Power plant hearth visual temperature field abnormity identification and early warning method fused with AI analysis

Abstract

The invention relates to the technical field of visual temperature fields, in particular to a power plant hearth visual temperature field anomaly identification and early warning method fused with AI analysis, which comprises the following steps: synchronously acquiring hearth radiation data, performing emissivity dynamic compensation by combining real-time coal quality parameters, and generating three-dimensional temperature field characteristic data; space-time fusion processing is carried out, space correlation characteristics are extracted through a graph convolution network coupled with hearth structure parameters, and space-time fusion data including temperature gradient distribution, heat flow evolution trend and coal type adaptability are generated; constructing a space-time transformer model fused with a memory enhancement mechanism, and outputting a dynamic risk tensor including an instability probability, a structural stress hot spot and a combustion efficiency loss coefficient; intelligent decision making is carried out, when the instability probability exceeds a load self-adaptive dynamic threshold value, a visual early warning instruction set is generated, and real-time closed-loop control is executed. Real-time closed-loop control is achieved through the edge controller, and the stability, safety and combustion efficiency of boiler operation are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of visualized temperature field technology, and in particular to a method for identifying and warning abnormalities in a power plant furnace visualized temperature field by integrating AI analysis. Background Art

[0002] As large-scale coal-fired power plants develop towards high parameters and high efficiency, the complex combustion behavior and uneven heat field distribution in the furnace have become key factors affecting the operating stability and combustion efficiency of the boiler. Traditional temperature monitoring methods based on thermocouples, infrared thermal imaging and other means have significant deficiencies in spatial resolution, response speed and data fusion capabilities, and are unable to meet the needs of fine perception and control of the combustion state under dynamic load conditions; at the same time, the coupling characteristics of multiple physical fields inside the furnace increase the technical difficulty of identifying and warning abnormal temperature field behaviors, making control strategies based on empirical rules and static models prone to response lags or misjudgments, posing a greater safety risk.

[0003] In recent years, the application of artificial intelligence technology in industrial process perception and cognitive control has continued to deepen, promoting the development of visual monitoring methods based on multimodal data fusion and graph structure modeling. However, in the actual power plant furnace operation environment, how to achieve high-precision collaborative acquisition of multi-source sensor information, how to integrate coal quality, structure and thermal field dynamic data to construct a spatiotemporal expression model, and how to use deep learning networks such as Transformer to achieve dynamic risk assessment of structural instability, thermal stress concentration and combustion efficiency degradation still face challenges such as complex modeling, poor real-time response and lack of control closed loop. Summary of the Invention

[0004] The present invention provides a method for identifying and warning abnormalities in the visual temperature field of a power plant furnace by integrating AI analysis.

[0005] A method for identifying and warning abnormalities in a power plant furnace's visualized temperature field by integrating AI analysis includes the following steps:

[0006] S1: The furnace radiation data is collected synchronously by using a dual-color spectral temperature measurement array and an ultrasonic phased array. Dynamic emissivity compensation is performed in combination with real-time coal quality parameters to generate three-dimensional temperature field characteristic data with spatial coordinate markers.

[0007] S2: Perform spatiotemporal fusion processing on the three-dimensional temperature field feature data, use a graph convolutional network coupled with furnace structure parameters to extract spatial correlation features, and generate spatiotemporal fusion data including temperature gradient distribution, heat flow evolution trend and coal type adaptability;

[0008] S3: Build a spatiotemporal transformer model that integrates a memory enhancement mechanism, input the spatiotemporal fusion data into a pre-trained risk assessment network, and output a dynamic risk tensor including instability probability, structural stress hotspots, and combustion efficiency loss coefficient;

[0009] S4: Intelligent decision-making is performed based on the dynamic risk tensor. When the probability of instability exceeds the dynamic threshold of load adaptation, a visual warning instruction set including burner swing angle adjustment instructions, wall wind control strategy and three-dimensional thermal stress cloud map is generated, and real-time closed-loop control is performed through the edge controller.

