Boiler heating surface over-temperature prediction method and device based on spatial correlation

Abstract

The application provides a boiler heating surface over-temperature prediction method and device based on spatial correlation, comprising: acquiring current external operating parameters of a to-be-predicted boiler and current wall temperature data of multiple wall temperature measuring points on a target heating surface of the to-be-predicted boiler; based on the current wall temperature data, a temperature field matrix representing the spatial correlation relationship between the wall temperature measuring points is constructed; based on the current external operating parameters, the current wall temperature data, and the temperature field matrix, endogenous variables and exogenous variables of a temperature prediction model are determined; the endogenous variables and the exogenous variables are input into the temperature prediction model, and wall temperature prediction values of the target heating surface of the to-be-predicted boiler in a future period output by the temperature prediction model are obtained. Through the temperature field matrix, the spatial correlation characteristics between the pipelines contained in the temperature field matrix are mined, and the wall temperature prediction values are predicted in combination with the temperature field matrix, so that synchronous prediction of all wall temperature measuring points of the boiler heating surface is realized, and the prediction accuracy is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of boiler condition monitoring technology in coal-fired power plants, and in particular to a method and device for predicting boiler heating surface overheating based on spatial correlation. Background Technology

[0002] As a core energy conversion device in thermal power generation, heating, and industrial production processes, the safety and economy of boiler operation are of paramount importance. Heating surfaces (such as water-cooled walls, superheaters, and reheaters) operate under harsh conditions of high temperature and high pressure for extended periods, making them highly susceptible to failures such as overheating, creep, and even tube rupture, directly threatening the safe and stable operation of the unit. Therefore, accurately predicting the temperature of boiler heating surfaces is of significant engineering importance for real-time monitoring of equipment status, optimizing soot blowing strategies, preventing overheating and tube rupture, and extending equipment lifespan.

[0003] According to relevant technologies, current technical solutions often assume that the temperature at a point on the heating surface is only mapped to the current operating adjustment variables of the boiler, simplifying the complex boiler system into a one-dimensional mapping from the overall operating state within the furnace to the temperature at a point on the heating surface. This approach ignores the significant impact of the spatial location information of the pipes within the furnace on the temperature distribution, resulting in limited accuracy in temperature prediction.

[0004] Therefore, finding a method to improve the accuracy of boiler heating surface temperature prediction has become a current research hotspot. Summary of the Invention

[0005] This invention provides a method and apparatus for predicting boiler heating surface overheating based on spatial correlation, thereby improving the prediction accuracy of boiler heating surface temperature.

[0006] This invention provides a method for predicting boiler heating surface overheating based on spatial correlation. The method includes: acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted; constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; determining endogenous and exogenous variables of a temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix, wherein the temperature prediction model is obtained through pre-training and is used to predict the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables; and inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model.

[0007] According to the present invention, a method for predicting boiler heating surface overheating based on spatial correlation is provided. The boiler to be predicted includes a multi-screen superheater, and each screen superheater includes multiple pipes. The current wall temperature data of the wall temperature measuring points includes the current wall temperature data of the pipe wall temperature measuring points of the multiple pipes. The step of determining the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix includes: determining the endogenous variables based on the current wall temperature data of the pipe wall temperature measuring points of the multiple pipes; and determining the exogenous variables based on the current external operating parameters and the temperature field matrix.

[0008] According to the present invention, a method for predicting boiler heating surface overheating based on spatial correlation is provided. The step of constructing a temperature field matrix characterizing the spatial correlation between wall temperature measuring points based on the current wall temperature data includes: determining the physical structure of the target heating surface; converting the target heating surface into a two-dimensional matrix with P rows and B columns based on the physical structure of the target heating surface, where P represents the number of superheater screens in the boiler to be predicted; and B represents the number of pipes in each superheater screen; mapping the current wall temperature data of the wall temperature measuring points according to the two-dimensional matrix in the order of screens first, then pipes, to obtain a temperature measuring point matrix; and embedding position codes into the temperature measuring point matrix to construct a temperature field matrix characterizing the spatial correlation between wall temperature measuring points.

[0009] According to the present invention, a method for predicting overheating of boiler heating surface based on spatial correlation is provided. The method for embedding position encoding of the temperature measurement point matrix is ​​implemented in the following way: based on sine and cosine functions, the spatial position of the temperature measurement point matrix in the temperature field matrix is ​​encoded to realize the embedding position encoding of the temperature measurement point matrix.

[0010] According to the present invention, a method for predicting overheating of boiler heating surfaces based on spatial correlation is provided. The method for embedding position encoding of the temperature measurement point matrix is ​​implemented in the following way: based on learnable position encoding, learnable position encoding with random initial values ​​is embedded for measurement points at different spatial locations in the temperature measurement point matrix to realize the embedding position encoding of the temperature measurement point matrix. The learnable position encoding learns the spatial relationship between wall temperature measurement points gradually during the training process of the temperature prediction model.

[0011] According to the present invention, a method for predicting boiler heating surface overheating based on spatial correlation is provided. The temperature prediction model includes a cross-attention layer, a self-attention layer, and a feedforward neural network layer. The step of inputting the endogenous variables and the exogenous variables into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler in a future time period, as output by the temperature prediction model, includes: inputting the exogenous variables into the cross-attention layer to obtain a vector output by the cross-attention layer; inputting the vector output by the cross-attention layer and the endogenous variables together into the self-attention layer to obtain a vector output by the self-attention layer; and mapping the vector output by the self-attention layer through the feedforward neural network layer to obtain the predicted wall temperature of the target heating surface of the boiler in a future time period.

[0012] According to the present invention, a method for predicting boiler heating surface overheating based on spatial correlation is provided. The temperature prediction model is trained in the following manner: acquiring historical external operating parameters of the boiler to be predicted, historical wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted, and wall temperature values ​​of the target heating surface of the boiler to be predicted in subsequent time periods of the historical time period; constructing a training dataset based on the historical external operating parameters, the historical wall temperature data, and the wall temperature values ​​of the target heating surface of the boiler to be predicted in subsequent time periods of the historical time period, wherein the wall temperature values ​​of the target heating surface of the boiler to be predicted in subsequent time periods of the historical time period are used as data labels for the training data; and based on the training dataset... Historical wall temperature data is used to construct a temperature field matrix training dataset representing the spatial correlation between wall temperature measurement points. Based on the historical wall temperature data in the training dataset, endogenous variable training data is constructed. Based on the historical external operating parameters and the temperature field matrix training data in the training dataset, exogenous variable training data is constructed. The endogenous and exogenous variable training data are input into a temperature prediction model to obtain the predicted wall temperature training data of the target heating surface of the boiler output by the temperature prediction model. Based on the predicted wall temperature training data and the data labels of the training data, a loss function is constructed, and the temperature prediction model is trained based on the loss function to obtain a trained temperature prediction model.

