Artificial intelligence abnormal heat supply troubleshooting method and system

Abstract

The invention provides an artificial intelligence abnormal heat supply troubleshooting method and system which are applied to the technical field of data processing, and the method comprises the steps: obtaining heat supply data; based on the heat supply data and a preset artificial intelligence algorithm, determining high-order spatial-temporal characteristics of the heat supply data; acquiring a sparse coupling factor matrix corresponding to thermal expansion of the heat supply pipeline; based on the high-order spatial-temporal characteristics of the heat supply data and the sparse coupling factor matrix corresponding to the thermal expansion of the heat supply pipeline, obtaining an abnormal probability matrix of the heat supply pipeline and a propagation path set of the heat supply pipeline; and obtaining the damaged position of the heat supply pipeline based on the abnormal probability matrix of the heat supply pipeline and the propagation path set of the heat supply pipeline. According to the invention, the abnormity positioning accuracy is improved, the troubleshooting time is shortened, the operation and maintenance cost is reduced, and the troubleshooting efficiency and effect are improved.

Description

Technical Field

[0001] The present invention relates to data processing technologies, and in particular, to a method and system for detecting abnormal heat supply in artificial intelligence. Background Art

[0002] During the operation of a centralized heat supply system, due to factors such as complex pipe network structures, environmental interference, and equipment aging, problems of abnormal heat supply often occur, such as thermal imbalance, pipeline leakage, air blockage, or dirt blockage. These problems not only lead to energy waste (such as hot water loss caused by pipeline leakage) and uneven room temperatures for users, but may also cause equipment failures and even safety accidents. For example, the "high hysteresis" characteristic of the heat supply system (i.e., the adjustment response lags behind the actual demand change) can cause data deviation normalization, masking systematic abnormalities and making it difficult for traditional detection methods to accurately identify the root cause of the problem. In addition, factors such as corrosion of old pipe networks, unauthorized water discharge by users, and design and construction defects further exacerbate the complexity of abnormal heat supply.

[0003] Currently, existing methods for detecting abnormal heat supply have a high degree of manual dependence and low automation, resulting in inaccurate positioning and low efficiency, thereby affecting the detection efficiency and effect. Summary of the Invention

[0004] Based on the above problems, the present invention is proposed to provide a method and system for detecting abnormal heat supply in artificial intelligence that overcomes the above problems or at least partially solves the above problems.

[0005] According to one aspect of the present invention, there is provided a method for detecting abnormal heat supply in artificial intelligence, including the following steps: Obtain heat supply data; Based on the heat supply data and a preset artificial intelligence algorithm, determine the high-order spatio-temporal characteristics of the heat supply data; Obtain a sparse coupling factor matrix corresponding to the thermal expansion of the heat supply pipeline; Based on the high-order spatio-temporal characteristics of the heat supply data and the sparse coupling factor matrix corresponding to the thermal expansion of the heat supply pipeline, obtain an abnormal probability matrix of the heat supply pipeline and a set of propagation paths of the heat supply pipeline; Based on the abnormal probability matrix of the heat supply pipeline and the set of propagation paths of the heat supply pipeline, obtain the damaged location of the heat supply pipeline.

[0006] Optionally, based on the heat supply data and a preset artificial intelligence algorithm, determining the high-order spatio-temporal characteristics of the heat supply data includes: Input the heat supply data into the preset artificial intelligence algorithm to obtain a spatio-temporal encoding matrix and an adjacency matrix; Based on the spatio-temporal encoding matrix and the adjacency matrix, generate a sequence of dynamic spatio-temporal propagation graphs; Based on the sequence of dynamic spatio-temporal propagation graphs, determine the high-order spatio-temporal characteristics of the heat supply data.

[0007] Optionally, input the heating data into a preset artificial intelligence algorithm to obtain a spatio-temporal coding matrix and an adjacency matrix, including: Use the preset artificial intelligence algorithm to process the heating data, establish a topological graph of the heating pipe network, and generate an adjacency matrix based on the topological graph of the heating pipe network; Align the heating data in time series according to a time window to obtain aligned data; Concatenate the historical fault labels with the aligned data to obtain a spatio-temporal coding matrix.

[0008] Optionally, generate a dynamic spatio-temporal propagation graph sequence based on the spatio-temporal coding matrix and the adjacency matrix, including: Calculate the similarity between the data in the spatio-temporal coding matrix; Determine the fusion weight based on the similarity between the data in the spatio-temporal coding matrix and the historical propagation pattern; Determine the target edge set of the adjacency matrix based on the fusion weight and a preset threshold; Generate a dynamic spatio-temporal propagation graph sequence based on the target edge set.

[0009] Optionally, determine the high-order spatio-temporal features of the heating data based on the dynamic spatio-temporal propagation graph sequence, including: Obtain the spatial graph convolution and temporal convolution of the dynamic spatio-temporal propagation graph sequence; Perform gated fusion on the spatial graph convolution and temporal convolution of the dynamic spatio-temporal propagation graph sequence to obtain high-order spatio-temporal features.

[0010] Optionally, obtain a sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, including: Perform coupling calculation on the high-order spatio-temporal features of the heating data to obtain an initial coupling factor matrix; Perform topological correction on the initial coupling factor matrix through the thermal expansion of the heating pipeline to obtain a corrected coupling factor matrix; Normalize the corrected coupling factor matrix to obtain a sparse coupling factor matrix.

[0011] Or, Calculate the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data; Extract the diagonal elements from the spatio-temporal coding matrix, and multiply the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data by the diagonal elements in the spatio-temporal coding matrix to obtain a set of first quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data; Calculate the second quantile corresponding to the covariance matrix corresponding to the thermal expansion of the heating pipeline; Add the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data to the second quantile corresponding to the covariance matrix of the thermal expansion of the heating pipeline to obtain the target quantile; Multiply the target quantile by each quantile in the first quantile set of the covariance matrix of the high-order spatio-temporal features of the heating data to obtain the second quantile set of the covariance matrix of the high-order spatio-temporal features of the heating data; Replace the central element in the transposed matrix of the spatio-temporal encoding matrix and the central element in the covariance matrix corresponding to the thermal expansion of the heating pipeline with the maximum quantile in the second quantile set, and add the transposed matrix of the spatio-temporal encoding matrix after replacement to the covariance matrix corresponding to the thermal expansion of the heating pipeline after replacement to obtain the sparse coupling factor matrix.

[0012] Optionally, based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, obtain the target anomaly probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline, including: Perform temporal anomaly detection on the high-order spatio-temporal features of the heating data to obtain the initial anomaly probability matrix of the heating pipeline; Add the initial anomaly probability matrix of the heating pipeline to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the target anomaly probability matrix of the heating pipeline; Perform spatial propagation enhancement and path backtracking on the target anomaly probability matrix of the heating pipeline to obtain the set of propagation paths of the heating pipeline.

[0013] Or, Extract the high-order spatio-temporal features of the heating pipeline from the high-order spatio-temporal features of the heating data; Obtain the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline, and obtain the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; Merge the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline with the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the covariance feature set of the heating pipeline; Select the set composed of the elements greater than the maximum value element in the transfer matrix of the high-order spatio-temporal features of the heating pipeline from the covariance feature set of the heating pipeline as the initial anomaly probability matrix of the heating pipeline; Correct the probability values in the initial anomaly probability matrix of the heating pipeline through the target elements in the covariance feature set of the heating pipeline to obtain the target anomaly probability matrix of the heating pipeline, where the target elements in the covariance feature set of the heating pipeline refer to the elements in the middle value of the covariance feature set of the heating pipeline; Perform spatial propagation enhancement and path backtracking on the target abnormal probability matrix of the heating pipeline to obtain the set of propagation paths of the heating pipeline.

[0014] Optionally, based on the target abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline, obtain the damaged location of the heating pipeline, including: Search for the probability in the abnormal probability matrix of the heating pipeline along the propagation paths in the set of propagation paths of the heating pipeline. When the maximum probability is reached, obtain the target node, and thus determine the damaged location of the heating pipeline from the target node.

