A Method and System for Optimizing the Temperature Field of an Incinerator Based on CFD Numerical Simulation

Abstract

This invention provides a method and system for optimizing the temperature field of an incinerator based on CFD numerical simulation. First, dynamic disturbance events during incinerator operation are identified and their characteristic information is extracted. A furnace space correlation diagram is constructed, and a graph neural network is used to learn the correlation between disturbance features and temperature field anomalies. The temperature anomaly correlation link is output and input into the CFD simulation system to generate and execute the target simulation task. Dynamic response data of the temperature field is acquired, and the data is analyzed to determine the direction of simulation correction. Parameter adjustment instructions are generated to update the simulation parameters, and the task is re-executed to obtain corrected data. The differences in the data are compared to determine whether the anomaly has been alleviated. Finally, the final simulation parameters are extracted as the optimization scheme after the anomaly is eliminated, thereby accurately optimizing the temperature field of the incinerator.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method and system for optimizing the temperature field of an incinerator based on CFD numerical simulation. Background Technology

[0002] In the field of incinerator operation, the stability and optimization of the temperature field plays a crucial role in combustion efficiency, pollutant emission control, and equipment lifespan. During actual operation, incinerators can be affected by various dynamic disturbances, such as sudden changes in fuel composition, fluctuations in feed rate, and abnormal adjustments in combustion air volume. These dynamic disturbances are random and uncertain, and can disrupt the original thermal equilibrium state within the incinerator, leading to abnormal fluctuations in the temperature field.

[0003] Traditional methods for controlling the temperature field in incinerators primarily rely on empirical formulas and simple feedback mechanisms. Empirical formulas are often derived from specific operating conditions and are difficult to accurately adapt to complex and ever-changing real-world operating environments. Simple feedback mechanisms typically only adjust after a significant temperature deviation occurs, resulting in a lag and an inability to promptly and effectively address the impact of dynamic disturbances on the temperature field. Furthermore, traditional methods lack in-depth analysis and utilization of the heat conduction relationships within the incinerator furnace, making it difficult to accurately grasp the root causes and propagation paths of temperature field anomalies. This leads to poor temperature field optimization and the potential for localized overheating or underheating, negatively impacting the overall operational performance of the incinerator. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for optimizing the temperature field of an incinerator based on CFD numerical simulation, the method comprising:

[0005] Identify dynamic interference events during the operation of the incinerator, and extract event feature information of the dynamic interference events, including interference type, interference occurrence time and interference impact range;

[0006] A spatial relationship graph of the incinerator furnace is constructed, with each spatial unit of the furnace as a graph node and the heat conduction relationship between each spatial unit of the furnace as the graph edge weight. The event feature information is embedded into the corresponding graph node. The graph neural network is used to learn the relationship between the interference features and the temperature field anomaly, and the temperature anomaly relationship link corresponding to the interference event is output.

[0007] The temperature anomaly correlation link is input into the CFD simulation system to generate a target simulation task for the correlation link. The target simulation task includes the target simulation area, simulation duration and key monitored temperature indicators.

[0008] The target simulation task is executed to obtain dynamic response data of the temperature field of the associated links. The graph node embedding changes corresponding to the dynamic response data of the temperature field are analyzed by graph neural network to determine the correlation strength between the temperature field response and the disturbance event, and then the simulation correction direction that needs to be adjusted is determined.

[0009] Based on the simulation correction direction, a CFD simulation parameter adjustment instruction is generated. The CFD simulation parameter adjustment instruction is input into the CFD simulation system to update the simulation parameters, and the target simulation task is re-executed to obtain the corrected temperature field response data.

[0010] By comparing the differences in graph node embeddings corresponding to the temperature field response data before and after correction, it is determined whether the abnormal temperature field mode has been alleviated. If it has not been alleviated, the dynamic interference event characteristics are re-identified and the furnace space correlation graph is updated until the abnormal temperature field mode is eliminated. Finally, the simulation parameters are extracted as the incinerator temperature field optimization scheme.

[0011] In another aspect, embodiments of the present invention also provide a CFD numerical simulation-based incinerator temperature field optimization system, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0012] Based on the above, this invention identifies dynamic disturbance events during incinerator operation and extracts event feature information, constructs a spatial correlation diagram of the incinerator furnace, embeds event feature information into graph nodes, and utilizes graph neural networks to learn the correlation between disturbance features and temperature field anomalies. This reveals the intrinsic mechanism and propagation path of temperature field anomalies, enabling precise location of temperature anomaly correlation links corresponding to disturbance events. These temperature anomaly correlation links are input into a CFD simulation system to generate target simulation tasks, allowing for targeted simulation analysis of key links, thus improving simulation efficiency and accuracy. The target simulation task is executed to obtain dynamic temperature field response data, and graph neural network analysis is used to determine the correlation strength between the temperature field response and the disturbance event, as well as the simulation correction direction. Based on the simulation correction direction, CFD simulation parameter adjustment instructions are generated and parameters are updated. The simulation task is re-executed to obtain corrected temperature field response data. The differences between the data before and after correction are compared to determine whether the temperature field anomaly mode has been alleviated. If not, the anomaly is re-identified and updated until it is eliminated. The finally extracted simulation parameters serve as an optimization scheme, effectively eliminating temperature field anomalies, improving the stability and uniformity of the incinerator temperature field, increasing incineration efficiency, reducing pollutant emissions, and enhancing the incinerator's adaptability under complex operating conditions. Attached Figure Description

[0013] Figure 1This is a schematic diagram of the execution flow of the incinerator temperature field optimization method based on CFD numerical simulation provided in an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of exemplary hardware and software components of the incinerator temperature field optimization system based on CFD numerical simulation provided in an embodiment of the present invention. Detailed Implementation

[0015] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for optimizing the temperature field of an incinerator based on CFD numerical simulation, according to an embodiment of the present invention. The following is a detailed description of this method for optimizing the temperature field of an incinerator based on CFD numerical simulation.

[0016] Step S110: Identify dynamic interference events during the operation of the incinerator, and extract event feature information of the dynamic interference events, including interference type, interference occurrence time and interference impact range.

[0017] This embodiment uses a dedicated incinerator for medical waste from municipal solid waste as an application scenario. This incinerator is used to treat special wastes such as infectious and pathological waste, and its operational stability directly affects the treatment effect and environmental safety. During operation, dynamic disturbances can arise from various sources, such as uneven feeding due to ruptured containers of medical waste, minor leaks in the combustion aid supply pipeline affecting combustion efficiency, and localized refractory material shedding within the furnace altering heat reflection characteristics. These events disrupt the original thermal balance within the furnace, causing abnormal temperature fields. To effectively identify dynamic disturbances, a monitoring system covering key areas of the incinerator is needed to continuously collect and analyze various operational data, capture abnormal signals, and then extract the core characteristic information of the events.

[0018] Step S111: Collect operating data at a preset frequency through each monitoring point in the incinerator operation status monitoring network. The operating data includes feed data, burner data, flue gas data, and furnace wall data.

[0019] The waste incinerator's operational status monitoring network comprises multiple monitoring points distributed according to functional areas. Near the feed inlet, weight sensors, infrared spectrometers, and conveyor belt speed monitors are installed to collect feed data, including the feed weight per unit time, the spectral characteristics of the waste composition (reflecting the proportion of organic and inorganic matter), and the conveyor belt speed. The burner area is equipped with thermocouples, pressure sensors, and flow meters to collect burner data, specifically covering the combustion flame temperature distribution, fuel supply pressure, combustion air flow rate, and oxygen content. At the flue gas outlet, gas analyzers, temperature sensors, and pressure transmitters are arranged to collect flue gas data, including the concentration of various pollutants in the flue gas, flue gas temperature, and flow pressure. Infrared thermometers are installed at different heights and angles on the furnace inner wall to acquire furnace wall data, i.e., the real-time temperature values ​​of different areas of the wall.

[0020] The data acquisition frequency at each monitoring point is set based on the dynamic characteristics of the data. Weight and velocity information in the feed data are collected at short intervals, while spectral characteristic analysis is performed at slightly longer intervals. Temperature and pressure data in the burner data are collected at higher frequencies to capture instantaneous changes in the flame. Pollutant concentration in the flue gas data is measured at set intervals, while temperature and pressure are collected continuously. Furnace wall temperature data is collected at medium time intervals to balance data volume and timeliness. All collected data is transmitted via wired connection to a central data storage unit, forming a continuous operational data sequence.

[0021] Step S112: Using each monitoring point as the initial node, and the standardized spatial distance and data correlation between monitoring points as the graph edge weights, construct a spatial correlation graph of monitoring points. Then, use a graph neural network to perform node anomaly detection on the spatial correlation graph of monitoring points and capture abnormal nodes.

[0022] Each monitoring point is defined as an initial node in the spatial association graph of monitoring points. Each node carries a unique identifier, which includes its installation location and the type of monitoring data. For example, the node corresponding to the weight sensor at the feed inlet is marked as the feed weight monitoring point, and the node corresponding to the thermocouple of burner 1 is marked as the burner 1 temperature monitoring point.

[0023] Step S1121: Distribute the monitoring points in the wandering incinerator operation status monitoring network, take each monitoring point as the initial node of the monitoring point spatial association diagram, and record the spatial coordinates and monitoring data types of each initial node.

[0024] Using pre-stored installation drawings of monitoring points, the spatial coordinates of each monitoring point in the three-dimensional coordinate system of the incinerator are obtained, with the origin set as the center of the incinerator feed inlet. Simultaneously, the data type of the monitoring points is clearly labeled in the node attributes, such as "feed weight," "combustion temperature," and "flue gas pollutant concentration." For example, the spatial coordinates of the infrared spectrometer at the feed inlet are (x1, y1, z1), and the data type of the monitoring point is "spectral characteristics of feed composition"; the spatial coordinates of the infrared thermometer in the middle layer of the east wall of the furnace are (x2, y2, z2), and the data type of the monitoring point is "wall temperature." This information is stored in the node attribute table as the basic data for constructing the association graph.

[0025] Step S1122: Calculate the standardized spatial distance between any two initial nodes, and combine it with the historical correlation coefficient of the monitoring data of the two initial nodes. Use the product of the inverse of the standardized spatial distance and the historical correlation coefficient as the graph edge weight between the two initial nodes.

