A method and device for locating air leakage points of a condenser of a thermal power plant

Abstract

This invention proposes a method and apparatus for locating air leaks in condensers of thermal power plants. Specifically, the method is a condenser air leak location method based on a sparse latent source Bayesian factor graph. The method includes: acquiring DCS measurement point data and equipment connection relationships of the condenser vacuum system; constructing an observation matrix by arranging the measurement point data in time sequence; identifying candidate leaks, retrieving reachable propagation paths from these leaks to the measurement points, and merging the measurement points into combinations to construct a constraint graph; decomposing the observation matrix under the constraints of the constraint graph, separating low-rank common-mode components and sparse anomaly components, and extracting the anomaly intensity of the measurement point combinations to obtain a vector; constructing a sparse latent source Bayesian factor graph, setting latent variables, and performing joint posterior inference based on operating conditions and noise constraints to obtain a probability and intensity sequence; posteriorly ranking the candidate leaks according to the above sequence; calculating the path and measurement point consistency index, verifying the posterior ranking, and obtaining the target leak or the leak to be verified.

Description

Technical Field

[0001] This invention belongs to the field of detection technology, specifically relating to a method and device for locating air leaks in a condenser of a thermal power plant. Background Technology

[0002] The condenser vacuum system, as a core component for ensuring the thermal economy of thermal power units, is widely used to maintain low back pressure operation. With the increasing size of power units, related technologies have constructed a leak diagnosis system through the collaborative operation of distributed control system measurement point acquisition, equipment topology mapping, and statistical characteristic analysis. Specifically, this system covers the entire process from multi-source data acquisition to anomaly area screening, including key aspects such as trend curve comparison, residual monitoring, and expert rule judgment.

[0003] However, existing location methods directly extract anomalies using single-point deviations or general statistical residuals, without embedding equipment connection constraints and leakage propagation mechanisms into the feature generation process. This results in severe mixing of operational disturbances and actual leakage signals at the feature level. Due to the lack of explicit modeling of reachable propagation paths, traditional solutions struggle to distinguish between noise and faults within a unified inference framework. Consequently, location results often remain at the fuzzy region level, failing to generate point-level conclusions consistent with the actual physical path, severely impacting maintenance efficiency and unit safety. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the first objective of this invention is to provide a method for locating air leaks in condensers of thermal power plants.

[0006] The second objective of this invention is to provide a device for locating air leaks in condensers of thermal power plants.

[0007] The third objective of this invention is to provide a computer device.

[0008] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0009] To achieve the above objectives, a first aspect of the present invention provides a method for locating air leaks in a condenser of a thermal power plant, comprising:

[0010] Acquire the distributed control system measurement point data and equipment connection relationships of the condenser vacuum system during unit operation, and arrange the distributed control system measurement point data in time sequence according to the measurement point identification and equipment location to construct an observation matrix; Candidate leaks are determined based on the device connection relationship. The reachable propagation paths from each candidate leak to each measurement point are retrieved. The measurement points are merged into measurement point combinations according to the reachable propagation paths. An influence constraint diagram characterizing the connection relationship between candidate leaks and measurement point combinations is constructed. Robust principal component analysis is performed on the observation matrix under the constraints of the influence constraint diagram. The overall linkage change is separated into low-rank common mode components, and the local anomalies propagating along the reachable propagation path are separated into sparse anomaly components. The distribution of the sparse anomaly components on unreachable measurement points is restricted according to the influence constraint diagram. The anomaly injection vector is obtained by extracting the anomaly intensity of each measurement point combination from the sparse anomaly components. Based on the candidate leak points, the influence constraint diagram and the abnormal injection vector, a sparse latent source Bayesian factor diagram is constructed. The candidate leak point number is set as a discrete latent variable and the leak intensity is set as a continuous latent variable. Under the constraints of operating condition fluctuation and measurement point noise, joint posterior inference is performed to obtain the posterior probability sequence and intensity estimation sequence. A posterior ranking is generated based on the posterior probability sequence and the intensity estimation sequence; By combining the influence constraint diagram and the anomaly injection vector to calculate the path consistency index and the measurement point consistency index, the posterior ranking is verified to obtain the target leak point or the leak point to be verified.

[0011] In one embodiment of the present invention, constructing the observation matrix includes: Determine the measurement point identifier, sampling time, and location of the equipment associated with each measurement point, and extract continuous time-series data of each measurement point within the same runtime period; Establish a position mapping for each measuring point based on the location of the equipment to which it belongs, so that each measuring point corresponds to a unique installation position in the condenser vacuum system; The continuous time series data is divided into columns according to a preset time window, and the distributed control system measurement point data is arranged with the measurement point as the row dimension and the time window as the column dimension to obtain the observation matrix.

[0012] In one embodiment of the present invention, the construction of an influence constraint graph characterizing the connection relationship between candidate leak points and test point combinations includes: Based on the position of the measurement point row in the observation matrix, locate the corresponding device node in the device connection relationship, and determine the device connection position in the device connection relationship as a candidate leak point; Starting from each candidate leak point, the propagation path from the candidate leak point to each measurement point is determined by searching along the propagation direction in the device connection relationship to the corresponding device node. Based on the reachable propagation path corresponding to each candidate leak point, the measurement points that are in the same reachable propagation path and are consecutively corresponding in the propagation order are aligned and merged to form a measurement point combination corresponding to the reachable propagation path. Using candidate leak points as row objects and measurement point combinations as column objects, determine whether there are reachable propagation paths from candidate leak points to each measurement point combination. If a reachable propagation path exists, establish a connection to obtain an influence constraint graph.

[0013] In one embodiment of the present invention, the step of extracting the anomaly intensity of each measurement point combination from the sparse anomaly components to obtain the anomaly injection vector includes: Map the connection relationship of measurement point combination in the influence constraint diagram to the corresponding measurement point position in the observation matrix. Open the sparse anomaly component allocation for the corresponding measurement point position with connection, and restrict the sparse anomaly component to zero value for the corresponding measurement point position without connection. The observation matrix is ​​input into robust principal component analysis and decomposed under allocation constraints. The synchronous changes caused by unit load, circulating water conditions and vacuum equipment regulation are attributed to low-rank common-mode components. Local variations in the decomposition results that are located at open allocation positions and occur continuously along the same reachable propagation path are classified as sparse outlier components. The sparse anomaly components are aligned according to the measurement point combination. The sparse anomaly components of each measurement point combination within the same time window are extracted and merged according to the order of reachable propagation paths to obtain the anomaly intensity of each measurement point combination. The anomaly intensity of each measurement point combination is arranged in the order of the measurement point combination column in the influence constraint diagram to obtain the anomaly injection vector.

[0014] In one embodiment of the present invention, the step of constructing a sparse latent source Bayesian factor graph based on the candidate leak points, the influence constraint graph, and the anomaly injection vector, setting the candidate leak point numbers as discrete latent variables and the leak intensity as continuous latent variables, and performing joint posterior inference to obtain a posterior probability sequence and an intensity estimation sequence under operating condition fluctuation constraints and measurement point noise constraints includes: Based on the connection relationship between candidate leak points and measurement point combinations in the influence constraint diagram, a sparse latent source Bayesian factor diagram is constructed, retaining only the factor connections with reachable propagation paths. Set up observation units and observation factors that correspond one-to-one with the measurement point combination, and write the anomaly injection vector into the corresponding observation factors in the order of the measurement point combination. Set up candidate leak point state units and leak intensity units that correspond one-to-one with candidate leak points, set the candidate leak point number as a discrete latent variable, and set the leak intensity as a continuous latent variable. Based on the influence constraint diagram, path constraint units are set between the observed factors and the candidate leak point state units, so that the anomaly injection vector participates in the interpretation of candidate leak points along the corresponding reachable propagation path. Operating condition fluctuation constraints are introduced on each candidate leak point state unit and leak intensity unit, and measurement point noise constraints are introduced between each observation factor and the candidate leak point state unit. Under the combined influence of operating condition fluctuation constraints and measurement point noise constraints, joint posterior inference is performed on discrete and continuous latent variables to obtain posterior probability sequences and intensity estimation sequences.

[0015] In one embodiment of the present invention, generating a posterior ranking based on the posterior probability sequence and the intensity estimation sequence includes: Align the posterior probability sequence and intensity estimation sequence according to the candidate leak point number, so that each candidate leak point corresponds to a unique posterior probability and intensity estimate. The posterior probability of candidate leak points is used as the primary ranking criterion, and the intensity estimate is used as the secondary ranking criterion to generate a posterior ranking.