[0013] According to the present invention, a method for predicting boiler heating surface overheating based on spatial correlation, after obtaining the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model, the method further includes: obtaining a wall temperature alarm value; and initiating an alarm reminder when the predicted wall temperature value of any pipe wall exceeds the wall temperature alarm value, wherein the alarm reminder is used to remind the user that the predicted wall temperature value of the target heating surface of the boiler to be predicted exceeds the wall temperature alarm value in a future time period.

[0014] This invention also provides a boiler heating surface overheating prediction device based on spatial correlation. The device includes: an acquisition module for acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted; a construction module for constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; a determination module for determining the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix, wherein the temperature prediction model is obtained through pre-training and is used to predict the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables; and a prediction module for inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the spatial correlation-based boiler heating surface overheat prediction method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the spatial correlation-based boiler heating surface overheat prediction method as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the spatial correlation-based boiler heating surface overheat prediction method as described above.

[0018] This invention provides a method and apparatus for predicting boiler heating surface overheating based on spatial correlation. The method includes: acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler; constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; determining the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix, wherein the temperature prediction model is obtained through pre-training; and inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in the future time period. By exploiting the spatial correlation characteristics between pipes contained in the temperature field matrix and combining it with the predicted wall temperature value, the prediction accuracy of boiler heating surface temperature is improved. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the boiler heating surface overheat prediction method based on spatial correlation provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the process provided by the present invention for constructing a temperature field matrix characterizing the spatial correlation between wall temperature measuring points based on the current wall temperature data.

[0022] Figure 3 This is a schematic diagram of the physical structure of the target heating surface provided by the present invention.

[0023] Figure 4 This is a temperature distribution data diagram of the heated surface provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the training process for the temperature prediction model provided by the present invention.

[0025] Figure 6 This is a schematic diagram of the structure of the boiler heating surface overheat prediction device based on spatial correlation provided by the present invention.

[0026] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] Figure 1 This is a flowchart illustrating the boiler heating surface overheat prediction method based on spatial correlation provided by the present invention. The following will combine... Figure 1 The process of the boiler heating surface overheat prediction method based on spatial correlation provided by the present invention is described.

[0029] In an exemplary embodiment of the present invention, combined with Figure 1 As can be seen, the boiler heating surface overheat prediction method based on spatial correlation can include steps 110 to 140, and each step will be described below.

[0030] In step 110, the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted are obtained.

[0031] In one embodiment, the current external operating parameters of the boiler to be predicted, as well as the current wall temperature data of multiple wall temperature measuring points on the target heating surface, can be collected in real time from the distributed control system (DCS) of the power plant and a historical database. The collected current external operating parameters may include unit operating and control parameters and monitoring parameters symmetrically arranged on the left and right sides of the boiler furnace. The unit operating and control parameters may include unit load, main feedwater flow rate, total coal feed rate, coal feed rate of each coal mill, total air volume, and flue gas oxygen content, etc. The monitoring parameters symmetrically arranged on the left and right sides of the boiler furnace may include cold primary air flow rate, hot primary air flow rate, secondary air flow rate, flue gas temperature at the inlet of the target heating surface, steam temperature at the inlet of the target heating surface, and the signal from the steam soot blower corresponding to the target heating surface. In this embodiment, a total of 30 operating parameters are collected.

[0032] In another embodiment, the target heated surface wall temperature data can be collected synchronously, such as the real-time wall temperature values ​​of 235 thermocouple measuring points on the right half of the final stage superheater. For outliers that may appear in the sensors, such as "NaN" (Not Integer), sudden drops to zero, or physically impossible spikes, this embodiment uses linear interpolation for replacement to ensure the real-time nature and continuity of the data sequence. Specifically, for data points marked as outliers, linear interpolation is used for replacement. Let t0 be the time of the outlier, and the preceding valid data point... The next valid data point is Then the interpolation calculation formula is formula (1): (1) This method can smoothly reconstruct missing or anomalous data, ensuring the continuity of the data sequence over time and providing a high-quality data foundation for subsequent modeling.

[0033] In step 120, a temperature field matrix characterizing the spatial correlation between wall temperature measuring points is constructed based on the current wall temperature data.

[0034] In one embodiment, a temperature field matrix characterizing the spatial relationship between 235 wall temperature measuring points on the target heated surface can be constructed based on the physical spatial arrangement of these points.

[0035] Figure 3 This is a schematic diagram illustrating an application scenario of constructing a temperature field matrix representing the spatial correlation between wall temperature measuring points, as provided by this invention. Combined with... Figure 3It can be seen that the thermocouples of the final stage superheater of this unit are arranged in a regular spatial pattern. From a top-view perspective, this heating surface can be abstracted as a two-dimensional matrix of 27 (number of screens) × 14 (number of pipes per screen). In this embodiment, following the rule of "screens first, then pipes," all 235 wall temperature measuring points are numbered sequentially. According to this arrangement, at each sampling time t... i The current wall temperature data of all measuring points are filled into the corresponding positions of the 27×14 matrix, thereby generating a distribution map of the heated surface temperature measuring points (i.e., the temperature field matrix). Figure 4 The diagram shows the temperature distribution data of the heated surface. The rows of the matrix represent the spatial relationship between different screens, and the columns represent the spatial relationship between different pipes within the same screen, thus fully preserving the spatial correlation characteristics between measuring points.

[0036] In another embodiment, the model can also introduce a position encoding mechanism to identify the spatial location of each measuring point in the temperature field matrix, which helps the model learn the spatial relationships between pipes. This embodiment will be described in detail below.

[0037] In step 130, based on the current external operating parameters, current wall temperature data, and temperature field matrix, the endogenous and exogenous variables of the temperature prediction model are determined. The temperature prediction model is obtained through pre-training and is used to predict the wall temperature of the target heating surface of the boiler to be predicted in the future time period based on the endogenous and exogenous variables.

[0038] In one embodiment, the pre-trained temperature prediction model can employ an improved time-series prediction network as its infrastructure. The time step of the model input can be 120 steps, corresponding to 10 minutes of historical data (sampling interval of 5 seconds). When determining the model input variables, endogenous variables can be values ​​taken from the current wall temperature data, for example, the time-series wall temperature data of all 235 pipes. Exogenous variables can include current external operating parameters and the temperature field matrix.

[0039] In step 140, endogenous and exogenous variables are input into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period.

[0040] In another embodiment, endogenous and exogenous variables can be input into a pre-trained temperature prediction model, which outputs predicted wall temperatures of the target heated surface in future time periods. In one example, the model can output predicted wall temperatures for all 235 pipes within the next 2 minutes, with a prediction step size of 24 steps. During training, the model uses the mean squared error (MSE) of all pipe predictions as a loss function to optimize the consistency of the model's predictions for all pipes on the heated surface, ultimately obtaining optimal model parameters for real-time prediction.