[0026] For any two initial nodes, calculate the straight-line distance based on their spatial coordinates, and then divide this distance by the maximum diagonal length of the incinerator furnace to obtain the standardized spatial distance, which ranges from 0 to 1. For example, if the actual spatial distance between node A and node B is D, and the maximum diagonal length of the furnace is L, then the standardized spatial distance is the ratio of D to L.

[0027] Simultaneously, historical monitoring data sequences of the two nodes over a past period were retrieved, and their correlation coefficient was calculated. The calculation process is as follows: after standardizing the two sets of data sequences to eliminate the influence of dimensions, the ratio of their covariance to the product of their respective standard deviations is calculated to obtain the historical correlation coefficient, which ranges from -1 to 1.

[0028] The graph edge weight between two initial nodes is obtained by multiplying the inverse of the standardized spatial distance by the historical correlation coefficient. The closer the two nodes are and the more consistent their data trends are, the larger the graph edge weight, indicating a stronger spatial and functional correlation between them.

[0029] Step S1123: Based on the initial nodes and graph edge weights, construct a spatial association graph of monitoring points, and use a graph autoencoder in a graph neural network to pre-train the spatial association graph of monitoring points to learn the embedding vectors of each initial node under normal operating conditions.

[0030] Based on the attribute information of the initial nodes and the calculated graph edge weights, a spatial association graph of monitoring points is constructed. This spatial association graph uses nodes to represent monitoring points and weighted edges to represent the association strength between monitoring points.

[0031] A graph autoencoder is used to pre-train the association graph. The graph autoencoder consists of an encoder and a decoder. The encoder contains multiple graph convolutional layers, each of which aggregates the features of a node and the features of its neighboring nodes. Specifically, the features of each node are weighted and combined with the features of its neighboring nodes according to the graph edge weights, and then non-linearly transformed through an activation function to obtain a more abstract feature representation. After processing by multiple graph convolutional layers, the encoder outputs the low-dimensional embedding vectors of each node.

[0032] The decoder receives these embedding vectors and attempts to reconstruct the original node features and graph edge weights through operations such as transpose and convolution. During pre-training, historical data from the incinerator's normal operation is used as training samples, and the network parameters are adjusted using the backpropagation algorithm to minimize the error between the reconstructed result and the original input. After a sufficient number of iterations, the graph autoencoder learns stable embedding vectors for each initial node under normal operating conditions. These stable embedding vectors contain the node's own features and its association information with other nodes.

[0033] Step S1124: Collect the operation data of each monitoring point in real time, convert the operation data into real-time node feature vectors, input them into the pre-trained graph autoencoder, and calculate the reconstruction error between the real-time node embedding vector and the initial node embedding vector under normal operation.

[0034] The real-time acquired operational data first undergoes preprocessing, including outlier removal (removing data that clearly exceeds the range of physical significance), missing value imputation (using interpolation of data from adjacent time points), and data standardization (converting data of different magnitudes to the same numerical range). The processed data from each monitoring point is organized into a vector, namely the real-time node feature vector. The dimension of the vector is determined by the number of indicators in the monitored data. For example, the feature vector of a burner monitoring point may contain standardized values ​​of multiple indicators such as flame temperature, fuel pressure, and airflow.

[0035] The real-time node feature vectors are input into the encoder part of a pre-trained graph autoencoder to obtain real-time node embedding vectors. These embedding vectors are then input into the decoder to obtain the reconstructed node features and graph edge weights. The difference between the reconstructed result and the real-time input data is calculated; this difference is the reconstruction error. The reconstruction error calculation encompasses both node feature reconstruction error and graph edge weight reconstruction error, and the two are weighted and summed according to a predetermined ratio to obtain the total reconstruction error.

[0036] Step S1125: Set a reconstruction error threshold, mark nodes whose reconstruction errors exceed the reconstruction error threshold as candidate abnormal nodes, perform neighbor node verification on the candidate abnormal nodes, and check whether the reconstruction errors of the neighboring nodes of the candidate abnormal nodes also exceed the reconstruction error threshold.

[0037] The reconfiguration error threshold is determined by analyzing historical reconfiguration error data during normal operation of the incinerator. The statistical distribution of historical reconfiguration errors is calculated, and a suitable quantile is selected as the threshold, ensuring that only a very small number of nodes exceed this threshold during normal operation.

[0038] The real-time calculated reconstruction error of each node is compared with this threshold. Nodes with reconstruction errors greater than the threshold are marked as candidate anomalous nodes. For each candidate anomalous node, all neighboring nodes with direct edges connected to it are found in the monitoring point spatial association graph, and it is checked whether the real-time reconstruction errors of these neighboring nodes also exceed the threshold.

[0039] Step S1126: If the reconstruction error of more than a preset proportion of the neighboring nodes of a candidate abnormal node exceeds the reconstruction error threshold, then the candidate abnormal node is determined to be a final abnormal node, and all final abnormal nodes are extracted to form an abnormal node set.

[0040] The preset ratio is determined based on the connection density of the spatial correlation diagram of the monitoring points. The denser the connection, the lower the preset ratio can be. For example, if a candidate abnormal node has multiple neighboring nodes, and the reconstruction error of more than one-third of the neighboring nodes also exceeds the threshold, it indicates that the abnormality of this node is not an isolated phenomenon, but is affected by the abnormality of surrounding nodes, or is itself the source of the abnormality spread. In this case, the candidate abnormal node is determined as the final abnormal node.

[0041] All the final abnormal nodes that meet the above conditions are aggregated to form an abnormal node set.

[0042] Step S113: Statistically determine the duration of data fluctuations at abnormal nodes, and in conjunction with the incinerator's operating mechanism, determine the start and end times of the interference corresponding to the abnormal nodes to form the interference occurrence period.

[0043] For each node in the set of abnormal nodes, retrieve its monitoring data time series, locate the point in the series where the data began to deviate from the normal range, and record it as the initial start time. Then find the point in time when the data recovers to the normal range and remains stable without fluctuation, and record it as the initial end time.

[0044] The two time points are verified in conjunction with the incinerator's operating mechanism. For example, if the data at a burner monitoring point is abnormal, according to the response characteristics of the combustion system, the temperature fluctuation caused by changes in fuel supply has a certain lag. If the initial start time is earlier than the time of change in fuel supply system parameters, the start time needs to be adjusted to the time of change in fuel supply parameters. By correcting the initial start and end times in the above way, the accurate start and end times of the disturbance are obtained, and the time period between the two is the disturbance occurrence period corresponding to the abnormal node. For multiple abnormal nodes, if their disturbance occurrence periods overlap, the earliest start time and the latest end time are taken as the disturbance occurrence period of the entire dynamic disturbance event.

[0045] Step S114: Determine the spatial boundary of the interference range based on the distribution location of the abnormal nodes and the data fluctuation amplitude, and simultaneously determine the interference type based on the monitoring data type corresponding to the abnormal nodes.

[0046] Mark the spatial coordinates of the abnormal nodes in the 3D model of the incinerator and observe their distribution. If the abnormal nodes are concentrated in a local area, such as near the feed inlet, then expand the range outward from that area as the interference range, taking into account the direction of airflow and heat transfer within the furnace. The determination of the spatial boundary should refer to the structural drawings of the incinerator to ensure that the boundary does not cross obvious structural partitions (such as refractory brick partitions within the furnace).

[0047] Data fluctuation amplitude is measured by calculating the degree of deviation between abnormal data and the average range of normal data. For each abnormal node, the extent to which its monitored data exceeds the upper limit of normal or falls below the lower limit of normal is calculated, and the average fluctuation amplitude of multiple abnormal nodes is taken as a reference for the overall impact intensity of the interference.

[0048] Based on the data types of the monitoring corresponding to the abnormal nodes, determine the type of interference. If the abnormal nodes are mainly feed-related monitoring points, the interference type may be "feed abnormality", including feed rate fluctuations, sudden changes in feed composition, etc. If the abnormal nodes are concentrated in the burner area, the interference type may be "burner malfunction", such as flame instability, insufficient fuel supply, etc. If the flue gas monitoring points are abnormal, it may be "flue gas emission system abnormality".

[0049] Step S115: Establish dynamic interference event classification rules. Based on the number of covered nodes in the interference impact range, the duration of the interference occurrence period, and the data fluctuation amplitude of abnormal nodes, dynamic interference events are classified into different priority levels.

[0050] The dynamic interference event classification rules include three evaluation indicators: the number of covered nodes, the duration, and the data fluctuation amplitude. Each indicator is divided into multiple levels. For example, the number of covered nodes is divided into three levels: "few", "medium", and "large"; the duration is divided into three levels: "short", "medium", and "long"; and the data fluctuation amplitude is divided into three levels: "slight", "moderate", and "significant".

[0051] Each indicator level is assigned a corresponding score; the more nodes covered, the longer the duration, and the greater the fluctuation, the higher the score. The scores of the three indicators are added together to obtain the total score. Based on the range of the total score, dynamic interference events are divided into three levels: "low priority," "medium priority," and "high priority." For example, a lower total score indicates low priority, meaning the interference has a minor impact and can be handled according to standard procedures; a higher total score indicates high priority, meaning the interference may have a serious impact on the incinerator operation and requires immediate attention.

[0052] Step S116: Integrate the interference type, interference occurrence time, interference impact range, and interference priority level to form an event feature information set containing priority information.

[0053] The information identified in the previous steps, including the interference type (e.g., "feeding anomaly"), the interference occurrence period (e.g., from start time T1 to end time T2), the spatial boundary coordinates of the interference's impact range, and the interference priority level (e.g., "medium priority"), is compiled into a structured dataset, namely, the event feature information set. This event feature information set is stored in a standardized data format and contains multiple fields, each corresponding to a feature information item, facilitating subsequent steps for retrieval and processing. For example, the value of the "interference type" field in the set is "feeding anomaly - compositional mutation," the value of the "interference occurrence period" field is [T1, T2], the value of the "spatial boundary" field is a polygonal boundary formed by a set of three-dimensional coordinate points, and the value of the "priority" field is "medium priority."

[0054] Step S120: Construct a spatial association graph of the incinerator furnace chamber, with each spatial unit of the furnace chamber as a graph node and the heat conduction relationship between units as the graph edge weight. Embed the event feature information into the corresponding graph node, learn the association relationship between interference features and temperature field anomalies through a graph neural network, and output the temperature anomaly association link corresponding to the interference event.

