Intelligent diagnosis method for pressure abnormity of directly-buried steam pipeline

Abstract

The invention discloses an intelligent diagnosis method for pressure abnormity of a directly-buried steam pipeline, and the method comprises the following steps: collecting original data of a plurality of pressure monitoring nodes, and generating a standardized pressure data sequence through time synchronization and interpolation completion; then double-path parallel processing is adopted, one path identifies local abnormal fluctuation by applying a Hampel algorithm and forms candidate abnormal points, and the other path keeps original pressure data; based on the candidate abnormal points, extracting data of a corresponding time period, and deducing a change point posterior probability by using a Bayesian online change point detection method; by analyzing the probability change trend, identifying abnormal event points and combining the topological structure and the flow direction of the steam pipeline to construct an abnormal propagation path, the position of an abnormal source and the first abnormal occurrence time are finally determined, and a complete abnormal diagnosis result is output. According to the invention, early recognition and traceability of potential faults of the steam pipeline are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial process monitoring and fault diagnosis technology, and in particular to an intelligent diagnostic method for abnormal pressure in directly buried steam pipelines. Background Art

[0002] As urban heating systems continue to expand in scale and increase in operational complexity, steam pipelines are prone to leaks, blockages, or bursts under long-term high-temperature and high-pressure conditions, severely impacting heating efficiency and operational safety. Currently, condition monitoring of directly buried steam pipelines mainly relies on pressure sensors deployed at key nodes, combined with fixed-threshold alarm mechanisms or statistical models based on overall trend analysis to identify anomalies.

[0003] However, existing technologies have significant shortcomings in dealing with localized sudden anomalies and weak signal fluctuations. On the one hand, traditional anomaly detection methods lack sensitivity to localized anomaly fluctuations, easily missing short-term, weak but significant precursory signs, leading to delayed diagnosis or misjudgment. On the other hand, existing change point detection methods rely heavily on static prior information, making it difficult to accurately identify the anomaly propagation path and its source in pressure data, and failing to effectively achieve joint reasoning and precise localization among multi-point data.

[0004] Therefore, how to provide an intelligent diagnostic method for abnormal pressure in directly buried steam pipelines is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent diagnostic method for pressure anomalies in directly buried steam pipelines. This invention performs dual-path processing on standardized pressure data sequences, constructs a dynamic prior probability structure by combining candidate anomaly points, and identifies the anomaly propagation path and anomaly source location based on the enhanced posterior probability of the anomaly points. It has the advantages of high diagnostic accuracy, fast response speed, and strong adaptability.

[0006] The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to an embodiment of the present invention includes the following steps: Raw pressure data from multiple pressure monitoring nodes were collected, and information on the topology of the steam pipeline and the direction of steam flow was obtained. Time synchronization and missing interpolation are performed on the raw pressure data to generate a standardized pressure data sequence; A dual-path processing is performed on the standardized stress data sequence. The first path uses the Hampel algorithm with an adaptive sliding window to detect local abnormal fluctuations and generate a set of candidate anomaly indexes. The second path preserves the standardized stress data sequence. Based on the candidate outlier index set, standardized stress data for the corresponding time period is extracted. When the cumulative amplitude of the change point probability exceeds the preset threshold, the Bayesian online change point detection method is used for inference to obtain the change point posterior probability sequence. Based on the continuous probability increasing segment of the posterior probability sequence of the change point, the intersection with the candidate outlier index set is processed to generate an outlier event point set. The set of abnormal event points is mapped to the standardized pressure data sequence of each pressure monitoring node. The difference in the posterior probability of the change point of adjacent pressure monitoring nodes in the same time period is calculated. The abnormal propagation path is constructed by combining the steam pipeline topology information and steam flow direction information. Based on the direction and time sequence of the posterior probability difference of the change points in the abnormal propagation path, the pressure monitoring node of the abnormal source and the time of its first abnormal occurrence are determined, and the abnormal diagnosis results are output.

[0007] Optional, steam pipeline topology information and steam flow direction information include: The steam pipeline topology information includes pressure monitoring node identification information, spatial coordinate information of pressure monitoring nodes, connection relationship information between pressure monitoring nodes, and physical length information of each connecting pipe segment. The connection relationship information indicates whether there is a pipeline connection between adjacent pressure monitoring nodes, and the physical length information indicates the axial distance between the connecting pipe segments. The steam flow direction information includes steam source identification, preset steam flow direction information in each pipe section, and flow direction relationship matrix between nodes. The flow direction information represents the transmission path of steam from upstream to downstream in the pipeline network. The flow direction relationship matrix is ​​a directed relationship matrix, and each element of it represents whether the steam flow direction exists and its directionality between the corresponding node pairs.

[0008] Optional, the generation of standardized stress data sequences includes: Time synchronization processing is performed on the raw pressure data collected from multiple pressure monitoring nodes, a unified time index set is constructed, a unified sampling period is set as the time base, the raw sampling time of each pressure monitoring node is aligned with the unified sampling period, and a preliminary synchronized pressure data sequence is generated. Identify time points in the synchronous pressure data sequence that are not mapped by the original pressure data, and generate a set of missing data points; For each missing data point in the set of missing data points, find its adjacent previous and next effective pressure data points in the synchronous pressure data sequence of the corresponding pressure monitoring node. If the time interval between the two effective pressure data points is less than the preset interpolation threshold and the pressure difference between the two effective pressure data points is less than the preset fluctuation threshold, then perform interpolation completion processing on the missing data point according to the trend between the adjacent effective pressure data points, and write the completion result into the synchronous pressure data sequence. A consistency check is performed on the interpolated and completed synchronous pressure data sequence, and local abrupt data points caused by interpolation are deleted to obtain the verified pressure data sequence. The verified pressure data sequence is reordered according to a unified time index set and combined into a unified format with each pressure monitoring node as the dimension, forming a standardized pressure data sequence.

