can further improve the accuracy and stability of the solution. The reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix output help to gain a deeper understanding of the heat transfer process of the boiler heating surface.

[0071] The GRU prediction model based on the attention mechanism of this invention can make full use of historical data and time series characteristics to accurately predict temperature evolution trends, providing a scientific basis for boiler operation and management.

[0072] Example 3: See Figure 1 This embodiment provides a method for monitoring the temperature of a boiler heating surface based on dynamic heat flux density monitoring, including the following steps: Step S1: Divide the boiler heating surface into N monitoring areas, deploy thermocouple and infrared thermal imager sensor arrays, and collect temperature field data. The specific implementation of step S1 is as follows: Divide the boiler heating surface into 8-12 monitoring areas, and arrange a sensor array in each monitoring area. The arrangement of the sensor array is as follows: Near wall layer (0.5m): Arrange a 16×16 miniature thermocouple array, using high-temperature resistant materials (such as ceramic matrix composites), to capture the transient temperature of the outer wall of the heating surface; Outer layer (1.2m): Configure an 8×8 infrared thermal imager sensor array to monitor the spatial distribution of heat flux density.

[0073] Dividing the boiler's heating surface into multiple monitoring zones allows for more detailed and comprehensive monitoring. Thermocouples and infrared thermal imagers are common temperature measurement devices, and their advantages complement each other. Thermocouples can accurately measure the temperature at localized points, while infrared thermal imagers can acquire temperature distribution information over large areas, thus collecting comprehensive temperature field data and providing a foundation for subsequent heat flow field analysis and temperature prediction.

[0074] Step S2: Based on the temperature field data, the improved Tikhonov regularization algorithm is used to solve the heat flow field inversely. The input is temperature field data, and the output is the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix. The specific implementation steps are as follows: Step S2.1: Establish a three-dimensional mathematical model for the unsteady-state inverse heat conduction problem: , where K is the heat conduction kernel matrix (constructed by the finite element / finite difference method). Let be the heat flux density distribution vector to be determined. The temperature field distribution vector (obtained based on temperature field data, using sensors such as thermocouples and infrared thermal imagers to collect temperature data at different locations and times in three-dimensional space, arranged as a vector according to node numbers) clarifies the mathematical expression of the problem and provides a basic framework for subsequent analysis and solution. Without this model, there would be no basis for subsequent regularization processing and solution. Thermal conduction kernel matrix The construction specifically includes:

[0075] in, For time step, Let n be the stiffness matrix, and let n be the dimension. Its elements are defined as follows: Describes the thermal conductivity characteristics of the system, and the thermal conductivity of the material. And related to the gradient of the shape function, where, These are elements of the stiffness matrix. The three-dimensional spatial domain for solving the heat conduction problem. The thermal conductivity of the material. For gradient operators; Let the quality matrix be of dimension n×n, and its elements be defined as follows: Describes the heat capacity distribution of the system, in relation to the material density. Specific heat capacity (Represents the heat required to raise the temperature per unit mass of material by one unit) and is related to the product of shape functions, controlling the time accuracy and stability during the discretization process. and These are all finite element shape functions used to interpolate temperature distributions within finite element units; they are typically piecewise linear functions. It is a volume infinitesimal element; Step S2.2: Construct the improved Tikhonov regularization generalized function: ,in, For regularization terms, For a modified operator based on the characteristics of thermal conduction gradient, This is a priori estimate of the heat flux distribution. For regularization parameters; because It may be ill-posed, and direct solution may yield inaccurate or unreasonable results. Therefore, an improved Tikhonov regularized generalized function is constructed. The introduction of the regularization term is to constrain and correct the solution, making the solution process more stable and reliable. The goal is to find the heat flux density distribution vector q that minimizes this generalized function. Modification operator The calculation method is as follows: Where β is the gradient weighting coefficient, with a value range of 0.1 ≤ β ≤ 0.3. To measure the gradient magnitude matrix of the temperature field, This is a function used to create a diagonal matrix, with 1 as the base value; In a further optimized scheme, the regularization parameter The time decay factor is introduced for correction, as expressed below: ,in, The time decay factor (a constant that determines the regularization parameter) (The rate of decay over time) The setting method is based on the sensor noise level, and the initial parameters are set accordingly. (0.1 is used when the sensor noise level is ≤0.5, and 0.3 is used when the sensor noise level is >0.5). Through regularization parameter correction and kernel matrix optimization, the error of the reconstructed heat flow field distribution is effectively reduced. Step S2.3: Solve the regularized optimization equation using the preprocessed conjugate gradient method, and set the residual threshold ε = 1 × 10⁻⁵ as the convergence criterion (to measure whether the difference between the current solution and the optimal solution is small enough):