[0015] According to another aspect of the present invention, there is provided an abnormal heating troubleshooting system for artificial intelligence, including: A first acquisition module for acquiring heating data; A feature determination module for determining the high-order spatio-temporal features of the heating data based on the heating data and a preset artificial intelligence algorithm; A second acquisition module for acquiring the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; A third acquisition module for acquiring the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the heating pipeline; A location determination module for obtaining the damaged location of the heating pipeline based on the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline.

[0016] According to the solution of the present invention, in the present invention, first acquire heating data, then determine the high-order spatio-temporal features of the heating data based on the heating data and a preset artificial intelligence algorithm, and then acquire the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, so as to obtain the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, and further obtain the damaged location of the heating pipeline based on the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline. The present invention improves the abnormal positioning accuracy, shortens the troubleshooting time, reduces the operation and maintenance cost, and further improves the troubleshooting efficiency and effect. Description of the Drawings

[0017] Figure 1 Shows a flowchart of the abnormal heating troubleshooting method for artificial intelligence according to an embodiment of the present invention; Figure 2 Shows a structural block diagram of the abnormal heating troubleshooting system for artificial intelligence according to an embodiment of the present invention. Detailed Embodiment

[0018] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0019] To solve the problems existing in the above-mentioned prior art, the solution of the present invention is proposed. An embodiment of the present invention provides a method and device for patent document management based on artificial intelligence.

[0020] As Figure 1 shown, an abnormal heating troubleshooting method based on artificial intelligence proposed in this embodiment includes the following steps: Step S101: Obtain heating data.

[0021] Among them, the heating data includes but is not limited to supply water temperature, return water pressure, flow rate, time, fault label, etc.

[0022] Step S102: Based on the heating data and a preset artificial intelligence algorithm, determine the high-order spatio-temporal features of the heating data.

[0023] Optionally, based on the heating data and a preset artificial intelligence algorithm, determining the high-order spatio-temporal features of the heating data includes: inputting the heating data into the preset artificial intelligence algorithm to obtain a spatio-temporal encoding matrix and an adjacency matrix; generating a dynamic spatio-temporal propagation graph sequence based on the spatio-temporal encoding matrix and the adjacency matrix; determining the high-order spatio-temporal features of the heating data based on the dynamic spatio-temporal propagation graph sequence.

[0024] The preset artificial intelligence algorithm is spatio-temporal propagation graph convolution.

[0025] Among them, inputting the heating data into the preset artificial intelligence algorithm to obtain a spatio-temporal encoding matrix and an adjacency matrix includes: using the preset artificial intelligence algorithm to process the heating data, establishing a heating pipe network topology graph, and generating an adjacency matrix based on the heating pipe network topology graph; performing time series alignment on the heating data according to a time window to obtain aligned data; splicing the historical fault label with the aligned data to obtain a spatio-temporal encoding matrix.

[0026] Exemplarily, first establish a heating pipe network topology graph and an adjacency matrix: Taking the heating pipe network of a certain community as an example, the pipe network includes 100 heat exchange station nodes, and the connection relationship of the pipe network is drawn through a geographic information system (GIS) to form a heating pipe network topology graph. Among them, if the nodes are directly connected, the corresponding position of the adjacency matrix is set to 1 (for example, if heat exchange station B and heat exchange station C are directly connected by a pipeline, the value of the p-th row and q-th column in the adjacency matrix is 1), otherwise it is 0.

[0027] Then, align the heating data in time series according to the time window to obtain the aligned data: Set the time window to 15 minutes, collect the real-time data of each node (such as supply water temperature, return water pressure, flow rate, etc.), and align them according to the timestamps to form a time series, that is, the aligned data. For example, from 0:00 to 24:00 on January 1, 2024, there are 96 time points in total, and each time point corresponds to 10-dimensional data (temperature, pressure, flow rate, etc.) of 100 nodes, forming a three-dimensional dataset of 96×100×10. After that, splice the historical fault labels with the aligned data to obtain the spatio-temporal coding matrix: If a pipeline leakage fault occurs at a heat exchange station at 8:15 on January 1, 2024, splice the fault label (such as the code "leakage" corresponding to 01) at this time point to the time series data of the corresponding node, forming a spatio-temporal coding matrix with a dimension of 96×100×(10 + 2) (2 dimensions are the fault label codes).

[0028] The embodiments of this application clearly present the pipe network structure: Establish a heating pipe network topology map and an adjacency matrix, which can intuitively display the connection relationship of each heat exchange station node in the heating pipe network, facilitate the staff to quickly understand the pipe network layout, and provide basic network structure information for subsequent fault location and analysis. Unify the data format and time standard: Align the heating data in time series by setting the time window, and organize various heating data (such as supply water temperature, return water pressure, flow rate, etc.) at different times and different nodes into a unified three-dimensional dataset format, making the data consistent in time and dimension, facilitating subsequent analysis and processing, and improving the availability and accuracy of the data. Integrate fault information with data: Splice the historical fault labels with the aligned data to form a spatio-temporal coding matrix, closely combine the fault information with the heating data at the corresponding time points and nodes, provide rich feature information for fault prediction, diagnosis, etc. based on data mining and machine learning, help improve the accuracy of fault identification and location, and better ensure the stable operation of the heating system.

[0029] Among them, based on the spatio-temporal coding matrix and the adjacency matrix, generate a sequence of dynamic spatio-temporal propagation graphs, including: calculating the similarity between the data in the spatio-temporal coding matrix; determining the fusion weight based on the similarity between the data in the spatio-temporal coding matrix and the historical propagation mode; determining the target edge set of the adjacency matrix based on the fusion weight and a preset threshold; generating a sequence of dynamic spatio-temporal propagation graphs based on the target edge set.

[0030] Exemplarily, calculate the similarity between the data in the spatio-temporal coding matrix: For the node data at adjacent time points in the spatio-temporal coding matrix (such as the temperature and pressure data of a heat exchange station at time t and time t + 1), calculate the similarity using the Euclidean distance:

[0031] Among them, They are the feature vectors of nodes i and j respectively, and n is the feature dimension (for example, 12 dimensions, including 10 - dimensional operation data + 2 - dimensional fault labels).

[0032] Then determine the fusion weight: Combine the historical fault propagation records (for example, when heat exchange station A fails in the past year, there is an 80% probability that heat exchange station B will show abnormalities subsequently), and correct the similarity weight: Fusion weight = , where, = 0.6 is the empirical coefficient to balance the influence of real - time data and historical patterns.

[0033] Finally, screen the target edge set and generate a sequence of dynamic spatio - temporal propagation graphs: Set the preset threshold to 0.7, and only retain the edges with a fusion weight ≥ 0.7 (for example, if the weight between heat exchange station A and B is 0.8, then retain edge A - B). Generate a sequence of dynamic spatio - temporal propagation graphs according to the time window (15 minutes), and each time point corresponds to a spatio - temporal propagation graph containing node connection relationships.

[0034] The embodiments of this application accurately capture data associations: By calculating the similarity between data in the spatio - temporal encoding matrix, it can accurately measure the similarity between data at different time points and different nodes, discover the potential connections between the operating states of each node in the heating system in terms of time and space, and provide quantitative association information for subsequent analysis. Combine historical and real - time data: Determine the fusion weight based on data similarity and historical propagation patterns, which not only considers the current system state reflected by real - time data but also incorporates the empirical knowledge of historical fault propagation, making the analysis results more reliable and forward - looking, and can more comprehensively reflect the operating rules and fault propagation characteristics of the heating system. Effectively screen key connections: Determine the target edge set of the adjacency matrix based on the fusion weight and preset threshold, which can filter out those connections with weak associations and less impact on system operation and fault propagation, highlight the connections between key nodes, simplify the network structure, and at the same time focus more on the parts that have an important impact on the system, helping to improve the efficiency of fault location and analysis. Intuitively display system dynamics: The generated sequence of dynamic spatio - temporal propagation graphs visually presents the node connection relationships and state changes of the heating system at different time points, enabling staff to intuitively observe the dynamic evolution process of the system, quickly discover abnormal changes and potential fault propagation paths, providing a powerful decision - making support tool for the monitoring and management of the heating system, and thus improving the monitoring efficiency and effect.