[0009] Optional, two-path processing of standardized stress data sequences includes: The standardized pressure data sequence is input into the first path and the second path respectively for parallel processing; In the first path, the standardized pressure data sequence of each pressure monitoring node is processed by the Hampel algorithm based on an adaptive sliding window to detect local abnormal fluctuations in the standardized pressure data sequence. Construct a sliding window structure, set the sliding window length and sliding step size, and extract the data subsequences corresponding to the sliding window sequentially along a unified time index set; Within each sliding window, the median of the corresponding data subsequence is calculated, and the absolute deviation of the median of the corresponding data subsequence is obtained based on the absolute deviation of each data point from the median. Determine the data point corresponding to the center position of the sliding window, and determine whether the deviation between it and the median is greater than the product of the absolute deviation of the median and a set multiple threshold. If the judgment condition is met, the corresponding data point is identified as having local abnormal fluctuations and is marked as a candidate abnormal point. The location index records of all data points marked as candidate outliers in the standardized stress data sequence are recorded and summarized to form a candidate outlier index set; In the second path, the standardized pressure data sequence of each pressure monitoring node is written into the data buffer structure in chronological order; The candidate outlier index set formed in the first path is passed to the second path. Based on the time index in the candidate outlier index set, the corresponding standardized stress data segment in the second path is extracted.

[0010] Optionally, the generation of the posterior probability sequence at the change point includes: Obtain the time index corresponding to each candidate anomaly in the candidate anomaly index set. With the corresponding time index as the center, extract the standardized pressure data symmetrical before and after the corresponding time index in the standardized pressure data sequence corresponding to the pressure monitoring node to which the corresponding candidate anomaly belongs, according to the preset fixed window length, and form a standardized pressure data segment. The standardized pressure data segment is passed as input to the Bayesian online change point detection method. The Bayesian online change point detection method outputs a change point posterior probability sequence corresponding to each time point based on the pressure value at each time point in the standardized pressure data segment. The posterior probability sequence of the change point is analyzed to identify the amplitude change between the maximum and minimum probability values ​​in the posterior probability sequence of the change point, and the amplitude change is used as the cumulative amplitude of the change point probability. The cumulative amplitude of the change point probability is compared with a preset change point probability amplitude judgment threshold. If the cumulative amplitude of the change point probability exceeds the change point probability amplitude judgment threshold, the time index corresponding to the candidate anomaly point is confirmed as a valid anomaly trigger point.

[0011] Optional Bayesian online change point detection methods include: Obtain the time index corresponding to each candidate anomaly in the candidate anomaly index set, and extract the standardized pressure data symmetrical before and after the corresponding time index in the standardized pressure data sequence corresponding to the pressure monitoring node associated with the corresponding candidate anomaly, with the time index as the center and according to the preset fixed time window length, to form a standardized pressure data segment. For the standardized stress data segments, a dynamic prior probability structure containing candidate outlier information is constructed. Each time index in the standardized stress data segments is traversed sequentially. It is determined whether each time index belongs to the candidate outlier index set. If it does, the change point prior probability of the time index is set to the enhanced value of the basic prior probability. If it does not, the change point prior probability of the corresponding time index is kept at the default value of the basic prior probability. The standardized stress data segment containing the dynamic prior probability structure is input into the Bayesian online change point detection method, and a recursive joint posterior inference operation is performed to output the change point posterior probability sequence that corresponds one-to-one with each time index in the standardized stress data segment. The output change point posterior probability sequence is enhanced by identifying each time index of the candidate outlier index set in the change point posterior probability sequence. A local symmetric response window is constructed with the identified time index as the center and a preset length. Within the response window, confidence enhancement operations are performed on the change point posterior probability values ​​at each time index, including weighted smoothing and trend interpolation, to generate the enhanced change point posterior probability sequence.

[0012] Optionally, the generation of the set of exception event points includes: Obtain the enhanced variable point posterior probability sequence corresponding to each pressure monitoring node, and perform time-by-time index scanning on the variable point posterior probability sequence based on the unified time index set; In the posterior probability sequence of each pressure monitoring node, the time period in which the posterior probability value of the change point continuously increases with the time index is identified and defined as the continuous probability increasing segment. The continuous probability increasing segment must meet the following conditions: the posterior probability value of the change point corresponding to adjacent time indices shows a monotonically increasing or fluctuating increase within a limited range, and the difference between the posterior probability of the change point at the beginning and end of the corresponding time period is not less than the set minimum increase threshold, and the time span is not less than the set minimum duration threshold. Extract all time indices that meet the conditions and summarize them to form a set of suspected abnormal time indices; Obtain the candidate anomaly point index set, perform intersection processing based on the unified time index set on the suspected anomaly time index set and the candidate anomaly point index set, and extract each time index in the intersection; Each time index in the intersection is bound to the identifier of its respective pressure monitoring node to form a pressure monitoring node-time index tuple, and all tuples are aggregated to generate a set of abnormal event points.