[0016] In one embodiment of the present invention, verifying the posterior ranking to obtain the target leak or the leak to be verified includes: Based on the posterior ranking, the candidate leak points with the highest ranking are determined, and the reachable propagation path and measurement point combination connection relationship corresponding to the candidate leak point are extracted from the influence constraint diagram. The measurement points with abnormal intensity responses in the abnormal injection vector are matched one by one with the reachable propagation path, and the path consistency index is calculated according to the matching of the propagation order. The connection relationship between the combination of measurement points with abnormal intensity response in the abnormal injection vector and the combination of measurement points is checked one by one, and the measurement point consistency index is calculated according to the connection coverage and abnormal intensity distribution. The posterior ranking is verified based on the path consistency index and the measurement point consistency index. The verification result is compared with the preset positioning threshold. When the preset positioning threshold is met, the candidate leak point at the top of the posterior ranking is determined as the target leak point; otherwise, it is determined as a leak point to be verified.

[0017] To achieve the above objectives, a second aspect of the present invention provides a device for locating air leaks in a condenser of a thermal power plant, comprising: The data acquisition module is used to acquire the distributed control system measurement point data and equipment connection relationships of the condenser vacuum system during unit operation, and to arrange the distributed control system measurement point data in time sequence according to the measurement point identification and equipment location to construct an observation matrix; The influence constraint graph construction module is used to determine candidate leaks based on the device connection relationship, retrieve the reachable propagation path from each candidate leak to each measurement point, merge the measurement points according to the reachable propagation path to form a measurement point combination, and construct an influence constraint graph that represents the connection relationship between candidate leaks and measurement point combinations. An anomaly injection vector extraction module is used to perform robust principal component analysis on the observation matrix under the constraints of the influence constraint diagram, separate the overall linkage change into low-rank common mode components, separate the local anomalies propagating along the reachable propagation path into sparse anomaly components, restrict the distribution of the sparse anomaly components on unreachable measurement points according to the influence constraint diagram, and extract the anomaly intensity of each measurement point combination from the sparse anomaly components to obtain the anomaly injection vector. The posterior inference module is used to construct a sparse latent source Bayesian factor graph based on the candidate leak points, the influence constraint graph and the anomaly injection vector, set the candidate leak point number as a discrete latent variable and the leak intensity as a continuous latent variable, and perform joint posterior inference under the constraints of operating condition fluctuation and measurement point noise to obtain the posterior probability sequence and intensity estimation sequence. The location output module is used to generate a posterior ranking based on the posterior probability sequence and the intensity estimation sequence; and to calculate the path consistency index and the measurement point consistency index by combining the influence constraint diagram and the anomaly injection vector, and to verify the posterior ranking to obtain the target leak point or the leak point to be verified.

[0018] The present invention provides a method and apparatus for locating air leaks in a condenser of a thermal power plant, which, compared with the prior art, has the following advantages: This invention constructs the connection relationship of the condenser vacuum system equipment into an influence constraint diagram and embeds a robust principal component analysis decomposition process. In the feature generation stage, physical propagation constraints are incorporated, which can classify the linkage changes of operating conditions into low-rank components and the local anomalies that conform to the leakage propagation law into sparse components. The extracted measurement point combination-level anomaly intensity is highly matched with the leakage point propagation effect. The key features for point-level location can be formed using only existing DCS data.

[0019] This invention proposes a sparse latent source Bayesian factor graph method to fix the connection relationship of model units in the influence constraint graph. It jointly processes leakage, operating condition fluctuations and measurement point noise under a unified framework, avoids cross-path interpretation anomalies of irrelevant leak points, and can simultaneously output the probability of candidate leak point occurrence and leakage intensity, thereby improving the consistency and interpretability of the localization results.

[0020] This invention forms a complete technical chain for condenser air leakage location, which puts path constraints in advance, centralizes the inference of uncertainties, and adds double consistency verification. It can output the target leak point and the leak point to be verified, which can meet the engineering application requirements of unit non-stop operation and minimal disturbance, and realize accurate leakage location from trend judgment to point location and verifiable location.

[0021] To achieve the above objectives, a third aspect of this application provides a computer device comprising a processor and a memory; wherein the processor runs a program corresponding to the executable program code stored in the memory, for implementing a method for locating air leaks in a condenser of a thermal power plant as described in the first aspect embodiment.

[0022] To achieve the above objectives, the fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for locating air leaks in a condenser of a thermal power plant as described in the first aspect embodiment.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for locating air leaks in a condenser of a thermal power plant according to an embodiment of the present invention; Figure 2 This is a flowchart of the observation matrix construction process according to an embodiment of the present invention; Figure 3 This is a flowchart of the influence constraint graph construction process according to an embodiment of the present invention; Figure 4 This is a flowchart of the anomaly injection vector generation process according to an embodiment of the present invention; Figure 5 This is a flowchart of the sparse latent source Bayesian factor graph inference process according to an embodiment of the present invention; Figure 6 This is a flowchart of the posterior sorting generation process according to an embodiment of the present invention; Figure 7 This is a flowchart of the positioning verification process according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the equipment connection and reachable propagation path of the condenser vacuum system according to an embodiment of the present invention; Figure 9 This is a structural diagram of a device for locating air leaks in a condenser of a thermal power plant, according to an embodiment of the present invention. Figure 10 It is a computer device according to an embodiment of the present invention. Detailed Implementation

[0025] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] The following description, with reference to the accompanying drawings, describes a method and apparatus for locating air leaks in a condenser of a thermal power plant, according to an embodiment of the present invention.

[0028] Example 1 Figure 1 This is a flowchart illustrating a method for calculating the isotope irradiation channel temperature of the side reflector layer in a pebble bed type high-temperature gas-cooled reactor according to an embodiment of the present invention. Figure 1 As shown, it includes: S1, acquire the distributed control system measurement point data and equipment connection relationships of the condenser vacuum system during unit operation, and arrange the distributed control system measurement point data in time sequence according to the measurement point identification and equipment location to construct an observation matrix. For example... Figure 2 shown, specifically.

[0029] Data from the distributed control system of the condenser vacuum system during grid-connected operation of the unit is collected. The measurement point identifier, sampling time and equipment location of each measurement point are determined, and continuous time-series data of each measurement point are extracted within the same operating period. Establish a position mapping for each measuring point based on the location of the equipment to which it belongs, so that each measuring point corresponds to a unique installation position in the condenser vacuum system, and that the installation position can be directly located in the equipment connection relationship; Organize the equipment connection relationships based on the physical connections of the equipment and the direction of medium transfer in the condenser vacuum system, so that the equipment connection relationships characterize the sequence and reachable channels of leakage propagation within the system. The continuous time series data is divided into columns according to the preset time window. The data is arranged with the measurement point as the row dimension and the time window as the column dimension. The measurement point data of the distributed control system is arranged according to the measurement point order determined by the position mapping and the time sequence determined by the sampling time, so as to obtain the observation matrix in which the row position corresponds one-to-one with the equipment connection relationship.

[0030] The observation matrix is ​​a data matrix arranged with measurement points as the row dimension and time windows as the column dimension. Each row represents the time-series observation of the same measurement point within a continuous time window, and each column represents the synchronous observation of each measurement point within the same time window. The observation matrix is ​​used as the input for robust principal component analysis and to maintain a consistent correspondence between the measurement point order and the subsequent measurement point combination construction. The device connection relationship is a connection description established between various device entities in the condenser vacuum system according to the direction of medium transmission. Each connection description corresponds to an actual device connection channel and includes the connection start point, connection end point and propagation direction. The device connection relationship is used to determine the reachable propagation path from the candidate leak point to each measuring point and to generate the fixed connection relationship in the influence constraint diagram.

[0031] S2, based on the device connection relationship, candidate leaks are determined, reachable propagation paths from each candidate leak to each measurement point are retrieved, and measurement points are merged according to the reachable propagation paths to form measurement point combinations. An influence constraint graph characterizing the connection relationship between candidate leaks and measurement point combinations is constructed. For example... Figure 3 shown, specifically.

[0032] Based on the position of the measurement point row in the observation matrix, locate the corresponding device node in the device connection relationship, and determine the device connection position in the device connection relationship as a candidate leak point; Starting from each candidate leak point, the propagation direction in the device connection relationship is sequentially searched to the device node corresponding to each test point. When there is a connection sequence from the candidate leak point to the device node corresponding to the test point with a continuous and consistent connection direction, the connection sequence is determined as the reachable propagation path from the candidate leak point to the test point. When there are multiple connection sequences, the connection sequence with the fewest device nodes is determined as the reachable propagation path. When there is no connection sequence, the test point is determined as an unreachable test point. Based on the reachable propagation path corresponding to each candidate leak point, the measurement points that are in the same reachable propagation path and are consecutively corresponding in the propagation order are aligned and merged to form a measurement point combination corresponding to the reachable propagation path. Using candidate leak points as row objects and measurement point combinations as column objects, determine whether there are reachable propagation paths from candidate leak points to each measurement point combination. If a reachable propagation path exists, establish a connection; otherwise, do not establish a connection, thus obtaining an influence constraint graph. The reachable propagation path is an ordered connection sequence that starts from the candidate leak point and travels along the propagation direction to the device node corresponding to the measurement point in the device connection relationship. The reachable propagation path is used to limit the range of measurement points that sparse outliers can enter and to determine the composition order of the measurement point combination. The influence constraint graph is a sparse connection structure constructed based on whether there is a reachable propagation path from the candidate leak point to the measurement point combination. The establishment of a connection indicates that the corresponding candidate leak point can generate a propagation interpretation of the corresponding measurement point combination, and the absence of a connection indicates that the corresponding candidate leak point cannot generate a propagation interpretation of the corresponding measurement point combination. The influence constraint graph is used as the allocation boundary of sparse anomaly components during the decomposition of the observation matrix, and is used to generate fixed connection relationships in the sparse latent source Bayesian factor graph.