[0041] In this embodiment, by constructing a temperature field matrix based on the current wall temperature data according to the physical arrangement and inputting it as an exogenous variable into the prediction model, the model can not only rely on historical temperature data and operating parameters during prediction, but also directly learn and utilize the spatial correlation characteristics between pipes. This enables synchronous parallel prediction of all 235 pipes on the entire heating surface, expanding traditional single-point time-series prediction to spatiotemporal joint prediction, significantly improving prediction efficiency. Furthermore, this embodiment preserves the spatial arrangement information between measuring points through the temperature field matrix. When predicting the future temperature of a target pipe, the model's attention mechanism can simultaneously focus on the real-time temperature status of adjacent pipes on the same screen, thereby more accurately capturing the spatial correlation caused by the similarity of the working fluid flow and heat transfer process. This allows for a more realistic reflection of the non-uniform heat transfer characteristics in the three-dimensional space within the furnace, resulting in higher prediction accuracy under flexible peak-shaving conditions.

[0042] This invention provides a method for predicting boiler heating surface overheating based on spatial correlation. The method includes: acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler; constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; determining the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix; and inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period. By leveraging the temperature field matrix, the spatial correlation characteristics between pipes implied in the temperature field matrix are explored, and the predicted wall temperature value is predicted in conjunction with the temperature field matrix, thereby improving the prediction accuracy of boiler heating surface temperature.

[0043] In another exemplary embodiment of the present invention, the following description continues using the previously described embodiments as examples. The boiler to be predicted may include a multi-screen superheater, and each superheater may include multiple pipes. The current wall temperature data of the wall temperature measuring points may include the current wall temperature data of the pipe wall temperature measuring points of the multiple pipes. The endogenous and exogenous variables of the temperature prediction model are determined based on the current external operating parameters, the current wall temperature data, and the temperature field matrix, which can be achieved in the following manner: Based on the current wall temperature data of multiple pipe wall temperature measurement points, determine the endogenous variables; Based on the current external operating parameters and the temperature field matrix, the exogenous variables are determined.

[0044] In one embodiment, the high-temperature reheater heating surface of a 1000MW supercritical unit can be used as the prediction object. This heating surface is a typical multi-screen superheater structure. The high-temperature reheater contains 22 superheater screens, each of which consists of 18 parallel pipes, with a total of 396 wall temperature measuring points. Some of the pipes are equipped with thermocouples, and the temperatures of the remaining pipes can be obtained by interpolation.

[0045] In one embodiment, the wall temperature time series data of all 396 pipes in the multi-screen superheater structure, such as the wall temperature time series data within the current time period, for example, the current 10 minutes, can be used as an endogenous variable. This data reflects the historical evolution pattern of the temperature of each pipe and is the basis for the model to make time series predictions.

[0046] In another embodiment, the exogenous variable can be composed of two parts: the first part is the time-series data of multiple (e.g., the 32 mentioned above) unit operating parameters within the current time period; the second part is the temperature field matrix data for each time point corresponding to the current time period (i.e., the time-series data after flattening the 22×18 matrix). In yet another example, the model can also introduce a location encoding mechanism to identify the spatial location (screen number, pipe number within the screen) of each measuring point in the temperature field matrix, assisting the model in learning the spatial correlation characteristics between different screens and between different pipes within the same screen.

[0047] In this embodiment, pipe wall temperature data is treated as an endogenous variable, while operating parameters and the temperature field matrix are treated as exogenous variables, forming a clear variable partitioning logic. Endogenous variables are responsible for capturing the temporal inertia of the pipe temperature itself, while exogenous variables are responsible for introducing external operating condition disturbances and spatial correlation information. This partitioning makes the model's prediction logic clearer; future temperature is the evolution result of the current temperature (endogenous variable) under the combined influence of current operating conditions and the surrounding environment (exogenous variables).

[0048] In yet another exemplary embodiment of the present invention, continuing with the previously described embodiment as an example, the temperature field matrix characterizing the spatial correlation between wall temperature measuring points, based on the current wall temperature data, can be constructed in the following manner: Based on the physical spatial arrangement of multiple wall temperature measuring points on the target heated surface, a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points is constructed.

[0049] In one embodiment, the structure of the target heated surface can be analyzed to clarify the actual spatial relationship of the wall temperature measuring points. Taking a screen-type superheater suspended in the furnace as an example, its physical structure has characteristics such as inter-screen distribution, intra-screen distribution, and measuring point arrangement. The inter-screen distribution can be that the superheater contains a total of 15 tube screens (number of screens P=15); the intra-screen distribution can be that each tube screen consists of 12 parallel U-shaped tubes (number of tubes per screen B=12); the measuring point arrangement can be that 180 wall temperature measuring points are arranged according to the principle of "one tube per screen", that is, a thermocouple is installed near the outlet of each pipe to monitor the wall temperature of the pipe.

[0050] Furthermore, based on the aforementioned physical spatial arrangement rules, a mapping relationship between wall temperature measuring points and their actual spatial locations can be established. Specifically, each wall temperature measuring point is assigned two spatial coordinate indices: a screen index and a tube index. The screen index ranges from 1 to 15, identifying the screen number to which the measuring point belongs; the tube index ranges from 1 to 12, identifying the tube row position of the measuring point within its respective screen. Through these two indices, each measuring point is assigned a unique spatial coordinate (P, B), thus transforming the abstract measuring point number into a physically meaningful spatial location identifier.

[0051] Furthermore, at each sampling time t, according to the aforementioned spatial coordinate index, the real-time wall temperature data of each measuring point is filled into the corresponding matrix position, thereby constructing a temperature field matrix. The i-th row of the matrix corresponds to the i-th superheater panel, and the j-th column corresponds to the j-th pipe within that panel. Matrix element M t The value of (i,j) is the real-time temperature value of the wall temperature measuring point located at the j-th pipe of the i-th screen at time t. For pipes without measuring points, if a measuring point is missing, the value can be interpolated from the adjacent measuring point.

[0052] In this embodiment, by analyzing the actual physical arrangement of wall temperature measuring points on the heating surface (number of screens, number of tubes per screen), the originally discrete and disordered measuring point data is reconstructed into a two-dimensional temperature field matrix. This reconstruction makes the data organization highly consistent with the actual physical structure of the boiler heating surface, providing a structured data foundation for the subsequent model learning of spatial correlation characteristics.

[0053] Figure 2 This is a schematic diagram of the process provided by the present invention for constructing a temperature field matrix characterizing the spatial correlation between wall temperature measuring points based on the current wall temperature data.

[0054] The following will combine Figure 2 The process of constructing a temperature field matrix characterizing the spatial correlation between wall temperature measuring points based on the current wall temperature data provided by the present invention will be explained.

[0055] In yet another exemplary embodiment of the present invention, combined with Figure 2As can be seen, constructing a temperature field matrix representing the spatial correlation between wall temperature measurement points based on the current wall temperature data can include steps 210 to 240, which will be described in detail below.

