Park pipe network monitoring and early warning method and system based on digital twinning

Abstract

The invention provides a digital twinborn-based park pipe network monitoring and early warning method and system, and relates to the technical field of pipe network monitoring and early warning, and the method comprises the steps: constructing a park pipe network domain knowledge graph, collecting real-time monitoring data and historical operation and maintenance data, and forming a knowledge enhanced pipe network data set; extracting pipe network performance attenuation characteristics by using a bidirectional gating recurrent neural network, and deeply fusing the pipe network performance attenuation characteristics with expert empirical rules in the domain knowledge graph to construct a performance attenuation analysis model; a performance degradation curve and remaining useful life are calculated based on the model, potential fault risks are predicted, and maintenance suggestions are generated. According to the invention, precise evaluation of pipe network performance degradation, early warning of fault risks and intelligent generation of maintenance suggestions are realized, and the intelligent level and reliability of operation and maintenance of the park pipe network are improved.

Description

Technical Field

[0001] The present invention relates to pipe network monitoring and early warning technology, and in particular to a park pipe network monitoring and early warning method and system based on digital twins. Background Art

[0002] The park pipe network is an important part of the park infrastructure, and its safe and stable operation is crucial to the overall operation of the park. Traditional park pipe network monitoring mainly relies on manual inspections and regular maintenance, which is inefficient, costly, and difficult to detect potential faults in a timely manner, which can easily cause safety accidents and economic losses. With the development of the Internet of Things, big data, and artificial intelligence technologies, more and more parks have begun to adopt digital and intelligent pipe network monitoring and early warning systems. These systems are usually based on sensors to collect real-time data from the pipe network and use data analysis technology for fault diagnosis and prediction.

[0003] Existing methods do not make full use of pipeline network data. Most systems only focus on real-time monitoring data, while ignoring valuable information such as historical operation and maintenance data and expert experience, resulting in low accuracy of fault diagnosis and prediction. Existing methods lack in-depth analysis of pipeline network performance degradation. Traditional threshold alarm methods can only alarm for faults that have occurred, but cannot warn of potential fault risks, making it difficult to achieve proactive maintenance. Existing methods are not intelligent enough. Many systems still rely on manual fault diagnosis and maintenance recommendation formulation, lacking automated and intelligent decision support, which increases the burden on operation and maintenance personnel. Summary of the invention

[0004] The embodiments of the present invention provide a campus pipe network monitoring and early warning method and system based on digital twins, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] Provides a digital twin-based park pipe network monitoring and early warning method, including:

[0007] Construct a domain knowledge graph for the park pipe network, wherein the domain knowledge graph includes entity nodes of pipe network equipment and edges of pipe network relationships; collect real-time monitoring data of the park pipe network, wherein the real-time monitoring data includes pressure data, flow data, and temperature data; obtain historical operation and maintenance data of the park pipe network, wherein the historical operation and maintenance data includes maintenance record data and replacement record data; perform data association mapping between the real-time monitoring data and the historical operation and maintenance data and the domain knowledge graph to form a knowledge-enhanced pipe network data set;

[0008] The knowledge-enhanced pipe network data set is segmented and processed in time sequence to construct a sliding time window sequence; the sliding time window sequence is input into a bidirectional gated recurrent neural network, and the bidirectional gated recurrent neural network includes a forward feature extraction unit and a reverse feature extraction unit; the pipe network performance attenuation characteristics are extracted by the forward feature extraction unit and the reverse feature extraction unit respectively; the pipe network performance attenuation characteristics are deeply integrated with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model;

[0009] The real-time collected park network performance parameters are input into the performance decay analysis model; the performance decay curve and remaining service life of the network equipment are calculated based on the performance decay analysis model; key performance inflection points are identified according to the performance decay curve, and potential failure risks are predicted; when the decay rate of the performance decay curve exceeds a preset threshold, fault diagnosis analysis is performed in combination with the domain knowledge graph to generate equipment maintenance recommendations; the remaining service life, potential failure risks, and equipment maintenance recommendations are integrated into an intelligent evaluation report, and sent to the network operation and maintenance terminal.

[0010] The real-time monitoring data, historical operation and maintenance data and the domain knowledge graph are mapped to form a knowledge-enhanced pipe network data set including:

[0011] The domain knowledge graph includes pipe network equipment entity nodes and pipe network relationship edges, the pipe network entity nodes include pipe section nodes, equipment nodes and measuring point nodes, and the pipe network relationship edges include spatial relationship edges and functional relationship edges; the real-time monitoring data is vectorized according to the time series features to obtain a time series feature vector, and the historical operation and maintenance data is vectorized according to the semantic features to obtain a semantic feature vector;

[0012] Establishing a first mapping relationship between the time series feature vector and the physical node of the pipe network equipment, and establishing a second mapping relationship between the semantic feature vector and the physical node of the pipe network equipment; wherein the first mapping relationship includes mapping the measuring point information in the time series feature vector to the measuring point node, and the second mapping relationship includes mapping the maintenance record information in the semantic feature vector to the equipment node;

[0013] Building a data association between the first mapping relationship and the second mapping relationship based on the pipe network relationship edge, wherein the data association includes associating the measurement point node to the pipe segment node through the spatial relationship edge, and associating the equipment node to the pipe segment node through the functional relationship edge;

[0014] Based on the first mapping relationship, the second mapping relationship and the data association, the real-time monitoring data and the historical operation and maintenance data are semantically associated and integrated in the domain knowledge graph to form a knowledge-enhanced pipeline network dataset with knowledge association.

[0015] The forward feature extraction unit and the reverse feature extraction unit are used to extract the pipeline network performance attenuation features respectively; the pipeline network performance attenuation features are deeply integrated with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model, including:

[0016] The forward feature extraction unit extracts the forward feature of the pipeline network performance attenuation. The extraction process of the forward feature of the pipeline network performance attenuation includes: extracting the time series change feature based on the multi-layer perceptron network, using the attention mechanism to identify the performance inflection point to obtain the key time node feature, calculating the performance deviation feature based on residual learning, and combining the time series change feature, the key time node feature and the performance deviation feature to form the forward feature of the pipeline network performance attenuation;

[0017] The reverse feature extraction unit extracts the reverse feature of the pipeline network performance attenuation. The extraction process of the reverse feature of the pipeline network performance attenuation includes: establishing a fault-symptom association model based on a Bayesian network to obtain a propagation path feature, using causal inference to obtain a performance attenuation root cause feature, and obtaining a performance degradation evolution feature based on a time series reverse analysis. The propagation path feature, the performance attenuation root cause feature, and the performance degradation evolution feature are combined to form the reverse feature of the pipeline network performance attenuation;

[0018] The forward feature of the pipeline network performance attenuation and the reverse feature of the pipeline network performance attenuation are feature-fused to obtain the pipeline network performance attenuation feature, wherein the feature fusion process includes: constructing a feature association matrix to calculate feature relevance, using an autoencoder to reduce the dimension of highly correlated features, and using a graph convolutional network to combine features;

[0019] Expert experience rules are extracted from the domain knowledge graph, and the expert experience rules are converted into calculation rules based on the rule formal language. The calculation rules are quantified through the rule reasoning mechanism to obtain rule features. The neural symbolic fusion network is used to deeply fuse the pipeline network performance attenuation features with the rule features, and a knowledge-guided attention mechanism is constructed based on the rule features to optimize the pipeline network performance attenuation features to obtain fusion features. A performance attenuation analysis model is constructed based on the fusion features.