[0033] S3, under the constraints of the influence constraint diagram, robust principal component analysis is performed on the observation matrix to separate the overall linkage change into low-rank common-mode components, and local anomalies propagating along the reachable propagation path are separated into sparse anomaly components. The distribution of the sparse anomaly components on unreachable measurement points is restricted according to the influence constraint diagram. Anomaly injection vectors are obtained by extracting the anomaly intensity of each measurement point combination from the sparse anomaly components. Figure 4 shown, specifically.

[0034] Map the connection relationship of measurement point combination in the influence constraint diagram to the corresponding measurement point position in the observation matrix. Open the sparse anomaly component allocation for the corresponding measurement point position with connection, and restrict the sparse anomaly component to zero value for the corresponding measurement point position without connection. Input the observation matrix into robust principal component analysis and decompose the observation matrix under the aforementioned allocation constraints; Extract the synchronously changing parts from multiple measuring points and continuous time windows from the decomposition results, and classify the synchronous changes caused by unit load, circulating water conditions and vacuum equipment regulation into low-rank common-mode components. Local changes that occur continuously along the same reachable propagation path in the decomposition results and are located in open allocation positions are classified as sparse outliers, while isolated changes that do not satisfy the continuity relationship of reachable propagation paths are excluded from sparse outliers. The sparse anomaly components are aligned according to the measurement point combination. The sparse anomaly components of each measurement point combination within the same time window are extracted and merged according to the order of reachable propagation paths to obtain the anomaly intensity of each measurement point combination. The anomaly intensity of each measurement point combination is arranged in the order of the measurement point combination column in the influence constraint diagram to obtain the anomaly injection vector, so that each component of the anomaly injection vector uniquely corresponds to a measurement point combination. Align the anomaly injection vector with the fixed connectivity in the influence constraint graph, and use it as the input to the observation unit and observation factor in the sparse latent source Bayesian factor graph; The overall linkage refers to a set of changes that occur synchronously across multiple measuring points within the same time window and are driven by adjustments in unit load, circulating water conditions, and vacuum equipment. The low-rank common-mode component is the data component corresponding to the overall linkage separated from the observation matrix by robust principal component analysis. The dimension of the low-rank common-mode component is consistent with that of the observation matrix. The anomaly intensity of each measurement point combination is the combined-level anomaly quantity obtained by merging the sparse anomaly components of each measurement point in the same time window according to the order of reachable propagation paths. The method of restricting the distribution of sparse outlier components on unreachable measurement points based on the influence constraint diagram is to map the combination of measurement points that are not connected in the influence constraint diagram to the measurement point positions in the observation matrix, and to keep the sparse outlier components at the corresponding measurement point positions as zero during the robust principal component analysis process. The anomaly injection vector is a vector formed by arranging the anomaly intensity of each measurement point combination according to the order of the measurement point combination column in the influence constraint diagram. The dimension of the anomaly injection vector is the number of measurement point combinations, and it is used to write the observation factor into the sparse latent source Bayesian factor diagram.

[0035] S4. Construct a sparse latent source Bayesian factor graph based on the candidate leak points, the influence constraint graph, and the anomaly injection vector. Set the candidate leak point number as a discrete latent variable and the leak intensity as a continuous latent variable. Perform joint posterior inference under the constraints of operating condition fluctuations and measurement point noise to obtain the posterior probability sequence and intensity estimation sequence. Figure 5 shown, specifically.

[0036] Using candidate leak points, influence constraint graphs, and anomaly injection vectors as inputs, a sparse latent source Bayesian factor graph is constructed according to the connection relationship between candidate leak points and measurement point combinations in the influence constraint graph. In the sparse latent source Bayesian factor graph, only factor connections with reachable propagation paths are retained, and factor connections without reachable propagation paths are canceled. In the sparse latent source Bayesian factor graph, an observation unit and observation factor are set up that correspond one-to-one with the measurement point combination. The anomaly injection vector is written into the corresponding observation factor in the order of measurement point combination, so that each observation factor only receives the anomaly intensity of the corresponding measurement point combination. In the sparse latent source Bayesian factor graph, a candidate leak point state unit and a leak intensity unit are set to correspond one-to-one with the candidate leak points. The candidate leak point number is set as a discrete latent variable, the leak intensity is set as a continuous latent variable, and each leak intensity unit is aligned and connected with the corresponding candidate leak point state unit. Based on the influence constraint diagram, path constraint units are set between the observed factors and the candidate leak point state units, so that each path constraint unit corresponds to a reachable propagation path, and the anomaly injection vector can only participate in the candidate leak point interpretation along the corresponding reachable propagation path. Operating condition fluctuation constraints are introduced into each candidate leak point state unit and leak intensity unit. The constraint information of unit load change, circulating water operating condition change and disturbance behavior caused by vacuum equipment adjustment is written into the posterior inference process to distinguish between candidate leak point interpretation and operating condition disturbance interpretation. Introducing measurement point noise constraints between each observation factor and the candidate leak point state unit restricts the abnormal bias of a single measurement point combination, so that the abnormal bias of a single measurement point combination cannot dominate the posterior probability and leakage intensity estimation of the candidate leak point. Under the combined influence of operating condition fluctuation constraints and measurement point noise constraints, joint posterior inference is performed on discrete latent variables and continuous latent variables to obtain posterior probability sequences and intensity estimation sequences that correspond one-to-one with each candidate leak point. Furthermore, the posterior probability sequences and intensity estimation sequences share the same connection structure and observation factors of the sparse latent source Bayesian factor graph.

[0037] The observation factor is a factor unit that corresponds one-to-one with the combination of measurement points in the sparse latent source Bayesian factor graph. Each observation factor receives a component of the corresponding combination of measurement points in the anomaly injection vector, and aligns the component with the reachability propagation effect of the candidate leak point according to the connection relationship in the influence constraint graph. The sparse latent source Bayesian factor graph is a fixed connectivity factor graph constructed based on candidate leak points, influence constraint graphs, and anomaly injection vectors. It includes observation units corresponding to measurement point combinations, path constraint units corresponding to reachable propagation paths, candidate leak point state units corresponding to candidate leak points, leak intensity units corresponding to candidate leak points, observation factor units corresponding to measurement point combinations, operating condition fluctuation constraint units corresponding to candidate leak points, and measurement point noise constraint units. The fixed connectivity factor graph only retains the positions where connections exist in the influence constraint graph. The step of setting the candidate leak point number as a discrete latent variable and the leak intensity as a continuous latent variable is to limit the value of the candidate leak point state unit to the discrete value of the candidate leak point number in the sparse latent source Bayesian factor graph, and limit the value of the leak intensity unit aligned with each candidate leak point state unit to the continuous intensity value of the corresponding candidate leak point, so as to simultaneously complete the leak point location judgment and the leak intensity judgment in the same inference process. The posterior inference under the constraints of operating condition fluctuation and measurement point noise involves incorporating the operating condition fluctuation constraint into the inference process of candidate leak point state unit and leak intensity unit in the sparse latent source Bayesian factor graph, and incorporating the measurement point noise constraint into the inference process between the observation factor and the candidate leak point state unit. Under the fixed connection relationship, the posterior probability and corresponding leak intensity of each candidate leak point are updated by combining the abnormal injection vector to obtain the posterior probability sequence and intensity estimation sequence.

[0038] S5, generate a posterior ranking based on the posterior probability sequence and the intensity estimation sequence. For example... Figure 6 shown, specifically.

[0039] The posterior probability sequence and intensity estimation sequence are aligned one by one according to the candidate leak point number, and kept consistent with the order of candidate leak points in the output layer of the sparse latent source Bayesian factor graph, so that each candidate leak point corresponds to a unique posterior probability and a unique intensity estimate. For each candidate leak point, a ranking pair consisting of posterior probability and intensity estimate is constructed. The posterior probability is determined as the primary ranking criterion, and the intensity estimate is determined as the secondary ranking criterion, so that the posterior ranking simultaneously represents the probability of occurrence of the candidate leak point and the corresponding leakage intensity. Candidate leak points are sorted from high to low according to their posterior probabilities. When the difference in posterior probabilities between adjacent candidate leak points is less than the preset sorting and resolution threshold, the order is determined from high to low according to the corresponding intensity estimates. Based on the ranking results, a posterior ranking of each candidate leak point is generated, and the posterior ranking is used as the verification input for the path consistency index and the measurement point consistency index.