[0013] Optional, the construction of the anomaly propagation path includes: Obtain the pressure monitoring node-time index tuple in the abnormal event point set, and extract the corresponding change point posterior probability value in the enhanced change point posterior probability sequence based on the time index to form an abnormal feature set containing abnormal event points and change point posterior probabilities. Based on the topology information of the steam pipeline, the set of direct adjacent pressure monitoring nodes for each abnormal event node is determined, and for each pair of adjacent pressure monitoring nodes with a connection relationship, the enhanced posterior probability value of the change point at the same time index is extracted under a unified time index set. Based on the steam flow direction information, the set of downstream adjacent pressure monitoring nodes corresponding to each abnormal event point is obtained. In the unified time index set, the difference calculation is performed on the change point posterior probability value between the abnormal event point and its downstream adjacent nodes to obtain the change point posterior probability difference of each node at the corresponding time index. Node pairs with a difference in posterior probability greater than a preset threshold are selected, and the direction of the difference is confirmed to be consistent with the direction of steam flow based on the steam flow direction information. If they are consistent, the corresponding node pairs are defined as abnormal propagation edges. All abnormal propagation edges are concatenated according to their time index, and combined with the spatial coordinate information of each pressure monitoring node in the steam pipeline topology information, an abnormal propagation path composed of multiple abnormal event points in the propagation order is constructed, and all paths are summarized to form an abnormal propagation path set.

[0014] Optionally, the output of abnormal diagnostic results includes: Obtain each abnormal propagation path in the abnormal propagation path set, count the frequency of abnormal triggering of the pressure monitoring node corresponding to the starting point of the path under the unified time index set, and record the time index of the earliest time that the node was identified as an abnormal event point, forming a candidate set of starting nodes and its corresponding time index set. In the candidate set of starting nodes, the path starting points in the abnormal propagation path are selected such that the direction of the difference in the posterior probability of the change point is always consistent with the direction of steam flow, and the posterior probability of the change point of each pressure monitoring node on the path shows a monotonically decreasing trend, thus forming a set of suspected abnormal source nodes. For each node in the set of suspected anomaly source nodes, sort them according to the first appearance time index in the set of anomaly event points, and extract the suspected anomaly source node with the earliest time index as the anomaly source pressure monitoring node. The abnormal source pressure monitoring node and its corresponding first occurrence time index are defined as the abnormal source pressure monitoring node and the first abnormal occurrence time, respectively, and output together with the corresponding abnormal propagation path to form the abnormal diagnosis result.

[0015] The beneficial effects of the present invention are: (1) This invention combines the adaptive sliding window Hampel algorithm with the Bayesian online change point detection method to construct an intelligent diagnostic mechanism that can identify local pressure fluctuations and change point anomalies in buried steam pipelines in real time. It can effectively capture early abnormal signs and improve the system's response capability to complex dynamic pressure anomalies.

[0016] (2) This invention enhances the sensitivity and accuracy of abnormal signal detection by introducing a dynamic prior enhancement mechanism based on candidate anomalies into the Bayesian online change point detection method and performs confidence enhancement processing on the posterior probability sequence of change points, thereby reducing false alarms and missed alarms and improving diagnostic stability and repeatability.

[0017] (3) This invention constructs anomaly propagation paths by integrating steam pipeline topology information and steam flow direction information, and identifies anomaly source nodes and the time of first anomaly occurrence by combining the direction of the posterior probability difference of the change point with the time sequence, thereby enhancing the physical rationality of anomaly location and the interpretability of the results. It is suitable for the intelligent operation and maintenance needs of complex multi-node steam pipeline networks. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows the intelligent diagnostic method for abnormal pressure in directly buried steam pipelines proposed in this invention. Figure 2This is a flowchart of the Bayesian online change point detection method with dynamic prior enhancement introduced in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0020] refer to Figures 1-2 The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines includes the following steps: Raw pressure data from multiple pressure monitoring nodes were collected, and information on the topology of the steam pipeline and the direction of steam flow was obtained. Time synchronization and missing interpolation are performed on the raw pressure data to generate a standardized pressure data sequence; A dual-path processing is performed on the standardized stress data sequence. The first path uses the Hampel algorithm with an adaptive sliding window to detect local abnormal fluctuations and generate a set of candidate anomaly indexes. The second path preserves the standardized stress data sequence. Based on the candidate outlier index set, standardized stress data for the corresponding time period is extracted. When the cumulative amplitude of the change point probability exceeds the preset threshold, the Bayesian online change point detection method is used for inference to obtain the change point posterior probability sequence. Based on the continuous probability increasing segment of the posterior probability sequence of the change point, the intersection with the candidate outlier index set is processed to generate an outlier event point set. The set of abnormal event points is mapped to the standardized pressure data sequence of each pressure monitoring node. The difference in the posterior probability of the change point of adjacent pressure monitoring nodes in the same time period is calculated. The abnormal propagation path is constructed by combining the steam pipeline topology information and steam flow direction information. Based on the direction and time sequence of the posterior probability difference of the change points in the abnormal propagation path, the pressure monitoring node of the abnormal source and the time of its first abnormal occurrence are determined, and the abnormal diagnosis results are output.

[0021] In this embodiment, the steam pipeline topology information and steam flow direction information include: The steam pipeline topology information includes pressure monitoring node identification information, spatial coordinate information of pressure monitoring nodes, connection relationship information between pressure monitoring nodes, and physical length information of each connecting pipe segment. The connection relationship information indicates whether there is a pipeline connection between adjacent pressure monitoring nodes, and the physical length information indicates the axial distance between the connecting pipe segments. The steam flow direction information includes steam source identification, preset steam flow direction information in each pipe section, and flow direction relationship matrix between nodes. The flow direction information represents the transmission path of steam from upstream to downstream in the pipeline network. The flow direction relationship matrix is ​​a directed relationship matrix, and each element of it represents whether the steam flow direction exists between the corresponding node pairs and its directionality. The topology information of steam pipelines and the steam flow direction information constitute a set of static structural data used to characterize the structural connection mode and flow transmission relationship of steam in the pipeline network.