[0010] Optionally, the S1 includes:

[0011] S11: A dual-color spectral temperature measurement array arranged at multiple observation points in the furnace collects radiation intensity data at different wavelengths. Simultaneously, an ultrasonic phased array is used to obtain information on changes in the medium's sound velocity at corresponding spatial locations, achieving synchronous mapping of temperature measurement data with furnace structure information.

[0012] S12: Based on the measured coal quality parameters, the improved Wiener approximation model is used to dynamically compensate for the local emissivity in the furnace and calculate the temperature field data;

[0013] S13: Binding the calculated temperature field data to the spatial coordinates to generate a three-dimensional temperature field feature data tensor marked with the spatial coordinates.

[0014] Optionally, the S2 includes:

[0015] S21, constructing a structure-aware temperature field graph model: Based on the three-dimensional temperature field characteristic data and furnace structure parameters, a spatial node graph structure is constructed, where the nodes represent the measurement points in the furnace and the edge weights reflect the heat conduction or airflow coupling relationship, forming a temperature field graph model with structural priors;

[0016] S22, extracting spatiotemporal fusion features to characterize thermal dynamic evolution: inputting node features into the graph convolutional network of coupled structural parameters to extract spatial correlation features, and extracting heat flow change trends through time series modeling methods to generate spatiotemporal fusion data that includes temperature gradient distribution, heat flow evolution trend and coal type adaptability score.

[0017] Optionally, the S21 includes:

[0018] S211: Constructing a structural topology graph with the three-dimensional spatial coordinates of the furnace as nodes;

[0019] S212: Input feature tensor.

[0020] Optionally, the S22 includes:

[0021] S221: Introduce a graph convolutional network to extract spatial features from the feature map and obtain a spatial fusion representation;

[0022] S222: Introduce the LSTM module in the time dimension to extract the heat flow evolution trend and encode the historical temperature dynamics to generate the final spatiotemporal fusion data tensor, including the temperature gradient distribution of node i, the heat flow evolution function of node i, and the adaptability score for coal quality parameters.

[0023] Optionally, generating an adjacency matrix from the structural topology graph in S211 includes:

[0024] S2111, Node definition and spatial distance calculation: The temperature measurement points in the furnace are modeled as a node set V in the graph structure, and each node corresponds to a spatial coordinate p i =(x i ,y i ,z i ), calculate the spatial distance between any two nodes;

[0025] S2112, structural connection constraint judgment: introduce a structural connection indicator function to indicate whether there is a physical heat flow or air flow channel between nodes;

[0026] S2113, calculate the structural coupling weight factor: by fusing the structural topology and the ultrasonic phased array echo results, define the coupling weight coefficient to represent the connection strength between nodes i and j;

[0027] S2114, generate adjacency matrix: define the elements of the adjacency matrix based on spatial distance, structural connection constraint judgment and structural coupling weight factor.

[0028] Optionally, the S3 includes:

[0029] S31, building a spatiotemporal Transformer model with a memory-enhanced mechanism: embedding and encoding spatiotemporal fusion features, and introducing a multi-head self-attention mechanism that integrates historical states to build a time series modeling structure with memory capabilities;

[0030] S32, Dynamic Risk Tensor Generation and Output: Based on the temporal risk features output by Transformer, the pre-trained multi-task risk assessment network is input to generate a dynamic risk tensor including instability probability, structural stress hotspots, and combustion efficiency loss coefficient.

[0031] Optionally, the S31 includes:

[0032] S311: Input embedding and position encoding: Each node sequence in the generated spatiotemporal fusion feature tensor is used as the Transformer input, linear embedding is used to obtain the representation vector and time position encoding is added;

[0033] S312, introduces memory-enhanced self-attention mechanism: a multi-head self-attention module with memory mechanism is used to model node time series.

[0034] Optionally, the S32 includes:

[0035] S321: Feature aggregation and risk assessment network input construction: Temporal aggregation of Transformer output representations to obtain node-level risk representations;

[0036] S322, generating a dynamic risk tensor: inputting the node risk representation into a pre-trained multi-task risk assessment network, and outputting a dynamic risk tensor.