[0035] Among them, based on the sequence of dynamic spatio - temporal propagation graphs, determine the high - order spatio - temporal features of heating data, including: Obtain the spatial graph convolution and temporal convolution of the sequence of dynamic spatio - temporal propagation graphs; perform gated fusion on the spatial graph convolution and temporal convolution of the sequence of dynamic spatio - temporal propagation graphs to obtain high - order spatio - temporal features.

[0036] Exemplarily, for the spatio-temporal propagation graph at each time point, a graph convolutional network is used to extract spatial features: Among them, is the adjacency matrix + identity matrix, is the degree matrix, is the node feature of the l-th layer, is the weight matrix, which extracts the spatial dependencies between nodes (such as the pressure correlation between adjacent heat exchange stations). The temporal features of each node (such as the temperature sequence in the past 24 hours) are input into the temporal convolutional network to capture the dynamic changes in the temporal dimension, and the temporal feature vector is output (such as the temperature change trend of a certain node at time t).

[0037] The spatial and temporal features are fused through a gating mechanism:

[0038] Among them, is the output of GCN, is the output of TCN, is the weight of, is the weight of, is a constant, is element-wise multiplication, and finally, high-order features containing spatio-temporal coupling relationships are obtained.

[0039] The embodiments of the present application deeply capture spatio-temporal coupling relationships and enhance the feature representation ability: Spatial graph convolution (GCN) is based on the topological structure of the dynamic spatio-temporal propagation graph, which can effectively extract the spatial dependency relationships between nodes (such as the pressure and flow correlations between adjacent heat exchange stations), break the dependence on "Euclidean space data" of traditional methods, adapt to the physical connection characteristics of the non-regular graph structure of the heating pipe network, and accurately model the direct and indirect influences between nodes. Temporal convolution (TCN) focuses on the dynamic changes of temporal data (such as the fluctuation trends of temperature and pressure over time), captures the periodic and seasonal laws and short-term abnormal fluctuations in the operation of the heating system, and avoids the one-sidedness of relying only on data at a single time point. The gating fusion mechanism dynamically adjusts the importance of spatio-temporal features through adaptive weights (such as enhancing the weight of temporal features during sudden faults and emphasizing spatial correlations during the steady state of the pipe network), realizes the organic integration of spatio-temporal information, generates high-order features containing complex spatio-temporal coupling relationships, and avoids insufficient feature representation caused by the fragmentation of spatio-temporal information.

[0040] In addition, it is also adapted to dynamic complex systems to enhance the ability to recognize abnormal patterns: The operating state of the heating system is affected by multiple factors such as environmental temperature, user load, and equipment aging, and has strong dynamics and nonlinearity. High-order spatio-temporal features can capture the following key information: Spatial propagation patterns: For example, the influence path and intensity of the abnormal pressure at a certain node on the upstream and downstream nodes through the pipe network connection; Temporal evolution trends: For example, the abnormal downward trend of the temperature at the same node within consecutive time windows; Spatio-temporal interaction features: For example, the collective parameter deviation (which may indicate regional leakage or equipment failure) of a specific spatial area (such as a cluster of heat exchange stations in a certain area) within the same time window. These features provide richer inputs for subsequent anomaly detection and fault location, significantly improving the recognition accuracy of complex scenarios such as "multi-node collaborative anomalies" and "progressive fault development", and avoiding missed detections or misjudgments caused by single-dimensional features (such as only space or time).

[0041] Furthermore, it also supports multi-scale analysis to improve the system monitoring and decision-making efficiency: Spatial multi-scale: Graph convolution operations can aggregate neighbor node information layer by layer through multiple layers of networks (such as 1st-order neighbor direct connections and 2nd-order neighbor indirect connections), realizing multi-scale extraction from "single-node local features" to "pipe network global structure features", which is suitable for comprehensive analysis from microscopic node failures to macroscopic pipe network operating states. Temporal multi-scale: Temporal convolution can capture short-term fluctuations (minute level) and long-term trends (hour / day level) through different dilation rates (Dilated Convolution) or window sizes, adapting to the dual needs of the heating system for real-time monitoring (such as immediate leak detection) and trend prediction (such as equipment aging warning). The multi-scale characteristics of high-order features enable the system to quickly locate the specific node where the anomaly occurs (microscopic), while analyzing the propagation range and potential risks of the anomaly in the pipe network (macroscopic), providing accurate basis for dispatching maintenance resources and formulating emergency plans.

[0042] Even further, it provides high-quality inputs for intelligent algorithms to empower accurate fault location: The extracted high-order spatio-temporal features contain clear "fault-sensitive information" (such as the spatio-temporal feature patterns of historical fault nodes), which can be directly used as inputs for: Calculation of the anomaly probability matrix: By comparing the similarity between the current features and historical fault features, the anomaly probability of each node is quantified; Search for propagation paths: Combining the edge weights (fusion weights) of the dynamic spatio-temporal propagation graph, the key paths of anomaly propagation (such as the pipes corresponding to high-weight edges) are identified, realizing the accurate location logic of "searching for the maximum probability along the propagation path" (such as the method for determining the damaged location initially mentioned by the user). Compared with the original data or low-order features, high-order spatio-temporal features can significantly reduce the interference of data noise and improve the sensitivity of the algorithm to "weak anomaly signals" (such as the weak parameter changes in the early stage of slight pipeline leakage), achieving "early detection and early location" of faults.

[0043] In summary, through the extraction of high-order spatio-temporal features, the monitoring of the heating system has been upgraded from "single-node data monitoring" to "spatio-temporal correlation pattern analysis", providing more comprehensive and accurate information support for anomaly detection, fault location, and system optimization, and ultimately improving the reliability, operation efficiency, and operation and maintenance intelligence level of the heating system.

[0044] Step S103: Obtain the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline.

[0045] In one embodiment, obtaining the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline includes: performing coupling calculation on the high-order spatio-temporal features of the heating data to obtain an initial coupling factor matrix; performing topological correction on the initial coupling factor matrix through the thermal expansion of the heating pipeline to obtain a corrected coupling factor matrix; and performing normalization processing on the corrected coupling factor matrix to obtain the sparse coupling factor matrix.

[0046] Exemplarily, for the high-order spatio-temporal feature matrix (where N is the number of nodes and F is the feature dimension), calculate the coupling degree between nodes i and j:

[0047] Obtain the initial coupling factor matrix , which represents the feature similarity between nodes.

[0048] Then, use the thermal expansion A of the heating pipeline to filter the coupling of non-adjacent nodes (only retain the node pairs actually connected in the pipe network): If , ; otherwise, .

[0049] Finally, perform row normalization on the corrected matrix to obtain the sparse coupling factor matrix S:

[0050] Ensure that the sum of the elements in each row is 1, highlighting the influence weight between adjacent nodes.

[0051] The embodiments of this application integrate spatio-temporal features and physical topologies to improve the accuracy of coupling relationships: Combining data-driven and physical constraints: The initial coupling factor matrix calculates the feature similarity between nodes based on high-order spatio-temporal features (integrating the real-time operation data, historical failure modes, and spatio-temporal dynamic associations of nodes), reflecting the actual dependence relationship between nodes in the operating state; filtering non-adjacent nodes through the adjacency matrix to ensure that the coupling relationship only exists between physically connected nodes in the pipe network, avoiding non-real associations of nodes without actual connections, making the coupling relationship conform to both the dynamic characteristics of data-driven and strictly follow the physical topology structure of the heating pipe network, improving the physical interpretability and reliability of the model. Suppressing noise interference: Excluding the invalid coupling of non-adjacent nodes reduces the interference of irrelevant nodes on the propagation path analysis, focuses on the feature propagation on the real pipe connections, and makes the subsequent abnormal probability calculation and damage location positioning more accurate.

[0052] In addition, it highlights the influence weights of key nodes to enhance the robustness of the model: Normalization processing optimizes the weight distribution: By normalizing each row so that the sum of the elements in each row is 1, it clearly quantifies the relative influence degree between adjacent nodes (for example, when a node fails, the proportion of the coupling weight of its directly connected nodes is standardized), avoiding weight deviations caused by differences in feature dimensions, and facilitating subsequent propagation path search based on the probability matrix (such as giving priority to abnormal propagation of high-weight connections). Sparsification reduces the computational complexity: Only retaining the coupling relationships of physically connected nodes significantly reduces the number of non-zero elements in the matrix, reduces the data dimension while maintaining the key connection information, and improves the operation efficiency of subsequent algorithms (such as path search, probability propagation calculation), adapting to the high-efficiency requirements of real-time monitoring scenarios.