[0056] In step 210, the physical structure of the target heated surface is determined.

[0057] In one embodiment, the physical structure of the target heating surface, such as the final stage superheater, is analyzed. This final stage superheater is arranged in a screen-like configuration within the furnace. Its physical structure parameters include the number of screens, the number of pipes per screen, and the arrangement of measuring points. Regarding the number of screens, the heating surface comprises 27 tube screens arranged sequentially along the width of the furnace, numbered 1 to 27. Regarding the number of pipes per screen, each tube screen consists of 14 parallel U-shaped tubes, numbered 1 to 14 within each tube. Regarding the arrangement of measuring points, due to hardware limitations, not every tube is equipped with a thermocouple. The actual 235 measuring points are distributed according to the principle of "prioritizing placement on the sides and key pipes in the middle of each screen," but all measuring points have a clear screen affiliation and tube location. Therefore, the physical structure of this heating surface can be completely described by a regular grid of 27 screens × 14 tubes.

[0058] In step 220, based on the physical structure of the target heating surface, the target heating surface is converted into a two-dimensional matrix with P rows and B columns, where P represents the number of superheater screens in the boiler to be predicted; and B represents the number of pipes in each superheater screen.

[0059] In one embodiment, based on the above physical structure, the target heating surface can be abstractly transformed into a two-dimensional matrix with 27 rows and 14 columns. Where: P = 27, representing the number of rows in the matrix, corresponding to the number of screens in the heating surface (27 screens); B = 14, representing the number of columns in the matrix, corresponding to the number of tubes in each superheater screen (14 tubes). The physical meaning of this matrix is ​​that each row represents a tube screen, and each column represents a specific tube position within that screen. Through this transformation, the physical spatial structure of the heating surface is completely mapped to a mathematical two-dimensional matrix structure.

[0060] Figure 3 This is a schematic diagram illustrating an application scenario of constructing a temperature field matrix representing the spatial correlation between wall temperature measuring points, as provided by the present invention. Figure 4 This is a temperature distribution data diagram of the heated surface provided by the present invention. Combined with... Figure 3 It can be seen that, based on the regular spatial arrangement of the measuring points of the final stage superheater of the unit, it can be abstracted into a two-dimensional matrix of 27 (number of screens) × 14 (number of pipes)B. Due to the limitations of the unit's hardware facilities (temperature measuring thermocouples), it is impossible to obtain the temperature data of every single pipe. Figure 3 The black circles represent heat exchange tubes equipped with thermocouples. All wall temperature measuring points are numbered and ordered sequentially, totaling 253. (Combined with...) Figure 4It can be seen that, according to this arrangement, at each time t i Both methods can obtain a temperature distribution map of the heated surface, where the "white" pipe is obtained by interpolating the temperatures of the two "black" pipes on its left and right.

[0061] In step 230, the current wall temperature data of the wall temperature measuring points are mapped based on the two-dimensional matrix and ordered from screen to tube to obtain the temperature measuring point matrix.

[0062] In step 240, position codes are embedded in the temperature measurement point matrix to construct a temperature field matrix that characterizes the spatial correlation between wall temperature measurement points.

[0063] In one embodiment, the measurement points are sorted in a screen-first, tube-later order, i.e., screen-first sorting, firstly according to screen number (screen 1, screen 2, ..., screen 27); then tube-later sorting, within each screen, the measurement points are sorted in tube order (tube 1, tube 2, ..., tube 14). Following this rule, all 235 wall temperature measurement points are uniformly numbered, with the following numbering logic: Screen 1 Tube 1 → Screen 1 Tube 2 → ... → Screen 1 Tube 14 → Screen 2 Tube 1 → Screen 2 Tube 2 → ... → Screen 27 Tube 14.

[0064] At each sampling time t, the current wall temperature data of each wall temperature measuring point is sequentially filled into a pre-constructed 27×14 two-dimensional matrix according to the numbering order. Each wall temperature measuring point is a vector arranged in chronological order. Specifically, for the measuring point of screen i and pipe j, its temperature value is filled into the i-th row and j-th column of the matrix. For pipes without measuring points (i.e., certain locations in the matrix), linear interpolation between adjacent measuring points or leaving them empty (identified by a masking mechanism in the trailing edge processing) is used to ensure the integrity of the matrix, thus obtaining the temperature measuring point matrix. Therefore, a complete temperature measuring point matrix is ​​constructed at each time t. Furthermore, positional encoding can be embedded into the temperature measuring point matrix to construct a temperature field matrix representing the spatial correlation between wall temperature measuring points. It should be noted that, compared to the temperature measuring point matrix, the temperature field matrix can better represent the spatial correlation between wall temperature measuring points.

[0065] In this embodiment, by first determining the number of screens P and the number of tubes per screen B, and then converting them into a two-dimensional matrix of P rows and B columns, the physical structure of the boiler heating surface is completely mapped into a mathematical matrix structure. The number of screens corresponds to the number of matrix rows, and the number of tubes per screen corresponds to the number of matrix columns. This one-to-one mapping relationship ensures a high degree of consistency between the temperature field matrix and the actual physical structure of the heating surface, laying a solid foundation for subsequent spatial correlation modeling.

[0066] In another exemplary embodiment of the present invention, the temperature prediction model may include a cross-attention layer, a self-attention layer, and a feedforward neural network layer; wherein, inputting endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period can be achieved in the following manner: The exogenous variables are input into the cross-attention layer to obtain the vector output by the cross-attention layer; The vector output by the cross-attention layer and the endogenous variable are input together into the self-attention layer to obtain the vector output by the self-attention layer; The vector output by the self-attention layer is mapped through the feedforward neural network layer to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period.

[0067] It should be noted that in the cross-attention layer, variable-level attention can be used to learn the future trend of each exogenous variable on the wall temperature. In particular, the spatial relationship between pipe wall temperatures can be learned using the temperature field matrix after position encoding.

[0068] Furthermore, the output vector of the cross-attention layer and the endogenous variables can be input into the self-attention layer together, and the output vector of the self-attention layer can be directly mapped to the predicted wall temperature of the target heated surface through the feedforward neural network layer.

[0069] To further illustrate the present invention, the following description will be provided in conjunction with the embodiments described below.

[0070] In one embodiment, the endogenous variable can be the wall temperature time series data of all 235 pipes within 10 minutes (120 time steps) of the current time period, with an input shape of [120 time points, 235 wall temperature sequences]. The exogenous variable is the time series data of 30 operating parameters within 10 minutes of the current time period, which is concatenated with the corresponding temperature field matrix data, with an input shape of [120 time points, 265 (30 operating parameters + 235 wall temperature sequences)].