[0020] Calculating the performance decay curve and remaining service life of the pipe network equipment based on the performance decay analysis model; identifying key performance inflection points based on the performance decay curve, and predicting potential failure risks include:

[0021] Inputting the output characteristics of the performance decay analysis model into a performance status quantitative index system, wherein the performance status quantitative index system includes an overall equipment performance evaluation index, a performance loss rate index, a reliability evaluation index, and an efficiency utilization rate index, and constructing a multidimensional performance evaluation matrix based on the performance status quantitative index system;

[0022] Based on the multidimensional performance evaluation matrix, a local weighted regression algorithm is used to process performance data fluctuations, a continuous performance curve is constructed in combination with piecewise polynomial fitting, and the continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance attenuation curve;

[0023] Establishing a deep learning prediction network, inputting the performance decay curve into the deep learning prediction network, fusing physical model constraints to extract multi-scale life characteristics, the multi-scale life characteristics including short-term fluctuation characteristics, medium-term trend characteristics and long-term evolution characteristics, and predicting the remaining service life of the equipment based on the multi-scale life characteristics;

[0024] Calculating the first-order derivative of the performance attenuation curve to obtain the change rate inflection point, analyzing the second-order derivative to obtain the acceleration change point, extracting the local extreme point to obtain the mutation feature point, and combining the change rate inflection point, the acceleration change point and the mutation feature point to form a key performance inflection point feature;

[0025] A density clustering algorithm is used to classify the key performance inflection point features, and the common features of inflection points of different categories are extracted to establish an inflection point evolution law model, and a risk propagation network is constructed based on the inflection point evolution law model;

[0026] The risk propagation network is deeply integrated with historical failure data, expert experience rules and environmental influencing factors to construct a multi-source information fusion failure risk prediction model, and the failure probability and risk level are calculated based on the failure risk prediction model.

[0027] Based on the multi-dimensional performance evaluation matrix, a local weighted regression algorithm is used to process performance data fluctuations, and a continuous performance curve is constructed by combining piecewise polynomial fitting. The continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance attenuation curve, including:

[0028] Receive performance data in the multidimensional performance evaluation matrix, calculate bandwidth parameters based on the performance data using an adaptive bandwidth selection strategy, dynamically adjust the bandwidth parameters with data density, construct a three-core weight function based on the bandwidth parameters, introduce time series correlation weights to form a dynamic weight calculation model; use a local weighted regression algorithm to perform fluctuation processing on the performance data, input the dynamic weight calculation model into a local weighted least squares objective function, use QR decomposition to solve the weighted normal equations, improve fitting accuracy through an iterative reweighting mechanism, and determine the optimal polynomial order in combination with cross-validation to obtain denoised performance data;

[0029] Based on the denoised performance data, the data curvature change is calculated, the segmentation point position is optimized by a dynamic programming algorithm, a criterion for determining the number of segmentation points is constructed, and the optimal segmentation point set is determined by local adjustment of the segmentation points; in combination with the piecewise polynomial fitting method, a cubic Hermite interpolation polynomial is constructed based on the optimal segmentation point set, a piecewise function continuity constraint condition and an endpoint derivative matching condition are established, and a monotonicity preservation strategy is introduced to form a piecewise function construction model; the piecewise function construction model is input into the overall smoothness objective function, a piecewise function parameter joint optimization algorithm is designed, a constraint condition processing mechanism is constructed, and a continuous performance curve is obtained through iterative optimization and solution;

[0030] The continuous performance curve is processed by kernel function smoothing technology, a variable bandwidth Gaussian kernel function is constructed, a bandwidth adaptive adjustment strategy is designed, a directional weight adjustment mechanism is introduced, and a kernel function shape optimization method is established to obtain an adaptive kernel function; a multi-scale decomposition framework is constructed based on the adaptive kernel function, a scale weight adaptive allocation strategy is designed, and a feature transfer mechanism between scales is established to achieve multi-scale result fusion; the boundary effect of the multi-scale result fusion is processed, a boundary correction kernel function and a boundary extrapolation strategy are designed, boundary constraints are constructed, and the final performance attenuation curve is obtained through a gradual transition processing mechanism.

[0031] When the decay rate of the performance decay curve exceeds a preset threshold, a fault diagnosis analysis is performed in combination with the domain knowledge graph to generate equipment maintenance suggestions including:

[0032] A benchmark threshold matrix is ​​constructed by performing probability density estimation on the performance attenuation curve, a seasonal adjustment factor is introduced to dynamically modify the benchmark threshold matrix, and a personalized threshold adjustment coefficient is constructed according to the equipment operating years and load characteristics to obtain a dynamic threshold criterion;

[0033] The local decay rate of the performance decay curve is calculated by using a sliding window technique, the local decay rate is input into an exponential smoothing processing model to eliminate short-term fluctuations, and the future decay trend is predicted based on time series analysis to obtain a device state prediction result; the device state prediction result is compared with the dynamic threshold criterion, and when the device state prediction result exceeds the dynamic threshold criterion, a fault diagnosis process is triggered;

[0034] Construct an equipment ontology knowledge base including equipment structure and function parameters, establish a fault case knowledge base including fault features and solutions, formalize the expert diagnosis rules to form an expert experience knowledge base, and construct a probability association network between features and fault causes through association rule mining based on the equipment ontology knowledge base, the fault case knowledge base and the expert experience knowledge base;

[0035] The equipment state prediction result is input into the probability association network, a multi-path reasoning algorithm is designed, the uncertainty in the diagnosis process is processed by evidence theory, a diagnosis credibility evaluation model is constructed by comprehensively considering the knowledge support and rule matching, and the fault causes are sorted based on the multi-criteria decision-making method to obtain the fault diagnosis result;

[0036] Based on the fault diagnosis results, a hierarchical maintenance solution system including emergency treatment and planned maintenance is established, the economic feasibility of different maintenance solutions is evaluated, the impact of the remaining life of the equipment on the maintenance strategy is considered, and a maintenance condition evaluation model is constructed; the maintenance condition evaluation model is input into the maintenance resource availability and time window constraint analysis module, the maintenance priority is determined in combination with the fault risk and the importance of the equipment, and a combination optimization model of the solution for multi-equipment collaborative maintenance is established to obtain the maintenance solution combination result;

[0037] The maintenance scheme combination results are displayed in grades according to urgency and importance, a maintenance step flow chart is generated, and maintenance effect feedback data is input into the fault case knowledge base and the expert experience knowledge base for updating to form closed-loop optimized equipment maintenance recommendations.