[0055] After acquiring the set of event feature information, a spatial correlation graph of the incinerator furnace needs to be constructed to analyze the intrinsic relationship between disturbance events and temperature field anomalies. First, the furnace is divided into multiple spatial units according to a pre-defined spatial partitioning strategy, with each unit serving as a node in the correlation graph. Then, the heat transfer coefficient between units is calculated based on the principle of heat conduction, serving as the edge weight between nodes. Next, event feature information is embedded into the corresponding nodes in the correlation graph, and then a graph neural network is used to learn the correlation between these features and temperature field anomaly patterns, ultimately identifying the links in the temperature anomaly chain affected by the disturbance.

[0056] Step S121: Divide the incinerator furnace into multiple spatial units, using each spatial unit as the basic node of the furnace spatial association diagram, and record the spatial coordinates and geometric attributes of each basic node.

[0057] Based on the furnace structure of the waste incinerator, a structured mesh generation method was used to divide the internal space of the furnace into multiple spatial units. During the generation, the size of the spatial units was smaller in areas with drastic heat flow changes, such as near the burner and in the feed inlet area, to improve the accuracy of the analysis; while the size of the spatial units could be appropriately increased in areas with gentle heat flow changes, such as the top and corners of the furnace, to reduce the computational load.

[0058] Each spatial unit serves as a foundational node in the furnace space relational diagram, recording its spatial coordinates (such as the 3D coordinates of the unit's center point) and geometric attributes (such as the unit's length, width, height, volume, and surface area). For example, the spatial unit located directly in front of burner 1 has center point coordinates of (x3, y3, z3), geometric attributes including length a, width b, height c, volume a×b×c, and surface area the sum of the areas of all surfaces. This information is stored in the attribute database of the foundational node, serving as the basis for subsequent calculations.

[0059] Step S122: Based on the heat conduction mechanism of the incinerator, calculate the heat conduction coefficient between adjacent spatial units, and use the heat conduction coefficient as the edge weight between the corresponding basic nodes in the furnace space association graph to construct the initial furnace space association graph.

[0060] Adjacent spatial units refer to spatial units that share a common surface or edge. According to the heat conduction mechanism, the heat transfer coefficient between two adjacent spatial units is related to the contact area between the units, the thermal conductivity of the materials, and the distance between the units. In the calculation, the shared contact area between the two adjacent units is first determined. Then, based on the material properties of that region within the furnace (such as the thermal conductivity of refractory materials), and combined with the distance between the unit centers, the heat transfer coefficient is logically calculated using the textual description of the heat conduction formula.

[0061] The calculated thermal conductivity coefficient is used as the edge weight between two adjacent basic nodes. The larger the weight value, the stronger the heat transfer capacity between the two spatial units. Following this method, weighted edge connections are established for all adjacent basic nodes, forming an initial furnace spatial correlation graph. This initial furnace spatial correlation graph reflects the path and intensity of heat transfer between different spatial units within the furnace.

[0062] Step S123: Determine the node features corresponding to each temperature field anomaly mode to form an anomaly mode node feature library.

[0063] By analyzing cases of temperature field anomalies observed in the historical operation of waste incinerators, common temperature field anomaly patterns were summarized, such as localized high-temperature zones, abnormal temperature gradients, and frequent temperature fluctuations. For each anomaly pattern, the characteristics of the corresponding nodes in the furnace spatial correlation diagram were extracted. For example, the node characteristics corresponding to localized high-temperature zones include the degree to which the node's temperature value exceeds the normal range, the duration of the high temperature, and the temperature difference with surrounding nodes; the node characteristics corresponding to abnormal temperature gradients include the rate of temperature change between adjacent nodes and the distribution range of the abnormal gradient.

[0064] These features are categorized and stored according to anomaly patterns, forming an anomaly pattern node feature library. Each anomaly pattern corresponds to a feature set in the library, containing multiple feature items. Each feature item has a clear definition and description, facilitating comparison with node features extracted in real time.

[0065] Step S124: The event feature information is used as an interference feature vector and embedded into the corresponding basic node in the initial furnace space association graph. A graph convolutional neural network is used to train the furnace space association graph with the embedded interference feature vector to learn the mapping relationship between the interference feature vector and the features in the abnormal mode node feature library, thus forming an interference temperature association learning model.

[0066] The event feature information set is converted into a numerical interference feature vector, with the vector dimension matching the number of feature terms. For example, the interference type is encoded into a numerical value, the interference occurrence period is converted into a numerical value for duration, the interference impact range is converted into a numerical value for the number of spatial units covered, and the interference priority level is also converted into a corresponding numerical value.

[0067] Based on the spatial boundary of the interference's influence range, the basic nodes related to the interference are determined in the initial furnace space association diagram, and the interference feature vector is embedded into the attributes of these nodes. If the interference's influence range involves multiple nodes, the same interference feature vector is embedded into each related node.

[0068] A graph convolutional neural network is used to train the furnace spatial correlation graph with embedded disturbance feature vectors to learn the mapping relationship between disturbance features and temperature field anomaly pattern features, thus forming a disturbance temperature correlation learning model.

[0069] Step S1241: Construct a training dataset, which includes interference feature vectors of historical interference events, corresponding furnace space correlation diagrams, and a set of manually labeled temperature anomaly nodes.

[0070] Collect cases of various dynamic disturbance events that occurred during the past operation of the incinerator. Each case includes the disturbance feature vector at that time (extracted from historical data), the corresponding furnace space correlation diagram (constructed according to the furnace structure and heat conduction characteristics at that time), and a set of manually marked temperature anomaly nodes (determined by technicians based on the temperature monitoring data and operation records at that time).

[0071] These cases were compiled into a training dataset, with each sample containing three parts: a disturbance feature vector, a furnace spatial correlation graph, and a set of labeled temperature anomaly nodes. The training dataset was divided into a training subset and a validation subset according to a set ratio for model training and performance validation.

[0072] Step S1242: Input the furnace space association graph in the training dataset into the graph convolutional neural network, aggregate the node features through the graph convolutional layer to generate node embedding vectors. The process of aggregating node features through the graph convolutional layer combines the weighted summation of the node's own features and the features of adjacent nodes.

[0073] Graph convolutional neural networks consist of multiple graph convolutional layers. The furnace space association graph from the training dataset is input into the network's first graph convolutional layer. The features of each node (including basic attribute features and embedded perturbation feature vectors) are aggregated with the features of neighboring nodes. During aggregation, the features of neighboring nodes are weighted according to graph edge weights (thermal conductivity coefficients); the higher the weight of a neighboring node, the greater the proportion of its features in the aggregation.

[0074] The aggregated features undergo a non-linear transformation using an activation function to obtain the node features output by this layer. These features are then fed into the next graph convolutional layer for deeper aggregation and transformation. After processing through multiple graph convolutional layers, the features of each node are compressed into a low-dimensional node embedding vector, which contains the node's own features and its association information with surrounding nodes.

[0075] Step S1243: In the fully connected layer of the graph convolutional neural network, the node embedding vector is compared with the features in the abnormal pattern node feature library, the similarity loss between the node embedding vector and the features in the abnormal pattern node feature library is calculated, and the gradient descent algorithm is used to optimize the network parameters of the graph convolutional neural network.

[0076] The fully connected layer of a graph convolutional neural network receives the node embedding vectors output by the graph convolutional layer and compares them with various features in an anomaly pattern node feature library. The comparison process is achieved by calculating the similarity between the node embedding vector and each feature in the feature library. The higher the similarity, the more likely the node belongs to the corresponding temperature anomaly pattern.

[0077] The similarity loss is calculated between the actual output similarity and the manually labeled set of temperature anomaly nodes. A gradient descent algorithm is used to adjust the parameters of each network layer (such as weight matrices and bias terms) based on the loss value, gradually reducing the loss. Through multiple iterations of training, the network parameters are continuously optimized to improve the model's accuracy in identifying temperature anomaly nodes.

[0078] Step S1244: Introduce an attention mechanism layer to assign attention weights to the node embedding vectors output by the graph convolutional layer, so that the graph convolutional neural network pays more attention to node features related to temperature anomalies.

[0079] An attention mechanism layer is introduced between the graph convolutional layer and the fully connected layer. This attention mechanism layer assigns an attention weight to each node by calculating the correlation between the embedding vector of each node and the temperature anomaly pattern features. Nodes with higher correlation receive larger attention weights and are given greater importance in subsequent feature processing.

[0080] Specifically, the attention mechanism layer processes the node embedding vectors output by the graph convolutional layer. It calculates an attention score for each node using a small neural network, and then converts the score into attention weights using a softmax function. The node embedding vector is multiplied by its corresponding attention weight before being input into the fully connected layer for further processing. In this way, the network can automatically focus on the node features most critical for temperature anomaly identification, improving model performance.

[0081] Step S1245: Divide the training dataset into a training subset and a validation subset. Use the training subset to train the network parameters of the graph convolutional neural network. Use the validation subset to verify the matching degree between the set of temperature anomaly nodes output by the graph convolutional neural network and the set of temperature anomaly nodes labeled manually.

[0082] The training dataset is divided proportionally into a training subset and a validation subset, for example, the majority of the data is used as the training subset and a small portion as the validation subset. The graph convolutional neural network is trained using the training subset, and the model is validated using the validation subset after each training iteration.

[0083] During the validation process, the furnace space correlation graph from the validation subset is input into the model to obtain the set of temperature anomaly nodes output by the model. This set is then compared with the manually labeled set to calculate the matching degree. The matching degree is calculated as follows: the number of intersections between the anomaly nodes identified by the model and the manually labeled anomaly nodes, divided by the number of unions of the two. A higher matching degree indicates better generalization ability of the model.

[0084] Step S1246: If the matching degree is lower than the preset matching degree standard, adjust the number of layers, kernel size and attention weight calculation method of the graph convolutional neural network, and retrain the graph convolutional neural network.

[0085] The preset matching degree standard is determined based on actual application requirements, for example, set to a relatively high proportion. If the matching degree of the validation subset is lower than this standard, it indicates that the model performance does not meet the requirements, and the network structure parameters need to be adjusted.

[0086] Adjustments can be made by: increasing or decreasing the number of graph convolutional layers, changing the size of the convolutional kernel (i.e., the range of adjacent nodes considered in each aggregation), and modifying the calculation function of attention weights in the attention mechanism layer (e.g., using different relevance metrics). After adjustments, the model is retrained using a training subset and validated again using a validation subset until the matching degree reaches the preset standard.