[0053] Furthermore, it provides a structured input for abnormal propagation analysis to support accurate positioning: Clarifying the influence strength between nodes: The sparse coupling factor matrix clearly depicts the association strength between adjacent nodes in the pipe network in spatio-temporal features (such as the conduction efficiency of temperature and pressure changes), providing a quantitative basis for modeling the propagation path of abnormal probability in the pipe network. For example, when an abnormality occurs at a certain node, the high-correlation adjacent nodes can be quickly located through the matrix weights, and the maximum value of the probability can be preferentially searched along the propagation path, shortening the positioning time of the damage location. Supporting dynamic propagation modeling: Combining the dynamic spatio-temporal propagation graph sequence and high-order spatio-temporal features, the sparse coupling factor matrix provides key parameters for constructing a dynamic model of abnormal propagation in the pipe network, enabling the model to capture the spatio-temporal evolution law of faults on physical connections (such as the conduction of pressure and flow changes from upstream to downstream nodes in a leakage fault), thereby more accurately locking the source of the abnormality or the damage location.

[0054] In summary, the sparse coupling factor matrix provides a key intermediate representation for the accurate location of the damage position of the heating pipeline by integrating spatio-temporal features and physical topology, optimizing the weight distribution, and reducing the computational complexity, supports the effective implementation from data-driven analysis to engineering practical applications, and improves the accuracy of location.

[0055] In another embodiment, obtaining the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline includes: calculating the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data; extracting the diagonal elements in the spatio-temporal coding matrix, and multiplying the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data by the diagonal elements in the spatio-temporal coding matrix to obtain a set of first quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data; calculating the second quantile corresponding to the covariance matrix corresponding to the thermal expansion of the heating pipeline; adding the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data to the second quantile corresponding to the covariance matrix corresponding to the thermal expansion of the heating pipeline to obtain the target quantile; performing dot multiplication on the target quantile and each quantile in the set of first quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data to obtain a set of second quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data; replacing the central element in the transposed matrix of the spatio-temporal coding matrix and the central element in the covariance matrix corresponding to the thermal expansion of the heating pipeline with the maximum quantile in the set of second quantiles, and adding the transposed matrix of the spatio-temporal coding matrix after replacement and the covariance matrix corresponding to the thermal expansion of the heating pipeline after replacement to obtain the sparse coupling factor matrix.

[0056] Exemplarily, first, for the high-order feature matrix of the heating data calculate the covariance matrix ΣH = Cov(H), where ΣH[i,j] represents the covariance between the i-th and j-th spatio-temporal features, reflecting the correlation of the fluctuations between the two. Calculate the first quantile q1 (such as the 0.25 quantile, representing the critical value of 25% of the data) for all non-zero elements (or global elements) in ΣH, which is used to mark the boundary between "weak correlation" and "strong correlation".

[0057] Then, extract the diagonal elements {E11, E22, …, Emm} of the spatio-temporal coding matrix E, and its physical meaning is the self-attention weight of each spatio-temporal point (such as the weight of the central node is higher). Multiply q1 by each diagonal element to obtain a set of first quantiles .

[0058] Next, calculate the covariance matrix ΣY = Cov(Y) for the thermal expansion data Y, which reflects the correlation of the thermal expansion amounts at different spatio-temporal points. Calculate the second quantile q2 (such as the 0.75 quantile, representing the critical value of 75% of the data) of ΣY, which is used to mark the boundary of strong correlation of thermal expansion.

[0059] Directly add q1 and q2 to obtain qtarget = q1 + q2. Through linear superposition, qtarget simultaneously contains the "weak correlation intensity of heating characteristics" and the "strong correlation intensity of thermal expansion", providing a fused threshold for subsequent dot product.

[0060] After that, perform a dot product (element-wise multiplication) of qtarget and the obtained Q1 to get .

[0061] Finally, locate the central element: Assume the spatio-temporal encoding transposed matrix E T and the central elements of the thermal expansion covariance matrix ΣY are E T [c, c] and ΣY[c, c] (corresponding to the core spatio-temporal nodes, such as the center point of the heat source).

[0062] Element replacement: Replace the above two central elements with the maximum value max(Q2) in Q2 to highlight the dominant role of the core node, obtaining the replaced spatio-temporal encoding transposed matrix E T 1 and the thermal expansion covariance matrix ΣY1.

[0063] Matrix addition: Calculate S = E T 1 + ΣY1, and obtain the final sparse coupling factor matrix through sparsification processing (such as setting a threshold to filter elements below a certain value).

[0064] The embodiments of this application accurately capture the coupling characteristics of spatio-temporal correlation and thermal expansion: By calculating the first quantile of the high-order spatio-temporal feature covariance matrix and the second quantile of the thermal expansion covariance matrix, the low-frequency key correlations (such as long-term temperature change trends, regional heat transfer laws) of heating data in the spatio-temporal dimension and the core influencing factors of thermal expansion phenomena (such as the deformation sensitivity threshold of key pipeline nodes) are respectively extracted. The use of quantiles can filter noise, retain the statistically significant features in the data distribution, and avoid interference from redundant information. Multiply the diagonal elements of the spatio-temporal encoding matrix (representing the characteristics of each node itself) by the first quantile, and then fuse the key thresholds of spatio-temporal features and thermal expansion through the target quantile, so that the coupling matrix simultaneously contains the node's own attributes (such as pipeline material, pipe diameter) and spatio-temporal interaction relationships (such as the heat conduction effect between adjacent pipelines). This design strengthens the combination of "local features" and "global associations", avoiding the limitations of only considering a single dimension in traditional methods.

[0065] In addition, this application also constructs a sparsified matrix to improve the calculation efficiency and model generalization ability: By replacing the central element of the matrix and adding the obtained sparsely coupled factor matrix, only the key connections (non-zero elements) that have a significant impact on thermal expansion are retained, and irrelevant or weakly related node couplings are eliminated. This operation greatly reduces the matrix dimension and the amount of calculation, is applicable to the anomaly detection of large-scale heating networks, avoids the "curse of dimensionality", and improves the accuracy of anomaly localization. By replacing the "central elements" of the spatio-temporal encoding matrix and the thermal expansion covariance matrix (usually corresponding to key pipeline nodes or hub positions), it is more sensitive to the thermal expansion effect of core nodes. For example, in a heating pipe network, the deformation of the main pipeline has a greater impact on the overall system stability. By strengthening the weight of the central element, the prediction accuracy of such key nodes can be improved, avoiding the interference of noise from edge nodes, and thus improving the accuracy and precision of anomaly localization.

[0066] Furthermore, this application enhances the adaptability to complex scenarios: Through the covariance matrix and quantile operation, it naturally accommodates the spatio-temporal dimensions of heating data (such as temperature and flow rate at different times and regions) and the physical characteristics of thermal expansion (such as temperature-deformation function, material thermal expansion coefficient), and can automatically extract cross-modal correlations without manual feature engineering. It has stronger adaptability to the complex coupling relationships in the heating system that are non-linear and time-varying (such as periodic expansion caused by diurnal temperature difference, material property changes in different seasons), can accurately locate anomalies, and avoid mislocation. The quantile operation is essentially a robust statistics of data distribution. Compared with directly using the original covariance or correlation matrix, the introduction of quantiles reduces the interference of extreme values on the coupling relationship, improves the stability of the model in a noisy environment, improves the accuracy of the sparsely coupled factor matrix, provides reliable support for anomaly detection, and improves the accuracy and precision of anomaly detection.

[0067] Therefore, through the three core mechanisms of noise filtering by quantile statistics, complexity reduction by sparse matrix, and deep coupling of spatio-temporal and physical features, this embodiment enhances the robustness to the thermal expansion phenomenon of heating pipes while ensuring the calculation efficiency, provides efficient and reliable technical support for the fault detection of intelligent heating systems, and thus improves the accuracy and precision of anomaly detection.