[0071] In another embodiment, endogenous variables can be input into the self-attention layer of the temperature prediction model. The input to the self-attention layer is the time-series wall temperature data of all 235 pipes. This layer uses a multi-head self-attention mechanism to extract temporal features from the wall temperature sequence of each pipe: for the j-th pipe (j=1~235), the self-attention mechanism calculates the attention weights between each time step within 120 time steps, capturing the temporal evolution of the pipe's temperature itself. Through self-attention calculation, the model can identify the inertial characteristics, periodic patterns, and trend information of the temperature changes in each pipe. After the self-attention layer completes its processing, it outputs the initial prediction of endogenous variables, which represents the preliminary prediction made by the model based solely on the historical temperature data of each pipe.

[0072] Furthermore, the endogenous and exogenous variables from the initial prediction are input together into the cross-attention layer of the temperature prediction model. The core function of the cross-attention layer is to realize the information interaction and fusion between endogenous and exogenous variables. Through the cross-attention mechanism, the model "injects" relevant information from exogenous variables into the preliminary prediction results. For example, the model will pay attention to the operating parameters in the exogenous variables and determine the driving effect of the current operating conditions (such as changes in coal feed rate and air volume adjustment) on the pipeline temperature. After the cross-attention layer completes the processing, it outputs the final predicted wall temperature value of the target heating surface for the future time period. This output contains the predicted wall temperature values ​​of all 235 pipelines within 24 future time steps (corresponding to the next 2 minutes).

[0073] In another embodiment, a unique two-dimensional spatial coordinate (i, j) can be assigned to each wall temperature measuring point based on the constructed temperature field matrix (27 rows × 14 columns), where i represents the screen number, ranging from 1 to 27, and j represents the pipe number within the screen, ranging from 1 to 14. For matrix positions where no measuring points are located, virtual spatial coordinates are also assigned to ensure the integrity of the matrix. Furthermore, the spatial positions of the wall temperature measuring points in the temperature field matrix can be encoded to obtain position encoding information.

[0074] In another embodiment, after generating the location encoding vector, it can be fused with the corresponding exogenous variable. The wall temperature time series data is fused with the location encoding vector before being input into the model. The fusion method can be vector addition, where the temperature scalar value at each time step is expanded into a vector of the same dimension as the location encoding and then added to the location encoding vector. Alternatively, a more efficient method can be used: after mapping the temperature value to a d-dimensional feature vector at the model embedding layer, the value is added element-wise to the location encoding vector at that location.

[0075] After the above processing, the temperature field matrix in the exogenous variables is no longer just pure numerical temperature time-series data, but a temperature feature sequence containing spatial location identifiers. The data from each temperature measurement point carries its unique spatial identity information, i.e., which screen and pipe it comes from. After location encoding enhancement, the temperature field matrix data in the exogenous variables becomes a feature tensor containing spatial location information. This tensor, together with the operating parameter data, constitutes the final exogenous variables containing location-encoded information.

[0076] Furthermore, the endogenous variables and exogenous variables containing location coding information after the initial prediction are input into the cross-attention layer in the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period output by the cross-attention layer.

[0077] Since the temperature field matrix in the exogenous variables already carries location encoding information, the cross-attention mechanism, when calculating attention weights, not only focuses on "which pipe has a high temperature" but also on "where that pipe is located spatially." Specifically, when the model predicts the future temperature of the j-th pipe in the i-th screen, it searches for spatially close measurement points in the exogenous variables (adjacent pipes in the same screen, pipes in different screens in the same column) and assigns higher attention weights to these points. Location encoding enables the model to distinguish between "adjacent pipes in the same screen" and "distant pipes in different screens," even though their temperature values ​​may be similar, their different spatial locations result in different spatial correlations with the target pipe. After the cross-attention layer completes the processing, it outputs the final predicted wall temperature of the target heated surface for the future time period.

[0078] In this embodiment, by introducing position coding, each measuring point is given a unique spatial identity, and the model can clearly know that the current input temperature value comes from the j-th tube of the i-th screen. This spatial position awareness enables the model to establish a "cognition" of the spatial structure of the heated surface, thereby achieving a leap from "numerical processing" to "spatial understanding".

[0079] In yet another exemplary embodiment of the present invention, the embedding of position encoding for the temperature measurement point matrix can be achieved in the following manner: Based on sine and cosine functions, the spatial position of the temperature measurement point matrix in the temperature field matrix is ​​encoded to achieve the embedding of position codes into the temperature measurement point matrix.

[0080] In one embodiment, a unique dmodel-dimensional position encoding vector PE(pos) can be generated for each spatial position pos of the temperature measurement point matrix in the temperature field matrix based on the sine and cosine functions, thereby obtaining position encoding information to realize the embedding of position encoding into the temperature measurement point matrix. The calculation formula is as follows: Formula (2): (2) Where i is the dimension index; after encoding, the position code is added to the feature vector of the corresponding variable, so that the model can identify the order and relative position of different exogenous variables in the sequence.

[0081] In one embodiment, the time-series encoding method in the standard Transformer can be referenced. This method arranges the temperature sequences of all pipes on the entire heated surface, treating each wall temperature measurement sequence as a "word" in the overall temperature matrix. Then, the native sine and cosine functions of the Transformer model are used to generate a fixed, periodic, high-dimensional encoding vector for each position. This encoding method can naturally express the relative positional relationships in the sequence and has a certain degree of extrapolation. Its advantages are simplicity and no need for additional learning parameters.

[0082] In yet another exemplary embodiment of the present invention, embedding position encoding into the temperature measurement point matrix can also be achieved in the following manner: Based on learnable position coding, learnable position coding with random initial values ​​is embedded for measuring points at different spatial locations in the temperature measuring point matrix, so as to realize the embedding of position coding into the temperature measuring point matrix. In this process, the learnable position coding gradually learns the spatial relationship between wall temperature measuring points during the temperature prediction model training.

[0083] In another embodiment, the encoding method for "image patches" in the standard Vision Transformer can be referenced. During the initialization phase of the temperature prediction model, this method pre-defines randomly initialized position encoding vectors for each possible location in the temperature field matrix (corresponding to learnable position encodings with random initial values ​​embedded in different spatial locations within the temperature measurement point matrix). The initial values ​​of these position encoding vectors can be randomly sampled from a standard normal or uniform distribution and used as part of the model parameters in subsequent training. These randomly initialized position encoding vectors lack any prior spatial structure information; the encoding vectors of adjacent locations may be completely unrelated in their initial state. The model needs to automatically learn, through a data-driven approach, encoding representations that reflect the actual spatial relationships during training.

[0084] Understandably, the learnable location encoding used in Vision Transformer can be referenced to obtain location encoding information. The temperature monitoring matrix of the entire heated surface is viewed as a two-dimensional image, with each wall temperature measuring point analogous to a patch in the image. The system randomly initializes a learnable, fixed-dimensional vector as its unique location identifier for each specific location in the matrix (i.e., each pipe). During model training, these location encoding vectors, along with the network's weight parameters, are optimized through gradient descent, thereby automatically learning the encoding representation that best characterizes the spatial relationships (such as proximity and directionality) between pipes. This method is flexible and expressive, capable of autonomously learning spatial structure from data.