[0087] Step S1247: If the matching degree reaches the preset matching degree standard, stop training and form a stable interference temperature correlation learning model.

[0088] When the matching degree of the validation subset reaches the preset standard, it indicates that the model has good recognition ability and generalization performance. At this point, training is stopped, the current network parameters are saved, and a stable interference temperature correlation learning model is formed. This interference temperature correlation learning model can receive the furnace spatial correlation graph with embedded interference feature vectors and output the corresponding temperature anomaly node set prediction results.

[0089] Step S125: Iteratively optimize the interference temperature association learning model using historical interference temperature data, so that the interference temperature association learning model can output a set of temperature anomaly nodes corresponding to the interference event, and determine the furnace functional link corresponding to the set of temperature anomaly nodes as the temperature anomaly association link; wherein, when the interference temperature association learning model outputs the temperature anomaly association link, it simultaneously outputs the association confidence score, which is calculated by the similarity of graph node embedding. If the association confidence score is lower than the preset association confidence score threshold, feature embedding and model training are re-executed.

[0090] Collect more historical disturbance events and their corresponding temperature field data to form new training samples, and periodically iterate and optimize the disturbance-temperature correlation learning model. The optimization process is similar to the initial training: input new samples into the model, adjust the network parameters, and further improve the model's accuracy.

[0091] The model calculates the association confidence score while outputting the set of temperature anomaly nodes. The association confidence score is obtained by calculating the average similarity between the node embedding vector output by the model and the corresponding features in the feature library of anomaly pattern nodes. The higher the similarity, the greater the association confidence score, indicating that the model has higher reliability for the prediction result.

[0092] Set a preset association confidence threshold. If the output association confidence is lower than the threshold, it means that the model's recognition reliability for the current interference event is insufficient. It is necessary to re-examine the embedding process of event feature information. It may be that the range of embedded nodes is inaccurate or there is a deviation in the feature vector transformation. After adjustment, retrain and predict the model until the association confidence reaches above the threshold.

[0093] The furnace functional elements corresponding to the set of temperature anomaly nodes output by the model are identified as temperature anomaly associated elements. For example, if the temperature anomaly nodes are concentrated in the spatial units near burner 2, the corresponding temperature anomaly associated element is "burner 2 working area".

[0094] Step S130: Input the temperature anomaly correlation link into the CFD simulation system to generate a target simulation task for the correlation link. The target simulation task includes the target simulation area, simulation duration, and key monitoring temperature indicators.

[0095] The identified temperature anomaly correlation is input into the CFD (Computational Fluid Dynamics) simulation system. The system generates a targeted simulation task based on the characteristics and relevant parameters of the correlation. The target simulation task needs to clearly define the spatial scope, time span, and key temperature indicators to be monitored to ensure that the simulation results accurately reflect the temperature field changes of the correlation under the influence of the disturbance event.

[0096] Step S131: Call the CFD simulation system to receive the temperature anomaly correlation link and the corresponding correlation confidence level. If the correlation confidence level is higher than the preset correlation confidence level threshold, start the target simulation task generation process.

[0097] The CFD simulation system receives information about temperature anomaly correlation links and their corresponding correlation confidence levels via an interface. The CFD simulation system internally presets a correlation confidence threshold, which is determined based on the accuracy requirements of the simulation results. When the received correlation confidence level is higher than this CFD simulation threshold, it indicates that the identification of the temperature anomaly correlation link is reliable, and the target simulation task generation process is initiated; if it is lower than the threshold, a prompt message is returned, requiring the temperature anomaly correlation link to be re-identified.

[0098] Step S132: The nodes corresponding to the temperature anomaly correlation links in the furnace space correlation diagram are ranked by importance using a graph neural network. The top K nodes and their adjacent nodes are extracted to determine the spatial range of the target simulation area. The spatial range of the target simulation area includes the spatial units corresponding to potential nodes where the interference may spread.

[0099] The nodes corresponding to temperature anomalies in the furnace space correlation diagram are input into a graph neural network. This network analyzes factors such as the connection strength and feature importance of the nodes in the correlation diagram to rank their importance. The ranking criteria include the correlation between the node and the disturbance feature vector, and the node's position in the heat conduction path. Nodes with higher importance require more refined processing in the simulation.

[0100] Several nodes with the highest priority are extracted, and their neighboring nodes are also considered to cover potential areas where interference may spread. The spatial units corresponding to these nodes are combined to form the spatial range of the target simulation area. For example, if the nodes with the highest priority are concentrated around burner 3, the target simulation area includes burner 3 and the spatial units within a defined range around it, ensuring that the diffusion process of interference in that area can be captured.

[0101] Step S133: Based on the interference occurrence period and interference duration characteristics in the event feature information, set the time span of the simulation duration. The time span includes the baseline stable period before the interference occurs, the interference occurrence period, and the recovery period after the interference is eliminated.

[0102] Referring to the interference occurrence time periods (start time T1 and end time T2) in the event characteristic information, and combining the duration characteristics of the interference (such as whether the interference is quickly eliminated or lasts for a long time after its occurrence), the simulation duration is set to a specific time span. The time span includes three parts: a baseline stabilization period before the interference occurs (a period of time preceding T1), used to obtain baseline temperature field data under normal operating conditions; the interference occurrence period (from T1 to T2), simulating the temperature field changes during the interference event; and a recovery period after the interference is eliminated (a period of time following T2), observing whether the temperature field can recover to a normal state.

[0103] For example, if the disturbance occurs for 10 minutes, the baseline stabilization period can be set to 5 minutes before the disturbance occurs, and the recovery period can be set to 5 minutes after the disturbance ends. The total simulation time is 20 minutes to fully capture the impact of the disturbance on the temperature field.

[0104] Step S134: Based on the node characteristics of the temperature field anomaly mode, determine the key monitoring temperature indicators. The key monitoring temperature indicators include the core temperature parameters corresponding to the temperature field anomaly mode and the associated derived temperature parameters. At the same time, set the monitoring frequency for different simulation periods, with the monitoring frequency during the interference period being higher than that during other simulation periods.

[0105] Based on the node characteristics of the temperature field anomaly pattern, key monitoring temperature indicators are determined. Core temperature parameters include parameters that directly reflect temperature anomalies, such as the temperature value of the anomaly node and the temperature difference between adjacent nodes; derived temperature parameters include parameters that indirectly reflect the stability of the temperature field, such as the rate of temperature change and the frequency of temperature fluctuations. For example, for an anomaly pattern of localized high temperature, the core temperature parameter is the temperature value of the high-temperature area, and the derived temperature parameter is the rate of temperature increase.

[0106] The monitoring frequency is set for different simulation periods. During the interference period, the monitoring frequency is set to a higher level to capture subtle temperature changes due to the drastic changes in the temperature field. During the baseline stabilization period and the recovery period, the temperature field changes relatively gently, and the monitoring frequency can be appropriately reduced to reduce the amount of data.

[0107] Step S135: Allocate CFD simulation resources based on interference priority levels to generate resource-optimized target simulation tasks.

[0108] Based on the priority level of dynamic disturbance events, the CFD simulation system allocates corresponding computing resources. High-priority disturbance events are allocated more computing cores and memory resources, and a finer computing grid is used to improve simulation accuracy and speed; low-priority disturbance events can have their resource allocation reduced and a relatively simplified computing setup used.

[0109] The target simulation area, simulation duration, key monitoring temperature indicators, and resource allocation information are integrated to generate a complete target simulation task. This target simulation task is stored in the form of a structured task description file, which contains all the necessary parameter settings for the CFD simulation system to execute.

[0110] Step S140: Execute the target simulation task, obtain the dynamic response data of the temperature field of the associated links, analyze the changes in the embedding of the graph nodes corresponding to the dynamic response data of the temperature field through a graph neural network, determine the correlation strength between the temperature field response and the disturbance event, and then determine the simulation correction direction that needs to be adjusted.

[0111] The CFD simulation system executes simulation calculations according to the settings of the target simulation task, acquiring dynamic temperature field response data of temperature anomaly-related factors at different times. By analyzing the changes in graph node embeddings caused by these data using a graph neural network, the system assesses the correlation between the temperature field response and the disturbance event. Based on the assessment results, it determines the direction of simulation corrections to optimize the temperature field simulation effect.

[0112] Step S141: Call the CFD simulation system to load the geometric model of the target simulation region corresponding to the target simulation task, import the initial running parameters, and start the simulation operation according to the set time step.

[0113] The CFD simulation system reads the geometric model file of the target simulation region specified in the target simulation task. This geometric model file contains geometric information such as the furnace structure and equipment layout within the region. At the same time, it imports the initial operating parameters of the incinerator, including the initial temperature distribution, the initial flow rate and pressure of the fluids (air and flue gas), and the initial fuel supply rate. These parameters are set based on historical data from the incinerator's normal operation.

[0114] Set the time step for the simulation. The size of the time step is determined based on the simulation accuracy requirements and computing resources. The time step during periods of disturbance can be set to a smaller value to improve the ability to capture rapidly changing processes. After starting the simulation, the system performs numerical calculations on the temperature field, flow field, etc., within each time step according to the basic equations of fluid mechanics and heat conduction, gradually advancing the simulation process.

[0115] Step S142: During the simulation, temperature data corresponding to each node in the target simulation area is collected according to the monitoring frequency to form a dynamic response data sequence of the temperature field, and the disturbance simulation status of each time node is recorded synchronously.

[0116] During the simulation, the CFD simulation system collects temperature data from all spatial unit nodes within the target simulation area at each monitoring time according to a preset monitoring frequency. This includes the instantaneous temperature value and temperature gradient of each node. These data are arranged in chronological order to form a dynamic response data sequence of the temperature field, reflecting the change of the temperature field over time.

[0117] Simultaneously, the system records the disturbance simulation status at each time point, such as whether the disturbance is occurring and the intensity level of the disturbance, in order to analyze the correspondence between the temperature field response and the disturbance status later. For example, at the time point during the disturbance occurrence period, the specific manifestation of the disturbance (such as a sudden increase in the feed rate) is recorded, while during the baseline stabilization period and the recovery period, it is recorded as a state of no disturbance or disturbance elimination.