[0076] in, The modification term introduced for regularization (enhancing the condition number of the matrix). for The transpose of the matrix, The Laplace matrix (which describes the spatial second derivative properties of heat diffusion and corresponds to a discretized Laplace operator, for example, in the finite difference method, the second derivative is approximated by the central difference formula to generate a tridiagonal or sparse symmetric matrix) is used to quantify the smoothness or gradient change of heat in space. for The transpose matrix; the preprocessing conjugate gradient method is an existing method, and the process of solving the regularized optimization equation using the preprocessing conjugate gradient method is known, so it will not be described in detail.

[0077] Step S2.4: Based on the solution obtained in step S2.3, we get: Reconstructed heat flow field distribution:

[0078] in, To reconstruct the heat flow field distribution, Indicates the first discrete points The heat flux density at that location, =1, 2... , The number of discrete points; Optimized heat conduction kernel matrix: .

[0079] Based on the collected temperature field data, an improved Tikhonov regularization algorithm is used to solve the heat flow field inversely. The Tikhonov regularization algorithm can improve the ill-posedness of the problem by introducing a regularization term, while the improved algorithm may further improve the accuracy and stability of the solution. The reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix output help to gain a deeper understanding of the heat transfer process of the boiler heating surface.

[0080] Step S3: Construct a GRU prediction model based on the attention mechanism to generate temperature evolution trends. The specific implementation is as follows: Step S3.1: Normalize the reconstructed heat flow field distribution data and the regularized optimized heat conduction kernel matrix, and perform spatiotemporal feature fusion to generate a multidimensional time-series input matrix by combining historical temperature field data; wherein, the normalization process adopts Min-Max (Min-Max Normalization) or Z-Score (Standard Deviation Standardization) normalization to avoid the influence of dimensional differences on model convergence; Step S3.2: Construct a GRU prediction model based on a multi-head attention mechanism, wherein the encoder consists of two layers of bidirectional GRU units, the decoder includes a unidirectional GRU layer and an attention computation layer, the encoder receives the multi-dimensional temporal input matrix, and the decoder generates a context vector containing prior knowledge of heat conduction. Step S3.3: Concatenate the context vector with the heat conduction kernel matrix using tensors to generate the initial hidden state of the decoder; expressed as follows: ,in, For context vectors, This is a compressed characteristic representation of the heat conduction kernel matrix. This is the initial hidden state of the decoder. This is the initial hidden state generator; Step S3.4: At each time step t of the decoder, the context vector... With the current hidden state Gated fusion is performed to generate physically guided prediction output:

[0081] in, The predicted output value at time step t (representing the local heat flux density or temperature field characteristics of the outer wall of the boiler heating surface). This is the weight matrix of the output layer (its function is to map the gated and fused feature vectors to the target output dimension, completing the linear transformation of the predicted values). This represents the gated weight vector (a control coefficient generated by the Sigmoid function in the [0,1] interval, whose function is to dynamically adjust the contribution weight of the physical kernel matrix and balance the coupling strength between data-driven and physical models). The gate weight matrix is ​​used to encode the joint features of the hidden state and the context vector; the hidden state of the decoder at time step t. It carries current time-series dynamic information; This is element-wise multiplication; This is the optimized heat conduction kernel matrix.