[0068] Step S104: Based on the high-order spatio-temporal features of heating data and the sparsely coupled factor matrix corresponding to the thermal expansion of heating pipes, obtain the anomaly probability matrix of heating pipes and the set of propagation paths of heating pipes.

[0069] In one embodiment, based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, an abnormal probability matrix of the heating pipeline and a set of propagation paths of the heating pipeline are obtained, including: performing temporal anomaly detection on the high-order spatio-temporal features of the heating data to obtain the abnormal probability matrix of the heating pipeline; adding the initial abnormal probability matrix of the heating pipeline and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the target abnormal probability matrix of the heating pipeline; performing spatial propagation enhancement and path backtracking on the abnormal probability matrix of the heating pipeline to obtain the set of propagation paths of the heating pipeline.

[0070] Exemplarily, first use the Isolation Forest algorithm to detect the high-order spatio-temporal feature sequence of each node, and output the abnormal probability of node i at time t (taking values from 0 to 1, such as 0.9 indicating a high abnormal probability), to form an abnormal probability matrix (T is the number of time points).

[0071] Then, based on the sparse coupling factor matrix S, the abnormal probability is diffused (added):

[0072] To reflect the influence of adjacent node anomalies on the current node.

[0073] Finally, starting from the node where the abnormal probability mutates (such as greater than 0.8), backtrack the historical propagation path according to the adjacency matrix (such as within the past 3 time windows, nodes A → B → C appear abnormal in sequence), to form a set of propagation paths including nodes and timestamps (such as {(A, t - 2), (B, t - 1), (C, t)}).

[0074] The spatio-temporal correlation of the anomaly detection in the embodiments of this application is enhanced, improving the detection accuracy: Combining high-order spatio-temporal features: By performing temporal anomaly detection on high-order spatio-temporal features (deep features that fuse spatial graph convolution and temporal convolution), complex patterns of spatio-temporal coupling during the operation of the heating pipeline can be captured (such as the pressure linkage of adjacent nodes, the time-series trend of temperature changes), avoiding the one-sidedness of single-dimensional detection. For example, the Isolation Forest algorithm can more accurately identify abnormal states deviating from the normal pattern based on the distribution law of the node historical feature sequence (such as a sudden drop in the temperature of a certain node and synchronous anomalies in the pressure of adjacent nodes). Spatial propagation and diffusion of anomaly probabilities: Using a sparse coupling factor matrix to diffuse the initial anomaly probabilities, essentially combining the physical connection relationship of the pipe network with the node feature similarity, making the calculation of anomaly probabilities not only depend on the state of a single node but also consider the abnormal influence of adjacent nodes (such as the anomaly probability of node B in the example will be enhanced by the abnormal state of adjacent node A). This conforms to the physical law of fault propagation in the heating pipe network (such as a pipeline leak will cause chain changes in upstream and downstream pressure and flow), reducing misjudgments in isolated detection and improving the reliability of the anomaly probability matrix.

[0075] In addition, the physical interpretability and fault location efficiency of propagation path backtracking: Path modeling based on the pipe network topology: By backtracking the propagation path through the adjacency matrix, ensuring that the path set strictly follows the actual connection relationship of the heating pipeline (such as only including the node sequences directly connected in the pipe network), avoiding the generation of "virtual paths" without physical meaning. For example, starting from the node with a sudden change in anomaly probability (such as above the threshold of 0.8), the sequence of abnormal nodes within the historical 3 time windows is traced back according to the timestamp (such as A→B→C), which directly corresponds to the real fault propagation path in the pipe network, providing clear spatial location clues for maintenance personnel. Path integrity in the dynamic time dimension: Combining timestamp records to propagate the path (such as {(A, t-2), (B, t-1), (C, t)}), can clearly present the time evolution order of the fault from the initial node to the downstream nodes, helping to analyze the speed and scope of fault diffusion (such as judging whether it is a rapidly spreading pressure anomaly or a slowly developing pipeline aging), providing data support for emergency response strategies (such as preferentially isolating the source node).

[0076] Furthermore, sparsity and normalization processing improve computational efficiency and model robustness. The dimensionality reduction advantage of the sparse coupling factor matrix: the coupling relationship of non-adjacent nodes is filtered through the adjacency matrix (only the weights of the actually connected nodes are retained), and the weights are normalized (the sum of each row is 1), avoiding interference from irrelevant nodes and reducing computational complexity. For example, in a pipe network containing 100 nodes, the sparse matrix only retains the actual pipe connection edges, reducing invalid calculations, while highlighting the direct influence weights of adjacent nodes, so that the abnormal diffusion process is more focused on the real pipe network structure. Enhanced noise resistance: Normalization processing ensures that the diffusion intensity of the abnormal probability is within a reasonable range (the sum of the weights is 1), avoiding excessive amplification or attenuation of the abnormal probability due to accidental fluctuations of individual nodes, and improving the robustness of the model to data noise. For example, a node has a falsely high abnormal probability due to a short-term sensor failure. After the weights of the adjacent nodes are normalized, its abnormal impact will be reasonably diluted to avoid misjudgment as a real fault.

[0077] Furthermore, it provides precise spatial positioning and propagation chain for maintenance decisions: Rapid target node locking: By searching for the maximum value in the abnormal probability matrix to determine the target node (such as the node with a probability of 0.9 in the example), combined with the node with the earliest abnormality in the propagation path set (such as node A with timestamp t-2), the initial location of the fault can be quickly located, rather than just detecting the affected nodes downstream, reducing the scope of investigation (such as from 100 nodes to 3-5 nodes on the path). Full chain fault analysis: The propagation path set not only provides the current abnormal node, but also records the historical trajectory of fault propagation, which is convenient for operation and maintenance personnel to analyze the fault type (such as whether it is a single-point sudden leakage or systematic aging of the pipeline network), and evaluate the scope of fault impact (such as nodes that may be affected in the future), providing a basis for preventive maintenance (such as checking weak nodes on the path in advance) and reducing operation and maintenance costs.

[0078] In summary, through the deep fusion of spatiotemporal features, modeling of abnormal diffusion under the constraints of pipeline network topology, and joint backtracing of time series and spatial paths, high-precision, strong interpretability and efficient positioning of heating pipeline anomaly detection are achieved, which solves the problems of traditional single-point detection ignoring spatial correlation and fault tracing relying on manual experience, and provides a scientific and feasible technical solution for real-time monitoring and precise operation and maintenance of smart heating pipelines.

[0079] In another embodiment, based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, an abnormal probability matrix of the heating pipeline and a set of propagation paths of the heating pipeline are obtained, including: extracting the high-order spatio-temporal features of the heating pipeline in the high-order spatio-temporal features of the heating data; obtaining the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline, and obtaining the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; combining the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline with the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain a set of covariance features of the heating pipeline; screening out, from the set of covariance features of the heating pipeline, the set composed of the elements greater than the maximum value element in the transfer matrix of the high-order spatio-temporal features of the heating pipeline as the initial abnormal probability matrix of the heating pipeline; correcting the probability values in the initial abnormal probability matrix of the heating pipeline through the target elements in the set of covariance features of the heating pipeline to obtain the target abnormal probability matrix of the heating pipeline, where the target elements in the set of covariance features of the heating pipeline refer to the elements at the median value in the set of covariance features of the heating pipeline; performing spatial propagation enhancement and path backtracking on the target abnormal probability matrix of the heating pipeline to obtain a set of propagation paths of the heating pipeline.

[0080] Exemplarily, the high-order spatio-temporal features of the heating data contain a large amount of information about the heating system in the time and space dimensions, such as the temperature, pressure, flow rate, etc. at different positions in different time periods. Through specific feature selection methods, according to the physical location, function and other attributes of the pipeline, the high-order spatio-temporal features only related to the heating pipeline are screened out. For example, using information such as the pipeline number and the area where it is located, the temperature change trend, pressure fluctuation and other features related to the pipeline are extracted from the multi-dimensional feature data.