[0118] Step S143: Embed the dynamic response data sequence of the temperature field into the corresponding node of the furnace space correlation graph to generate a dynamic node embedding vector. Calculate the change in the dynamic node embedding vector before and after the disturbance by using the node embedding update algorithm of the graph neural network.

[0119] Temperature data at each time point in the dynamic response data sequence of the temperature field is embedded into the corresponding basic nodes of the furnace space correlation graph according to the node correspondence relationship, and the feature information of the nodes is updated. A node embedding and update algorithm of graph neural network is used to process the correlation graph after embedding the new temperature data to generate dynamic node embedding vectors for each node at different time points.

[0120] This update algorithm is similar to the encoding process of a graph autoencoder. It aggregates the temperature features of a node and the features of its neighboring nodes through graph convolution operations to generate a new embedding vector. The difference between the dynamic node embedding vector of the same node before the interference occurs (baseline stable period) and after the interference occurs (interference period) is calculated to obtain the change. The larger the change, the more significantly the temperature features of the node are affected by the interference.

[0121] Step S144: Construct the association strength analysis dimension, which includes a time dimension, a spatial dimension, and a feature dimension. The time dimension analyzes the synchronicity between the occurrence of interference and the change of dynamic node embedding vector. The spatial dimension analyzes the overlap between abnormal nodes and nodes affected by interference. The feature dimension analyzes the matching degree between the change of dynamic node embedding vector and interference features.

[0122] The correlation strength analysis includes three dimensions: time, space, and features. In the time dimension, the synchronicity is analyzed by comparing the time point of interference occurrence with the time point when the dynamic node embedding vector begins to change significantly. If the time point of vector change coincides with or is close to the time point of interference occurrence, the correlation in the time dimension is high.

[0123] In the spatial dimension, the number of overlaps between abnormal nodes in the statistical temperature anomaly correlation link and nodes whose embedding vectors have changed due to interference is counted. The greater the number of overlaps, the higher the degree of correlation in the spatial dimension.

[0124] In terms of feature dimension, the degree of matching between the changing features of dynamic node embedding vectors (such as the magnitude and trend of change) and interference feature vectors (such as the type and intensity of interference) is analyzed. For example, for the interference feature of increased feed rate, if the corresponding node's embedding vector shows an upward trend in temperature, then the feature dimension matching degree is high.

[0125] Step S145: Calculate the association score of each association strength analysis dimension through graph neural network, standardize and transform the association score of each association strength analysis dimension, and then sum them up according to preset weights to obtain the comprehensive association strength. If the comprehensive association strength is lower than the preset comprehensive association strength threshold, re-examine the parameter settings of the target simulation task.

[0126] The graph neural network calculates a correlation score for each dimension of correlation strength analysis. The score for the time dimension is determined based on the degree of synchronicity, with higher scores for better synchronicity; the score for the spatial dimension is calculated based on the percentage of overlap, with higher percentages for higher scores; and the score for the feature dimension is based on the degree of matching, with higher degrees of matching for higher scores.

[0127] The association scores across the three dimensions are standardized to conform to a uniform numerical range (e.g., 0 to 1). Preset weights are assigned based on the importance of each dimension; for example, time and space dimensions are given higher weights, while feature dimensions are given slightly lower weights. The standardized scores are then summed according to their weights to obtain the overall association strength.

[0128] The preset comprehensive correlation strength threshold is determined based on historical data statistics. If the calculated comprehensive correlation strength is lower than this threshold, it indicates that the correlation between the temperature field response and the disturbance event is not significant enough. This may be due to unreasonable parameter settings in the target simulation task (such as the simulation area being too small or the time step being too large). It is necessary to re-examine and adjust the parameter settings and re-execute the simulation task.

[0129] Step S146: If the overall correlation strength is higher than the preset overall correlation strength threshold, based on the temperature data characteristics corresponding to the change in the dynamic node embedding vector, and combined with the incinerator temperature field optimization target, the simulation correction direction is determined. The simulation correction direction includes the heat source adjustment direction, the fluid flow adjustment direction, and the heat exchange adjustment direction.

[0130] When the overall correlation strength exceeds a preset threshold, it indicates a significant correlation between the temperature field response and the disturbance event, which can be used to determine the direction of simulation correction. The temperature data characteristics corresponding to the dynamic node embedding vector change include the distribution of temperature anomaly nodes, the magnitude of temperature changes, and the temperature gradient. The optimization objectives for the incinerator temperature field include temperature field uniformity (temperature differences between regions are within a reasonable range), temperature stability (small temperature fluctuations), and temperature range compliance (temperature remains within a set reasonable range).

[0131] Based on the discrepancy between the temperature data characteristics and the optimization target, the specific direction of simulation correction is determined. If the temperature data shows that the temperature in a local area exceeds the reasonable range, the heat source needs to be adjusted; if the temperature gradient is too large, the fluid flow needs to be optimized; if the temperature changes too rapidly, the heat exchange components need to be adjusted.

[0132] Step S1461: Extract the temperature data features corresponding to the change in the dynamic node embedding vector. The temperature data features include the location of the temperature anomaly node, the magnitude of temperature change, the frequency of temperature fluctuation, and the temperature gradient distribution between nodes.

[0133] Temperature data features are extracted from the changes in the embedding vectors of dynamic nodes. The location of nodes with abnormal temperatures is determined by their spatial coordinates; the temperature change amplitude is the difference between the abnormal node's temperature and the normal range; the temperature fluctuation frequency is the number of times the temperature exceeds the normal range per unit time; and the temperature gradient distribution between nodes is the spatial distribution of the temperature difference between adjacent nodes. For example, if a node in a certain area has a large temperature change amplitude, a high fluctuation frequency, and a significant temperature gradient with its surrounding nodes, these are all temperature data features that need to be considered.

[0134] Step S1462: Obtain the optimization target of the incinerator temperature field, which includes the temperature field uniformity target, the temperature stability target, and the temperature range compliance target.

[0135] The optimization targets for the incinerator temperature field are set by technicians based on the process requirements for medical waste treatment. The temperature field uniformity target requires that the temperature difference between different areas within the furnace be controlled within the set range to avoid local overcooling or overheating; the temperature stability target requires that the amplitude and frequency of temperature fluctuations be within the allowable range to ensure stable combustion; and the temperature range compliance target requires that the temperature within the furnace be maintained within a range suitable for the complete combustion of medical waste without generating excessive harmful pollutants.

[0136] Step S1463: If the temperature data features show that the temperature of a local node is higher than the temperature range corresponding to the temperature range compliance target, and the correlation strength analysis shows that the embedding change of the local node is significantly correlated with the embedding change of the heat source input node, then the simulation correction direction is determined to be the heat source adjustment direction, specifically, reducing the heat source input intensity corresponding to the heat source input node.

[0137] When temperature data characteristics show that the temperature of a certain local node is consistently higher than the upper limit of the temperature range compliance target, and the correlation strength analysis shows that the dynamic embedding vector change of this local node has significant synchronicity and high matching degree with the embedding vector change of the heat source input node (such as the node corresponding to the burner), it indicates that the local high temperature is caused by excessive heat source input. In this case, the simulation correction direction is determined to be the heat source adjustment direction. The specific measure is to reduce the fuel supply or combustion intensity corresponding to the heat source input node to reduce heat input and reduce the local temperature to a reasonable range.

[0138] Step S1464: If the temperature data features show that the temperature gradient between nodes exceeds the temperature gradient range corresponding to the temperature field uniformity target, and the correlation strength analysis shows that the temperature gradient change is strongly correlated with the node embedding change corresponding to the fluid flow path, then the simulation correction direction is determined to be the fluid flow adjustment direction, specifically, optimizing the flow velocity parameters of the node region corresponding to the fluid flow path.

[0139] When the temperature gradient between nodes exceeds the allowable range for temperature field uniformity, and correlation strength analysis shows a strong correlation between the temperature gradient change and the node embedding vector change along the fluid (e.g., combustion air, flue gas) flow path (e.g., temperature gradient changes synchronously with flow velocity changes), it indicates that the uneven temperature distribution is caused by poor fluid flow. In this case, the simulation correction direction is determined to be fluid flow adjustment. Specific measures include adjusting the flow velocity parameters of relevant node regions along the fluid flow path, such as increasing or decreasing the airflow in certain regions, to improve fluid mixing and heat transfer, and reduce the temperature gradient.

[0140] Step S1465: If the temperature data characteristics show that the node temperature change rate exceeds the temperature change rate range corresponding to the temperature stability target, and the correlation strength analysis shows that the temperature change rate is strongly correlated with the embedding change of the heat exchange node, then the simulation correction direction is determined to be the heat exchange adjustment direction, specifically, adjusting the heat exchange component parameters corresponding to the heat exchange node.

[0141] When the rate of change of node temperature exceeds the limit set by the temperature stability target, and the correlation strength analysis shows that the rapid change is strongly correlated with the change of the embedding vector of the heat exchange node (such as the node corresponding to the cooling water pipe on the furnace wall), it indicates that the temperature instability is caused by abnormal heat exchange efficiency. At this time, the simulation correction direction is determined to be the heat exchange adjustment direction. The specific measures are to adjust the parameters of the heat exchange components corresponding to the heat exchange node, such as changing the cooling water flow rate or flow rate, adjusting the heat exchange efficiency, reducing the rate of temperature change, and making the temperature field tend to stabilize.

[0142] Step S1466: Verify the feasibility of the determined simulation correction direction. Combine the parameter adjustment range of the CFD simulation system and the actual operating limitations of the incinerator to ensure that the simulation correction direction is within the achievable range. If it is not feasible, re-analyze the temperature data characteristics corresponding to the change in the dynamic node embedding vector and adjust the simulation correction direction.

[0143] The feasibility of the initially determined simulation correction direction was verified. The adjustment range of the corresponding parameters in the CFD simulation system was checked to ensure that the parameter changes required by the correction direction were within the system's allowable range. At the same time, the actual operating limitations of the incinerator, such as the maximum and minimum output power of the burner and the speed range of the blower, were considered to ensure that the correction measures could be implemented in actual operation.

[0144] If the simulated correction direction exceeds the system parameter adjustment range or actual operating limitations, it is deemed infeasible and requires re-analysis of temperature data characteristics to find other possible correction directions. For example, if reducing the heat source input intensity would cause the combustion temperature to fall below the minimum requirement for medical waste treatment, this direction needs to be abandoned, and the local high temperature caused by other factors needs to be re-analyzed to adjust the correction direction.