[0022] In this embodiment, the generation of standardized pressure data sequences includes: Time synchronization processing is performed on the raw pressure data collected from multiple pressure monitoring nodes, a unified time index set is constructed, a unified sampling period is set as the time base, the raw sampling time of each pressure monitoring node is aligned with the unified sampling period, and a preliminary synchronized pressure data sequence is generated. Identify time points in the synchronous pressure data sequence that are not mapped by the original pressure data, and generate a set of missing data points; For each missing data point in the set of missing data points, find its adjacent previous and next effective pressure data points in the synchronous pressure data sequence of the corresponding pressure monitoring node. If the time interval between the two effective pressure data points is less than the preset interpolation threshold and the pressure difference between the two effective pressure data points is less than the preset fluctuation threshold, then perform interpolation completion processing on the missing data point according to the trend between the adjacent effective pressure data points, and write the completion result into the synchronous pressure data sequence. A consistency check is performed on the interpolated and completed synchronous pressure data sequence, and local abrupt data points caused by interpolation are deleted to obtain the verified pressure data sequence. The verified pressure data sequence is reordered according to a unified time index set and combined into a unified format with each pressure monitoring node as the dimension, forming a standardized pressure data sequence.

[0023] In this embodiment, the dual-path processing of the standardized pressure data sequence includes: The standardized pressure data sequence is input into the first path and the second path respectively for parallel processing; In the first path, the standardized pressure data sequence of each pressure monitoring node is processed by the Hampel algorithm based on an adaptive sliding window to detect local abnormal fluctuations in the standardized pressure data sequence. Construct a sliding window structure, set the sliding window length and sliding step size, and extract the data subsequences corresponding to the sliding window sequentially along a unified time index set; Within each sliding window, the median of the corresponding data subsequence is calculated, and the absolute deviation of the median of the corresponding data subsequence is obtained based on the absolute deviation of each data point from the median. Determine the data point corresponding to the center position of the sliding window, and determine whether the deviation between it and the median is greater than the product of the absolute deviation of the median and a set multiple threshold. If the judgment condition is met, the corresponding data point is identified as having local abnormal fluctuations and is marked as a candidate abnormal point. The location index records of all data points marked as candidate outliers in the standardized stress data sequence are recorded and summarized to form a candidate outlier index set; In the second path, the standardized pressure data sequence of each pressure monitoring node is written into the data buffer structure in chronological order; The candidate outlier index set formed in the first path is passed to the second path. Based on the time index in the candidate outlier index set, the corresponding standardized stress data segment in the second path is extracted.

[0024] In this embodiment, the generation of the posterior probability sequence at the change point includes: Obtain the time index corresponding to each candidate anomaly in the candidate anomaly index set. With the corresponding time index as the center, extract the standardized pressure data symmetrical before and after the corresponding time index in the standardized pressure data sequence corresponding to the pressure monitoring node to which the corresponding candidate anomaly belongs, according to the preset fixed window length, and form a standardized pressure data segment. The standardized pressure data segment is passed as input to the Bayesian online change point detection method. The Bayesian online change point detection method outputs a change point posterior probability sequence corresponding to each time point based on the pressure value at each time point in the standardized pressure data segment. The posterior probability sequence of the change point is analyzed to identify the amplitude change between the maximum and minimum probability values ​​in the posterior probability sequence of the change point, and the amplitude change is used as the cumulative amplitude of the change point probability. The cumulative amplitude of the change point probability is compared with a preset change point probability amplitude judgment threshold. If the cumulative amplitude of the change point probability exceeds the change point probability amplitude judgment threshold, the time index corresponding to the candidate anomaly point is confirmed as a valid anomaly trigger point.

[0025] In this embodiment, the Bayesian online change point detection method includes: Obtain the time index corresponding to each candidate anomaly in the candidate anomaly index set, and extract the standardized pressure data symmetrical before and after the corresponding time index in the standardized pressure data sequence corresponding to the pressure monitoring node associated with the corresponding candidate anomaly, with the time index as the center and according to the preset fixed time window length, to form a standardized pressure data segment. For the standardized stress data segments, a dynamic prior probability structure containing candidate outlier information is constructed. Each time index in the standardized stress data segments is traversed sequentially. It is determined whether each time index belongs to the candidate outlier index set. If it does, the change point prior probability of the time index is set to the enhanced value of the basic prior probability. If it does not, the change point prior probability of the corresponding time index is kept at the default value of the basic prior probability. The standardized stress data segment containing the dynamic prior probability structure is input into the Bayesian online change point detection method, and a recursive joint posterior inference operation is performed to output the change point posterior probability sequence that corresponds one-to-one with each time index in the standardized stress data segment. The output change point posterior probability sequence is enhanced by identifying each time index of the candidate outlier index set in the change point posterior probability sequence. A local symmetric response window is constructed with the identified time index as the center and a preset length. Within the response window, confidence enhancement operations are performed on the change point posterior probability values ​​at each time index, including weighted smoothing and trend interpolation, to generate the enhanced change point posterior probability sequence. A dynamic prior probability structure containing candidate outlier information is constructed, including establishing a mapping relationship between the time indices in the standardized stress data segment and the corresponding prior probability values. This structure consists of a prior probability vector, where each element corresponds to a time index in the standardized stress data segment and records the prior probability value of the change point corresponding to that index. The prior probability value is set according to the candidate outlier index set. If a certain time index belongs to the candidate outlier index set, its corresponding prior probability value is increased by a fixed enhancement ratio from the basic prior probability. If it does not belong, it remains the basic prior probability. The basic prior probability is calculated based on the default crisis function in the Bayesian online change point detection method. The overall dynamic prior probability structure is a linear array structure with the same length as the standardized stress data segment. Performing recursive joint posterior inference refers to, given a standardized stress data segment and a corresponding dynamic prior probability structure, recursively calculating the posterior probability of a change point at the current time index, following the temporal processing method of the Bayesian online change point detection method. Specifically, this includes: initializing the change point probability when the time index is zero; for each subsequent time index, calculating the joint posterior probability of the change point location based on the previous time index's posterior distribution, the standardized stress value at the current time index, and the corresponding prior probability value; updating and normalizing the joint posterior probabilities at all time indices within the entire standardized stress data segment, ultimately obtaining a sequence of change point posterior probabilities corresponding one-to-one with each time index in the standardized stress data segment.