[0037] Optionally, the S4 includes:

[0038] S41, intelligent determination of instability risk status: dynamic threshold determination of the instability probability in the dynamic risk tensor;

[0039] S42, generating a multi-dimensional executable warning instruction set: For node areas that meet the trigger conditions, the structural stress score and combustion efficiency loss are combined to generate a visual warning instruction set including burner swing angle adjustment instructions, wall wind control strategy, and three-dimensional thermal stress cloud map;

[0040] S43, issuing warning instructions and closed-loop execution: the warning instruction set is encoded and issued to the execution unit through the industrial edge controller to complete closed-loop control.

[0041] Beneficial effects of the present invention:

[0042] The present invention integrates a dual-color spectral temperature measurement array with an ultrasonic phased array to achieve high-precision synchronous acquisition of the radiation temperature field and structural acoustic field in the furnace, and dynamically compensates for the local emissivity in combination with real-time coal quality parameters to obtain three-dimensional temperature field characteristic data with spatial positioning capability. Compared with traditional single sensing methods, this method significantly improves the spatial resolution and accuracy of temperature measurement, providing a reliable data basis for subsequent multi-source feature extraction and risk identification.

[0043] The present invention can deeply extract multi-dimensional correlation features such as temperature gradient, heat flow evolution trend and coal type adaptability to form a dynamic risk tensor that expresses the thermal behavior of the furnace; on this basis, the system is based on the load adaptive threshold judgment mechanism of the instability probability, and intelligently generates multi-dimensional early warning instruction sets such as burner swing angle adjustment, wall wind strategy optimization and thermal stress visualization, and realizes real-time closed-loop control through the edge controller, effectively improving the stability, safety and combustion efficiency of boiler operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0046] Figure 2 This is a data collection diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0048] It should be noted that references in the specification to "one embodiment," "an embodiment," "an exemplary embodiment," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. In addition, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0049] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0050] like Figure 1-Figure 2 As shown, a method for identifying and warning abnormal temperature fields in a power plant furnace by visualizing the temperature field and integrating AI analysis includes the following steps:

[0051] S1: The furnace radiation data is collected synchronously by using a dual-color spectral temperature measurement array and an ultrasonic phased array. Dynamic emissivity compensation is performed in combination with real-time coal quality parameters to generate three-dimensional temperature field characteristic data with spatial coordinate markers.

[0052] S2: Perform spatiotemporal fusion processing on the three-dimensional temperature field feature data, use a graph convolutional network coupled with furnace structure parameters to extract spatial correlation features, and generate spatiotemporal fusion data including temperature gradient distribution, heat flow evolution trend and coal type adaptability;

[0053] S3: Build a spatiotemporal transformer model that integrates a memory enhancement mechanism, input the spatiotemporal fusion data into a pre-trained risk assessment network, and output a dynamic risk tensor including instability probability, structural stress hotspots, and combustion efficiency loss coefficient;

[0054] S4: Intelligent decision-making is performed based on the dynamic risk tensor. When the probability of instability exceeds the dynamic threshold of load adaptation, a visual warning instruction set including burner swing angle adjustment instructions, wall wind control strategy and three-dimensional thermal stress cloud map is generated, and real-time closed-loop control is performed through the edge controller.

[0055] S1 includes:

[0056] S11: Radiation intensity data at different wavelengths are collected through a dual-color spectral temperature measurement array arranged at multiple observation points in the furnace. and At the same time, the ultrasonic phased array is used to obtain the medium sound velocity change information v at the corresponding spatial position (x, y, z) s (x, y, z, t), realizing the synchronous mapping of temperature measurement data and furnace structure information;

[0057] S12: Based on the measured coal quality parameters C(t) = {A ash ,M moist ,V volatile ,F fixed}, the improved Wiener approximation model is used to dynamically compensate the local emissivity ε(x, y, z, t) in the furnace and calculate the temperature field data T(x, y, z, t), which is expressed as:

[0058]

[0059] in, is the radiation intensity at wavelengths λ1 and λ2, (x, y, z) are the coordinates of the measurement point in the three-dimensional space of the furnace, t is the sampling time, ε(x, y, z, t) is the dynamic emissivity, which is affected by the coal quality parameters, C(t) is the coal quality parameter vector, representing ash, moisture, volatile matter and fixed carbon respectively, c2 is Planck's second radiation constant, f -1 (·) is the function of inversely calculating the temperature from the two-color temperature measurement formula;

[0060] S13: Bind the calculated temperature field data T(x, y, z, t) to the spatial coordinates (x, y, z) to generate a three-dimensional temperature field feature data tensor T with spatial coordinate markers 3D(x, y, z, t), which is used for subsequent spatiotemporal fusion processing and risk assessment.