[0038] According to a second aspect of the embodiments of the present invention,

[0039] Provide a digital twin-based park pipe network monitoring and early warning system, including:

[0040] The first unit is used to construct a domain knowledge graph of the park pipe network, wherein the domain knowledge graph includes entity nodes of pipe network equipment and edges of pipe network relationships; collect real-time monitoring data of the park pipe network, wherein the real-time monitoring data includes pressure data, flow data, and temperature data; obtain historical operation and maintenance data of the park pipe network, wherein the historical operation and maintenance data includes maintenance record data and replacement record data; perform data association mapping between the real-time monitoring data and the historical operation and maintenance data and the domain knowledge graph to form a knowledge-enhanced pipe network data set;

[0041] The second unit is used to segment the knowledge-enhanced pipe network data set in time sequence to construct a sliding time window sequence; input the sliding time window sequence into a bidirectional gated recurrent neural network, the bidirectional gated recurrent neural network includes a forward feature extraction unit and a reverse feature extraction unit; extract the pipe network performance attenuation characteristics through the forward feature extraction unit and the reverse feature extraction unit respectively; deeply integrate the pipe network performance attenuation characteristics with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model;

[0042] The third unit is used to input the real-time collected park pipeline performance parameters into the performance decay analysis model; calculate the performance decay curve and remaining service life of the pipeline equipment based on the performance decay analysis model; identify key performance inflection points according to the performance decay curve, and predict potential failure risks; when the decay rate of the performance decay curve exceeds a preset threshold, perform fault diagnosis analysis in combination with the domain knowledge graph and generate equipment maintenance suggestions; integrate the remaining service life, potential failure risks, and equipment maintenance suggestions into an intelligent evaluation report, and send it to the pipeline operation and maintenance terminal.

[0043] According to a third aspect of the embodiments of the present invention,

[0044] An electronic device is provided, comprising:

[0045] processor;

[0046] a memory for storing processor-executable instructions;

[0047] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0048] A fourth aspect of the embodiments of the present invention is:

[0049] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0050] The beneficial effects of this application are as follows:

[0051] 1. Improve the accuracy of pipeline network monitoring and early warning: This method constructs a knowledge-enhanced pipeline network dataset by combining real-time monitoring data, historical operation and maintenance data, and domain knowledge graphs, and uses a bidirectional gated recurrent neural network to extract pipeline network performance degradation characteristics. It can more comprehensively and accurately analyze the performance degradation of pipeline network equipment, thereby improving the accuracy of fault prediction and diagnosis.

[0052] 2. Extend the service life of pipeline network equipment: By constructing a performance degradation analysis model, the remaining service life of pipeline network equipment can be calculated, and key performance inflection points can be identified based on the performance degradation curve to predict potential failure risks, so that maintenance measures can be taken in advance to avoid equipment failure and extend the service life of equipment.

[0053] 3. Improve the intelligence level of pipeline network operation and maintenance: This method can automatically generate intelligent evaluation reports, provide equipment maintenance suggestions, and send them to the pipeline network operation and maintenance terminal to realize intelligent management of pipeline network operation and maintenance, reduce manual intervention, and improve operation and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of a park pipe network monitoring and early warning method based on digital twins according to an embodiment of the present invention;

[0055] Figure 2 It is a structural schematic diagram of a campus pipe network monitoring and early warning system based on digital twins according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0058] Figure 1 Schematic diagram of the process of the park pipe network monitoring and early warning method based on digital twins according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] S11. Construct a domain knowledge graph for the park pipe network, wherein the domain knowledge graph includes entity nodes of pipe network equipment and edges of pipe network relationships; collect real-time monitoring data of the park pipe network, wherein the real-time monitoring data includes pressure data, flow data, and temperature data; obtain historical operation and maintenance data of the park pipe network, wherein the historical operation and maintenance data includes maintenance record data and replacement record data; perform data association mapping between the real-time monitoring data and the historical operation and maintenance data and the domain knowledge graph to form a knowledge-enhanced pipe network data set;

[0060] S12. The knowledge-enhanced pipe network data set is segmented and processed in time sequence to construct a sliding time window sequence; the sliding time window sequence is input into a bidirectional gated recurrent neural network, wherein the bidirectional gated recurrent neural network includes a forward feature extraction unit and a reverse feature extraction unit; the pipe network performance attenuation characteristics are extracted by the forward feature extraction unit and the reverse feature extraction unit respectively; the pipe network performance attenuation characteristics are deeply integrated with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model;

[0061] S13. Input the real-time collected park network performance parameters into the performance decay analysis model; calculate the performance decay curve and remaining service life of the network equipment based on the performance decay analysis model; identify key performance inflection points according to the performance decay curve, and predict potential failure risks; when the decay rate of the performance decay curve exceeds a preset threshold, perform fault diagnosis analysis in combination with the domain knowledge graph, and generate equipment maintenance recommendations; integrate the remaining service life, potential failure risks, and equipment maintenance recommendations into an intelligent evaluation report, and send it to the network operation and maintenance terminal.

[0062] In an optional implementation, the real-time monitoring data, the historical operation and maintenance data and the domain knowledge graph are data-associated and mapped to form a knowledge-enhanced pipe network data set, including:

[0063] The domain knowledge graph includes pipe network equipment entity nodes and pipe network relationship edges, the pipe network entity nodes include pipe section nodes, equipment nodes and measuring point nodes, and the pipe network relationship edges include spatial relationship edges and functional relationship edges; the real-time monitoring data is vectorized according to the time series features to obtain a time series feature vector, and the historical operation and maintenance data is vectorized according to the semantic features to obtain a semantic feature vector;

[0064] Establishing a first mapping relationship between the time series feature vector and the physical node of the pipe network equipment, and establishing a second mapping relationship between the semantic feature vector and the physical node of the pipe network equipment; wherein the first mapping relationship includes mapping the measuring point information in the time series feature vector to the measuring point node, and the second mapping relationship includes mapping the maintenance record information in the semantic feature vector to the equipment node;

[0065] Building a data association between the first mapping relationship and the second mapping relationship based on the pipe network relationship edge, wherein the data association includes associating the measurement point node to the pipe segment node through the spatial relationship edge, and associating the equipment node to the pipe segment node through the functional relationship edge;

[0066] Based on the first mapping relationship, the second mapping relationship and the data association, the real-time monitoring data and the historical operation and maintenance data are semantically associated and integrated in the domain knowledge graph to form a knowledge-enhanced pipeline network dataset with knowledge association.

[0067] A knowledge-enhanced pipeline network dataset construction method is used to associate and map real-time monitoring data, historical operation and maintenance data with domain knowledge graphs to form a knowledge-enhanced pipeline network dataset.

[0068] First, construct a knowledge graph in the field of pipe network. The graph contains pipe network equipment entity nodes and pipe network relationship edges. Pipe network entity nodes include pipe segment nodes, such as "pipe segment A", "pipe segment B"; equipment nodes, such as "valve 1", "pump station 2"; measurement point nodes, such as "pressure sensor P1", "flow meter F2". Pipe network relationship edges include spatial relationship edges, such as "pipe segment A connects to valve 1", "pipe segment B connects to pump station 2", indicating that the equipment is located on the pipe segment; and functional relationship edges, such as "valve 1 controls pipe segment A", "pump station 2 supplies water to pipe segment B", indicating the functional effect of the equipment on the pipe segment.