[0026] In this embodiment, the generation of the abnormal event point set includes: Obtain the enhanced variable point posterior probability sequence corresponding to each pressure monitoring node, and perform time-by-time index scanning on the variable point posterior probability sequence based on the unified time index set; In the posterior probability sequence of each pressure monitoring node, the time period in which the posterior probability value of the change point continuously increases with the time index is identified and defined as the continuous probability increasing segment. The continuous probability increasing segment must meet the following conditions: the posterior probability value of the change point corresponding to adjacent time indices shows a monotonically increasing or fluctuating increase within a limited range, and the difference between the posterior probability of the change point at the beginning and end of the corresponding time period is not less than the set minimum increase threshold, and the time span is not less than the set minimum duration threshold. Extract all time indices that meet the conditions and summarize them to form a set of suspected abnormal time indices; Obtain the candidate anomaly point index set, perform intersection processing based on the unified time index set on the suspected anomaly time index set and the candidate anomaly point index set, and extract each time index in the intersection; Bind each time index in the intersection to its corresponding pressure monitoring node identifier to form a pressure monitoring node-time index tuple, and summarize all the tuples to generate a set of abnormal event points; The minimum increase threshold is a criterion used to determine whether an upward trend in the posterior probability sequence at a change point is significant. It represents the minimum cumulative increase in posterior probability values ​​within a continuous time index. If, within a continuous time index interval, the posterior probability values ​​monotonically increase, and the probability difference between the first and last ends exceeds the minimum increase threshold, then a significant upward trend is considered to exist. The minimum duration threshold is a criterion used to limit the shortest duration required to maintain this upward trend. It represents the minimum length of the continuous time index interval corresponding to the aforementioned upward trend. Only when a continuous upward segment of the posterior probability simultaneously satisfies both the minimum increase threshold and the minimum duration threshold is it identified as a valid continuous upward probability segment.

[0027] In this embodiment, the construction of the anomaly propagation path includes: Obtain the pressure monitoring node-time index tuple in the abnormal event point set, and extract the corresponding change point posterior probability value in the enhanced change point posterior probability sequence based on the time index to form an abnormal feature set containing abnormal event points and change point posterior probabilities. Based on the topology information of the steam pipeline, the set of direct adjacent pressure monitoring nodes for each abnormal event node is determined, and for each pair of adjacent pressure monitoring nodes with a connection relationship, the enhanced posterior probability value of the change point at the same time index is extracted under a unified time index set. Based on the steam flow direction information, the set of downstream adjacent pressure monitoring nodes corresponding to each abnormal event point is obtained. In the unified time index set, the difference calculation is performed on the change point posterior probability value between the abnormal event point and its downstream adjacent nodes to obtain the change point posterior probability difference of each node at the corresponding time index. Node pairs with a difference in posterior probability greater than a preset threshold are selected, and the direction of the difference is confirmed to be consistent with the direction of steam flow based on the steam flow direction information. If they are consistent, the corresponding node pairs are defined as abnormal propagation edges. All abnormal propagation edges are concatenated according to their time index, and combined with the spatial coordinate information of each pressure monitoring node in the steam pipeline topology information, an abnormal propagation path composed of multiple abnormal event points in the propagation order is constructed, and all paths are summarized to form an abnormal propagation path set.

[0028] In this embodiment, the output of the abnormal diagnosis results includes: Obtain each abnormal propagation path in the abnormal propagation path set, count the frequency of abnormal triggering of the pressure monitoring node corresponding to the starting point of the path under the unified time index set, and record the time index of the earliest time that the node was identified as an abnormal event point, forming a candidate set of starting nodes and its corresponding time index set. In the candidate set of starting nodes, the path starting points in the abnormal propagation path are selected such that the direction of the difference in the posterior probability of the change point is always consistent with the direction of steam flow, and the posterior probability of the change point of each pressure monitoring node on the path shows a monotonically decreasing trend, thus forming a set of suspected abnormal source nodes. For each node in the set of suspected anomaly source nodes, sort them according to the first appearance time index in the set of anomaly event points, and extract the suspected anomaly source node with the earliest time index as the anomaly source pressure monitoring node. The abnormal source pressure monitoring node and its corresponding first occurrence time index are defined as the abnormal source pressure monitoring node and the first abnormal occurrence time, respectively, and output together with the corresponding abnormal propagation path to form the abnormal diagnosis result.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a city energy group to deploy and verify the intelligent pressure anomaly diagnosis system on a directly buried steam main pipeline with a total length of approximately 13.7 kilometers within its jurisdiction. This steam pipeline connects multiple heat exchange stations, serving areas including residential areas, office areas, and industrial parks. It has a high overall heat load and a complex structure, and poses potential hazards such as steam leakage, insulation aging, or weld fatigue during year-round operation. Because the pipeline is completely buried underground, traditional inspection methods rely on manual labor and periodic temperature measurement points, resulting in problems such as response delays, inaccurate fault location, and inability to identify anomalies in real time.