[0145] Step S150: Generate a CFD simulation parameter adjustment instruction according to the simulation correction direction, input the CFD simulation parameter adjustment instruction into the CFD simulation system to update the simulation parameters, re-execute the target simulation task, and obtain the corrected temperature field response data.

[0146] Based on the determined simulation correction direction, specific CFD simulation parameter adjustment instructions are generated, specifying the parameters to be adjusted, the adjustment direction, and the magnitude. These instructions are input into the CFD simulation system to update the relevant simulation parameters. Then, the simulation calculation is re-executed according to the original target simulation task settings to obtain the corrected temperature field response data, thereby observing the improvement effect of the correction measures on the temperature field.

[0147] Step S151: Establish a mapping relationship library between simulation correction directions and CFD parameters. The mapping relationship library records the core parameter types and parameter adjustment constraints corresponding to different simulation correction directions. The constraints are classified into physical constraints and numerical constraints. Physical constraints include the range of safe operating parameters of the equipment, and numerical constraints include simulation stability requirements.

[0148] The mapping library between simulation correction directions and CFD parameters was established by analyzing the physical model of the incinerator and the numerical model of CFD simulation. For the heat source adjustment direction, the corresponding core parameter types include the burner's fuel flow rate and combustion temperature setpoint; for the fluid flow adjustment direction, the corresponding parameters are the fan's air volume, air velocity, and fluid inlet angle; and for the heat exchange adjustment direction, the corresponding parameters are the cooling medium's flow rate and temperature.

[0149] The mapping database also records the constraints for parameter adjustments. Physical constraints are based on the safe operation requirements of the incinerator equipment, such as the burner's fuel flow rate not exceeding its maximum design value and the cooling medium pressure not falling below a safety threshold. Numerical constraints are based on the stability requirements of CFD simulations, such as the flow rate change not being too drastic to cause numerical calculation divergence, and the temperature gradient adjustment needing to ensure the numerical stability of the energy conservation equation.

[0150] Step S152: Based on the determined simulation correction direction, extract the corresponding core parameter type from the mapping relationship library, obtain the current value of the core parameter type in the CFD simulation system, and determine the reasonable range for adjusting the core parameter type by combining the change of the dynamic node embedding vector corresponding to the dynamic response data of the temperature field.

[0151] Based on the determined simulation correction direction (such as heat source adjustment direction), the corresponding core parameter type (such as burner fuel flow rate) is searched from the mapping database. The current value of the core parameter is obtained through the parameter query interface of the CFD simulation system.

[0152] Analyze the changes in the dynamic node embedding vectors within the dynamic response data of the temperature field. Larger changes indicate potentially greater adjustments are needed. Combined with parameter constraints, determine a reasonable range for adjusting core parameters. For example, if the current fuel flow rate is a certain value and the vector change indicates a need for a significant temperature reduction, then within the limits of physical constraints, determine the range for lowering the fuel flow rate to ensure that the adjusted parameters effectively improve the temperature field without violating the constraints.

[0153] Step S153: Generate CFD simulation parameter adjustment instructions. The CFD simulation parameter adjustment instructions include target parameter identifier, adjustment direction, adjustment range, and verification rules for the adjusted parameter values, so that the adjusted parameter values ​​meet the constraints.

[0154] CFD simulation parameter adjustment instructions are generated using a standardized format and contain multiple fields. The target parameter identifier clearly identifies the specific parameter that needs to be adjusted (e.g., "burner 1 fuel flow"); the adjustment direction indicates whether the parameter is increased or decreased; the adjustment range is determined based on a reasonable range and can be expressed as a percentage or absolute change relative to the current value; the adjusted parameter value verification rules are used to check whether the parameter adjustment meets the constraints, such as verifying whether the adjusted fuel flow is within the range of physical constraints.

[0155] For example, an adjustment instruction might be: target parameter identifier "burner 1 fuel flow", adjustment direction "reduction", adjustment range "ten percent", verification rule "the adjusted value must be greater than the burner minimum fuel flow and less than the current value".

[0156] Step S154: Import the CFD simulation parameter adjustment command through the parameter update interface of the CFD simulation system, automatically replace the original parameter values, and back up the parameter configuration before adjustment.

[0157] The CFD simulation system provides a dedicated parameter update interface to receive generated parameter adjustment instructions. The system parses the instructions, locates the corresponding parameters based on the target parameter identifier, and automatically updates the parameter values ​​according to the adjustment direction and magnitude. During the update process, the system automatically executes verification rules to check whether the adjusted parameters meet the constraints; if not, the update is rejected and an error message is displayed.

[0158] After the parameters are updated, the system automatically backs up the parameter configuration before the adjustment and stores it in a designated backup file so that the system can restore the state before the adjustment if needed. The backup file contains information such as the original values ​​of all parameters and the adjustment time.

[0159] Step S155: Call the basic configuration of the target simulation task, keep the target simulation area, simulation duration and monitoring frequency unchanged, start the re-simulation calculation, collect temperature data at the set frequency, form the corrected temperature field response data, and synchronously record the embedding vector of each node after correction.

[0160] The basic configuration of the target simulation task is retrieved from the task storage of the CFD simulation system to ensure that the spatial range of the target simulation area, the time span of the simulation duration, and the monitoring frequency of each time period are consistent with the previous simulation task, so as to ensure that the simulation results before and after the correction are comparable.

[0161] For example, step S1551: retrieve the basic configuration of the target simulation task from the task management process of the CFD simulation system, and confirm the spatial coordinate range of the target simulation area, the time nodes of the simulation duration, and the monitoring frequency of each simulation period.

[0162] In the task management module of the CFD simulation system, locate and retrieve the corresponding basic configuration file for the target simulation task. This basic configuration file records in detail the spatial coordinate range of the target simulation area (such as the boundary defined by multiple three-dimensional coordinate points), the time nodes of the simulation duration (the start and end times of the baseline stabilization period, the start and end times of the disturbance occurrence period, and the start and end times of the recovery period), and the monitoring frequency of each simulation period (such as monitoring once at a relatively long interval during the baseline stabilization period and once at a relatively short interval during the disturbance occurrence period).

[0163] Confirm the retrieved configuration information to ensure it is consistent with the original simulation task. If there are any discrepancies, correct them to ensure that the conditions for the re-simulation are the same as the original simulation conditions.

[0164] Step S1552: Check the computing resource status of the CFD simulation system to determine whether the current computing resources meet the requirements for resimulation. If the computing resources do not meet the requirements for resimulation, queue them according to the interference priority level or adjust the resource allocation of non-critical simulation tasks.

[0165] The resource monitoring module of the CFD simulation system allows you to view currently available computing resources, including the number of CPU cores, memory capacity, and GPU utilization. These resources are then compared to the resources required for resimulation to determine if they are sufficient. The resource requirements for resimulation are determined based on factors such as the size of the target simulation region, simulation duration, and mesh refinement.

[0166] If computing resources meet the requirements, the simulation will be restarted immediately. If not, the dynamic interference events will be handled according to their priority levels. High-priority events can be queued to wait for resource release, or the currently executing non-critical simulation task can be paused and its resources allocated to the resimulation. Low-priority events will be queued and wait for other tasks to complete before acquiring resources.

[0167] Step S1553: Load the adjusted simulation parameters, confirm that the adjusted simulation parameter values ​​meet the constraints, start the re-simulation operation, and monitor the operation status in real time during the re-simulation operation.

[0168] The updated simulation parameters are loaded into the computation module of the CFD simulation system, and the system verifies again whether the adjusted parameter values ​​meet the physical and numerical constraints. After successful verification, the simulation is restarted.

[0169] During the simulation, the computational status is monitored in real time, including computation progress, convergence at each time step, and whether numerical instability occurs (such as abnormal jumps in temperature or pressure). If the computation is interrupted or convergence fails, the computation is stopped immediately, parameter settings and model configuration are checked, and the problem is resolved before restarting.

[0170] Step S1554: Collect temperature data of each node in the target simulation area according to the set monitoring frequency, record the temperature value, temperature change rate and temperature gradient data between nodes at each time node, and form the corrected temperature field response data.

[0171] During the resimulation, the system collects temperature data from all spatial unit nodes within the target simulation area at each monitoring time node according to a preset monitoring frequency. The collected data includes the instantaneous temperature value of each node, the rate of temperature change per unit time (the ratio of temperature difference to time interval), and the temperature gradient between adjacent nodes (the ratio of temperature difference to spatial distance).

[0172] These data are organized in chronological order to form corrected temperature field response data, which uses the same data format as the uncorrected temperature field response data to facilitate subsequent comparative analysis.

[0173] Step S1555: Embed the corrected temperature field response data into the corresponding node of the furnace space correlation graph, generate the corrected node embedding vector through the node embedding algorithm of graph neural network, and label and associate it with the node embedding vector before correction.

[0174] The corrected temperature field response data is embedded into the basic nodes of the furnace space association graph according to the node correspondence. The same graph neural network node embedding algorithm as before (such as the graph convolution-based encoding process) is used to generate the corrected node embedding vector of each node at different time points.

[0175] Add a marker to the corrected node embedding vector to indicate the corresponding parameter adjustment (e.g., "fuel flow of burner 1 is reduced by 10%), and establish a relationship with the node embedding vector before correction to ensure that the two can be compared one by one by node and time point.

[0176] Step S1556: After the re-simulation is completed, the corrected temperature field response data and the corrected node embedding vector are classified and stored, a simulation process report is generated, and the parameter adjustment, calculation status and data acquisition status are recorded.

[0177] After the resimulation is completed, the corrected temperature field response data and the corrected node embedding vectors are classified and stored in the system's database according to the disturbance event number and simulation task number, so as to facilitate subsequent query and analysis.

[0178] Generate a simulation process report, which details the parameter adjustments (adjusted parameter names, original values, new values, and adjustment range), computational status (start time, end time, any anomalies, and convergence status), and data acquisition details (number of acquisition time points, total data volume, and data quality assessment). The report is saved as a document as a record of the temperature field optimization process.

[0179] Step S160: Compare the differences in graph node embeddings corresponding to the temperature field response data before and after correction, determine whether the abnormal temperature field mode has been alleviated, and if not, re-identify the dynamic interference event characteristics and update the furnace space association graph until the abnormal temperature field mode is eliminated, and extract the final simulation parameters as the incinerator temperature field optimization scheme.