[0069] Next, the real-time monitoring data and historical operation and maintenance data are vectorized. Real-time monitoring data, such as the pressure value from "pressure sensor P1" and the flow value from "flow meter F2", are organized according to time series. For example, the pressure value and flow value per hour are combined into a vector to form a time series feature vector, such as "[1.2MPa,2.5m 3 / s]" represents the pressure and flow values ​​at a certain moment. Historical operation and maintenance data, such as "Valve 1 repaired and replaced the seal ring on January 1, 2024" and "Pump station 2 repaired and replaced the motor on March 5, 2024", are organized according to semantic features to form a semantic feature vector. For example, the maintenance record text can be segmented and encoded to form a vector representation, such as "[0.2, 0.5, 0.8, ...]" represents the semantic features of a maintenance record.

[0070] Then, a first mapping relationship between the time series feature vector and the physical node of the pipeline network equipment, and a second mapping relationship between the semantic feature vector and the physical node of the pipeline network equipment are established. The first mapping relationship maps the measurement point information in the time series feature vector to the corresponding measurement point node. For example, the time series feature vector "[1.2MPa,...]" from "pressure sensor P1" is mapped to the "pressure sensor P1" node in the knowledge graph. The second mapping relationship maps the maintenance record information in the semantic feature vector to the corresponding device node. For example, the semantic feature vector "[0.2,0.5,0.8,...]" of "Valve 1 maintenance and replacement of sealing ring on January 1, 2024" is mapped to the "Valve 1" node in the knowledge graph.

[0071] Subsequently, the data association between the first mapping relationship and the second mapping relationship is constructed based on the pipe network relationship edge. For example, "pressure sensor P1" is associated with "pipe segment A" through the spatial relationship edge "pressure sensor P1 is located in pipe segment A", and "valve 1" is also associated with "pipe segment A" through the functional relationship edge "valve 1 controls pipe segment A". In this way, real-time monitoring data and historical operation and maintenance data are associated with the same pipe segment.

[0072] Finally, based on the first mapping relationship, the second mapping relationship and data association, the real-time monitoring data and the historical operation and maintenance data are semantically associated and integrated in the domain knowledge graph to form a knowledge-enhanced pipeline network data set with knowledge association. For example, the real-time monitoring data of "pipe section A" (from "pressure sensor P1") and the historical operation and maintenance data (from the maintenance record of "valve 1") are integrated together to form the complete information of the pipe section.

[0073] The solution of this application can:

[0074] Data fusion: Integrate data from different sources and different types (real-time monitoring data and historical operation and maintenance data) into a unified knowledge graph to form a more comprehensive data set, avoid data silos, and facilitate comprehensive analysis and utilization. Knowledge enhancement: Use domain knowledge graphs to associate data with knowledge such as pipeline equipment, spatial relationships, and functional relationships, giving data richer semantic information and improving data value and interpretability. For example, by associating the maintenance record of "Valve 1" and the pressure data of "Pipeline Section A", the impact of maintenance operations on the pressure of the pipeline section can be analyzed. Decision support: Knowledge-enhanced pipeline network data sets can provide more accurate data support for pipeline operation and maintenance, fault diagnosis, prediction and early warning, etc., and assist managers in making more scientific decisions. For example, combining the historical maintenance records and real-time pressure data of a pipeline section can predict the possible failure risk of the pipeline section in the future.

[0075] In an optional implementation, the forward feature extraction unit and the reverse feature extraction unit are used to extract the pipeline network performance attenuation features respectively; the pipeline network performance attenuation features are deeply integrated with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model, including:

[0076] The forward feature extraction unit extracts the forward feature of the pipeline network performance attenuation. The extraction process of the forward feature of the pipeline network performance attenuation includes: extracting the time series change feature based on the multi-layer perceptron network, using the attention mechanism to identify the performance inflection point to obtain the key time node feature, calculating the performance deviation feature based on residual learning, and combining the time series change feature, the key time node feature and the performance deviation feature to form the forward feature of the pipeline network performance attenuation;

[0077] The reverse feature extraction unit extracts the reverse feature of the pipeline network performance attenuation. The extraction process of the reverse feature of the pipeline network performance attenuation includes: establishing a fault-symptom association model based on a Bayesian network to obtain a propagation path feature, using causal inference to obtain a performance attenuation root cause feature, and obtaining a performance degradation evolution feature based on a time series reverse analysis. The propagation path feature, the performance attenuation root cause feature, and the performance degradation evolution feature are combined to form the reverse feature of the pipeline network performance attenuation;

[0078] The forward feature of the pipeline network performance attenuation and the reverse feature of the pipeline network performance attenuation are feature-fused to obtain the pipeline network performance attenuation feature, wherein the feature fusion process includes: constructing a feature association matrix to calculate feature relevance, using an autoencoder to reduce the dimension of highly correlated features, and using a graph convolutional network to combine features;

[0079] Expert experience rules are extracted from the domain knowledge graph, and the expert experience rules are converted into calculation rules based on the rule formal language. The calculation rules are quantified through the rule reasoning mechanism to obtain rule features. The neural symbolic fusion network is used to deeply fuse the pipeline network performance attenuation features with the rule features, and a knowledge-guided attention mechanism is constructed based on the rule features to optimize the pipeline network performance attenuation features to obtain fusion features. A performance attenuation analysis model is constructed based on the fusion features.

[0080] In order to achieve accurate analysis of pipeline network performance degradation, a pipeline network performance degradation analysis model construction method based on domain knowledge graph and deep learning is proposed. This method combines the advantages of data-driven and knowledge-guided, and can more comprehensively and accurately identify and analyze the causes of pipeline network performance degradation.

[0081] First, the historical operation data of the pipeline network is collected and preprocessed. For example, the pipeline network flow, pressure, water quality and other data are collected, and the data is cleaned, denoised, normalized and processed for subsequent feature extraction. Assume that the pipeline network flow data for a period of time is collected, such as the daily flow value, and a time series is obtained after preprocessing.

[0082] Then, the forward feature extraction unit extracts the positive feature of the pipeline network performance attenuation. The multi-layer perceptron network is used to extract the time series change feature. For example, the preprocessed time series is input into a three-layer multi-layer perceptron network, and the output of the network can represent the time series change trend of the flow. Then the attention mechanism is used to identify the performance inflection point to obtain the key time node feature. For example, by analyzing the change trend of the flow time series, the time points at which the flow rate drops significantly can be identified, and these time points can be used as key time node features. After that, the performance deviation feature is calculated based on residual learning. For example, the difference between the actual flow value and the predicted flow value is used as the performance deviation feature. Finally, the time series change feature, the key time node feature and the performance deviation feature are combined to form the positive feature of the pipeline network performance attenuation. For example, the above three features are spliced ​​into a vector as the positive feature of the pipeline network performance attenuation.

[0083] Next, the reverse feature extraction unit extracts the reverse feature of the pipeline network performance degradation. A fault-symptom association model is established based on the Bayesian network to obtain the propagation path feature. For example, a Bayesian network is constructed, in which the nodes of the network represent different components in the pipeline network, and the edges between the nodes represent the dependencies between the components. Through this network, the propagation path of the fault in the pipeline network can be inferred. The root cause feature of performance degradation is obtained by causal inference. For example, by analyzing the historical operation data of the pipeline network and the fault-symptom association model, the root cause of performance degradation, such as pipeline corrosion and leakage, can be inferred. The performance degradation evolution feature is obtained based on the reverse analysis of the time series. For example, by reversely analyzing the time series, the evolution process of performance degradation can be understood. Finally, the propagation path feature, the root cause feature of performance degradation, and the evolution feature of performance degradation are combined to form the reverse feature of pipeline network performance degradation. For example, the above three features are spliced ​​into a vector as the reverse feature of pipeline network performance degradation.