[0081] Covariance calculation is performed on the extracted high-order spatio-temporal features of the heating pipeline. Covariance measures the overall error between two variables. In the case of multiple features, the calculated first covariance matrix F reflects the correlation between the features. For example, if the covariance between two features is positive and the value is large, it means that these two features tend to increase or decrease simultaneously; if the covariance is negative, it means that when one feature increases, the other feature tends to decrease.

[0082] Covariance calculation is performed on the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the second covariance matrix G. The sparse coupling factor matrix describes the coupling degree between thermal expansion and other related factors. Calculating its covariance can further analyze the stability and change trend of these coupling relationships.

[0083] Combine the first covariance feature and the second covariance feature. The combination method can be simple splicing, combining the two covariance matrices into a larger matrix according to certain rules, or using a weighted combination method, where corresponding weights are assigned according to the importance of different covariance features and then combined.

[0084] Obtain the transfer matrix J of the high-order spatio-temporal features of the heating pipeline. This matrix describes the transfer of features between different states, and the maximum value element represents the maximum degree of a certain feature transfer. Then, select the elements greater than this maximum value element from the covariance feature set of the heating pipeline. The set composed of these elements is the initial abnormal probability matrix of the heating pipeline.

[0085] Find the element at the median value in the covariance feature set of the heating pipeline as the target element. The median value element can reflect a typical feature of the data, and use the target element to correct the probability values in the initial abnormal probability matrix. The correction method can be through a certain functional relationship, such as linear adjustment or non-linear transformation, to make the initial probability values more reasonable.

[0086] Spatial propagation enhancement: Considering that anomalies may spread in space, expand the influence range of anomaly information in space through a certain algorithm. For example, according to the physical connection relationship and heat conduction characteristics of the pipeline, spread the anomaly information to adjacent pipeline nodes to enhance the spatial coverage of the anomaly information.

[0087] Path backtracking: According to the enhanced anomaly information, reverse-search for the paths of anomaly occurrence and propagation. Algorithms in graph theory, such as depth-first search or breadth-first search, can be used to start from nodes with higher anomaly probabilities and backtrack to possible anomaly sources.

[0088] Multi-source data fusion and precise modeling in the embodiments of this application: By extracting the high-order spatio-temporal features of the heating pipeline (such as the spatio-temporal fluctuation rules of temperature, pressure, and flow) and the thermal expansion sparse coupling factor matrix (reflecting the correlation between pipeline physical deformation and environmental factors), combine the "operation state data" and "physical property data" of the heating system. The combination of their covariance features realizes a double constraint on the abnormal causes - not only capturing the abnormal fluctuations of operation parameters, but also considering the coupling effects of physical properties such as pipeline thermal expansion and contraction, avoiding the one-sidedness of single-dimensional analysis (such as only relying on temperature data and ignoring missed judgments caused by pipeline structure changes). By screening the initial abnormal probability matrix with the "maximum value of the high-order spatio-temporal feature transfer matrix" as the threshold, features significantly deviating from the normal state are preferentially located; further, the probability is corrected by the "median value of the covariance feature" to balance the interference of extreme values and the representativeness of typical features. This mechanism of "first roughly screening extreme anomalies and then refining and correcting based on data distribution characteristics" avoids the problem of poor adaptability of traditional threshold methods to data fluctuations and improves the robustness of abnormal probability assessment.

[0089] In addition, this application can also efficiently locate the source of anomalies and their propagation paths: Based on the target anomaly probability matrix, through spatial propagation enhancement (considering the physical connections and heat conduction laws of pipelines) and path backtracking algorithms, the diffusion path of anomalies in the pipeline network can be clearly presented. For example, when there is a slight leak in a certain section of the pipeline due to thermal expansion, the system can not only detect the temperature anomaly at this node but also, through propagation analysis, predict the downstream areas that may be affected by the anomaly or trace back to the source of the upstream pressure anomaly, providing a complete view of "from a single-point anomaly to global impact" for maintenance personnel and significantly shortening the troubleshooting time. The target anomaly probability matrix quantifies the anomaly risks of each pipeline node in a numerical form. Combining with the set of propagation paths, an "anomaly risk map" of the heating pipeline can be constructed. The maintenance department can formulate differential maintenance strategies based on this: Prioritize the inspection of pipelines with high-probability anomalies and at key nodes in the propagation path, and continuously monitor nodes with low probability but at the end of the propagation, realizing the transformation from "post-fault handling" to "pre-event risk control" and reducing the probability of sudden accidents.

[0090] Therefore, through the closed-loop of "data feature fusion → anomaly probability quantification → propagation path deduction", the "precision, efficiency, and visualization" of anomaly troubleshooting for heating pipelines are achieved. Its core value lies in deeply integrating the operation data and physical characteristics of the heating system, not only solving the problem of insufficient adaptability of traditional methods to complex working conditions but also providing a feasible technical path for intelligent maintenance, ultimately achieving multiple goals of "cost reduction, efficiency improvement, and stability guarantee".

[0091] Step S105: Based on the anomaly probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline, obtain the damaged location of the heating pipeline.

[0092] Among them, obtaining the damaged location of the heating pipeline based on the anomaly probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline includes: Searching for the probability in the anomaly probability matrix of the heating pipeline along the propagation paths in the set of propagation paths of the heating pipeline. When the maximum probability is reached, the target node is obtained, and thus the damaged location of the heating pipeline is determined by the target node.

[0093] Exemplarily, for each propagation path (such as the node sequence → ...→ ), search the anomaly probability matrix in reverse order of time (retracing from the current anomaly node to historical nodes), and record the anomaly probability of each node at the corresponding time point.

[0094] If the anomaly probability of the node in the path at time is the maximum value in the path (and greater than 0.9), and if this node triggers an abnormality in a downstream node at a subsequent time point, it is determined that it is the root node.

[0095] For example, in the path ← ← the abnormality probability of is 0.95, which is the maximum value in the path and appears abnormal before B and C. Then the target node is A. Therefore, it is very likely that the damaged location of the heating pipeline is at node A. Subsequently, maintenance personnel can focus on inspecting and repairing the heating pipeline at and near node A to solve the possible damage problem of the heating pipeline.

[0096] The embodiment of the present application accurately locks the source of the fault and avoids misjudging the diffusion node: Time reverse search + probability maximization: Trace back along the propagation path in reverse time order (from the current abnormal node to the historical node), and give priority to identifying the root node that appears abnormal earliest and has the highest probability (such as the abnormality probability of node A in the example is 0.95, which is the maximum value in the path and triggers an abnormality before B and C), rather than only focusing on the current downstream nodes with high probability. This conforms to the physical law of the fault propagation of the heating pipe network (such as leakage usually starts from a certain node and then spreads to adjacent nodes), avoiding misjudging the "downstream nodes affected by the fault" as the damaged location and greatly improving the positioning accuracy. Dual constraints filter out noise: Screen the target node through the dual conditions of "abnormality probability threshold (such as ≥0.9)" and "triggering downstream abnormality", effectively excluding non-real abnormalities caused by accidental fluctuations or sensor noise, ensuring that the positioning result is based on significant abnormal signals and conforms to the causal logic of fault propagation (the upstream node abnormality appears before the downstream node).

[0097] In addition, focusing on the physical propagation path improves the positioning efficiency and interpretability: Path constraint narrows the search range: Only search for the target node in the set of identified propagation paths (such as the node sequence A→B→C within the historical 3 time windows), rather than traversing the entire pipe network, narrowing the investigation scope from "all network nodes" to "key nodes on the path" (such as 3 nodes in the example), significantly reducing the manual inspection cost. Physical interpretability supports operation and maintenance decisions: The propagation path is strictly based on the pipe network adjacency matrix (actual pipe connection relationship), and the positioning result of the target node can be directly mapped to the specific location in the geographic information system (GIS) (such as the pipe connection at heat exchange station A). Maintenance personnel do not need to rely on complex algorithm derivation and can quickly lock the inspection area through the pipe network drawing, improving the operation and maintenance efficiency.