[0061] S2 includes:

[0062] S21, constructing a structure-aware temperature field graph model: Based on the three-dimensional temperature field characteristic data and furnace structure parameters, a spatial node graph structure is constructed, where the nodes represent the measurement points in the furnace and the edge weights reflect the heat conduction or airflow coupling relationship, forming a temperature field graph model with structural priors;

[0063] S22, extracting spatiotemporal fusion features to characterize thermal dynamic evolution: inputting node features into the graph convolutional network of coupled structural parameters to extract spatial correlation features, and extracting heat flow change trends through time series modeling methods to generate spatiotemporal fusion data that includes temperature gradient distribution, heat flow evolution trend and coal type adaptability score.

[0064] S21 includes:

[0065] S211: Construct a structural topology graph G with the three-dimensional spatial coordinates (x, y, z) of the furnace as nodes, expressed as:

[0066] G=(V,E,A);

[0067] Where V = {v i} is a node set, each node v i Corresponding to a position (x i ,y i ,z i ), E is the edge set, representing the spatial adjacency relationship between nodes, A∈R N×N is the adjacency matrix, element a ij Represents node v i With v j The heat conduction or airflow connection weight is calculated by combining the furnace structure parameters (heating surface distribution, air duct layout) with the ultrasonic phased array echo data;

[0068] S212: Input feature tensor X, expressed as:

[0069] X∈R N×F×T , where N is the number of nodes, and F is the characteristic dimension of the temperature field of each node, including the local temperature T(x,y,z,t), gradient The direction of heat flow, T is the number of time steps.

[0070] S22 includes:

[0071] S221: Introduce the graph convolutional network (GCN) to extract spatial features from the feature tensor X and obtain a spatial fusion representation Expressed as:

[0072]

[0073] in, is the adjacency matrix with self-connection added, yes The degree matrix of is the node feature representation of the lth layer, W (l) is the learnable weight matrix of layer l, σ(·) is the nonlinear activation function (ReLU);

[0074] S222: Introduce the LSTM module in the time dimension to extract the heat flow evolution trend and encode the historical temperature dynamics to generate the final spatiotemporal fusion data tensor F spatiotemporal ∈R N×D , including the temperature gradient distribution of node i Heat flow evolution function φ of node i i (t), adaptability score α for coal quality parameter C(t) i (C).

[0075] The adjacency matrix generated by the topological graph in S211 includes:

[0076] S2111, Node definition and spatial distance calculation: The temperature measurement points in the furnace are modeled as a node set V in the graph structure (V = {v1, v2, ..., v N}), each node corresponds to the spatial coordinate p i =(x i ,y i ,z i ), calculate the spatial distance d between any two nodes ij , expressed as:

[0077] d ij =||p i -p j || 2 ;

[0078] Among them, ||·|| represents the Euclidean norm;

[0079] S2112, structural connection constraint judgment: introduce the structural connection indicator function δ(i,j) to indicate whether there is a physical heat flow or air flow channel between nodes, expressed as:

[0080]

[0081] Structural connections can be determined based on the furnace CAD model, duct layout drawings, or maintenance manuals;

[0082] S2113, Calculate the structural coupling weight factor: Define the coupling weight coefficient η by fusing the structural topology with the ultrasonic phased array echo results s(i, j) represents the connection strength between nodes i and j, expressed as:

[0083] η s (i,j)=ω T ·η T (i,j)+ω F ·η F (i,j);