[0084] Subsequently, the forward features and reverse features are fused to obtain the pipeline network performance attenuation features. A feature association matrix is ​​constructed to calculate feature correlation. For example, the correlation coefficient between the forward feature vector and the reverse feature vector is calculated to obtain a feature association matrix. An autoencoder is used to reduce the dimension of highly correlated features. For example, by inputting the feature association matrix into an autoencoder, high-dimensional features can be reduced to a low-dimensional space. A graph convolutional network is used for feature combination. For example, by inputting the reduced-dimensional features into a graph convolutional network, the complex relationship between the features can be learned and combined into a new feature vector as the pipeline network performance attenuation feature.

[0085] At the same time, expert experience rules are extracted from the domain knowledge graph. For example, the rule "pipeline corrosion will cause flow rate reduction" is extracted from the knowledge graph. Expert experience rules are converted into calculation rules based on rule formalization language. For example, the above rule is converted into "if the pipeline corrosion degree is greater than a certain threshold, the probability of flow rate reduction will increase". The calculation rules are quantified through the rule reasoning mechanism to obtain rule features. For example, a rule feature value is calculated based on the pipeline corrosion degree and the probability of flow rate reduction. The neural symbolic fusion network is used to deeply fuse the pipeline network performance attenuation feature with the rule feature. For example, the pipeline network performance attenuation feature and the rule feature are input into a neural symbolic fusion network. The network can learn the relationship between data features and knowledge rules and fuse them into a new feature vector. Based on the rule feature, a knowledge-guided attention mechanism is constructed to optimize the pipeline network performance attenuation feature to obtain a fusion feature. For example, according to the rule feature value, the pipeline network performance attenuation feature is weighted to highlight the important features. Finally, a performance attenuation analysis model is constructed based on the fusion feature. For example, the fusion feature is input into a classifier or regressor to predict the future performance attenuation of the pipeline network.

[0086] The solution of this application can:

[0087] Improved the accuracy of performance degradation analysis: This method combines the advantages of data-driven and knowledge-guided methods, and can more comprehensively consider the factors that affect the performance degradation of the pipeline network, thereby improving the accuracy of the analysis. Enhanced the interpretability of performance degradation analysis: This method uses domain knowledge graphs and expert experience rules to explain the causes and mechanisms of performance degradation, and enhances the interpretability of the analysis results. Improved the efficiency of performance degradation analysis: This method uses technologies such as deep learning and graph convolutional networks to automatically extract features and learning rules, reduce manual intervention, and improve analysis efficiency.

[0088] In an optional implementation, calculating the performance decay curve and remaining service life of the pipe network equipment based on the performance decay analysis model; identifying key performance inflection points according to the performance decay curve, and predicting potential failure risks include:

[0089] Inputting the output characteristics of the performance decay analysis model into a performance status quantitative index system, wherein the performance status quantitative index system includes an overall equipment performance evaluation index, a performance loss rate index, a reliability evaluation index, and an efficiency utilization rate index, and constructing a multidimensional performance evaluation matrix based on the performance status quantitative index system;

[0090] Based on the multidimensional performance evaluation matrix, a local weighted regression algorithm is used to process performance data fluctuations, a continuous performance curve is constructed in combination with piecewise polynomial fitting, and the continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance attenuation curve;

[0091] Establishing a deep learning prediction network, inputting the performance decay curve into the deep learning prediction network, fusing physical model constraints to extract multi-scale life characteristics, the multi-scale life characteristics including short-term fluctuation characteristics, medium-term trend characteristics and long-term evolution characteristics, and predicting the remaining service life of the equipment based on the multi-scale life characteristics;

[0092] Calculating the first-order derivative of the performance attenuation curve to obtain the change rate inflection point, analyzing the second-order derivative to obtain the acceleration change point, extracting the local extreme point to obtain the mutation feature point, and combining the change rate inflection point, the acceleration change point and the mutation feature point to form a key performance inflection point feature;

[0093] A density clustering algorithm is used to classify the key performance inflection point features, and the common features of inflection points of different categories are extracted to establish an inflection point evolution law model, and a risk propagation network is constructed based on the inflection point evolution law model;

[0094] The risk propagation network is deeply integrated with historical failure data, expert experience rules and environmental influencing factors to construct a multi-source information fusion failure risk prediction model, and the failure probability and risk level are calculated based on the failure risk prediction model.

[0095] The performance degradation and remaining service life prediction method of pipeline equipment predicts the future performance degradation trend and remaining service life of the equipment by analyzing the equipment operation data, so as to take maintenance measures in time to avoid equipment failure.

[0096] First, collect the operating data of the pipeline network equipment, such as flow, pressure, temperature, vibration, etc. This data can come from sensors, monitoring systems or databases. For example, collect the pressure data of a certain section of pipeline for one year, collect it once a day, and a total of 365 data points.

[0097] Then, select a suitable performance decay analysis model to analyze the collected data. Commonly used models include exponential model, power function model, logarithmic model, etc. The choice of model should be determined according to the characteristics of the data and the actual situation. For example, according to the analysis of historical data, the pressure decay of the pipeline conforms to the exponential model.

[0098] The output characteristics of the performance degradation analysis model are input into the performance status quantitative index system. The system includes the overall performance evaluation index of the equipment, the performance loss rate index, the reliability evaluation index and the efficiency utilization index. For example, the performance loss rate of the pipeline is calculated to be 5% per year based on the pressure data.

[0099] Next, a multi-dimensional performance evaluation matrix is ​​constructed based on the performance status quantitative index system. For example, the performance loss rates of indicators such as pressure, flow, and temperature are combined into a matrix.

[0100] The local weighted regression algorithm is used to handle performance data fluctuations. This algorithm can effectively smooth out noise and outliers in the data. For example, the local weighted regression algorithm is used to smooth the collected pressure data.

[0101] Combine piecewise polynomial fitting to build a continuous performance curve. Divide the smoothed data into several segments, fit each segment with a polynomial, and finally get a continuous performance curve. For example, divide a year's pressure data into 12 segments, and fit each segment with a quadratic polynomial.

[0102] The continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance decay curve. For example, a Gaussian kernel function is used to smooth the fitted curve to obtain a final performance decay curve.

[0103] A deep learning prediction network is established and the performance decay curve is input into the network. The network can integrate physical model constraints and extract multi-scale life characteristics, including short-term fluctuation characteristics, medium-term trend characteristics, and long-term evolution characteristics. For example, convolutional neural networks are used to extract features at different time scales.

[0104] Predict the remaining service life of equipment based on multi-scale life features. For example, based on the extracted features, predict the remaining service life of the pipeline in the next 5 years.