[0040] S6, combining the influence constraint diagram and the anomaly injection vector, calculate the path consistency index and the measurement point consistency index, and verify the posterior ranking to obtain the target leak point or the leak point to be verified. For example... Figure 7 shown, specifically.

[0041] Based on the posterior ranking, the candidate leak points with the highest ranking are determined, and the reachable propagation path and measurement point combination connection relationship corresponding to the candidate leak point are extracted from the influence constraint diagram. The measurement points with abnormal intensity responses in the abnormal injection vector are matched one by one with the aforementioned reachable propagation paths, and the path consistency index is calculated according to the matching of the propagation order. The connection relationship between the combination of measurement points with abnormal intensity response in the abnormal injection vector and the aforementioned combination of measurement points is checked one by one, and the measurement point consistency index is calculated according to the connection coverage and abnormal intensity distribution. The posterior ranking is verified based on the path consistency index and the measurement point consistency index. The verification result is compared with the preset positioning threshold. When the verification result meets the preset positioning threshold, the candidate leak point at the top of the posterior ranking is determined as the target leak point. When the verification result does not meet the preset positioning threshold, the candidate leak point at the top of the posterior ranking is determined as the leak point to be verified.

[0042] The path consistency index is calculated by matching the combination of measurement points in the anomaly injection vector corresponding to the reachable propagation path of the first candidate leak point in the posterior ranking one by one according to the propagation order, and calculating the index based on the coverage of the matched measurement point combination in all measurement point combinations corresponding to the reachable propagation path and whether the propagation order is consistent. The consistency index of the measurement points is an index obtained by establishing a connection between the measurement point combination of the first candidate leak point in the posterior ranking in the influence constraint diagram, summarizing the anomaly intensity of the corresponding measurement point combination in the anomaly injection vector, and comparing the summarization result with the anomaly intensity distribution of all measurement point combinations in the anomaly injection vector. The preset positioning threshold is a combination of judgment thresholds pre-set for the path consistency index and the measurement point consistency index. When the path consistency index is not lower than the corresponding threshold and the measurement point consistency index is not lower than the corresponding threshold, the verification result is determined to meet the preset positioning threshold. When the path consistency index is lower than the corresponding threshold or the measurement point consistency index is lower than the corresponding threshold, the verification result is determined to not meet the preset positioning threshold.

[0043] Example 2 The following is a specific example: In this embodiment, step S1 specifically includes: Select the same operating period during the grid-connected operation of the unit As the data acquisition interval, the total number of all measuring points connected to the distributed control system within the condenser vacuum system is denoted as . , will the Each measuring point is denoted as ,in The measurement point number is used; the measurement point identifier is recorded as... Record the location of the device as For each measuring point Read continuous time-series data according to the original sampling order in the distributed control system, and then... Each sampling time is recorded as , measure point At sampling time The measured value is recorded as ,in This is the sampling sequence number. This forms the basis for each measurement point. In the runtime segment A continuous sequence of measurements, each measurement in the continuous sequence of measurements All are related to the unique sampling time Corresponding measurement point markings Used to distinguish different measuring points and the location of the equipment they belong to. Used to locate the installation position of the measuring point.

[0044] After obtaining the measurement point identification Location of the equipment Sampling time and measured values Then, based on the location of the equipment. Establish a location mapping. The location mapping uses measurement point identifiers. Corresponding unique installation location Established in this way. Unique installation location. The location is determined by both the equipment name and the installation location. The equipment name comes from the distributed control system's measurement point configuration, and the installation location comes from the measurement point layout record. After the location mapping is established, each measurement point... Each corresponds to a unique installation location in the condenser vacuum system. Different measuring points do not share the same installation location; there is only one installation location. It can be used directly as the name of the connection node in the device connection relationship, so that the position of the measurement point row can directly correspond to the position of the device node.

[0045] After establishing the location mapping, organize the equipment connection relationships. The equipment interface locations, pipe section connection locations, and measuring point installation locations in the condenser vacuum system are uniformly defined as connection nodes, and the total number of connection nodes is recorded as follows: , will the The number of connected nodes is denoted as . ,in Let be the node sequence number. Based on the system piping connection diagram, equipment interface records, and measurement point installation locations, establish a connection description for any two actually connected nodes, and record the total number of connection descriptions as . , will the The link description is denoted as ,in This refers to the sequence number of the connection description. Each connection description... Each connection includes a start point, a finish point, and a propagation direction. The start and finish points are represented by connection nodes. This indicates that the propagation direction is determined according to the actual direction of medium transmission within the condenser vacuum system. When a device has multiple interface positions, each interface position is treated as a different connection node. (This is described in the complete connection description.) The device connection relationships are formed. Each connection description in the device connection relationship corresponds to an actual device connection channel. Therefore, the device connection relationship not only indicates whether the device entities are connected, but also clearly indicates the order in which leakage disturbances propagate within the system.

[0046] After the device connection relationships are determined, the continuous time-series data is divided into columns according to a preset time window. The preset window length is denoted as... ,in This represents the number of consecutive sampling points contained in each time window; the total number of time windows is denoted as... , will the Each time window is recorded as ,in This refers to the time window number. Each time window... From continuous Each sampling time Composition. For each measuring point and each time window Select time window The measurement value at the end time is used as the time window. The window value is determined and written into the observation matrix. The corresponding positions. Construct the observation matrix. At that time, the rows are arranged according to the measurement points in the location mapping, and the columns are arranged according to the time window. The columns are arranged in the order of their positions. This gives us the dimension. observation matrix Observation matrix The Fixed corresponding measuring points and the only installation location Observation matrix The Columns correspond to fixed time windows Observation matrix Each element of the matrix is ​​the original measurement value of the specified measuring point at the end of the specified time window; therefore, the observation matrix... The row positions correspond one-to-one with the connection node positions in the device connection relationship, and the observation matrix It can be directly used as input for robust principal component analysis.

[0047] The observation matrix , is a data matrix arranged with measurement points as the row dimension and time windows as the column dimension, observation matrix The dimension is ,in The total number of measurement points. Total number of time windows; observation matrix Each row represents the time-series observations of the same measuring point within a continuous time window; the observation matrix... Each column represents the synchronous observations of various measuring points within the same time window; the observation matrix... Each element of the matrix represents the measurement value of the corresponding measuring point at the end of the corresponding time window; the observation matrix... Used as input for robust principal component analysis and to maintain a consistent correspondence between the sequence of measurement points and the construction of measurement point combinations.

[0048] The device connection relationship is as follows: Link description A set of directed connections, where each connection describes This corresponds to a single actual device connection channel, and includes the connection start point, connection end point, and propagation direction; the connection start point and connection end point are determined by the connection node. Confirm, connect node This includes the location of equipment interfaces, pipe connection locations, and measurement point installation locations. The propagation direction is determined according to the actual transmission direction of the medium within the condenser vacuum system. Equipment connection relationships are used to determine the reachable propagation paths from candidate leak points to each measurement point and to generate fixed connection relationships in the influence constraint diagram.

[0049] In this embodiment, step S2 specifically includes: Using the observation matrix Candidate leaks and path constraints are constructed based on the row position correspondence and device connection relationships. Observation matrix The Row corresponding measurement points The only installation location Read all connection start and end points from the device connection relationships, and record each connection position as a connection node. ,in This refers to the connection node number. For each unique installation location... In the connection node Search for nodes with the same name in the middle, and determine the retrieved connected nodes as test points. Corresponding device nodes Then, all connected nodes... Node type filtering is performed, retaining only connection nodes belonging to equipment interface locations or equipment-pipeline connection locations, excluding connection nodes used solely as measurement point installation locations. The retained connection nodes are then deduplicated to determine candidate leak points. The total number of candidate leak points is recorded as follows: , will the Each candidate leak point is denoted as... Therefore, the observation matrix row position and device node A fixed correspondence is established, and the actual device connection locations in the device connection relationship are specified as candidate leak points. .

[0050] At candidate leaks Once identified, for each candidate leak point... With each device node Perform reachability propagation path retrieval. During the retrieval, candidate missed points are identified. Write the current search node table and scan the connection descriptions one by one. When a certain connection description When the starting point of the connection is consistent with a node in the current search node table, the connection description is... The endpoint of the connection is written to the next-level retrieval node table, and the predecessor node and corresponding connection description of the endpoint are also recorded. After completing one layer of scanning, the next layer's search node table is replaced with the current search node table, and the scanning continues until a device node appears in the current search node table. Or, the current search node table no longer generates new nodes. When a device node appears in the current search node table... At that time, according to the recorded predecessor node and corresponding connection description Self-device node Backtracking step by step to candidate leaks , obtain from candidate leak points To device node An ordered sequence of connections is identified, and this ordered sequence is denoted as a reachable propagation path. When multiple ordered connection sequences are obtained through backtracking, the number of connection nodes in each ordered connection sequence is counted, and the ordered connection sequence with the fewest connection nodes is taken as the reachable propagation path. When the current search node table no longer generates new nodes and no device node has yet appeared. At that time, the measuring point Relative to candidate leaks The point has been determined to be unreachable.