[0084] Among them, η T (i, j) indicates whether they are in the same heated structural section. is the airflow connectivity based on the sound wave propagation delay difference, Δt ij Ultrasonic waves in v i With v j The propagation delay difference between T ,ω F is the weight coefficient of the heat channel and the wind channel, satisfying ω T +ω F =1, κ>0 is the adjustment coefficient of the sound speed difference on the connectivity strength;

[0085] S2114, generate adjacency matrix: Based on spatial distance, structural connection constraint judgment and structural coupling weight factor, define adjacency matrix A = [a ij ] element a ij , expressed as:

[0086]

[0087] Among them, σ>0 is the spatial influence attenuation control coefficient, a ij Represents node v i With v j The structure-aware connection strength between them.

[0088] S3 includes:

[0089] S31, building a spatiotemporal Transformer model with a memory-enhanced mechanism: embedding and encoding spatiotemporal fusion features, and introducing a multi-head self-attention mechanism that integrates historical states, to build a time series modeling structure with memory capabilities, improving the ability to represent the evolution of furnace thermal states;

[0090] S32, Dynamic Risk Tensor Generation and Output: Based on the temporal risk features output by Transformer, the pre-trained multi-task risk assessment network is input to generate a dynamic risk tensor including instability probability, structural stress hotspots, and combustion efficiency loss coefficient, providing a basis for subsequent early warning decisions.

[0091] S31 includes:

[0092] S311: Input embedding and position encoding: The generated spatiotemporal fusion feature tensor Fspatiotemporal Each node sequence in As the Transformer input, linear embedding is used to obtain the representation vector And add time position coding, expressed as:

[0093]

[0094] Among them, f i (t) is the fused feature vector of node i at time t, Embed(·) is the linear projection layer, PosEnc(t) is the temporal position encoding, which uses sine and cosine forms to encode time step information;

[0095] S312, introduces a memory-enhanced self-attention mechanism: a multi-head self-attention module with a memory mechanism is used to model the node time series, improving the perception of the dynamic evolution of the historical temperature field, expressed as:

[0096]

[0097] Among them, Q, K, and V are query, key, and value matrices respectively, which come from the current input M is an external memory matrix that records historical key states (high temperature instability section, rapid combustion change area) and is updated through a sliding window. k is the key vector dimension, l is the current Transformer layer number;

[0098] Multiple attention heads are executed in parallel and then concatenated to obtain the next layer representation through residual connections and feedforward networks.

[0099] S32 includes:

[0100] S321: Feature aggregation and risk assessment network input construction: Temporal aggregation of Transformer output representation to obtain node-level risk representation r i , as the input of the risk assessment network, is expressed as:

[0101]

[0102] Where L is the total number of layers of Transformer;

[0103] S322, generate dynamic risk tensor: input the node risk representation into the pre-trained multi-task risk assessment network, and output the dynamic risk tensor R i , expressed as:

[0104] R i =[p i ,σ i , i ];

[0105] Among them, p i is the local instability probability of node i, which is defined as the classification output of the combination of structural temperature fluctuation and heat flux mutation trend, σ i is the structural stress hotspot score, combined with the local temperature gradient and prediction of furnace wall heating structure response, i Combustion efficiency loss coefficient, based on coal type adaptability α i (C(t)) and heat flow trend φ i The joint regression estimate of (t) is i ∈[0,1];

[0106] The final output node-level risk tensor set R risk , expressed as:

[0107]

[0108] S4 includes:

[0109] S41, intelligent determination of instability risk state: dynamic risk tensor R risk The instability probability p in i Perform dynamic threshold judgment and trigger warning conditions as follows:

[0110] p i >θ dyn (L);

[0111]

[0112] Among them, p i is the probability of local instability of node i, θ dyn (L) is the load adaptive dynamic threshold, which is adjusted based on the current unit load L, θ0 is the basic risk threshold 0.5, μ is the load adjustment coefficient, L is the current boiler load, L max is the maximum design load;

[0113] S42, generate a multi-dimensional executable warning instruction set: for the node area that meets the trigger conditions, the structural stress score σ i and combustion efficiency loss i , generating a visual warning instruction set including burner swing angle adjustment instructions, wall wind control strategy, and three-dimensional thermal stress cloud map Expressed as:

[0114] (1) Burner swing angle adjustment instruction Adjust the burner swing angle β according to the deviation of heat flow direction and temperature concentration area b , which minimizes the tilt angle in the high temperature area Expressed as:

[0115]

[0116] Among them, β b is the swing angle of the bth burner, is the target heat flow direction angle, (σ i ∈[0,1]),( i ∈0,1]);

[0117] (2) Wall wind control strategy If the node is close to the furnace wall area and ∈ i Significantly increased, generating a local enhanced wind gain factor λ i >1, the strategy is expressed as:

[0118] λ i =1+α·∈ i ;

[0119] Among them, α is the air volume adjustment gain coefficient, α∈[0.5,2.0];

[0120] (3) 3D thermal stress cloud map Based on the structural stress hot spot score σ i ,(σ i ∈[0,1]) draws the thermal stress distribution of high-risk areas in three-dimensional space and realizes the binding visualization with the physical structure;

[0121] S43, early warning instruction issuance and closed-loop execution: the early warning instruction set After encoding, it is sent to the execution unit through the industrial edge controller to complete closed-loop control, including:

[0122] (1) Burner execution module according to Adjust the swing angle;

[0123] (2) The secondary air execution module is based on λ i Adjust the opening of local air valve;

[0124] (3) Real-time presentation of the operator interface Used to assist manual intervention or decision confirmation.

[0125] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0126] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for identifying and warning abnormalities in the visual temperature field of a power plant furnace by integrating AI analysis, characterized in that: The following steps are involved: S1: The furnace radiation data is collected synchronously by using a dual-color spectral temperature measurement array and an ultrasonic phased array. Dynamic emissivity compensation is performed in combination with real-time coal quality parameters to generate three-dimensional temperature field characteristic data with spatial coordinate markers. S2: Perform spatiotemporal fusion processing on the three-dimensional temperature field feature data, use a graph convolutional network coupled with furnace structure parameters to extract spatial correlation features, and generate spatiotemporal fusion data including temperature gradient distribution, heat flow evolution trend and coal type adaptability; S3: Build a spatiotemporal transformer model that integrates a memory enhancement mechanism, input the spatiotemporal fusion data into a pre-trained risk assessment network, and output a dynamic risk tensor including instability probability, structural stress hotspots, and combustion efficiency loss coefficient; S4: Intelligent decision-making is performed based on the dynamic risk tensor. When the probability of instability exceeds the dynamic threshold of load adaptation, a visual warning instruction set including burner swing angle adjustment instructions, wall wind control strategy and three-dimensional thermal stress cloud map is generated, and real-time closed-loop control is performed through the edge controller.

2. The method for identifying and warning abnormalities in the furnace temperature field of a power plant by integrating AI analysis according to claim 1 is characterized in that: Said S1 comprises: S11: A dual-color spectral temperature measurement array arranged at multiple observation points in the furnace collects radiation intensity data at different wavelengths. Simultaneously, an ultrasonic phased array is used to obtain information on changes in the medium's sound velocity at corresponding spatial locations, achieving synchronous mapping of temperature measurement data with furnace structure information. S12: Based on the measured coal quality parameters, the improved Wiener approximation model is used to dynamically compensate for the local emissivity in the furnace and calculate the temperature field data; S13: Binding the calculated temperature field data to the spatial coordinates to generate a three-dimensional temperature field feature data tensor marked with the spatial coordinates.

3. The method for identifying and warning abnormalities in the furnace temperature field of a power plant by integrating AI analysis according to claim 2 is characterized in that: The S2 includes: S21, constructing a structure-aware temperature field graph model: Based on the three-dimensional temperature field characteristic data and furnace structure parameters, a spatial node graph structure is constructed, where the nodes represent the measurement points in the furnace and the edge weights reflect the heat conduction or airflow coupling relationship, forming a temperature field graph model with structural priors; S22, extracting spatiotemporal fusion features to characterize thermal dynamic evolution: inputting node features into the graph convolutional network of coupled structural parameters to extract spatial correlation features, and extracting heat flow change trends through time series modeling methods to generate spatiotemporal fusion data that includes temperature gradient distribution, heat flow evolution trend and coal type adaptability score.