[0105] The first-order derivative of the performance attenuation curve is calculated to obtain the inflection point of the rate of change, the second-order derivative is analyzed to obtain the acceleration change point, and the local extreme point is extracted to obtain the mutation feature point. For example, by calculating the first-order derivative of the performance attenuation curve, it is found that the pressure drop rate suddenly accelerates at a certain time point.

[0106] The change rate inflection point, acceleration change point and mutation feature point are combined to form the key performance inflection point feature. For example, the time point when the pressure drop speed increases is marked as a key performance inflection point.

[0107] The density clustering algorithm is used to classify the key performance inflection point features, extract the common features of inflection points of different categories, and establish an inflection point evolution law model. For example, all inflection points with accelerated pressure drop speed are classified into one category and their common features are analyzed.

[0108] A risk propagation network is constructed based on the inflection point evolution law model. The network can describe the mutual influence and evolution relationship between different inflection points. For example, an accelerated pressure drop may lead to a pipeline rupture.

[0109] The risk propagation network is deeply integrated with historical fault data, expert experience rules and environmental influencing factors to build a multi-source information fusion fault risk prediction model. For example, combining historical fault data and expert experience to determine whether the accelerated pressure drop will lead to an increased risk of pipeline rupture.

[0110] The probability of failure and risk level are calculated based on the failure risk prediction model. For example, the probability of the pipeline rupturing in the next year is calculated to be 10%, and the risk level is medium.

[0111] The solution of this application can:

[0112] Improve prediction accuracy: This method takes into account a variety of factors, including performance degradation curves, multi-scale life characteristics, key performance inflection points, historical failure data, expert experience, and environmental factors, and can more accurately predict the remaining service life and failure risk of equipment. Enhance the reliability of prediction: This method uses local weighted regression, piecewise polynomial fitting, kernel function smoothing and other technologies to effectively deal with data fluctuations and noise, and improve the reliability of prediction results. Improve the practicality of prediction: This method can identify key performance inflection points, predict potential failure risks, and provide a basis for formulating reasonable maintenance strategies, which helps to extend the service life of equipment and reduce maintenance costs.

[0113] In an optional implementation, based on the multidimensional performance evaluation matrix, a local weighted regression algorithm is used to process performance data fluctuations, a continuous performance curve is constructed in combination with piecewise polynomial fitting, and the continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance attenuation curve, including:

[0114] Receive performance data in the multidimensional performance evaluation matrix, calculate bandwidth parameters based on the performance data using an adaptive bandwidth selection strategy, dynamically adjust the bandwidth parameters with data density, construct a three-core weight function based on the bandwidth parameters, introduce time series correlation weights to form a dynamic weight calculation model; use a local weighted regression algorithm to perform fluctuation processing on the performance data, input the dynamic weight calculation model into a local weighted least squares objective function, use QR decomposition to solve the weighted normal equations, improve fitting accuracy through an iterative reweighting mechanism, and determine the optimal polynomial order in combination with cross-validation to obtain denoised performance data;

[0115] Based on the denoised performance data, the data curvature change is calculated, the segmentation point position is optimized by a dynamic programming algorithm, a criterion for determining the number of segmentation points is constructed, and the optimal segmentation point set is determined by local adjustment of the segmentation points; in combination with the piecewise polynomial fitting method, a cubic Hermite interpolation polynomial is constructed based on the optimal segmentation point set, a piecewise function continuity constraint condition and an endpoint derivative matching condition are established, and a monotonicity preservation strategy is introduced to form a piecewise function construction model; the piecewise function construction model is input into the overall smoothness objective function, a piecewise function parameter joint optimization algorithm is designed, a constraint condition processing mechanism is constructed, and a continuous performance curve is obtained through iterative optimization and solution;

[0116] The continuous performance curve is processed by kernel function smoothing technology, a variable bandwidth Gaussian kernel function is constructed, a bandwidth adaptive adjustment strategy is designed, a directional weight adjustment mechanism is introduced, and a kernel function shape optimization method is established to obtain an adaptive kernel function; a multi-scale decomposition framework is constructed based on the adaptive kernel function, a scale weight adaptive allocation strategy is designed, and a feature transfer mechanism between scales is established to achieve multi-scale result fusion; the boundary effect of the multi-scale result fusion is processed, a boundary correction kernel function and a boundary extrapolation strategy are designed, boundary constraints are constructed, and the final performance attenuation curve is obtained through a gradual transition processing mechanism.

[0117] Receive performance data of a multi-dimensional performance evaluation matrix. Take a matrix containing time and three performance indicators (e.g., CPU utilization, memory usage, and disk IO rate) as an example. Assume that data at 100 time points are collected to form a 100x4 matrix, where the first column is time and the next three columns are data of the three performance indicators.

[0118] The bandwidth parameter is calculated based on the performance data using an adaptive bandwidth selection strategy. Considering that data density may vary in different time periods, an adaptive bandwidth selection strategy is used. For example, the k-nearest neighbor method is used to dynamically adjust the bandwidth parameter based on the local density of data points. In areas with dense data, the bandwidth parameter is small, while in areas with sparse data, the bandwidth parameter is large. Assuming that the data is dense around the 50th time point, the bandwidth parameter at this time is set to 0.5; and the data is sparse around the 90th time point, the bandwidth parameter at this time is set to 1.0.

[0119] The three-core weight function is constructed based on the bandwidth parameters, and the time-series correlation weight is introduced to form a dynamic weight calculation model. The three-core weight function is constructed using the Gaussian kernel function as the basis and combined with the calculated bandwidth parameters. At the same time, the time-series correlation weight is introduced. For example, an exponential decay function can be used so that the weight of the data point farther from the current time point is smaller. The three-core weight function and the time-series correlation weight are multiplied to form a dynamic weight calculation model.

[0120] A local weighted regression algorithm is used to handle fluctuations in performance data. The dynamic weight calculation model is input into the local weighted least squares objective function, and the weighted normal equations are solved using QR decomposition. The weights and regression coefficients are continuously updated through an iterative reweighting mechanism to improve fitting accuracy. A cross-validation method, such as 5-fold cross-validation, is used to determine the optimal polynomial order. Assuming that the final polynomial order is 3, the denoised performance data is obtained.

[0121] The data curvature change is calculated based on the denoised performance data. The data curvature change is calculated using a difference method, such as a second order difference.

[0122] The dynamic programming algorithm is used to optimize the segmentation point positions, and the criterion for determining the number of segmentation points is constructed. The optimal segmentation point set is determined by local adjustment of the segmentation points. According to the change of data curvature, the segmentation point positions are determined by the dynamic programming algorithm, and the criterion for determining the number of segmentation points is constructed, such as the AIC criterion or the BICC criterion. The optimal segmentation point set is determined by locally adjusting the segmentation point positions, for example, moving forward or backward by one time point. Assume that three segmentation points are finally determined, which are located at the 20th, 50th, and 80th time points respectively.

[0123] Combined with the piecewise polynomial fitting method, a cubic Hermite interpolation polynomial is constructed based on the optimal piecewise point set. The piecewise function continuity constraint and endpoint derivative matching conditions are established, and a monotonicity preservation strategy is introduced to form a piecewise function construction model. The piecewise function construction model is input into the overall smoothness objective function, a piecewise function parameter joint optimization algorithm is designed, a constraint processing mechanism is constructed, and a continuous performance curve is obtained through iterative optimization.