[0085] Figure 5 This is a schematic diagram of the training process for the temperature prediction model provided by the present invention.

[0086] The following will combine Figure 5 The process of training the temperature prediction model provided by this invention will be explained.

[0087] In an exemplary embodiment of the present invention, combined with Figure 5 As can be seen, training the temperature prediction model may include steps 510 to 570, which will be described in detail below.

[0088] In step 510, the historical external operating parameters of the boiler to be predicted during the historical time period, the historical wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted, and the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period are obtained.

[0089] In step 520, a training dataset is constructed based on the historical external operating parameters, the historical wall temperature data, and the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period, wherein the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period is used as the data label of the training data.

[0090] In one embodiment, historical external operating parameters of the boiler to be predicted and historical wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted can be obtained within a historical time period. The historical external operating parameters and historical wall temperature data correspond to the current external operating parameters and current wall temperature data mentioned earlier, and will not be elaborated further in this embodiment. Additionally, it is necessary to simultaneously collect the wall temperature values ​​of all 235 measuring points within a future time period corresponding to the aforementioned historical time period (e.g., 2 minutes after the historical time point) as data labels for subsequent training. Then, a training dataset can be constructed based on the historical external operating parameters, historical wall temperature data, and the wall temperature values ​​of the target heating surface of the boiler to be predicted in a later time period of the historical time period, where the wall temperature values ​​of the target heating surface of the boiler to be predicted in a later time period of the historical time period serve as the data labels for the training data.

[0091] In another embodiment, the complete dataset can also be divided into three non-overlapping subsets in chronological order: training set D. train Validation set D val and test set D test Subsequently, to eliminate the influence of different physical dimensions on model training and to prevent variables with large numerical ranges from dominating the optimization process, the data was Z-score standardized. This standardization process strictly used the training set D. train The system calculates its own statistics and uses these statistics to uniformly transform the entire dataset (including the validation and test sets). For any variable... x (Covering all unit operating parameters and wall temperature data), its standardized value calculation formula is as follows: Formula (3): (3) Where, μ train and σ train These are the variables x in the training set D. train The mean and standard deviation of the features are used. This process ensures the numerical stability of the model training and makes the weights of each feature's contribution to the model comparable.

[0092] Furthermore, based on the constructed training dataset, the temperature prediction model is trained to obtain a trained temperature prediction model. During training, the mean squared error (MSE) of all pipe predictions can be used as the loss function to optimize the consistency of the model's predictions for all pipes on the heated surface. The Adam optimizer can be used, with an initial learning rate set to 1e-4 and a weight decay coefficient set to 1e-5. A cosine annealing strategy is used to dynamically adjust the learning rate during training.

[0093] To prevent overfitting, an early stopping strategy is introduced. After each training epoch, the loss is calculated on the validation set. If the validation set loss does not decrease for 10 consecutive epochs, training is stopped, and the model parameters with the best performance on the validation set are retained as the final trained temperature prediction model.

[0094] In step 530, based on the historical wall temperature data in the training dataset, a temperature field matrix training data representing the spatial correlation between wall temperature measuring points is constructed.

[0095] In one embodiment, a temperature field matrix training data representing the spatial correlation between wall temperature measuring points can be constructed based on historical wall temperature data in the training dataset. The method of constructing the temperature field matrix training data is the same as the method of constructing the temperature field matrix described above, and will not be repeated in this embodiment.

[0096] In step 540, endogenous variable training data is constructed based on the historical wall temperature data in the training dataset.

[0097] In step 550, exogenous variable training data is constructed based on the historical external operating parameters and temperature field matrix training data in the training dataset.

[0098] In one embodiment, endogenous variable training data can be constructed based on historical wall temperature data in the training dataset, and exogenous variable training data can be constructed based on historical external operating parameters and temperature field matrix training data in the training dataset. The method of constructing endogenous variable training data is the same as the method of constructing endogenous variables described above, and the method of constructing exogenous variable training data is the same as the method of constructing exogenous variables described above. It will not be described again in this embodiment.

[0099] In step 560, the training data of endogenous variables and the training data of exogenous variables are input into the temperature prediction model to obtain the training data of the predicted wall temperature of the target heating surface of the boiler to be predicted output by the temperature prediction model.

[0100] In step 570, a loss function is constructed based on the wall temperature prediction training data and the data labels of the training data, and the temperature prediction model is trained based on the loss function to obtain a trained temperature prediction model.

[0101] In one embodiment, endogenous variable training data and exogenous variable training data can be input into the temperature prediction model to obtain the wall temperature prediction training data of the target heating surface of the boiler to be predicted output by the temperature prediction model. Further, based on the wall temperature prediction training data and the data labels of the training data, a loss function is constructed, and the temperature prediction model is trained based on the loss function to obtain the trained temperature prediction model.

[0102] In yet another exemplary embodiment of the present invention, after obtaining the predicted wall temperature of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model, the method further includes: Obtain the wall temperature alarm value; If the predicted wall temperature of any pipe wall exceeds the wall temperature alarm value, an alarm will be triggered. The alarm will remind the user that the predicted wall temperature of the target heating surface of the boiler will exceed the wall temperature alarm value in the future.

[0103] In one embodiment, a wall temperature alarm value is acquired to determine whether an overheating has occurred. The wall temperature alarm value is determined based on the metallic properties of the pipe material, its design life, and the boiler operating procedures. Furthermore, if the predicted wall temperature exceeds the wall temperature alarm value, an alarm notification can be initiated. This alarm notification alerts the user that the predicted wall temperature of the target heating surface of the boiler will exceed the wall temperature alarm value in a future time period.

[0104] In another embodiment, a wall temperature allowable value (corresponding to a wall temperature alarm value) T is set. allowable (Determined based on pipe material properties, such as 635℃). Define a complete over-temperature event E. k The time boundary is: from the moment the wall temperature first exceeds T allowable The moment (t) start,k (Starting from) until the wall temperature drops and stabilizes below T allowable The moment (t) end,k End. To handle multiple brief overheating events caused by temperature fluctuations in the actual data, a merging time threshold t is set. threshold (The timeframe can be selected from 2 to 5 minutes based on the characteristics of the unit's interrogation data). If the interval Δt between two over-temperature events... gap <t threshold If so, they are combined into a single over-temperature event for processing and statistics.

[0105] When the model predicts that a certain pipeline will... t Wall temperature at any time At that moment, the system immediately issued an over-temperature warning for the pipeline.

[0106] To quantify the effectiveness of early warning, an early warning lead time is defined. .in, t actual This represents the actual onset time of overheating in the pipeline. t predict This is the moment when the model issues a warning for the pipeline. t alarm A positive value indicates an early warning.