[0030] A total of 23 high-precision pressure monitoring nodes, model HEC-P01, were installed along the pipeline. The sampling period was set to 10 seconds, and data was transmitted back to the dispatch master station in real time via a LoRa wireless network. The data processing server was deployed in a local IDC data center, equipped with an industrial computing platform with a 16-core CPU and 64GB of memory, running the diagnostic system software proposed in this invention.

[0031] In the data preprocessing stage, the system first performed unified time alignment on the raw data collected from 23 pressure monitoring nodes. All node data were completed based on a unified time index set. For missing data points, the system used local linear trend interpolation based on the interval between adjacent valid data points and the pressure difference to complete the missing data. The completed data underwent consistency verification and local mutation removal, ultimately generating a standardized pressure data sequence as input for subsequent analysis.

[0032] During the diagnostic analysis phase, standardized stress data sequences are fed into a dual-path analysis module in parallel. The first path uses an adaptive sliding window Hampel algorithm to detect local anomalies in the sequence at each node. The system sets the sliding window length to 20 sampling points and the sliding step size to 5 sampling points, and uses the median and median absolute deviation (MAD) for judgment, marking center points with deviations greater than 3 times the MAD as candidate anomalies. The second path synchronously caches all node data and extracts data segments for the corresponding time periods based on the candidate anomaly indices detected by the first path.

[0033] The extracted data segments are input into the Bayesian online change point detection model. During model execution, the system constructs a dynamic prior structure based on candidate outliers, enhances the prior probabilities at candidate indices, and simultaneously performs a recursive joint posterior probability update operation. The output is the posterior probability value of the change point at each time point. The system further enhances this probability sequence by performing weighted smoothing and trend interpolation within a locally symmetric window centered on the candidate point, generating an enhanced posterior probability sequence of the change point.

[0034] Based on the posterior probability sequence of the change points, the system identifies multiple consecutive probability-increasing segments. The system sets a minimum increase threshold of 0.15 and a minimum duration threshold of 120 seconds, extracts suspected abnormal time indices that meet the conditions, and performs intersection processing with the candidate abnormal point index set to ultimately identify a set of abnormal event points. Each abnormal event point consists of a monitoring node identifier and a time index, accurately reflecting the location and time of the abnormality.

[0035] The system combines the actual topology of the steam pipeline and the steam flow direction to calculate the probability difference between abnormal event points and construct anomaly propagation paths. Starting from the abnormal event point, it progressively matches nodes with lower posterior probabilities downstream and confirms whether the path direction conforms to the steam flow direction. Only anomaly propagation paths that meet the conditions are retained, ultimately constructing multiple anomaly propagation path subgraphs.

[0036] The system automatically sorts and summarizes all abnormal propagation paths, identifies the first abnormal node in the propagation path as the abnormal source pressure monitoring node, records the time of its first abnormal occurrence, and generates a structured abnormal diagnosis report.

[0037] In this embodiment, the abnormal fluctuation was first detected at node D-21, and the initial posterior probability value quickly rose to 0.78. Subsequently, nodes D-20 and D-19 detected similar characteristics 4 minutes and 6 minutes later, respectively. The system successfully constructed an abnormal path propagating from D-21 to D-19, and ultimately confirmed node D-21 as the source of the anomaly. The following is a data illustration of this abnormal event: Table 1. Posterior probability values ​​of some abnormal propagation path nodes.

[0038] To verify the actual effectiveness of the method of the present invention, we selected typical leakage events recorded by the energy company in the past two years for backtracking tests. The same nodes and diagnostic systems were deployed in a total of 6 test pipe sections for reproduction verification, and compared with the traditional temperature sensing anomaly detection method.

[0039] Table 2 Comparison of detection performance of the new and old methods in typical anomaly detection scenarios.

[0040] In Table 1, node D-21 first exhibited a significant jump in posterior probability of change point at 14:35:20, reaching a value as high as 0.78, far exceeding the anomaly judgment threshold (generally set around 0.5), indicating that it was the earliest point where local pressure fluctuations occurred. Subsequently, downstream nodes D-20 and D-19 showed increases in change point probability approximately 4 minutes and 6 minutes later, respectively, with values ​​of 0.69 and 0.56, showing a decreasing trend, consistent with the propagation characteristics of steam anomalies in the actual physical flow direction. The probability values ​​of the following nodes D-18 and D-17 further decreased, to 0.34 and 0.22, respectively, indicating that the impact of the anomaly gradually weakened, further verifying the rationality and temporal continuity of the anomaly propagation path constructed by the system.

[0041] Table 2 shows that the system significantly outperforms traditional methods in six typical anomaly scenarios. The traditional temperature sensing detection method has an average latency of 33.83 minutes, while this invention's latency is only 5.5 minutes, representing an advance response time of over 85%. In terms of spatial positioning, the traditional method has an average error of 161.5 meters, while this invention's error is only 27.5 meters, a reduction of over 80%, greatly improving the accuracy of anomaly source location. These results fully demonstrate that this invention, combining the Hampel algorithm with a Bayesian online change point detection strategy, has significant advantages in extracting the temporal and spatial features of anomalies, making it particularly suitable for online monitoring and accurate diagnosis of complex directly buried steam pipeline systems.