[0124] The kernel function smoothing technology is used to process the continuous performance curve. A variable bandwidth Gaussian kernel function is constructed, a bandwidth adaptive adjustment strategy is designed, a directional weight adjustment mechanism is introduced, and a kernel function shape optimization method is established to obtain an adaptive kernel function.

[0125] A multi-scale decomposition framework is constructed based on an adaptive kernel function. A scale weight adaptive allocation strategy is designed, and a feature transfer mechanism between scales is established to achieve multi-scale result fusion.

[0126] The boundary effect is processed for the fusion of multi-scale results. The boundary correction kernel function and boundary extrapolation strategy are designed, the boundary constraint conditions are constructed, and the final performance attenuation curve is obtained through the gradual transition processing mechanism.

[0127] The solution of this application can:

[0128] Improve the prediction accuracy of the performance attenuation curve: Through technologies such as adaptive bandwidth selection, dynamic weight calculation, piecewise polynomial fitting and kernel function smoothing, the impact of data fluctuations and noise is effectively reduced, the prediction accuracy of the performance attenuation curve is improved, and the performance degradation trend of the equipment can be more accurately reflected. Enhance the robustness and adaptability of the model: Adopting an adaptive bandwidth selection strategy and a multi-scale decomposition framework, the model can adapt to situations with different data densities and noise levels, enhancing the robustness and adaptability of the model, and can be applied to a wider range of equipment performance evaluation scenarios. Improve performance evaluation efficiency: By optimizing the location of segmentation points through a dynamic programming algorithm and designing a joint optimization algorithm for segmentation function parameters, the computational efficiency of the model is improved, and the performance attenuation curve can be constructed more quickly, providing timely and effective decision support for equipment maintenance and replacement.

[0129] In an optional implementation, when the decay rate of the performance decay curve exceeds a preset threshold, fault diagnosis analysis is performed in combination with the domain knowledge graph to generate equipment maintenance suggestions including:

[0130] A benchmark threshold matrix is ​​constructed by performing probability density estimation on the performance attenuation curve, a seasonal adjustment factor is introduced to dynamically modify the benchmark threshold matrix, and a personalized threshold adjustment coefficient is constructed according to the equipment operating years and load characteristics to obtain a dynamic threshold criterion;

[0131] The local decay rate of the performance decay curve is calculated by using a sliding window technique, the local decay rate is input into an exponential smoothing processing model to eliminate short-term fluctuations, and the future decay trend is predicted based on time series analysis to obtain a device state prediction result; the device state prediction result is compared with the dynamic threshold criterion, and when the device state prediction result exceeds the dynamic threshold criterion, a fault diagnosis process is triggered;

[0132] Construct an equipment ontology knowledge base including equipment structure and function parameters, establish a fault case knowledge base including fault features and solutions, formalize the expert diagnosis rules to form an expert experience knowledge base, and construct a probability association network between features and fault causes through association rule mining based on the equipment ontology knowledge base, the fault case knowledge base and the expert experience knowledge base;

[0133] The equipment state prediction result is input into the probability association network, a multi-path reasoning algorithm is designed, the uncertainty in the diagnosis process is processed by evidence theory, a diagnosis credibility evaluation model is constructed by comprehensively considering the knowledge support and rule matching, and the fault causes are sorted based on the multi-criteria decision-making method to obtain the fault diagnosis result;

[0134] Based on the fault diagnosis results, a hierarchical maintenance solution system including emergency treatment and planned maintenance is established, the economic feasibility of different maintenance solutions is evaluated, the impact of the remaining life of the equipment on the maintenance strategy is considered, and a maintenance condition evaluation model is constructed; the maintenance condition evaluation model is input into the maintenance resource availability and time window constraint analysis module, the maintenance priority is determined in combination with the fault risk and the importance of the equipment, and a combination optimization model of the solution for multi-equipment collaborative maintenance is established to obtain the maintenance solution combination result;

[0135] The maintenance scheme combination results are displayed in grades according to urgency and importance, a maintenance step flow chart is generated, and maintenance effect feedback data is input into the fault case knowledge base and the expert experience knowledge base for updating to form closed-loop optimized equipment maintenance recommendations.

[0136] The method for generating equipment maintenance suggestions includes the following steps:

[0137] First, construct a dynamic threshold criterion. Collect historical performance data of the equipment, such as speed, temperature, pressure, etc., and draw a performance decay curve. Perform probability density estimation on the performance decay curve, such as using the kernel density estimation method, to construct a benchmark threshold matrix. This matrix reflects the normal fluctuation range of different performance indicators in different time periods. For example, taking temperature as an example, the temperature data of the same month in the past three years can be counted, its probability distribution can be calculated, and the 95% confidence interval can be determined as the benchmark temperature threshold range for that month. Introduce a seasonal adjustment factor to dynamically correct the benchmark threshold matrix. For example, considering that the temperature is generally higher in summer, the temperature threshold for summer months can be appropriately increased based on historical data. Construct a personalized threshold adjustment coefficient based on the operating years and load characteristics of the equipment. For example, for equipment that has been running for many years or equipment running at high load, its performance decay rate may be faster, so its threshold can be appropriately lowered. Combine the benchmark threshold matrix, seasonal adjustment factor, and personalized threshold adjustment coefficient to obtain a dynamic threshold criterion.

[0138] Next, perform equipment status prediction. Use sliding window technology to calculate the local decay rate of the performance decay curve. For example, use a sliding window with a width of 7 days to calculate the daily performance decay rate. Input the local decay rate into the exponential smoothing model to eliminate short-term fluctuations. For example, use a single exponential smoothing model to smooth the daily decay rate to obtain a relatively stable decay trend. Predict future decay trends based on time series analysis. For example, use the ARI MA model to predict the smoothed decay rate to obtain the decay trend prediction results for a period of time in the future. Compare the equipment status prediction results with the dynamic threshold criteria. When the prediction result exceeds the dynamic threshold criterion, the fault diagnosis process is triggered. For example, if it is predicted that the temperature in the next week will exceed the set dynamic threshold, the fault diagnosis process is triggered.

[0139] Then, fault diagnosis is performed. Construct an equipment ontology knowledge base containing equipment structure and functional parameters. For example, record the equipment model, component composition, functional parameters and other information. Establish a fault case knowledge base containing fault features and solutions. For example, record the fault phenomenon, cause, solution and other information of historical fault cases. Formalize the expert diagnosis rules to form an expert experience knowledge base. For example, convert the rules summarized by experts such as "If the motor temperature is too high and accompanied by abnormal noise, it may be bearing damage" into logical expressions that can be understood by computers. Based on the equipment ontology knowledge base, fault case knowledge base and expert experience knowledge base, a probabilistic association network of features and fault causes is constructed through association rule mining. For example, by analyzing historical data, the association probability between "temperature is too high" and "bearing damage" can be obtained. Input the equipment state prediction results into the probabilistic association network. For example, input the predicted "temperature is too high" into the probabilistic association network. Design a multi-path reasoning algorithm and use evidence theory to deal with the uncertainty in the diagnosis process. For example, use the Dempster-Shafer evidence theory to fuse evidence from different knowledge bases to obtain the confidence of the fault cause. Consider the knowledge support and rule matching to construct a diagnostic credibility evaluation model. The fault causes are ranked based on the multi-criteria decision-making method to obtain the fault diagnosis results. For example, the fault causes are ranked according to their confidence and severity to determine the most likely fault cause.