[0098] Furthermore, dynamically adapt to the fault propagation mode to enhance robustness: be compatible with progressive and burst faults. For progressive faults (such as the leakage gradually increasing due to pipeline aging), the continuous upward trend of the early anomaly probability can be captured through time-reverse search (such as the probability at node A being 0.8 at time t-2, 0.9 at time t-1, and 0.95 at time t), and the hidden trouble points that have not fully erupted but have begun to affect the downstream can be located in a timely manner. For burst faults (such as a sudden pipeline rupture), the source of the sudden anomaly can be quickly identified through the condition of "probability maximum + triggering downstream anomaly" (even if downstream nodes have simultaneous anomalies, the originating node can be distinguished through the time stamp order). Strong anti-interference ability: The anomaly probability matrix has been processed through spatio-temporal feature fusion and sparse coupling factor diffusion, filtering out the interference of irrelevant nodes. The search process is only carried out on the effective propagation path, reducing the positioning ambiguity in the case of multi-node anomalies in complex pipe networks (such as distinguishing independent faults from cascading faults).

[0099] Furthermore, form a closed-loop operation and maintenance process to support preventive maintenance: Linkage between fault location and propagation analysis: The determination of the target node not only provides the damage location but also reveals the fault diffusion speed and influence range through the time series of the propagation path (such as the anomaly order of A→B→C), helping the operation and maintenance personnel to judge the fault type (such as single-point leakage or pipe network pressure imbalance) and formulate targeted measures (such as strengthening the weak nodes on the path in advance). Data-driven optimization of maintenance strategies: Through the accumulation of historical location results (such as multiple faults concentrated near node A), the pipe network design or operation and maintenance plan can be optimized in reverse (such as increasing the sensor density in this area, shortening the inspection cycle), realizing the transformation from "passive maintenance" to "active prevention".

[0100] In summary, through the core logic of "searching for the maximum anomaly probability along the actual propagation path", the abstract data analysis is transformed into the positioning of specific physical nodes, with the following three advantages: High precision (source identification), high efficiency (path constraint), and high reliability (causal verification), effectively solving the problems of "relying on manual experience, wide troubleshooting scope, and high misjudgment rate" in traditional heating pipe network fault location, and providing key technical support for the intelligent operation and maintenance of the intelligent heating system.

[0101] In summary, in this embodiment, first obtain heating data, then determine the high-order spatio-temporal features of the heating data based on the heating data and a preset artificial intelligence algorithm, and then obtain the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline. Thus, based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, obtain the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline, and further obtain the damaged position of the heating pipeline based on the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline. The present invention improves the accuracy of abnormal positioning, shortens the troubleshooting time, reduces the operation and maintenance costs, and thus improves the troubleshooting efficiency and effect.

[0102] Figure 2 FIG. shows an abnormal heating troubleshooting system of artificial intelligence proposed by an embodiment of the present invention, as Figure 2 shown, the system includes: The first acquisition module 201 is used to acquire heating data; The feature determination module 202 is used to determine the high-order spatio-temporal features of the heating data based on the heating data and a preset artificial intelligence algorithm; The second acquisition module 203 is used to acquire the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; The third acquisition module 204 is used to obtain the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the heating pipeline; The position determination module 205 is used to obtain the damaged position of the heating pipeline based on the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline.

[0103] Optionally, the feature determination module 202 is further used to input the heating data into a preset artificial intelligence algorithm to obtain a spatio-temporal coding matrix and an adjacency matrix; Generate a sequence of dynamic spatio-temporal propagation graphs based on the spatio-temporal coding matrix and the adjacency matrix; Determine the high-order spatio-temporal features of the heating data based on the sequence of dynamic spatio-temporal propagation graphs.

[0104] Optionally, the feature determination module 202 is further used to process the heating data using a preset artificial intelligence algorithm, establish a heating pipe network topology graph, and generate an adjacency matrix based on the heating pipe network topology graph; Perform time series alignment on the heating data according to a time window to obtain aligned data; Stitch the historical fault labels with the aligned data to obtain a spatio-temporal coding matrix.

[0105] Optionally, generating a sequence of dynamic spatio-temporal propagation graphs based on the spatio-temporal coding matrix and the adjacency matrix includes: Calculate the similarity between the data in the spatio-temporal coding matrix; Determine a fusion weight based on the similarity between data in the spatio-temporal encoding matrix and the historical propagation pattern; Determine a target edge set of the adjacency matrix based on the fusion weight and a preset threshold; Generate a dynamic spatio-temporal propagation graph sequence based on the target edge set.

[0106] Optionally, the feature determination module 202 is further configured to obtain the spatial graph convolution and the temporal convolution of the dynamic spatio-temporal propagation graph sequence; Perform gated fusion on the spatial graph convolution and the temporal convolution of the dynamic spatio-temporal propagation graph sequence to obtain high-order spatio-temporal features.

[0107] Optionally, the second acquisition module 203 is further configured to perform coupled calculation on the high-order spatio-temporal features of the heating data to obtain an initial coupled factor matrix; Perform topological correction on the initial coupled factor matrix through the thermal expansion of the heating pipeline to obtain a corrected coupled factor matrix; Perform normalization processing on the corrected coupled factor matrix to obtain a sparse coupled factor matrix.

[0108] Or, Calculate the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data; Extract the diagonal elements in the spatio-temporal encoding matrix, and multiply the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data by the diagonal elements in the spatio-temporal encoding matrix to obtain a set of first quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data; Calculate the second quantile corresponding to the covariance matrix corresponding to the thermal expansion of the heating pipeline; Add the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data to the second quantile corresponding to the covariance matrix corresponding to the thermal expansion of the heating pipeline to obtain a target quantile; Perform dot multiplication on the target quantile and each quantile in the set of first quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data to obtain a set of second quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data; Use the maximum quantile in the set of second quantiles to replace the central element in the transposed matrix of the spatio-temporal encoding matrix and the central element in the covariance matrix corresponding to the thermal expansion of the heating pipeline, and add the transposed matrix of the spatio-temporal encoding matrix after replacement to the covariance matrix corresponding to the thermal expansion of the heating pipeline after replacement to obtain a sparse coupled factor matrix.

[0109] Optionally, the third acquisition module 204 is further configured to perform temporal anomaly detection on the high-order spatio-temporal features of the heating data to obtain an initial anomaly probability matrix of the heating pipeline; Add the initial abnormal probability matrix of the heating pipeline to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the target abnormal probability matrix of the heating pipeline; Perform spatial propagation enhancement and path backtracking on the target abnormal probability matrix of the heating pipeline to obtain the propagation path set of the heating pipeline.

[0110] Or, Extract the high-order spatio-temporal features of the heating pipeline from the high-order spatio-temporal features of the heating data; Obtain the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline, and obtain the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; Combine the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline with the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the covariance feature set of the heating pipeline; Select from the covariance feature set of the heating pipeline the set composed of the elements greater than the maximum value element in the transfer matrix of the high-order spatio-temporal features of the heating pipeline as the initial abnormal probability matrix of the heating pipeline; Correct the probability value in the initial abnormal probability matrix of the heating pipeline through the target element in the covariance feature set of the heating pipeline to obtain the target abnormal probability matrix of the heating pipeline, where the target element in the covariance feature set of the heating pipeline refers to the element in the middle value of the covariance feature set of the heating pipeline; Perform spatial propagation enhancement and path backtracking on the target abnormal probability matrix of the heating pipeline to obtain the propagation path set of the heating pipeline.

[0111] Optionally, the location determination module 205 is further configured to search for the probability in the abnormal probability matrix of the heating pipeline along the propagation paths in the propagation path set of the heating pipeline. When the maximum probability is reached, the target node is obtained, so as to determine the damage location of the heating pipeline from the target node.

[0112] In the specification provided herein, the algorithms and displays are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the examples of the present invention. Based on the above description, the structure required to construct such a system is obvious. In addition, the present invention is not directed to any particular programming language. It should be understood that the content of the present invention described herein can be implemented using various programming languages, and the description of the specific language above is for disclosing the preferred embodiments of the present invention.

[0113] In the description provided herein, numerous specific details are set forth. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail so as not to obscure an understanding of this description.

[0114] Those skilled in the art will appreciate that the modules or units or components of the devices in the examples disclosed herein may be arranged in the devices as described in that embodiment, or alternatively may be located in one or more devices different from those of the example. The modules in the foregoing examples may be combined into one module or further divided into multiple sub-modules.