[0107] Furthermore, to comprehensively evaluate the effectiveness of the model in a real power plant monitoring environment, this invention defines two statistical standards for over-temperature prediction: precise prediction and fuzzy prediction. For precise prediction, in this mode, the lead time for the warning is calculated only if the pipe from which the model issues a warning is the same pipe that actually experiences over-temperature. t alarm This mode is used to evaluate the model's ability to accurately predict overheating in specific pipes. For fuzzy predictions, considering the strong spatial correlation between pipes on the heated surface, overheating usually occurs in regional forms. In this mode, once an overheating warning is issued for any pipe, the system considers that there is a risk of overheating in that region. Therefore, when calculating the overheating prediction lead time, for an overheating event at time t0, the statistical model calculates the overheating time over a past time window Δ... T window (The prediction duration can be selected based on the model's forecast duration) This includes the over-temperature predictions for all pipelines, with the moment the first warning is issued used as the baseline. t predict To calculate t alarm This model better meets the actual needs of operators to conduct overall monitoring and early warning of "risk areas," emphasizing the systematic and timely nature of early warning, while tolerating a certain degree of location deviation.

[0108] As described above, the present invention provides a method for predicting boiler heating surface overheating based on spatial correlation, which realizes the spatiotemporal joint prediction of boiler heating surface temperature field. By using the spatial temperature field matrix as model input and introducing position coding, the model can directly learn and utilize the spatial correlation characteristics between pipes, effectively solving the fundamental defect of existing technologies that ignore the spatial position information of pipes, and providing an advanced and reliable technical means to ensure the safe and stable operation of boilers under flexible peak-shaving conditions.

[0109] The following describes the boiler heating surface overheat prediction device based on spatial correlation provided by the present invention. The boiler heating surface overheat prediction device based on spatial correlation described below can be referred to in correspondence with the boiler heating surface overheat prediction method based on spatial correlation described above.

[0110] Figure 6 This is a schematic diagram of the boiler heating surface overheat prediction device based on spatial correlation provided by the present invention. The following will be combined with... Figure 6 The structure of the boiler heating surface overheat prediction device based on spatial correlation provided by the present invention will be described.

[0111] In an exemplary embodiment of the present invention, combined with Figure 6As can be seen, the boiler heating surface overheat prediction device based on spatial correlation may include an acquisition module 610, a construction module 620, a determination module 630, and a prediction module 640. Each module will be described below.

[0112] The acquisition module 610 can be configured to acquire the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted. The construction module 620 can be configured to construct a temperature field matrix characterizing the spatial correlation between wall temperature measurement points based on the current wall temperature data; The determination module 630 can be configured to determine the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix. The temperature prediction model is obtained through pre-training and is used to predict the predicted wall temperature of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables. The prediction module 640 can be configured to input the endogenous variables and the exogenous variables into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model.

[0113] In an exemplary embodiment of the present invention, the boiler to be predicted includes a multi-screen superheater, each screen superheater including multiple pipes; the current wall temperature data of the wall temperature measuring point includes the current wall temperature data of the pipe wall temperature measuring points of the multiple pipes; The determination module 630 can determine the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix in the following manner: The endogenous variable is determined based on the current wall temperature data of multiple pipe wall temperature measurement points. Based on the current external operating parameters and the temperature field matrix, the exogenous variables are determined.

[0114] In an exemplary embodiment of the present invention, the construction module 620 can construct a temperature field matrix characterizing the spatial correlation between wall temperature measurement points based on the current wall temperature data in the following manner: Determine the physical structure of the target heated surface; Based on the physical structure of the target heating surface, the target heating surface is converted into a two-dimensional matrix with P rows and B columns, where P represents the number of superheater screens in the boiler to be predicted; and B represents the number of pipes in each superheater screen. Based on a two-dimensional matrix, the current wall temperature data of the wall temperature measuring points are mapped according to the order of screen first and tube second to obtain a temperature measuring point matrix. Position codes are embedded in the temperature measurement point matrix to construct a temperature field matrix that characterizes the spatial correlation between wall temperature measurement points.

[0115] In an exemplary embodiment of the present invention, the construction module 620 may embed position encoding into the temperature measurement point matrix in the following manner: Based on sine and cosine functions, the spatial positions of the temperature measurement point matrix in the temperature field matrix are encoded to embed position codes into the temperature measurement point matrix.

[0116] In an exemplary embodiment of the present invention, the construction module 620 may embed position encoding into the temperature measurement point matrix in the following manner: Based on learnable position coding, initial values ​​of random learnable position codes are embedded for measuring points at different spatial locations in the temperature measuring point matrix, so as to realize the embedding of position codes into the temperature measuring point matrix. The learnable position codes gradually learn the spatial relationship between wall temperature measuring points during the training process of the temperature prediction model.

[0117] In an exemplary embodiment of the present invention, the temperature prediction model includes a cross-attention layer, a self-attention layer, and a feedforward neural network layer; the prediction module 640 can input the endogenous variables and the exogenous variables into the temperature prediction model in the following manner to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period output by the temperature prediction model: The exogenous variables are input into the cross-attention layer to obtain the vector output by the cross-attention layer; The vector output by the cross-attention layer and the endogenous variable are input together into the self-attention layer to obtain the vector output by the self-attention layer; The vector output by the self-attention layer is mapped through the feedforward neural network layer to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period.

[0118] In an exemplary embodiment of the present invention, the prediction module 640 may train the temperature prediction model in the following manner: Acquire historical external operating parameters of the boiler to be predicted during a historical time period, historical wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted, and wall temperature values ​​of the target heating surface of the boiler to be predicted in the later time period of the historical time period. Based on the historical external operating parameters, the historical wall temperature data, and the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period, a training dataset is constructed, wherein the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period is used as the data label of the training data. Based on the historical wall temperature data in the training dataset, a temperature field matrix training dataset representing the spatial correlation between wall temperature measuring points is constructed. Based on the historical wall temperature data in the training dataset, construct endogenous variable training data; Based on the historical external operating parameters and the temperature field matrix training data in the training dataset, exogenous variable training data is constructed. The endogenous variable training data and the exogenous variable training data are input into the temperature prediction model to obtain the wall temperature prediction value training data of the target heating surface of the boiler to be predicted output by the temperature prediction model. Based on the wall temperature prediction training data and the data labels of the training data, a loss function is constructed, and the temperature prediction model is trained based on the loss function to obtain a trained temperature prediction model.

[0119] In an exemplary embodiment of the present invention, the prediction module 640 may further be configured to: Obtain the wall temperature alarm value; If the predicted wall temperature of any pipe wall exceeds the wall temperature alarm value, an alarm will be triggered, wherein the alarm is used to remind the user that the predicted wall temperature of the target heating surface of the boiler to be predicted will exceed the wall temperature alarm value in a future time period.