[0051] All reachable propagation paths Once obtained, it is propagated along the reachable path. The measurement points are merged to form measurement point combinations. For fixed candidate leak points... First, collect all existing reachable propagation paths. measuring points Then based on the device nodes In reachable propagation path The measurement points are sorted according to their order of appearance. After sorting, measurement point combinations are established starting from the measurement point with the highest propagation order. When two adjacent measurement points are located within the same reachable propagation path and there are no other measurement point corresponding to the device nodes between the two device nodes, the two measurement points are grouped into the same measurement point combination. If this condition is not met, the current measurement point combination is terminated and a new measurement point combination is established from the current measurement point. The total number of measurement point combinations formed is denoted as . , will the The combination of measurement points is denoted as Each measurement point combination It consists of one or more measurement point identifiers arranged in the propagation sequence, therefore each measurement point combination All of them have clearly defined measurement point members and a clear propagation order, and can be directly used to define column objects for anomaly injection vectors.

[0052] Figure 8The diagram illustrates the equipment connections between the condenser, vacuum pumping equipment, measuring points, and candidate leak points in a condenser vacuum system. Dashed lines represent the reachable propagation paths of air leakage anomalies within the system. The correspondence between candidate leak points 1, 2, and 3 and measuring points 1 to 4 visually demonstrates that the anomaly does not spread arbitrarily but is constrained by the physical connections of the equipment and the direction of propagation. The diagram highlights the advantages of this invention: first, it utilizes the system topology to limit the range of candidate sources and reachable measuring points, then it performs anomaly extraction and posterior inference, thereby reducing interference from irrelevant candidate points and improving the accuracy, interpretability, and adaptability to fluctuations in operating conditions.

[0053] In the combination of measuring points Once determined, construct the influence constraint diagram. Influence constraint diagram The influence constraint diagram is represented by a connection matrix where candidate leak points are row objects and measurement point combinations are column objects. The dimension is The influence constraint diagram The Line number Column elements are denoted as For any candidate leak point Combined with any measuring point Search for candidate leaks All corresponding reachable propagation paths When there exists a reachable propagation path that simultaneously contains a combination of measurement points. All measurement points within the path correspond to device nodes, and the order of these device nodes in the reachable propagation path is consistent with the combination of measurement points. When the propagation order is consistent, Set as When no reachable propagation path exists that satisfies this condition, Set as Composed of all elements Composition Influence Constraint Diagram The reachable propagation path To identify candidate leaks in device connectivity relationships Starting from the direction of propagation and arriving at the measuring point Corresponding device node An ordered connection sequence; the influence constraint graph Based on candidate leak points To the measurement point combination Does a sparse connection structure exist that allows for the construction of reachable propagation paths? This leads to a sequence of candidate leak points. and measurement point combination sequence Candidate leak sequence From all candidate leaks Composed in row order, with dimensions of Measurement point combination sequence Combination of all measuring points Composed in column order, with dimensions as follows: Influence constraint diagram The dimension is Candidate leak sequence Used to define candidate leak point state units in a sparse latent source Bayesian factor graph, and a sequence of measurement point combinations. Used to define the component order of the anomaly injection vector, affecting the constraint graph. Used to construct structural constraints for robust principal component analysis and fixed connectivity relationships for sparse latent source Bayesian factor graphs.

[0054] In this embodiment, step S3 specifically includes: Using observation matrix Influence constraint diagram and measurement point combination sequence As input for computation: observation matrix The dimension is ,in Indicates the total number of measuring points. Represents the total number of time windows; influence constraint diagram The dimension is ,in This represents the total number of candidate leaks. Represents the total number of measurement point combinations; measurement point combination sequence By the Combination of measurement points Composition. For each measurement point combination... First, read the combination of measuring points. The entire measurement point identifier is included in the observation matrix. The row position corresponding to the measurement point in the center is determined. This affects the constraint diagram. The If at least one position in the column has an established connection, then the measurement points will be combined. The corresponding positions of all measurement points are determined as the locations of open sparse anomaly components; if the influence constraint diagram is... The If all positions in the column are not connected, then combine the measurement points. The corresponding positions of all measurement points are determined as zero-value-limited positions. Open sparse outlier component allocation positions allow sparse outlier components to be written throughout the entire time window, while zero-value-limited positions maintain sparse outlier components at zero values ​​throughout the entire time window.

[0055] To enable the values ​​from different measuring points to be directly merged within the same measuring point combination, the observation matrix is... Perform scaling on each measurement point row. The upper limit of the measuring range at each measuring point is denoted as upper limit of measuring range The range definition values ​​are taken from the measurement point configuration in the distributed control system. For the observation matrix... The Divide each measurement value in each time window by the upper limit of the measurement range. This yields a dimensionless observation matrix. In the scaled observation matrix Perform robust principal component analysis and record the decomposition results as low-rank common-mode components. and sparse outlier components Low-rank common-mode components Dimensions and observation matrix Consistent, sparse outliers Dimensions and observation matrix Consistent. The decomposition employs an iterative update method with positional constraints, first reading the observation matrix. Then update the low-rank common-mode components using all matrix elements. To accommodate changes occurring simultaneously across multiple measurement points, and subsequently update sparse anomaly components. To accommodate local changes. Update sparse outlier components. At this time, non-zero values ​​are only allowed to be written at the allocation positions of open sparse outlier components, while zero values ​​are directly written at the zero-value-limited positions and remain unchanged throughout all iterations. The maximum number of iterations is denoted as... In this embodiment Pick The decomposition ends when the maximum number of iterations is reached or when the distribution of non-zero positions remains consistent between two adjacent iterations.

[0056] For low-rank common-mode components Perform overall linkage identification. Define the set of changes that simultaneously appear at multiple measurement point positions within the same time window and continue to appear within consecutive time windows as overall linkage. During identification, read the low-rank common-mode components column by column. The numerical distribution is analyzed, and the number of measurement points exhibiting synchronous changes within the same time window is counted. The duration of this synchronous change within adjacent time windows is also calculated. In this embodiment, the preset common-mode measurement point threshold is set to... For each measurement point, a preset continuous time window threshold is taken. A time window. When the number of synchronously changing measuring points is not less than [number missing] One, and the duration is not less than Within a specific time window, this change is incorporated into the overall linkage. Disturbances caused by unit load changes, circulating water condition changes, and vacuum equipment adjustments are thus preserved in the low-rank common-mode component. middle.

[0057] For sparse outliers Perform path continuity verification. During verification, use a sequence of measurement points. Each measurement point combination The measurement point combination is read sequentially according to the reachable propagation path within the same time window, using units of measurement. The sparse anomaly component values ​​of each measurement point included. In this embodiment, the threshold for continuous measurement points along the preset path is taken as... Each measuring point. When the same measuring point is combined... Measurement points with non-zero values ​​occupy consecutive positions in the reachable propagation path, and the number of consecutive measurement points is not less than [number missing]. At this time, the local change is preserved in the sparse outlier component. In the middle, when a non-zero value appears only at a single isolated measurement point, or when the measurement point corresponding to a non-zero value is discontinuous in the reachable propagation path, the corresponding position is reset to zero. Through this processing, sparse outlier components... The non-zero values ​​retained in the data only correspond to local changes that satisfy the continuity relationship of reachable propagation paths.

[0058] In sparse outlier components After completing the path continuity verification, the anomaly intensity of each measurement point combination is extracted. The observation matrix is ​​then... The The time window corresponding to the column is denoted as the current analysis time window. For each combination of measuring points Combine the measuring points The number of measuring points arranged in order of reachable propagation paths is denoted as , will the Each measurement point in the observation matrix The corresponding row number is denoted as First, based on the ratio of the allowable zero-point drift value to the upper limit of the measurement range at each measuring point, the largest ratio among all measuring points is taken as the preset sparse anomaly threshold. Then, using the current analysis time window... Calculation of sparse anomaly component values ​​within the first part Combination of measurement points abnormal intensity : ; in, Indicates the first Combination of measurement points In the current analysis time window Abnormal intensity within; Indicates the sequence number of the measuring point combination; Indicates the combination of measuring points The number of measuring points included; Indicates the combination of measuring points The measurement point numbers are arranged in order of reachable propagation paths. Indicates the combination of measuring points The Middle Each measurement point in the observation matrix The corresponding row number; Represents sparse outlier components The Middle Line number The values ​​in the column; Indicates the current analysis time window Column number in all time windows; Indicates a continuous path marker, when and Not less than The time value is ,when and Not less than And the first For each measurement point, at least one of its adjacent measurement points along the reachable propagation path has an absolute value of at least one sparse anomaly component that is not less than [value missing]. The time value is In other cases, the value is [value]. The first measuring point of the path only checks the next adjacent measuring point, the last measuring point of the path only checks the previous adjacent measuring point, and the middle measuring point of the path checks both the previous and next adjacent measuring points. This indicates the preset sparse anomaly threshold; Indicates taking the absolute value; Indicates the combination of measuring points All measurement points within the area are accumulated in order of reachable propagation paths; This indicates that when there are no valid continuous anomalies within the current analysis time window, the denominator is limited to 0. Due to the observation matrix Scale-up processing has been completed, sparse outlier components have been removed. Matrix elements, preset sparsity anomaly threshold and abnormal intensity All values ​​are dimensionless, and the dimensions of all participating quantities in the formula are consistent.