[0042] In summary, this embodiment successfully deployed and verified the effectiveness of the invention in a real urban steam pipeline network environment. It not only achieved rapid identification and propagation path tracing of abnormal events, but also demonstrated good versatility and engineering applicability. In the future, this method is expected to be extended to the diagnosis of anomalies in other urban steam systems, heating networks, and high-risk industrial pipelines.

[0043] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent diagnosis of abnormal pressure in directly buried steam pipelines, characterized in that: include: Raw pressure data from multiple pressure monitoring nodes were collected, and information on the topology of the steam pipeline and the direction of steam flow was obtained. Time synchronization and missing interpolation are performed on the raw pressure data to generate a standardized pressure data sequence; A dual-path processing is performed on the standardized stress data sequence. The first path uses the Hampel algorithm with an adaptive sliding window to detect local abnormal fluctuations and generate a set of candidate anomaly indexes. The second path preserves the standardized stress data sequence. Based on the candidate outlier index set, standardized stress data for the corresponding time period is extracted. When the cumulative amplitude of the change point probability exceeds the preset threshold, the Bayesian online change point detection method is used for inference to obtain the change point posterior probability sequence. Based on the continuous probability increasing segment of the posterior probability sequence of the change point, the intersection with the candidate outlier index set is processed to generate an outlier event point set. The set of abnormal event points is mapped to the standardized pressure data sequence of each pressure monitoring node. The difference in the posterior probability of the change point of adjacent pressure monitoring nodes in the same time period is calculated. The abnormal propagation path is constructed by combining the steam pipeline topology information and steam flow direction information. Based on the direction and time sequence of the posterior probability difference of the change points in the abnormal propagation path, the pressure monitoring node of the abnormal source and the time of its first abnormal occurrence are determined, and the abnormal diagnosis results are output.

2. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, Steam pipeline topology information and steam flow direction information include: The steam pipeline topology information includes pressure monitoring node identification information, spatial coordinate information of pressure monitoring nodes, connection relationship information between pressure monitoring nodes, and physical length information of each connecting pipe segment. The connection relationship information indicates whether there is a pipeline connection between adjacent pressure monitoring nodes, and the physical length information indicates the axial distance between the connecting pipe segments. The steam flow direction information includes steam source identification, preset steam flow direction information in each pipe section, and flow direction relationship matrix between nodes. The flow direction information represents the transmission path of steam from upstream to downstream in the pipeline network. The flow direction relationship matrix is ​​a directed relationship matrix, and each element of it represents whether the steam flow direction exists and its directionality between the corresponding node pairs.

3. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, The generation of standardized stress data sequences includes: Time synchronization processing is performed on the raw pressure data collected from multiple pressure monitoring nodes, a unified time index set is constructed, a unified sampling period is set as the time base, the raw sampling time of each pressure monitoring node is aligned with the unified sampling period, and a preliminary synchronized pressure data sequence is generated. Identify time points in the synchronous pressure data sequence that are not mapped by the original pressure data, and generate a set of missing data points; For each missing data point in the set of missing data points, find its adjacent previous and next effective pressure data points in the synchronous pressure data sequence of the corresponding pressure monitoring node. If the time interval between the two effective pressure data points is less than the preset interpolation threshold and the pressure difference between the two effective pressure data points is less than the preset fluctuation threshold, then perform interpolation completion processing on the missing data point according to the trend between the adjacent effective pressure data points, and write the completion result into the synchronous pressure data sequence. A consistency check is performed on the interpolated and completed synchronous pressure data sequence, and local abrupt data points caused by interpolation are deleted to obtain the verified pressure data sequence. The verified pressure data sequence is reordered according to a unified time index set and combined into a unified format with each pressure monitoring node as the dimension, forming a standardized pressure data sequence.

4. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, Two-path processing of standardized stress data sequences includes: The standardized pressure data sequence is input into the first path and the second path respectively for parallel processing; In the first path, the standardized pressure data sequence of each pressure monitoring node is processed by the Hampel algorithm based on an adaptive sliding window to detect local abnormal fluctuations in the standardized pressure data sequence. Construct a sliding window structure, set the sliding window length and sliding step size, and extract the data subsequences corresponding to the sliding window sequentially along a unified time index set; Within each sliding window, the median of the corresponding data subsequence is calculated, and the absolute deviation of the median of the corresponding data subsequence is obtained based on the absolute deviation of each data point from the median. Determine the data point corresponding to the center position of the sliding window, and determine whether the deviation between it and the median is greater than the product of the absolute deviation of the median and a set multiple threshold. If the judgment condition is met, the corresponding data point is identified as having local abnormal fluctuations and is marked as a candidate abnormal point. The location index records of all data points marked as candidate outliers in the standardized stress data sequence are recorded and summarized to form a candidate outlier index set; In the second path, the standardized pressure data sequence of each pressure monitoring node is written into the data buffer structure in chronological order; The candidate outlier index set formed in the first path is passed to the second path. Based on the time index in the candidate outlier index set, the corresponding standardized stress data segment in the second path is extracted.

5. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, The generation of the posterior probability sequence at the change point includes: Obtain the time index corresponding to each candidate anomaly in the candidate anomaly index set. With the corresponding time index as the center, extract the standardized pressure data symmetrical before and after the corresponding time index in the standardized pressure data sequence corresponding to the pressure monitoring node to which the corresponding candidate anomaly belongs, according to the preset fixed window length, and form a standardized pressure data segment. The standardized pressure data segment is passed as input to the Bayesian online change point detection method. The Bayesian online change point detection method outputs a change point posterior probability sequence corresponding to each time point based on the pressure value at each time point in the standardized pressure data segment. The posterior probability sequence of the change point is analyzed to identify the amplitude change between the maximum and minimum probability values ​​in the posterior probability sequence of the change point, and the amplitude change is used as the cumulative amplitude of the change point probability. The cumulative amplitude of the change point probability is compared with a preset change point probability amplitude judgment threshold. If the cumulative amplitude of the change point probability exceeds the change point probability amplitude judgment threshold, the time index corresponding to the candidate anomaly point is confirmed as a valid anomaly trigger point.

6. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 5, characterized in that, Bayesian online change point detection methods include: Obtain the time index corresponding to each candidate anomaly in the candidate anomaly index set, and extract the standardized pressure data symmetrical before and after the corresponding time index in the standardized pressure data sequence corresponding to the pressure monitoring node associated with the corresponding candidate anomaly, with the time index as the center and according to the preset fixed time window length, to form a standardized pressure data segment. For the standardized stress data segments, a dynamic prior probability structure containing candidate outlier information is constructed. Each time index in the standardized stress data segments is traversed sequentially. It is determined whether each time index belongs to the candidate outlier index set. If it does, the change point prior probability of the time index is set to the enhanced value of the basic prior probability. If it does not, the change point prior probability of the corresponding time index is kept at the default value of the basic prior probability. The standardized stress data segment containing the dynamic prior probability structure is input into the Bayesian online change point detection method, and a recursive joint posterior inference operation is performed to output the change point posterior probability sequence that corresponds one-to-one with each time index in the standardized stress data segment. The output change point posterior probability sequence is enhanced by identifying each time index of the candidate outlier index set in the change point posterior probability sequence. A local symmetric response window is constructed with the identified time index as the center and a preset length. Within the response window, confidence enhancement operations are performed on the change point posterior probability values ​​at each time index, including weighted smoothing and trend interpolation, to generate the enhanced change point posterior probability sequence.

7. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, The generation of the set of abnormal event points includes: Obtain the enhanced variable point posterior probability sequence corresponding to each pressure monitoring node, and perform time-by-time index scanning on the variable point posterior probability sequence based on the unified time index set; In the posterior probability sequence of each pressure monitoring node, the time period in which the posterior probability value of the change point continuously increases with the time index is identified and defined as the continuous probability increasing segment. The continuous probability increasing segment must meet the following conditions: the posterior probability value of the change point corresponding to adjacent time indices shows a monotonically increasing or fluctuating increase within a limited range, and the difference between the posterior probability of the change point at the beginning and end of the corresponding time period is not less than the set minimum increase threshold, and the time span is not less than the set minimum duration threshold. Extract all time indices that meet the conditions and summarize them to form a set of suspected abnormal time indices; Obtain the candidate anomaly point index set, perform intersection processing based on the unified time index set on the suspected anomaly time index set and the candidate anomaly point index set, and extract each time index in the intersection; Each time index in the intersection is bound to the identifier of its respective pressure monitoring node to form a pressure monitoring node-time index tuple, and all tuples are aggregated to generate a set of abnormal event points.

8. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, The construction of the anomaly propagation path includes: Obtain the pressure monitoring node-time index tuple in the abnormal event point set, and extract the corresponding change point posterior probability value in the enhanced change point posterior probability sequence based on the time index to form an abnormal feature set containing abnormal event points and change point posterior probabilities. Based on the topology information of the steam pipeline, the set of direct adjacent pressure monitoring nodes for each abnormal event node is determined, and for each pair of adjacent pressure monitoring nodes with a connection relationship, the enhanced posterior probability value of the change point at the same time index is extracted under a unified time index set. Based on the steam flow direction information, the set of downstream adjacent pressure monitoring nodes corresponding to each abnormal event point is obtained. In the unified time index set, the difference calculation is performed on the change point posterior probability value between the abnormal event point and its downstream adjacent nodes to obtain the change point posterior probability difference of each node at the corresponding time index. Node pairs with a difference in posterior probability greater than a preset threshold are selected, and the direction of the difference is confirmed to be consistent with the direction of steam flow based on the steam flow direction information. If they are consistent, the corresponding node pairs are defined as abnormal propagation edges. All abnormal propagation edges are concatenated according to their time index, and combined with the spatial coordinate information of each pressure monitoring node in the steam pipeline topology information, an abnormal propagation path composed of multiple abnormal event points in the propagation order is constructed, and all paths are summarized to form an abnormal propagation path set.

9. The intelligent diagnostic method for abnormal pressure in directly buried steam pipelines according to claim 1, characterized in that, The output of abnormal diagnostic results includes: Obtain each abnormal propagation path in the abnormal propagation path set, count the frequency of abnormal triggering of the pressure monitoring node corresponding to the starting point of the path under the unified time index set, and record the time index of the earliest time that the node was identified as an abnormal event point, forming a candidate set of starting nodes and its corresponding time index set. In the candidate set of starting nodes, the path starting points in the abnormal propagation path are selected such that the direction of the difference in the posterior probability of the change point is always consistent with the direction of steam flow, and the posterior probability of the change point of each pressure monitoring node on the path shows a monotonically decreasing trend, thus forming a set of suspected abnormal source nodes. For each node in the set of suspected anomaly source nodes, sort them according to the first appearance time index in the set of anomaly event points, and extract the suspected anomaly source node with the earliest time index as the anomaly source pressure monitoring node. The abnormal source pressure monitoring node and its corresponding first occurrence time index are defined as the abnormal source pressure monitoring node and the first abnormal occurrence time, respectively, and output together with the corresponding abnormal propagation path to form the abnormal diagnosis result.

Images