[0120] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communications bus 740, wherein the processor 710, the communications interface 720, and the memory 730 communicate with each other through the communications bus 740. The processor 710 can call logic instructions in the memory 730 to execute a spatial correlation-based method for predicting boiler heating surface overheating. This method includes: acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler; constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; determining endogenous and exogenous variables of a temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix. The temperature prediction model is obtained through pre-training and is used to predict the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables; and inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model.

[0121] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0122] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the spatial correlation-based boiler heating surface overheat prediction method provided by the above methods. The method includes: acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted; constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; determining the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix. The temperature prediction model is obtained through pre-training and is used to predict the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables; inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model.

[0123] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the spatial correlation-based boiler heating surface overheat prediction method provided by the above methods. The method includes: acquiring the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted; constructing a temperature field matrix characterizing the spatial correlation between the wall temperature measuring points based on the current wall temperature data; determining the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix. The temperature prediction model is obtained through pre-training and is used to predict the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables; inputting the endogenous and exogenous variables into the temperature prediction model to obtain the predicted wall temperature value of the target heating surface of the boiler to be predicted in a future time period output by the temperature prediction model.

[0124] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0125] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting boiler heating surface overheating based on spatial correlation, characterized in that, The method includes: Obtain the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted; Based on the current wall temperature data, a temperature field matrix representing the spatial correlation between wall temperature measurement points is constructed. Based on the current external operating parameters, the current wall temperature data, and the temperature field matrix, the endogenous and exogenous variables of the temperature prediction model are determined. The temperature prediction model is obtained through pre-training and is used to predict the wall temperature of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables. The endogenous and exogenous variables are input into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period.

2. The boiler heating surface overheat prediction method based on spatial correlation according to claim 1, characterized in that, The boiler to be predicted includes a multi-screen superheater, and each screen superheater includes multiple pipes; the current wall temperature data of the wall temperature measuring points includes the current wall temperature data of the pipe wall temperature measuring points of the multiple pipes; the determination of the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix includes: The endogenous variable is determined based on the current wall temperature data of multiple pipe wall temperature measurement points. Based on the current external operating parameters and the temperature field matrix, the exogenous variables are determined.

3. The method for predicting boiler heating surface overheating based on spatial correlation according to claim 1 or 2, characterized in that, The step of constructing a temperature field matrix characterizing the spatial correlation between wall temperature measuring points based on the current wall temperature data includes: Determine the physical structure of the target heated surface; Based on the physical structure of the target heating surface, the target heating surface is converted into a two-dimensional matrix with P rows and B columns, where P represents the number of superheater screens in the boiler to be predicted; and B represents the number of pipes in each superheater screen. Based on a two-dimensional matrix, the current wall temperature data of the wall temperature measuring points are mapped according to the order of screen first and tube second to obtain a temperature measuring point matrix. Position codes are embedded in the temperature measurement point matrix to construct a temperature field matrix that characterizes the spatial correlation between wall temperature measurement points.

4. The boiler heating surface overheat prediction method based on spatial correlation according to claim 3, characterized in that, The embedding of position encoding in the temperature measurement point matrix is ​​implemented in the following way: Based on sine and cosine functions, the spatial positions of the temperature measurement point matrix in the temperature field matrix are encoded to embed position codes into the temperature measurement point matrix.

5. The boiler heating surface overheat prediction method based on spatial correlation according to claim 3, characterized in that, The embedding of position encoding in the temperature measurement point matrix is ​​implemented in the following way: Based on learnable position coding, initial values ​​of random learnable position codes are embedded for measuring points at different spatial locations in the temperature measuring point matrix, so as to realize the embedding of position codes into the temperature measuring point matrix. The learnable position codes gradually learn the spatial relationship between wall temperature measuring points during the training process of the temperature prediction model.

6. The boiler heating surface overheat prediction method based on spatial correlation according to claim 1, characterized in that, The temperature prediction model includes a cross-attention layer, a self-attention layer, and a feedforward neural network layer; the step of inputting the endogenous variables and the exogenous variables into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler under future time periods output by the temperature prediction model includes: The exogenous variables are input into the cross-attention layer to obtain the vector output by the cross-attention layer; The vector output by the cross-attention layer and the endogenous variable are input together into the self-attention layer to obtain the vector output by the self-attention layer; The vector output by the self-attention layer is mapped through the feedforward neural network layer to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period.

7. The method for predicting boiler heating surface overheating based on spatial correlation according to claim 1, characterized in that, The temperature prediction model was trained using the following method: Acquire historical external operating parameters of the boiler to be predicted during a historical time period, historical wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted, and wall temperature values ​​of the target heating surface of the boiler to be predicted in the later time period of the historical time period. Based on the historical external operating parameters, the historical wall temperature data, and the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period, a training dataset is constructed, wherein the wall temperature value of the target heating surface of the boiler to be predicted in the later time period of the historical time period is used as the data label of the training data. Based on the historical wall temperature data in the training dataset, a temperature field matrix training dataset representing the spatial correlation between wall temperature measuring points is constructed. Based on the historical wall temperature data in the training dataset, construct endogenous variable training data; Based on the historical external operating parameters and the temperature field matrix training data in the training dataset, exogenous variable training data is constructed. The endogenous variable training data and the exogenous variable training data are input into the temperature prediction model to obtain the wall temperature prediction value training data of the target heating surface of the boiler to be predicted output by the temperature prediction model. Based on the wall temperature prediction training data and the data labels of the training data, a loss function is constructed, and the temperature prediction model is trained based on the loss function to obtain a trained temperature prediction model.

8. The method for predicting boiler heating surface overheating based on spatial correlation according to claim 1, characterized in that, After obtaining the predicted wall temperature of the target heating surface of the boiler under prediction for a future time period output by the temperature prediction model, the method further includes: Obtain the wall temperature alarm value; If the predicted wall temperature of any pipe wall exceeds the wall temperature alarm value, an alarm will be triggered, wherein the alarm is used to remind the user that the predicted wall temperature of the target heating surface of the boiler to be predicted will exceed the wall temperature alarm value in a future time period.

9. A boiler heating surface overheat prediction device based on spatial correlation, characterized in that, The device includes: The acquisition module is used to acquire the current external operating parameters of the boiler to be predicted and the current wall temperature data of multiple wall temperature measuring points on the target heating surface of the boiler to be predicted. The construction module is used to construct a temperature field matrix representing the spatial correlation between wall temperature measurement points based on the current wall temperature data; The determination module is used to determine the endogenous and exogenous variables of the temperature prediction model based on the current external operating parameters, the current wall temperature data, and the temperature field matrix. The temperature prediction model is obtained through pre-training and is used to predict the wall temperature of the target heating surface of the boiler to be predicted in a future time period based on the endogenous and exogenous variables. The prediction module is used to input the endogenous variables and the exogenous variables into the temperature prediction model to obtain the predicted wall temperature of the target heating surface of the boiler to be predicted in the future time period output by the temperature prediction model.

Images