4. The method for identifying and warning abnormalities in the furnace temperature field of a power plant by integrating AI analysis according to claim 3 is characterized in that: The S21 includes: S211: Constructing a structural topology graph with the three-dimensional spatial coordinates of the furnace as nodes; S212: Input feature tensor.

5. The method for identifying and warning abnormalities in the furnace temperature field of a power plant by integrating AI analysis according to claim 3 is characterized in that: The S22 includes: S221: Introduce graph convolutional network to extract spatial features from feature tensors and obtain spatial fusion representation; S222: Introduce the LSTM module in the time dimension to extract the heat flow evolution trend and encode the historical temperature dynamics to generate the final spatiotemporal fusion data tensor, including the temperature gradient distribution of node i, the heat flow evolution function of node i, and the adaptability score for coal quality parameters.

6. The method for identifying and warning abnormalities in the furnace temperature field of a power plant by integrating AI analysis according to claim 4 is characterized in that: Generating an adjacency matrix from the structural topology graph in S211 includes: S2111, Node definition and spatial distance calculation: The temperature measurement points in the furnace are modeled as a node set V in the graph structure, and each node corresponds to a spatial coordinate p i =(x i ,y i ,z i ), calculate the spatial distance between any two nodes; S2112, structural connection constraint judgment: introduce a structural connection indicator function to indicate whether there is a physical heat flow or air flow channel between nodes; S2113, calculate the structural coupling weight factor: by fusing the structural topology and the ultrasonic phased array echo results, define the coupling weight coefficient to represent the connection strength between nodes i and j; S2114, generate adjacency matrix: define the elements of the adjacency matrix based on spatial distance, structural connection constraint judgment and structural coupling weight factor.

7. The method for identifying and warning abnormalities in the visual temperature field of a power plant furnace integrated with AI analysis according to claim 6 is characterized in that: The S3 includes: S31, building a spatiotemporal Transformer model with a memory-enhanced mechanism: embedding and encoding spatiotemporal fusion features, and introducing a multi-head self-attention mechanism that integrates historical states to build a time series modeling structure with memory capabilities; S32, Dynamic Risk Tensor Generation and Output: Based on the temporal risk features output by Transformer, the pre-trained multi-task risk assessment network is input to generate a dynamic risk tensor including instability probability, structural stress hotspots, and combustion efficiency loss coefficient.

8. The method for identifying and warning abnormalities in the furnace temperature field of a power plant by visualizing the temperature field and integrating AI analysis according to claim 7 is characterized in that: The S31 includes: S311: Input embedding and position encoding: Each node sequence in the generated spatiotemporal fusion feature tensor is used as the Transformer input, linear embedding is used to obtain the representation vector and time position encoding is added; S312, introduces memory-enhanced self-attention mechanism: a multi-head self-attention module with memory mechanism is used to model node time series.

9. The method for identifying and warning abnormalities in the visual temperature field of a power plant furnace integrated with AI analysis according to claim 7 is characterized in that: The S32 includes: S321: Feature aggregation and risk assessment network input construction: Temporal aggregation of Transformer output representations to obtain node-level risk representations; S322, generating a dynamic risk tensor: inputting the node risk representation into a pre-trained multi-task risk assessment network, and outputting a dynamic risk tensor.

10. The method for identifying and warning abnormalities in the visual temperature field of a power plant furnace integrated with AI analysis according to claim 9 is characterized in that: The S4 includes: S41, intelligent determination of instability risk status: dynamic threshold determination of the instability probability in the dynamic risk tensor; S42, generating a multi-dimensional executable warning instruction set: For node areas that meet the trigger conditions, the structural stress score and combustion efficiency loss are combined to generate a visual warning instruction set including burner swing angle adjustment instructions, wall wind control strategy, and three-dimensional thermal stress cloud map; S43, issuing warning instructions and closed-loop execution: the warning instruction set is encoded and issued to the execution unit through the industrial edge controller to complete closed-loop control.

Images