[0180] The embedded vectors of the graph nodes corresponding to the temperature field response data before and after correction are compared to analyze the differences and assess whether the abnormal temperature field pattern has been mitigated. If the abnormal pattern has not been mitigated, the steps of dynamic interference event identification and furnace space correlation graph update need to be repeated, and the optimization process is repeated until the abnormal temperature field pattern is eliminated. Finally, the final simulation parameters are extracted as the optimization scheme for the incinerator temperature field.

[0181] Step S161: Calculate the difference in embedding vectors of corresponding nodes before and after correction at the same time point. The difference calculation uses a vector similarity measurement method, such as cosine similarity or Euclidean distance, to obtain a sequence of difference values.

[0182] For each node in the furnace spatial association diagram, at the same time point (e.g., the same moment after the disturbance occurs), the node embedding vector before and after correction are extracted respectively. Cosine similarity is used to calculate the similarity between the two vectors; the closer the cosine similarity is to 1, the more similar the vectors are. Alternatively, Euclidean distance is used to calculate the spatial distance between the two vectors; the smaller the distance, the more similar the vectors are.

[0183] The calculated similarity or distance values ​​are organized by node and time point to form a difference value sequence, which reflects the degree of influence of the correction measures on the embedding vector of each node.

[0184] Step S162: Based on the difference value sequence, analyze the changes in node embedding features corresponding to the abnormal temperature field pattern. If the difference value of the abnormal features exceeds the preset improvement threshold, it is determined that the abnormal temperature field pattern has been alleviated; otherwise, it is determined that it has not been alleviated.

[0185] The preset improvement threshold is set according to the temperature field optimization target. For example, for cosine similarity, a threshold is set. When the similarity between the corrected embedding vector and the normal pattern feature vector increases by more than the threshold, it indicates that the abnormal features have been improved.

[0186] Analyze the changes in the embedding characteristics of nodes related to the abnormal temperature field pattern (such as previously identified abnormal temperature nodes) in the difference value sequence. If the difference value of these nodes exceeds the preset improvement threshold, it indicates that the correction measures are effective and the abnormal temperature field pattern has been alleviated; if the difference value does not reach the threshold, it is determined that the abnormal pattern has not been alleviated.

[0187] Step S163: If the abnormal temperature field pattern is not alleviated, return to step S110 to re-identify the dynamic interference event characteristics, and update the edge weights (based on the latest heat conduction data) and node attributes of the furnace space association graph, and repeat the subsequent steps.

[0188] If the abnormal temperature field pattern does not improve, it indicates that the previous identification of the disturbance event may have been inaccurate, or the furnace space correlation diagram may not accurately reflect the current heat conduction state. In this case, return to step S110, re-collect operating data, and identify the characteristics of the dynamic disturbance event. It may be necessary to adjust the parameters of the anomaly detection (such as the reconstruction error threshold) to improve the identification accuracy.

[0189] Simultaneously, based on the latest temperature field data and operating status, the thermal conductivity coefficients (edge ​​weights) between adjacent nodes in the furnace space correlation diagram are updated, as well as the geometric attributes or basic characteristics of the nodes (such as changes in node attributes due to minor changes in the furnace structure), to make the correlation diagram more consistent with the current actual situation. Then, steps S120 to S150 are repeated until the abnormal temperature field pattern is alleviated.

[0190] Step S164: If the abnormal temperature field mode is determined to be eliminated, extract the simulation parameters in the current CFD simulation system, including heat source parameters, fluid flow parameters and heat exchange parameters, and organize them into an optimization scheme for the incinerator temperature field. The scheme includes the specific values ​​of the parameters, the applicable interference scenarios and the expected temperature field improvement effect.

[0191] Once the abnormal temperature field mode is eliminated (i.e., all node embedding features related to the abnormal mode reach the normal range), extract the various simulation parameters currently used in the CFD simulation system, including heat source parameters such as the burner's fuel flow rate and temperature setpoint; fluid flow parameters such as the fan's air volume and speed; and heat exchange parameters such as the cooling medium's flow rate and temperature.

[0192] These parameters are compiled into an incinerator temperature field optimization scheme. The scheme specifies the exact values ​​for each parameter, explains the applicable interference scenarios (such as temperature anomalies caused by sudden changes in feed composition), and describes the expected temperature field improvement effects based on simulation results (such as the degree of improvement in temperature field uniformity and the reduction in temperature fluctuation amplitude). This incinerator temperature field optimization scheme can serve as a basis for adjusting incinerator operating parameters and guide on-site operations.

[0193] Figure 2 The illustration shows exemplary hardware and software components of a CFD numerical simulation-based incinerator temperature field optimization system 100, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used on the CFD numerical simulation-based incinerator temperature field optimization system 100 and to perform the functions in this application.

[0194] The CFD numerical simulation-based incinerator temperature field optimization system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the CFD numerical simulation-based incinerator temperature field optimization method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0195] For example, the CFD numerical simulation-based incinerator temperature field optimization system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the CFD numerical simulation-based incinerator temperature field optimization system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The CFD numerical simulation-based incinerator temperature field optimization system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0196] For ease of explanation, only one processor is described in the CFD numerical simulation-based incinerator temperature field optimization system 100. However, it should be noted that the CFD numerical simulation-based incinerator temperature field optimization system 100 of this application may also include multiple processors. Therefore, the steps executed by one processor as described in this application may also be executed jointly by multiple processors or individually. For example, if the processor of the CFD numerical simulation-based incinerator temperature field optimization system 100 executes steps A and B, it should be understood that steps A and B may also be executed jointly by two different processors or individually by one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0197] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned incinerator temperature field optimization method based on CFD numerical simulation is implemented.

[0198] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for optimizing the temperature field of an incinerator based on CFD numerical simulation, characterized in that, The method includes: Identify dynamic interference events during the operation of the incinerator, and extract event feature information of the dynamic interference events, including interference type, interference occurrence time and interference impact range; A spatial relationship graph of the incinerator furnace is constructed, with each spatial unit of the furnace as a graph node and the heat conduction relationship between each spatial unit of the furnace as the graph edge weight. The event feature information is embedded into the corresponding graph node. The graph neural network is used to learn the relationship between the interference features and the temperature field anomaly, and the temperature anomaly relationship link corresponding to the interference event is output. The temperature anomaly correlation link is input into the CFD simulation system to generate a target simulation task for the correlation link. The target simulation task includes the target simulation area, simulation duration and key monitored temperature indicators. The target simulation task is executed to obtain dynamic response data of the temperature field of the associated links. The graph node embedding changes corresponding to the dynamic response data of the temperature field are analyzed by graph neural network to determine the correlation strength between the temperature field response and the disturbance event, and then the simulation correction direction that needs to be adjusted is determined. Based on the simulation correction direction, a CFD simulation parameter adjustment instruction is generated. The CFD simulation parameter adjustment instruction is input into the CFD simulation system to update the simulation parameters, and the target simulation task is re-executed to obtain the corrected temperature field response data. By comparing the differences in graph node embeddings corresponding to the temperature field response data before and after correction, it is determined whether the abnormal temperature field mode has been alleviated. If it has not been alleviated, the dynamic interference event characteristics are re-identified and the furnace space correlation graph is updated until the abnormal temperature field mode is eliminated. Finally, the simulation parameters are extracted as the incinerator temperature field optimization scheme.

2. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 1, characterized in that, The process of identifying dynamic disturbance events during the operation of the incinerator and extracting event feature information of the dynamic disturbance events includes: Operational data is collected at a preset frequency by each monitoring point in the incinerator operation status monitoring network. The operational data includes feed data, burner data, flue gas data, and furnace wall data. Using each monitoring point as the initial node, and the standardized spatial distance and data correlation between monitoring points as the graph edge weights, a spatial correlation graph of monitoring points is constructed. Anomaly detection of nodes is performed on the spatial correlation graph of monitoring points through a graph neural network to capture abnormal nodes. The duration of data fluctuations at abnormal nodes is statistically analyzed. Combined with the incinerator's operating mechanism, the start and end times of the interference corresponding to the abnormal nodes are determined to form the interference occurrence period. Based on the distribution location of abnormal nodes and the amplitude of data fluctuations, determine the spatial boundary of the interference impact range, and simultaneously determine the type of interference based on the data type of the monitoring data corresponding to the abnormal nodes; Establish dynamic interference event classification rules, and classify dynamic interference events into different priority levels based on the number of covered nodes in the interference impact range, the duration of the interference occurrence period, and the data fluctuation amplitude of abnormal nodes; By integrating the interference type, the time period of interference, the scope of interference impact, and the interference priority level, an event feature information set containing priority information is formed.

3. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 2, characterized in that, The process involves constructing a spatial association graph of monitoring points, using each monitoring point as an initial node and the standardized spatial distance and data correlation between monitoring points as graph edge weights. Anomaly detection is then performed on this spatial association graph using a graph neural network to capture anomalous nodes, including: The distribution of monitoring points in the mobile incinerator operation status monitoring network is determined by using each monitoring point as an initial node in the spatial association diagram of the monitoring points, and recording the spatial coordinates and monitoring data types of each initial node. Calculate the standardized spatial distance between any two initial nodes, and combine it with the historical correlation coefficient of the monitoring data of the two initial nodes. Use the product of the inverse of the standardized spatial distance and the historical correlation coefficient as the graph edge weight between the two initial nodes. Based on the initial nodes and graph edge weights, a spatial association graph of monitoring points is constructed. The graph autoencoder in the graph neural network is used to pre-train the spatial association graph of monitoring points to learn the embedding vector of each initial node under normal operating conditions. Real-time collection of operational data from each monitoring point, conversion of operational data into real-time node feature vectors, input of the pre-trained graph autoencoder, and calculation of the reconstruction error between the real-time node embedding vector and the initial node embedding vector under normal operating conditions; Set a reconstruction error threshold, mark nodes whose reconstruction errors exceed the threshold as candidate anomalous nodes, verify the neighboring nodes of the candidate anomalous nodes, and check whether the reconstruction errors of the neighboring nodes of the candidate anomalous nodes also exceed the reconstruction error threshold. If the reconstruction error of more than a preset proportion of the neighboring nodes of a candidate abnormal node exceeds the reconstruction error threshold, then the candidate abnormal node is determined to be a final abnormal node, and all final abnormal nodes are extracted to form an abnormal node set.

4. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 1, characterized in that, The construction of the incinerator furnace space relation graph uses each space unit in the furnace as a graph node and the heat conduction relationship between units as the graph edge weight. The event feature information is embedded into the corresponding graph nodes. A graph neural network is used to learn the correlation between disturbance features and temperature field anomalies, outputting the temperature anomaly correlation link corresponding to the disturbance event, including: The incinerator furnace is divided into multiple spatial units, and each spatial unit is used as the basic node of the furnace spatial relationship diagram. The spatial coordinates and geometric attributes of each basic node are recorded. Based on the heat conduction mechanism of the incinerator, the heat conduction coefficient between adjacent spatial units is calculated, and this heat conduction coefficient is used as the graph edge weight between the corresponding basic nodes in the furnace space association graph to construct the initial furnace space association graph. Determine the node features corresponding to each temperature field anomaly mode to form an anomaly mode node feature library; The event feature information is used as an interference feature vector and embedded into the corresponding basic node in the initial furnace space association graph. A graph convolutional neural network is used to train the furnace space association graph with the embedded interference feature vector to learn the mapping relationship between the interference feature vector and the features in the abnormal mode node feature library, thus forming an interference temperature association learning model. The interference temperature correlation learning model is iteratively optimized by using historical interference temperature data, so that the interference temperature correlation learning model can output a set of temperature anomaly nodes corresponding to the interference event. The furnace functional links corresponding to the set of temperature anomaly nodes are identified as temperature anomaly correlation links. When the interference temperature correlation learning model outputs temperature anomaly correlation links, it simultaneously outputs the correlation confidence score. The correlation confidence score is calculated by the similarity of graph node embedding. If the correlation confidence score is lower than the preset correlation confidence score threshold, feature embedding and model training are re-executed.

5. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 4, characterized in that, The method employs a graph convolutional neural network to train a furnace space correlation graph embedded with interference features, learning the mapping relationship between interference feature vectors and features in the abnormal pattern node feature library, forming an interference temperature correlation learning model, including: Construct a training dataset, which includes interference feature vectors of historical interference events, corresponding furnace space correlation diagrams, and a set of manually labeled temperature anomaly nodes; The furnace space association graph in the training dataset is input into a graph convolutional neural network. The node features are aggregated through the graph convolutional layer to generate node embedding vectors. The process of aggregating node features through the graph convolutional layer combines the weighted summation of the node's own features and the features of its neighboring nodes. In the fully connected layer of the graph convolutional neural network, the node embedding vector is compared with the features in the abnormal pattern node feature library, the similarity loss between the node embedding vector and the features in the abnormal pattern node feature library is calculated, and the gradient descent algorithm is used to optimize the network parameters of the graph convolutional neural network. An attention mechanism layer is introduced to assign attention weights to the node embedding vectors output by the graph convolutional layer, so that the graph convolutional neural network pays more attention to the node features related to temperature anomalies. The training dataset is divided into a training subset and a validation subset. The network parameters of the graph convolutional neural network are trained using the training subset, and the matching degree between the set of temperature anomaly nodes output by the graph convolutional neural network and the set of temperature anomaly nodes labeled by humans is verified using the validation subset. If the matching degree is lower than the preset matching degree standard, adjust the number of layers, kernel size and attention weight calculation method of the graph convolutional neural network, and retrain the graph convolutional neural network; If the matching degree reaches the preset matching degree standard, training stops, and a stable interference temperature correlation learning model is formed.

6. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 1, characterized in that, The step of inputting the temperature anomaly correlation link into the CFD simulation system and generating a target simulation task for the correlation link includes: The CFD simulation system is invoked to receive temperature anomaly correlation links and corresponding correlation confidence levels. If the correlation confidence level is higher than a preset correlation confidence threshold, the target simulation task generation process is initiated. The nodes corresponding to the temperature anomaly correlation link in the furnace space correlation diagram are ranked by importance using a graph neural network. The top K nodes and their adjacent nodes are extracted to determine the spatial range of the target simulation area. The spatial range of the target simulation area includes the spatial units corresponding to potential nodes where the interference may spread. Based on the interference occurrence time and interference duration characteristics in the event feature information, the simulation duration is set to a time span, which includes the baseline stability period before the interference occurs, the interference occurrence period, and the recovery period after the interference is eliminated. Based on the node characteristics of the temperature field anomaly pattern, key monitoring temperature indicators are determined. These key monitoring temperature indicators include the core temperature parameters corresponding to the temperature field anomaly pattern and the associated derived temperature parameters. At the same time, the monitoring frequency for different simulation periods is set, with the monitoring frequency during the interference period being higher than that during other simulation periods. CFD simulation resources are allocated based on interference priority levels to generate resource-optimized target simulation tasks.

7. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 1, characterized in that, The process of executing the target simulation task involves acquiring dynamic temperature field response data of the associated components, analyzing the changes in graph node embeddings corresponding to the dynamic temperature field response data using a graph neural network, determining the correlation strength between the temperature field response and the disturbance event, and then determining the simulation correction direction that needs to be adjusted, including: The CFD simulation system is invoked to load the geometric model of the target simulation region corresponding to the target simulation task, import the initial running parameters, and start the simulation operation according to the set time step. During the simulation, temperature data corresponding to each node in the target simulation area is collected according to the monitoring frequency to form a dynamic response data sequence of the temperature field, and the disturbance simulation state at each time node is recorded synchronously. The dynamic response data sequence of the temperature field is embedded into the corresponding node of the furnace space correlation graph to generate a dynamic node embedding vector. The change of the dynamic node embedding vector before and after the disturbance is calculated by the node embedding update algorithm of the graph neural network. A correlation strength analysis dimension is constructed, which includes a time dimension, a spatial dimension, and a feature dimension. The time dimension analyzes the synchronicity between the occurrence of interference and the change of dynamic node embedding vector. The spatial dimension analyzes the overlap between abnormal nodes and nodes affected by interference. The feature dimension analyzes the matching degree between the change of dynamic node embedding vector and interference features. The association scores of each association strength analysis dimension are calculated by graph neural network. The association scores of each association strength analysis dimension are standardized and then weighted and summed according to preset weights to obtain the comprehensive association strength. If the comprehensive association strength is lower than the preset comprehensive association strength threshold, the parameter settings of the target simulation task are re-examined. If the overall correlation strength is higher than the preset overall correlation strength threshold, the simulation correction direction is determined based on the temperature data characteristics corresponding to the change in the dynamic node embedding vector and the optimization target of the incinerator temperature field. The simulation correction direction includes the heat source adjustment direction, the fluid flow adjustment direction, and the heat exchange adjustment direction.

8. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 7, characterized in that, If the overall correlation strength is higher than a preset overall correlation strength threshold, based on the temperature data characteristics corresponding to the change in the dynamic node embedding vector, and combined with the incinerator temperature field optimization objective, the simulation correction direction is determined, including: Extract temperature data features corresponding to the changes in the embedding vector of dynamic nodes. The temperature data features include the location of temperature anomaly nodes, the magnitude of temperature changes, the frequency of temperature fluctuations, and the temperature gradient distribution between nodes. Obtain the optimization objectives for the incinerator temperature field, which include the objectives for temperature field uniformity, temperature stability, and temperature range compliance. If the temperature data characteristics show that the temperature of a local node is higher than the temperature range corresponding to the temperature range compliance target, and the correlation strength analysis shows that the embedding change of the local node is significantly correlated with the embedding change of the heat source input node, then the simulation correction direction is determined to be the heat source adjustment direction, specifically, reducing the heat source input intensity corresponding to the heat source input node. If the temperature data characteristics show that the temperature gradient between nodes exceeds the temperature gradient range corresponding to the temperature field uniformity target, and the correlation strength analysis shows that the temperature gradient change is strongly correlated with the node embedding change corresponding to the fluid flow path, then the simulation correction direction is determined to be the fluid flow adjustment direction, specifically to optimize the flow velocity parameters of the node region corresponding to the fluid flow path. If the temperature data characteristics show that the node temperature change rate exceeds the temperature change rate range corresponding to the temperature stability target, and the correlation strength analysis shows that the temperature change rate is strongly correlated with the embedding change of the heat exchange node, then the simulation correction direction is determined to be the heat exchange adjustment direction, specifically adjusting the heat exchange component parameters corresponding to the heat exchange node. The feasibility of the determined simulation correction direction is verified. The simulation correction direction is kept within the achievable range by combining the parameter adjustment range of the CFD simulation system and the actual operation limitations of the incinerator. If it is not feasible, the temperature data characteristics corresponding to the change in the dynamic node embedding vector are re-analyzed and the simulation correction direction is adjusted.

9. The incinerator temperature field optimization method based on CFD numerical simulation according to claim 1, characterized in that, The process of generating CFD simulation parameter adjustment instructions based on the simulation correction direction, inputting the CFD simulation parameter adjustment instructions into the CFD simulation system to update the simulation parameters, re-executing the target simulation task, and obtaining the corrected temperature field response data includes: Establish a mapping relationship library between simulation correction directions and CFD parameters. The mapping relationship library records the core parameter types and parameter adjustment constraints corresponding to different simulation correction directions. The constraints are classified into physical constraints and numerical constraints. Physical constraints include the range of safe operating parameters of the equipment, and numerical constraints include simulation stability requirements. Based on the determined simulation correction direction, the corresponding core parameter type is extracted from the mapping relationship library, the current value of the core parameter type in the CFD simulation system is obtained, and the reasonable range of core parameter type adjustment is determined by combining the change of dynamic node embedding vector corresponding to the dynamic response data of the temperature field. Generate a CFD simulation parameter adjustment instruction. The CFD simulation parameter adjustment instruction includes a target parameter identifier, adjustment direction, adjustment range, and verification rules for the adjusted parameter values, so that the adjusted parameter values ​​meet the constraints. Import CFD simulation parameter adjustment commands through the parameter update interface of the CFD simulation system, automatically replace the original parameter values, and back up the parameter configuration before adjustment. Call the basic configuration of the target simulation task, keep the target simulation area, simulation duration and monitoring frequency unchanged, start the simulation operation again, collect temperature data at the set frequency, form the corrected temperature field response data, and synchronously record the embedding vector of each node after correction.

10. A temperature field optimization system for an incinerator based on CFD numerical simulation, characterized in that, The device includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the incinerator temperature field optimization method based on CFD numerical simulation as described in any one of claims 1-9.

Images