[0059] Combine all measuring points abnormal intensity According to the influence constraint diagram The measurement point combination columns are arranged in sequence to obtain the anomaly injection vector. Anomaly injection vector Depend on to Sequential composition, anomaly injection vector The dimension is Each component uniquely corresponds to a combination of measurement points. Influence constraint diagram Keep The fixed connection structure, the exception injection vector of Each component is written into the sparse latent source Bayesian factor graph. Each observation unit, influence constraint diagram Locations where connections have been established are written into the fixed connection between the observation unit and the candidate leak point status unit; locations where no connection has been established remain disconnected. The overall linkage refers to a set of changes occurring synchronously across multiple measuring points within the same time window and driven by adjustments in unit load, circulating water conditions, and vacuum equipment; the low-rank common-mode component... For robust principal component analysis, from the observation matrix The data components separated from the overall linkage; the anomaly injection vector To determine the anomaly intensity of each measurement point combination according to the influence constraint diagram The vector formed by the sequential arrangement of the measurement points is used to write the observation factors into the sparse latent source Bayesian factor map.

[0060] In this embodiment, step S4 specifically includes: Using candidate leak sequence Influence constraint diagram Anomaly injection vector and the low-rank common-mode components corresponding to the current analysis time window As input for computation: Candidate leak sequence Depend on One candidate leak composition, Indicates the candidate leak point number; influence constraint diagram The dimension is , Represents the total number of measurement point combinations; anomaly injection vector The dimension is , No. Each component is denoted as , Indicates the measurement point combination number; low-rank common-mode component. The dimension is consistent with the observation matrix, and the column number of the current analysis time window is denoted as... According to the influence constraint diagram Construct a sparse latent source Bayesian factor graph at the locations where connections have already been established. Input layer settings Individual observation units; connection layer settings and influence constraint diagram The path constraint units with one-to-one correspondence between connection positions have been established; the hidden variable layer is set. One candidate leak state unit and One leakage strength element; constraint layer settings Each observed factor Individual operating condition fluctuation constraint unit and Noise constraint unit at each measurement point; Output layer settings Each posterior probability unit and Intensity estimation unit. Influence constraint diagram. No factor connections are set at positions where no connections are established.

[0061] Injecting the anomaly vector Write in sequence according to the measurement point combination order. Each observation unit transmits its corresponding component to the observation factor with the same ordinal number. The observation factor only receives the first observation factor. 1 abnormal injection vector component Discrete latent variables are denoted as... Discrete latent variables The range of values ​​is to This is used to indicate which candidate leak point the air leak occurred at; (using...) Each candidate leak point state unit represents a discrete hidden variable. The state when retrieving the candidate leak point number. The [number]th [missing information]. The continuous latent variables in each leakage intensity element are denoted as follows: , Indicates candidate leak points The leakage intensity, Take non-negative continuous values. Each leakage intensity unit is only aligned and connected to the state unit of the candidate leakage point with the same sequence number.

[0062] For any location where a connection has been established In the The observed factor and the first A path constraint unit is set between each candidate leak point status unit, and the corresponding candidate leak points are... To the The reachable propagation path of the nth measurement point combination is written into this path constraint unit. The path constraint unit only allows the nth measurement point combination to be written into this path constraint unit. The explanatory information generated by each observed factor is transmitted to... The corresponding candidate leak point state unit and leak strength unit are not allowed to be passed to unconnected candidate leak points. Therefore, the anomaly injection vector... Each component can only move along the influence constraint diagram. Locations with established connections are included in the candidate leak analysis.

[0063] Operating condition fluctuation constraints use low-rank common-mode components Dimensionless numerical values ​​are constructed within the current analysis time window. Noise constraints at measurement points are constructed using the ratio of the allowable error at each measurement point to the upper limit of the measurement range. The [number]th [measurement]... The allowable error for each measuring point is denoted as... Allowable error Take the absolute error value corresponding to the instrument accuracy parameter at that measuring point; and take the first... The upper limit of the measuring range at each measuring point is denoted as upper limit of measuring range The range definition values ​​are taken from the measurement point configuration in the distributed control system. Calculations are performed for each measurement point. Then, take the arithmetic mean of the ratios corresponding to all measuring points within each measuring point combination, and finally take the maximum value among the arithmetic means of all measuring point combinations to obtain the preset upper limit of a single combination. The weight of the operating condition fluctuation constraint is denoted as... In joint posterior inference, the following formula is used to update the first... The posterior probability of each candidate leak point : ; in, Indicates the first One candidate leak The posterior probability; Influence constraint diagram The Middle Line number The column join values ​​are retrieved when the join is established. Take when no connection is established ; Represents the anomaly injection vector The One component; This indicates the preset maximum limit for a single combination; Indicates the weight of the operating condition fluctuation constraint; Indicates the total number of measuring points; Represents low-rank common-mode components The Middle Line number The values ​​in the column; Indicates the measurement point number; Indicates the current analysis time window column number; This represents the total number of measurement point combinations; Indicates the total number of candidate leaks; Indicates the candidate leak number used when normalizing the denominator; Indicates the operation of natural exponents; This indicates the operation of taking the smaller value; This indicates the operation of taking the larger value; This represents an indicator function, which is selected when the condition within the parentheses is true. If the condition is not met, take ; Indicates taking the absolute value; This indicates that the summation is performed according to the corresponding sequence number. (Anomaly injection vector) Low-rank common-mode components And preset single combination limit All values ​​are dimensionless, and the dimensions of all participating quantities in the formula are consistent.

[0064] In each round of joint posterior inference, the observed factor first receives the anomaly injection vector component. Then the path constraint unit is determined according to the influence constraint diagram. The fixed connection relationship transmits observation support, and then the measurement point noise constraint unit performs anomaly constraint on the single measurement point combination. Cut-off, determined by the operating condition fluctuation constraint unit according to The observation support corresponding to the common-mode perturbation is attenuated, and finally the posterior probability of all candidate leak points is updated according to the above formula. After obtaining all Then, for each candidate leak point Search Influence Constraint Graph No. Read the corresponding exception injection vector component from all established connection positions in the row. For each implement Truncate, then sum the truncated values ​​and multiply by the current posterior probability. Then divide by the number of active connections to get the number of connections. The leakage intensity of each leakage intensity unit The number of activated connections is the influence constraint graph. No. Satisfaction in the line and The number of measurement point combinations; when the number of activated connections is At that time, Set as The maximum number of inference rounds is denoted as... When the candidate leak point rankings obtained from two adjacent rounds of inference are consistent, or when the number of inference rounds reaches a certain threshold... At that time, the joint posterior inference ends. All The candidate leak points are arranged in order to form a posterior probability sequence. ,in , dimension ; All The candidate leak points are arranged in order to form an intensity estimation sequence. ,in , dimension Posterior probability sequence With intensity estimation sequence They share the same sparse latent source Bayesian factor graph connection structure and observation factors.

[0065] In this embodiment, step S5 specifically includes: Using candidate leak sequence Posterior probability sequence and intensity estimation sequence As input. Candidate leak sequence Depend on One candidate leak composition, Indicates the candidate leak number; posterior probability sequence Recorded as ,in Indicates candidate leak points Posterior probability; intensity estimation sequence Recorded as ,in Indicates candidate leak points The estimated inflow intensity. Candidate leak sequence. Posterior probability sequence and intensity estimation sequence All candidate leak point numbering orders are maintained in the same order as the output layer of the sparse latent source Bayesian factor graph. For the... One candidate leak Read the posterior probability sequence Posterior probability of the same number position Then read the intensity estimation sequence. Intensity estimate of the same numbered position and will and Binding to the same candidate leak Up, thus making each candidate leak point A unique posterior probability and the only intensity estimate .

[0066] After completing the one-to-one alignment, for each candidate leak point Construct sorted pairs Sort pairs A candidate leak point is composed of three data points in a fixed order: candidate leak point number, posterior probability, and intensity estimate. The sorted records. Sort pairs In the mean, posterior probability As the primary ranking criterion, the intensity estimate This serves as the basis for secondary sorting. After adopting this construction method, any candidate leak point... The sorting position is no longer determined solely by the probability of occurrence, but rather by both the probability of occurrence and the leakage intensity. To ensure that the sorting process has clear discrimination criteria, a preset sorting discrimination threshold is set. Preset sorting and resolution threshold This is used to determine whether the difference in posterior probability between two candidate leaks is sufficient to directly distinguish their order. In this embodiment, a preset sorting discrimination threshold is used. Pick .