[0115] Those skilled in the art will understand that the modules in the devices of the embodiments can be adaptively changed and disposed in one or more devices different from those of the embodiment. The modules or units or components in the embodiments can be combined into one module or unit or component, and furthermore can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all the features disclosed in this specification and all the processes or units of any method or device so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification can be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0116] In addition, some of the embodiments herein are described as a combination of methods or method elements that can be implemented by a processor of a computer system or by other devices performing the functions. Accordingly, a processor having the necessary instructions for implementing the methods or method elements forms a device for implementing the methods or method elements. In addition, the elements described herein in the device embodiments are examples of such devices: the devices for implementing the functions performed by the elements for the purpose of implementing the invention.

[0117] As used herein, unless otherwise specified, the use of ordinal numbers "first", "second", "third", etc. to describe ordinary objects merely indicates different instances of similar objects and is not intended to imply that the objects so described must have a given order in terms of time, space, ranking, or in any other manner.

[0118] Although the present invention has been described in terms of a limited number of embodiments, those skilled in the art, having the benefit of the foregoing description, will appreciate that other embodiments can be contemplated within the scope of the invention as thus described. In addition, it should be noted that the language used in this specification has been principally selected for readability and instructional purposes and not to limit or circumscribe the inventive subject matter. Accordingly, many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the appended claims. For the scope of the present invention, the disclosure herein is illustrative and not restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. An abnormal heating detection method for artificial intelligence, characterized in that Including the following steps: Obtain heating data; Based on the heating data and a preset artificial intelligence algorithm, determine the high-order spatio-temporal features of the heating data; Obtain the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; Based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline, obtain the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline; Based on the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline, obtain the damaged location of the heating pipeline.

2. The method for troubleshooting abnormal heat supply of the artificial intelligence according to claim 1, wherein The determining the high-order spatio-temporal features of the heating data based on the heating data and a preset artificial intelligence algorithm includes: Input the heating data into the preset artificial intelligence algorithm to obtain a spatio-temporal encoding matrix and an adjacency matrix; Based on the spatio-temporal encoding matrix and the adjacency matrix, generate a dynamic spatio-temporal propagation graph sequence; Based on the dynamic spatio-temporal propagation graph sequence, determine the high-order spatio-temporal features of the heating data.

3. The abnormal heating troubleshooting method of the artificial intelligence according to claim 2, wherein, The inputting the heating data into the preset artificial intelligence algorithm to obtain a spatio-temporal encoding matrix and an adjacency matrix includes: Use the preset artificial intelligence algorithm to process the heating data, establish a heating pipe network topology graph, and generate an adjacency matrix based on the heating pipe network topology graph; Perform temporal alignment on the heating data according to a time window to obtain aligned data; Concatenate the historical fault labels with the aligned data to obtain the spatio-temporal encoding matrix.

4. The abnormal heat supply troubleshooting method of the artificial intelligence according to claim 2, characterized in that, The generating a dynamic spatio-temporal propagation graph sequence based on the spatio-temporal encoding matrix and the adjacency matrix includes: Calculate the similarity between data in the spatio-temporal encoding matrix; Based on the similarity between data in the spatio-temporal encoding matrix and the historical propagation pattern, determine the fusion weight; Based on the fusion weight and a preset threshold, determine the target edge set of the adjacency matrix; Based on the target edge set, generate the dynamic spatio-temporal propagation graph sequence.

5. The abnormal heat supply troubleshooting method of the artificial intelligence according to claim 2, wherein, The determining the high-order spatio-temporal features of the heating data based on the dynamic spatio-temporal propagation graph sequence includes: Obtain the spatial graph convolution and temporal convolution of the dynamic spatio-temporal propagation graph sequence; Perform gated fusion on the spatial graph convolution and temporal convolution of the dynamic spatio-temporal propagation graph sequence to obtain the high-order spatio-temporal features.

6. The method for troubleshooting abnormal heat supply of the artificial intelligence according to claim 1, characterized in that, The obtaining the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline includes: Perform coupling calculation on the high-order spatio-temporal features of the heating data to obtain an initial coupling factor matrix; Perform topological correction on the initial coupling factor matrix through the thermal expansion of the heating pipeline to obtain a corrected coupling factor matrix; Perform normalization processing on the corrected coupling factor matrix to obtain the sparse coupling factor matrix; Or, Calculate the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data; Extract the diagonal elements from the spatio-temporal encoding matrix, and multiply the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data by the diagonal elements in the spatio-temporal encoding matrix to obtain the set of first quantiles of the covariance matrix of the high-order spatio-temporal features of the heating data; Calculate the second quantile corresponding to the covariance matrix of the thermal expansion of the heating pipeline; Add the first quantile corresponding to the covariance matrix of the high-order spatio-temporal features of the heating data to the second quantile corresponding to the covariance matrix of the thermal expansion of the heating pipeline to obtain the target quantile; Perform dot multiplication on the target quantile and each quantile in the first quantile set of the covariance matrix of the high-order spatio-temporal features of the heating data to obtain the second quantile set of the covariance matrix of the high-order spatio-temporal features of the heating data; Replace the central element in the transpose matrix of the spatio-temporal encoding matrix and the central element in the covariance matrix corresponding to the thermal expansion of the heating pipeline with the maximum quantile in the second quantile set, and add the transpose matrix of the spatio-temporal encoding matrix after replacement to the covariance matrix corresponding to the thermal expansion of the heating pipeline after replacement to obtain the sparse coupling factor matrix.

7. The abnormal heating troubleshooting method of the artificial intelligence according to claim 1, wherein The obtaining of the target anomaly probability matrix of the heating pipeline and the propagation path set of the heating pipeline based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline includes: Perform temporal anomaly detection on the high-order spatio-temporal features of the heating data to obtain the initial anomaly probability matrix of the heating pipeline; Add the initial anomaly probability matrix of the heating pipeline to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the target anomaly probability matrix of the heating pipeline; Perform spatial propagation enhancement and path backtracking on the target anomaly probability matrix of the heating pipeline to obtain the propagation path set of the heating pipeline; Or, Extract the high-order spatio-temporal features of the heating pipeline from the high-order spatio-temporal features of the heating data; Obtain the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline, and obtain the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; Merge the first covariance feature corresponding to the high-order spatio-temporal features of the heating pipeline with the second covariance feature corresponding to the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline to obtain the covariance feature set of the heating pipeline; Select the set composed of the elements greater than the maximum value element in the transfer matrix of the high-order spatio-temporal features of the heating pipeline from the covariance feature set of the heating pipeline as the initial anomaly probability matrix of the heating pipeline; Correct the probability values in the initial anomaly probability matrix of the heating pipeline through the target elements in the covariance feature set of the heating pipeline to obtain the target anomaly probability matrix of the heating pipeline, where the target elements in the covariance feature set of the heating pipeline refer to the elements in the middle value in the covariance feature set of the heating pipeline; Perform spatial propagation enhancement and path backtracking on the target anomaly probability matrix of the heating pipeline to obtain the propagation path set of the heating pipeline.

8. The abnormal heating detection method of the artificial intelligence according to claim 1, characterized in that The obtaining of the damage location of the heating pipeline based on the target anomaly probability matrix of the heating pipeline and the propagation path set of the heating pipeline includes: Search for the probability in the abnormal probability matrix of the heating pipeline along the propagation paths in the set of propagation paths of the heating pipeline. When the maximum probability is reached, obtain the target node, and thus determine the damage location of the heating pipeline from the target node.

9. An abnormal heating detection system for artificial intelligence, characterized in that, Including: The first acquisition module is used to acquire heating data; The feature determination module is used to determine the high-order spatio-temporal features of the heating data based on the heating data and a preset artificial intelligence algorithm; The second acquisition module is used to acquire the sparse coupling factor matrix corresponding to the thermal expansion of the heating pipeline; The third acquisition module is used to obtain the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline based on the high-order spatio-temporal features of the heating data and the sparse coupling factor matrix corresponding to the heating pipeline; The location determination module is used to obtain the damage location of the heating pipeline based on the abnormal probability matrix of the heating pipeline and the set of propagation paths of the heating pipeline.

Images