[0082] Through the above settings, a multi-dimensional time-series input matrix is ​​generated by fusing reconstructed heat flow field distribution data, regularized optimized heat conduction kernel matrix, and historical temperature field data. The GRU prediction model uses the conduction kernel matrix to perform physical regularization calculation on the attention weights, generating a context vector containing prior knowledge of heat conduction, and generating a physically guided prediction output. This ensures the accuracy and rationality of the prediction results, conforms to the law of thermodynamic conservation, and avoids problems such as physical inaccuracies and noise sensitivity that exist in traditional prediction models.

[0083] The temperature evolution trend is generated as follows: the predicted output value at the current moment is fed back to the model input, and iterative prediction of the temperature field in the next 6-12 hours is performed to generate a continuous evolution curve and obtain the temperature evolution trend.

[0084] Through iterative feedback mechanisms (such as the sequential recursive structure of LSTM / GRU), the model can gradually capture the long-term dependence of temperature field changes, avoiding the accumulation of errors caused by ignoring historical prediction results in traditional single-step prediction.

[0085] A tiered warning is triggered when the real-time gradient or predicted temperature exceeds a threshold. The specific implementation is as follows: When the real-time gradient or predicted temperature exceeds the level 4 threshold, the corresponding warning signal is triggered. The real-time gradient is calculated by using time-series temperature data collected by thermocouples and infrared thermal imagers, and is calculated by the difference of the time step. The calculation method is: the difference between the temperature value of the monitoring point at the current moment and the temperature value of the previous time interval (e.g., 1 second), and then divided by the data sampling interval. Level 1 warning: Real-time gradient ≤ 5℃ / min; Level II warning: 5℃ / min < real-time gradient ≤ 10℃ / min; Level 3 warning: 10℃ / min < real-time gradient ≤ 20℃ / min; Level 4 warning: Real-time gradient > 20℃ / min or predicted temperature exceeds the material safety limit.

[0086] By following the steps above, the temperature evolution trend can be accurately predicted, and an early warning strategy can be generated based on the prediction, providing a scientific basis for the operation and management of the boiler.

[0087] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A method for monitoring boiler heating surface temperature based on dynamic heat flux density monitoring, characterized in that, Includes the following steps: Acquire temperature field data for several monitoring areas of the boiler's heating surface; Based on temperature field data, an improved Tikhonov regularization algorithm is used to solve the heat flow field inversely, and the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix are obtained. Based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, a GRU prediction model based on the attention mechanism is used to generate the temperature evolution trend.

2. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 1, characterized in that, The acquisition of temperature field data for several monitoring areas of the boiler heating surface is as follows: The boiler heating surface is divided into several monitoring areas, and temperature data collected by the sensor array in each monitoring area is obtained. The sensor array is arranged as follows: An a×a miniature thermocouple array is arranged in the near-wall layer of the boiler heating surface; A b×b infrared thermal imager sensor array is configured on the outer layer of the boiler heating surface.

3. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 1, characterized in that, The improved Tikhonov regularization algorithm, constructed based on temperature field data, is used to inversely solve the heat flow field, obtaining the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, including: A three-dimensional mathematical model of the unsteady heat conduction inverse problem is established based on temperature field data. An improved Tikhonov regularized generalized function is constructed based on a mathematical model of the three-dimensional unsteady heat conduction inverse problem, and a regularized optimization equation is obtained. The regularized optimization equations were solved using the preprocessed conjugate gradient method to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix.

4. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 3, characterized in that, The specific mathematical model for the three-dimensional unsteady heat conduction inverse problem is as follows: in, For heat conduction kernel matrix, Let be the heat flux density distribution vector to be determined. This is the temperature field distribution vector; The improved Tikhonov regularization generalized function is as follows: in, For regularization terms, For a modified operator based on the characteristics of thermal conduction gradient, This is a priori estimate of the heat flux distribution. For regularization parameters; The specific steps for solving the regularized optimization equation using the preprocessed conjugate gradient method are as follows: in, The modification term introduced for regularization. This is the transpose of the heat conduction kernel matrix. For Laplace matrix, This is the transpose of the Laplace matrix; The formula for obtaining the reconstructed heat flow field distribution is as follows: in, To reconstruct the heat flow field distribution, Indicates the first discrete points The heat flux density at that location, =1, 2... , The number of discrete points; The regularized optimized heat conduction kernel matrix is ​​as follows: in, This is the heat conduction kernel matrix after regularization optimization.

5. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 4, characterized in that, The regularization parameter incorporates a time decay factor to obtain a modified regularization parameter. The formula for obtaining the modified regularization parameter is as follows: in, These are the corrected regularization parameters; This refers to the decay time; It is an exponential function; This is the time decay factor; This is a set value, set according to the sensor noise level; The formula for obtaining the correction operator based on the thermal conduction gradient feature is as follows: in, For a modified operator based on the characteristics of thermal conduction gradient, These are the gradient weighting coefficients. To measure the gradient magnitude matrix of the temperature field, This is a function used to create a diagonal matrix; The formula for constructing the heat conduction kernel matrix is ​​as follows: in, For heat conduction core matrix; The quality matrix has dimensions n×n; For time step; Let be the stiffness matrix, with dimensions n×n.

6. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 1, characterized in that, The temperature evolution trend is generated using a GRU prediction model based on an attention mechanism, based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, including: The reconstructed heat flow field distribution data and the regularized optimized heat conduction kernel matrix are normalized and then fused with spatiotemporal features. A multidimensional time-series input matrix is ​​generated by combining historical temperature field data. A GRU prediction model based on an attention mechanism is constructed. The encoder of the GRU prediction model based on an attention mechanism consists of two layers of bidirectional GRU units, and the decoder of the GRU prediction model based on an attention mechanism includes a unidirectional GRU layer and an attention computation layer. The multidimensional temporal input matrix is ​​input into the encoder of the attention-based GRU prediction model, and the decoder of the attention-based GRU prediction model generates a context vector containing prior knowledge of heat conduction. The context vector is concatenated with the regularized heat conduction kernel matrix to generate the initial hidden state of the decoder, which is the current hidden state at the initial time step. At each time step of the decoder, the context vector is gated and fused with the current hidden state to generate the physics-guided prediction output for the current moment. The current predicted output value is fed back to the input of the attention-based GRU prediction model to perform iterative prediction of the future temperature field, generate a continuous evolution curve, and obtain the temperature evolution trend.

7. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 6, characterized in that, The formula for obtaining the prediction output at the current moment is as follows: in, This is the predicted output value for time step t; This is the weight matrix of the output layer; Represents the gating weight vector; This is the gate weight matrix; This represents the hidden state of the decoder at time step t; This is element-wise multiplication; This is the optimized heat conduction kernel matrix.

8. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 1, characterized in that, It also includes the following steps: Real-time gradients are obtained based on temperature evolution trends; Graded early warnings are issued based on real-time gradient, gradient threshold range, predicted temperature, and temperature threshold.

9. The boiler heating surface temperature monitoring method based on dynamic heat flux density monitoring according to claim 1, characterized in that, A tiered early warning system is implemented based on real-time gradient, gradient threshold range, predicted temperature, and temperature threshold, as detailed below: When the real-time gradient is less than or equal to 5℃ / min, a Level 1 warning is triggered; When the real-time gradient is greater than 5℃ / min and less than or equal to 10℃ / min, a level 2 warning is triggered; When the real-time gradient is greater than 10℃ / min and less than or equal to 20℃ / min, a level 3 warning is triggered. A Level 4 warning is triggered when the real-time gradient is greater than 20℃ / min and the predicted temperature is greater than the material's safety limit temperature.

10. A boiler heating surface temperature monitoring system based on dynamic heat flux density monitoring, characterized in that, include: The data acquisition module is used to acquire temperature field data of several monitoring areas on the boiler heating surface; The data optimization module is used to construct an improved Tikhonov regularization algorithm based on temperature field data, and to use the improved Tikhonov regularization algorithm to solve the heat flow field in reverse, so as to obtain the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix. The prediction output module is used to generate temperature evolution trends based on the reconstructed heat flow field distribution and the regularized optimized heat conduction kernel matrix, using an attention-based GRU prediction model.

Images