[0067] For all sorted pairs When performing sorting, first compare any two candidate leak points. and posterior probability and .when At that time, candidate leak points Listed as candidate leak points Previously; when At that time, candidate leak points Listed as candidate leak points Previously; when When the posterior probabilities of the two values ​​are determined to be in the same resolution interval, the intensity estimates are then compared. and Candidate leak points with larger intensity estimates are ranked higher; when and When sorting, candidate leak points are arranged in ascending order of their numbers. This sorting rule ensures a fixed execution order: first, posterior probabilities are compared; then, intensity estimates are compared; and finally, candidate leak point numbers are compared, thus avoiding uncertainty in the sorting results.

[0068] All candidate leaks After sorting is complete, a posterior sorted sequence is generated based on the sorting results. Posterior sorted sequence Recorded as ,in Indicates ranking in Candidate leak point number, Indicates the sort order number. Posterior sort sequence. The dimension is posterior sorted sequence Each component represents a candidate leak point number, and the component position indicates the posterior ranking of the corresponding candidate leak point among all candidate leak points. The posterior ranking sequence... During bit-by-bit reading, the posterior probability corresponding to the candidate leak point can be retrieved simultaneously. and intensity estimates This results in a ranking of candidate leak points that maintains consistent numbering and clear position. The posterior ranking sequence... Write the verification input position of the path consistency index and the posterior sort sequence. Write the verification input position of the measurement point consistency index so that both the path consistency index and the measurement point consistency index are verified in the same candidate leak point order.

[0069] In this embodiment, step S6 specifically includes: Using posterior ordering sequence Influence constraint diagram Anomaly injection vector and measurement point combination sequence As input for computation: posterior ordered sequence Recorded as ,in Indicates ranking in Candidate leak point number, Represents the total number of candidate leaks; Influence constraint diagram The matrix elements are denoted as ,in Indicates candidate leak points Combined with measuring points There is a connection. Indicates candidate leak points Combined with measuring points No connection exists; anomaly injection vector Recorded as ,in Indicates the first Combination of measurement points abnormal intensity, Represents the total number of measurement point combinations; measurement point combination sequence Recorded as Used to generate an influence constraint diagram Column objects and exception injection vectors The component object. First read the posterior sorted sequence. First number Numbered The candidate leaks were identified as the top-ranked candidate leaks. Read the influence constraint diagram again. The Okay, all will be satisfied. Measurement point combination Extract them and simultaneously retrieve candidate leak points. To each connected measurement point combination The records of reachable propagation paths are compiled. All reachable propagation path records are categorized according to the rule of consistent propagation origin and continuous propagation order to form candidate leak points. The corresponding path sequence, each path in the path sequence contains a list of measurement point combinations arranged in the propagation order.

[0070] At candidate leaks Once the corresponding path sequence is determined, the anomaly injection vector is... The combination of measuring points exhibiting abnormal intensity responses will be screened. Those that meet the criteria will be selected. Measurement point combination The response measurement point combinations were determined. Then, each response measurement point combination was... Mapping each one to a candidate leak point In the corresponding path sequence, check the combination of response measurement points. The path consistency index is determined by whether the position of the measured point in the reachable propagation path matches the propagation order in the path record. For each reachable propagation path, first count the total number of all measured point combinations in the path, then count the number of matched measured point combinations located on the path within the response measured point combinations. When matched measured point combinations appear sequentially according to their positions in the path record, and there is no reverse jump within the path, the reachable propagation path is identified as a path with consistent order. When matched measured point combinations undergo reverse jumps, the reachable propagation path is identified as a path with inconsistent order. For all path-consistent paths, calculate the coverage ratio of the number of matched measured point combinations to the total number of measured point combinations in that path, and then take the arithmetic mean of the coverage ratios of all path-consistent paths to obtain the path consistency index. When no sequentially consistent path exists, the path consistency index will be... Set as Path consistency index This directly characterizes the coverage and propagation sequence matching of the response measurement point combination to the reachable propagation path.

[0071] In path consistency metrics After the calculation is completed, the consistency index of the measuring points is... Perform the calculations. First, read the influence constraint diagram. No. All rows are satisfied Measurement point combination This combination of measurement points was identified as a candidate leak point. The connection measurement point combination. Then inject the abnormal vector. Read the anomaly intensity corresponding to all connected measurement point combinations. The anomaly intensities of this portion are summed one by one to obtain the total connection anomaly intensity; at the same time, the anomaly injection vector is read. Anomaly intensity of all measuring point combinations The total global anomaly intensity is obtained by summing up all the anomaly intensities one by one. Then, the combinations of connected measurement points that satisfy the following conditions are statistically analyzed. The number of response measurement point combinations is calculated, and the number of combinations that satisfy the requirements is counted. The number of response measurement point combinations is used to obtain the number of connected responses and the number of global responses. The ratio of the total connected anomaly strength to the total global anomaly strength is used to obtain the strength coverage value; the ratio of the number of connected responses to the number of global responses is used to obtain the connection coverage value; finally, the arithmetic mean of the strength coverage value and the connection coverage value is taken to obtain the measurement point consistency index. When the total global anomaly intensity is Or the global response count is At that time, the consistency index of the measuring points will be... Set as Consistency index of measurement points This directly reflects the candidate leak point. In the influence constraint diagram The connection of measurement points is established to inject anomaly vectors. The extent of the explanation coverage.

[0072] In path consistency metrics Consistency index with measuring points After obtaining the data, perform a post-hoc sorting check. Record the path consistency index threshold as... The threshold for the consistency index of measurement points is denoted as In this embodiment, Pick , Pick First, compare path consistency metrics. Path consistency index threshold Then compare the consistency index of the measurement points. Consistency index threshold with measurement points .when Not less than ,and Not less than When the verification result is determined to meet the preset positioning threshold, the candidate leak point ranked first in the posterior ranking will be identified. Identified as the target leak point; when Below ,or Below When the verification result is determined to be a candidate leak point that does not meet the preset positioning threshold, the candidate leak point that ranks first in the posterior ranking will be identified. This has been identified as a leak point to be verified. A path consistency index is then generated based on this. Consistency index of measurement points Verification results and target or pending leaks.

[0073] Example 3 To achieve the above embodiments, such as Figure 9As shown, this embodiment also provides a device 10 for locating air leaks in a condenser of a thermal power plant. The device 10 includes a data acquisition module 100, an influence constraint graph construction module 200, an anomaly injection vector extraction module 300, a posterior inference module 400, and a location output module 500.

[0074] The data acquisition module 100 is used to acquire the distributed control system measurement point data and equipment connection relationship of the condenser vacuum system during unit operation, and to arrange the distributed control system measurement point data in time sequence according to the measurement point identification and equipment location to construct an observation matrix; The influence constraint graph construction module 200 is used to determine candidate leaks based on the device connection relationship, retrieve the reachable propagation path from each candidate leak to each measurement point, merge the measurement points according to the reachable propagation path to form a measurement point combination, and construct an influence constraint graph that represents the connection relationship between the candidate leak and the measurement point combination. Anomaly injection vector extraction module 300 is used to perform robust principal component analysis on the observation matrix under the constraints of the influence constraint diagram, separate the overall linkage change into low-rank common mode components, separate the local anomalies propagating along the reachable propagation path into sparse anomaly components, restrict the distribution of the sparse anomaly components on unreachable measurement points according to the influence constraint diagram, and extract the anomaly intensity of each measurement point combination from the sparse anomaly components to obtain the anomaly injection vector. The posterior inference module 400 is used to construct a sparse latent source Bayesian factor graph based on the candidate leak points, the influence constraint graph and the anomaly injection vector, set the candidate leak point number as a discrete latent variable and the leak intensity as a continuous latent variable, and perform joint posterior inference under the constraints of operating condition fluctuation and measurement point noise to obtain the posterior probability sequence and intensity estimation sequence. The positioning output module 500 is used to generate a posterior ranking based on the posterior probability sequence and the intensity estimation sequence; and to calculate the path consistency index and the measurement point consistency index by combining the influence constraint diagram and the anomaly injection vector, and to verify the posterior ranking to obtain the target leak point or the leak point to be verified.

[0075] An embodiment of the present invention provides a device for locating air leaks in a condenser of a thermal power plant, which can effectively separate the linkage of operating conditions from abnormal leakage propagation, achieve precise location of air leaks in the condenser under complex fluctuations, and improve the interpretability and verifiability of diagnostic results.

[0076] To implement the methods of the above embodiments, the present invention also provides a computer device, such as... Figure 10As shown, the computer device 600 includes a memory 601 and a processor 602; wherein, the processor 602 reads executable program code stored in the memory 601 to run a program corresponding to the executable program code, so as to implement the various steps of the method described above.

[0077] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0078] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0079] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for locating air leak points in a condenser of a thermal power plant, characterized in that, include: Acquire the distributed control system measurement point data and equipment connection relationships of the condenser vacuum system during unit operation, and arrange the distributed control system measurement point data in time sequence according to the measurement point identification and equipment location to construct an observation matrix; Candidate leaks are determined based on the device connection relationship. The reachable propagation paths from each candidate leak to each measurement point are retrieved. The measurement points are merged into measurement point combinations according to the reachable propagation paths. An influence constraint diagram characterizing the connection relationship between candidate leaks and measurement point combinations is constructed. Robust principal component analysis is performed on the observation matrix under the constraints of the influence constraint diagram. The overall linkage change is separated into low-rank common mode components, and the local anomalies propagating along the reachable propagation path are separated into sparse anomaly components. The distribution of the sparse anomaly components on unreachable measurement points is restricted according to the influence constraint diagram. The anomaly injection vector is obtained by extracting the anomaly intensity of each measurement point combination from the sparse anomaly components. Based on the candidate leak points, the influence constraint diagram and the abnormal injection vector, a sparse latent source Bayesian factor diagram is constructed. The candidate leak point number is set as a discrete latent variable and the leak intensity is set as a continuous latent variable. Under the constraints of operating condition fluctuation and measurement point noise, joint posterior inference is performed to obtain the posterior probability sequence and intensity estimation sequence. A posterior ranking is generated based on the posterior probability sequence and the intensity estimation sequence; By combining the influence constraint diagram and the anomaly injection vector to calculate the path consistency index and the measurement point consistency index, the posterior ranking is verified to obtain the target leak point or the leak point to be verified.

2. The method as described in claim 1, characterized in that, The construction of the observation matrix includes: Determine the measurement point identifier, sampling time, and location of the equipment associated with each measurement point, and extract continuous time-series data of each measurement point within the same runtime period; Establish a position mapping for each measuring point based on the location of the equipment to which it belongs, so that each measuring point corresponds to a unique installation position in the condenser vacuum system; The continuous time series data is divided into columns according to a preset time window, and the distributed control system measurement point data is arranged with the measurement point as the row dimension and the time window as the column dimension to obtain the observation matrix.

3. The method as described in claim 1, characterized in that, The construction of the influence constraint graph characterizing the connection relationship between candidate leak points and measurement point combinations includes: Based on the position of the measurement point row in the observation matrix, locate the corresponding device node in the device connection relationship, and determine the device connection position in the device connection relationship as a candidate leak point; Starting from each candidate leak point, the propagation path from the candidate leak point to each measurement point is determined by searching along the propagation direction in the device connection relationship to the corresponding device node. Based on the reachable propagation path corresponding to each candidate leak point, the measurement points that are in the same reachable propagation path and are consecutively corresponding in the propagation order are aligned and merged to form a measurement point combination corresponding to the reachable propagation path. Using candidate leak points as row objects and measurement point combinations as column objects, determine whether there are reachable propagation paths from candidate leak points to each measurement point combination. If a reachable propagation path exists, establish a connection to obtain an influence constraint graph.

4. The method as described in claim 1, characterized in that, The step of extracting the anomaly injection vector from the anomaly intensity of each measurement point combination from the sparse anomaly components includes: Map the connection relationship of measurement point combination in the influence constraint diagram to the corresponding measurement point position in the observation matrix. Open the sparse anomaly component allocation for the corresponding measurement point position with connection, and restrict the sparse anomaly component to zero value for the corresponding measurement point position without connection. The observation matrix is ​​input into robust principal component analysis and decomposed under allocation constraints. The synchronous changes caused by unit load, circulating water conditions and vacuum equipment regulation are attributed to low-rank common-mode components. Local variations in the decomposition results that are located at open allocation positions and occur continuously along the same reachable propagation path are classified as sparse outlier components. The sparse anomaly components are aligned according to the measurement point combination. The sparse anomaly components of each measurement point combination within the same time window are extracted and merged according to the order of reachable propagation paths to obtain the anomaly intensity of each measurement point combination. The anomaly intensity of each measurement point combination is arranged in the order of the measurement point combination column in the influence constraint diagram to obtain the anomaly injection vector.

5. The method as described in claim 1, characterized in that, The step of constructing a sparse latent source Bayesian factor graph based on the candidate leak points, the influence constraint graph, and the anomaly injection vector, setting the candidate leak point numbers as discrete latent variables and the leak intensity as continuous latent variables, and performing joint posterior inference under operating condition fluctuation constraints and measuring point noise constraints to obtain the posterior probability sequence and intensity estimation sequence includes: Based on the connection relationship between candidate leak points and measurement point combinations in the influence constraint diagram, a sparse latent source Bayesian factor diagram is constructed, retaining only the factor connections with reachable propagation paths. Set up observation units and observation factors that correspond one-to-one with the measurement point combination, and write the anomaly injection vector into the corresponding observation factors in the order of the measurement point combination. Set up candidate leak point state units and leak intensity units that correspond one-to-one with candidate leak points, set the candidate leak point number as a discrete latent variable, and set the leak intensity as a continuous latent variable. Based on the influence constraint diagram, path constraint units are set between the observed factors and the candidate leak point state units, so that the anomaly injection vector participates in the interpretation of candidate leak points along the corresponding reachable propagation path. Operating condition fluctuation constraints are introduced on each candidate leak point state unit and leak intensity unit, and measurement point noise constraints are introduced between each observation factor and the candidate leak point state unit. Under the combined influence of operating condition fluctuation constraints and measurement point noise constraints, joint posterior inference is performed on discrete and continuous latent variables to obtain posterior probability sequences and intensity estimation sequences.

6. The method as described in claim 1, characterized in that, The step of generating a posterior ranking based on the posterior probability sequence and the intensity estimation sequence includes: Align the posterior probability sequence and intensity estimation sequence according to the candidate leak point number, so that each candidate leak point corresponds to a unique posterior probability and intensity estimate. The posterior probability of candidate leak points is used as the primary ranking criterion, and the intensity estimate is used as the secondary ranking criterion to generate a posterior ranking.

7. The method as described in claim 1, characterized in that, The posterior ranking is verified to obtain the target leak or the leak to be verified, including: Based on the posterior ranking, the candidate leak points with the highest ranking are determined, and the reachable propagation path and measurement point combination connection relationship corresponding to the candidate leak point are extracted from the influence constraint diagram. The measurement points with abnormal intensity responses in the abnormal injection vector are matched one by one with the reachable propagation path, and the path consistency index is calculated according to the matching of the propagation order. The connection relationship between the combination of measurement points with abnormal intensity response in the abnormal injection vector and the combination of measurement points is checked one by one, and the consistency index of measurement points is calculated according to the connection coverage and the distribution of abnormal intensity. The posterior ranking is verified based on the path consistency index and the measurement point consistency index. The verification result is compared with the preset positioning threshold. When the preset positioning threshold is met, the candidate leak point at the top of the posterior ranking is determined as the target leak point; otherwise, it is determined as a leak point to be verified.

8. A device for locating air leaks in a condenser of a thermal power plant, characterized in that, include: The data acquisition module is used to acquire the distributed control system measurement point data and equipment connection relationships of the condenser vacuum system during unit operation, and to arrange the distributed control system measurement point data in time sequence according to the measurement point identification and equipment location to construct an observation matrix; The influence constraint graph construction module is used to determine candidate leaks based on the device connection relationship, retrieve the reachable propagation path from each candidate leak to each measurement point, merge the measurement points according to the reachable propagation path to form a measurement point combination, and construct an influence constraint graph that represents the connection relationship between candidate leaks and measurement point combinations. An anomaly injection vector extraction module is used to perform robust principal component analysis on the observation matrix under the constraints of the influence constraint diagram, separate the overall linkage change into low-rank common mode components, separate the local anomalies propagating along the reachable propagation path into sparse anomaly components, restrict the distribution of the sparse anomaly components on unreachable measurement points according to the influence constraint diagram, and extract the anomaly intensity of each measurement point combination from the sparse anomaly components to obtain the anomaly injection vector. The posterior inference module is used to construct a sparse latent source Bayesian factor graph based on the candidate leak points, the influence constraint graph and the anomaly injection vector, set the candidate leak point number as a discrete latent variable and the leak intensity as a continuous latent variable, and perform joint posterior inference under the constraints of operating condition fluctuation and measurement point noise to obtain the posterior probability sequence and intensity estimation sequence. The location output module is used to generate a posterior ranking based on the posterior probability sequence and the intensity estimation sequence; and to calculate the path consistency index and the measurement point consistency index by combining the influence constraint diagram and the anomaly injection vector, and to verify the posterior ranking to obtain the target leak point or the leak point to be verified.

9. A computer device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement a method for locating air leaks in a condenser of a thermal power plant as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a method for locating air leaks in a condenser of a thermal power plant as described in any one of claims 1-7.

Images