[0140] Finally, equipment maintenance recommendations are generated. A hierarchical maintenance solution system including emergency treatment and planned maintenance is established based on the fault diagnosis results. For example, for the fault of "bearing damage", an emergency treatment solution for bearing replacement and a planned maintenance solution for regular lubrication can be formulated. The economic feasibility of different maintenance solutions is evaluated, and the impact of the remaining life of the equipment on the maintenance strategy is considered to build a maintenance condition evaluation model. For example, the cost of bearing replacement and regular lubrication is evaluated, and the most appropriate maintenance solution is selected considering the remaining life of the equipment. The maintenance condition evaluation model is input into the maintenance resource availability and time window constraint analysis module. For example, the availability of maintenance personnel and the available time window are considered. The maintenance priority is determined by combining the fault risk and the importance of the equipment. For example, a higher priority is given to important equipment or high-risk faults. A scheme combination optimization model for multi-equipment collaborative maintenance is established to obtain the combination results of maintenance solutions. For example, the maintenance requirements and resource constraints of multiple equipment are considered to formulate the optimal maintenance plan. The combination results of maintenance solutions are displayed in a hierarchical manner according to the urgency and importance, and a maintenance step flowchart is generated. For example, the steps of replacing bearings are decomposed into specific steps such as disassembly, cleaning, and installation, and displayed in the form of a flowchart. Feedback data on maintenance effects is input into the fault case knowledge base and the expert experience knowledge base for updating, forming closed-loop optimized equipment maintenance recommendations. For example, the equipment operation status after bearing replacement is recorded and the results are fed back into the knowledge base to improve future maintenance strategies.

[0141] The solution of this application can:

[0142] Improve maintenance efficiency: By predicting and diagnosing the status of equipment, potential faults can be discovered in advance, and unexpected equipment downtime can be avoided, thereby improving maintenance efficiency. For example, by predicting the temperature increase trend, bearing wear problems can be discovered in advance, avoiding equipment downtime caused by bearing damage. Reduce maintenance costs: By optimizing maintenance plans, unnecessary maintenance work can be reduced and maintenance costs can be reduced. For example, by analyzing the remaining life of the equipment and the risk of failure, the most appropriate maintenance plan can be selected to avoid excessive or insufficient maintenance. Extend equipment life: By formulating a reasonable maintenance strategy, the service life of the equipment can be extended. For example, through regular lubrication and cleaning, equipment wear can be reduced and the service life of the equipment can be extended.

[0143] Figure 2 Schematic diagram of the structure of the park pipe network monitoring and early warning system based on digital twins according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0144] The first unit is used to construct a domain knowledge graph of the park pipe network, wherein the domain knowledge graph includes entity nodes of pipe network equipment and edges of pipe network relationships; collect real-time monitoring data of the park pipe network, wherein the real-time monitoring data includes pressure data, flow data, and temperature data; obtain historical operation and maintenance data of the park pipe network, wherein the historical operation and maintenance data includes maintenance record data and replacement record data; perform data association mapping between the real-time monitoring data and the historical operation and maintenance data and the domain knowledge graph to form a knowledge-enhanced pipe network data set;

[0145] The second unit is used to segment the knowledge-enhanced pipe network data set in time sequence to construct a sliding time window sequence; input the sliding time window sequence into a bidirectional gated recurrent neural network, the bidirectional gated recurrent neural network includes a forward feature extraction unit and a reverse feature extraction unit; extract the pipe network performance attenuation characteristics through the forward feature extraction unit and the reverse feature extraction unit respectively; deeply integrate the pipe network performance attenuation characteristics with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model;

[0146] The third unit is used to input the real-time collected park pipeline performance parameters into the performance decay analysis model; calculate the performance decay curve and remaining service life of the pipeline equipment based on the performance decay analysis model; identify key performance inflection points according to the performance decay curve, and predict potential failure risks; when the decay rate of the performance decay curve exceeds a preset threshold, perform fault diagnosis analysis in combination with the domain knowledge graph and generate equipment maintenance suggestions; integrate the remaining service life, potential failure risks, and equipment maintenance suggestions into an intelligent evaluation report, and send it to the pipeline operation and maintenance terminal.

[0147] According to a third aspect of the embodiments of the present invention,

[0148] An electronic device is provided, comprising:

[0149] processor;

[0150] a memory for storing processor-executable instructions;

[0151] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0152] A fourth aspect of the embodiments of the present invention is:

[0153] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0154] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A park pipe network monitoring and early warning method based on digital twins is characterized by: include: Construct a knowledge graph for the park pipe network domain, wherein the knowledge graph includes pipe network equipment entity nodes and pipe network relationship edges; Collect real-time monitoring data of the park pipe network, the real-time monitoring data includes pressure data, flow data, and temperature data; obtain historical operation and maintenance data of the park pipe network, the historical operation and maintenance data includes maintenance record data and replacement record data; perform data association mapping between the real-time monitoring data and the historical operation and maintenance data and the domain knowledge graph to form a knowledge-enhanced pipe network data set; The knowledge-enhanced pipe network data set is segmented and processed in time sequence to construct a sliding time window sequence; the sliding time window sequence is input into a bidirectional gated recurrent neural network, and the bidirectional gated recurrent neural network includes a forward feature extraction unit and a reverse feature extraction unit; the pipe network performance attenuation characteristics are extracted by the forward feature extraction unit and the reverse feature extraction unit respectively; the pipe network performance attenuation characteristics are deeply integrated with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model; Inputting the real-time collected park pipe network performance parameters into the performance decay analysis model; Calculate the performance decay curve and remaining service life of the pipe network equipment based on the performance decay analysis model; Identify key performance inflection points based on the performance decay curve and predict potential failure risks; when the decay rate of the performance decay curve exceeds a preset threshold, perform fault diagnosis analysis in combination with the domain knowledge graph and generate equipment maintenance recommendations; The remaining service life, potential failure risks, and equipment maintenance recommendations are integrated into an intelligent assessment report and sent to the pipeline network operation and maintenance terminal.

2. The method according to claim 1, characterized in that The real-time monitoring data, historical operation and maintenance data and the domain knowledge graph are mapped to form a knowledge-enhanced pipe network data set including: The domain knowledge graph includes pipe network equipment entity nodes and pipe network relationship edges, the pipe network entity nodes include pipe section nodes, equipment nodes and measuring point nodes, and the pipe network relationship edges include spatial relationship edges and functional relationship edges; the real-time monitoring data is vectorized according to the time series features to obtain a time series feature vector, and the historical operation and maintenance data is vectorized according to the semantic features to obtain a semantic feature vector; Establishing a first mapping relationship between the time series feature vector and the physical node of the pipe network equipment, and establishing a second mapping relationship between the semantic feature vector and the physical node of the pipe network equipment; wherein the first mapping relationship includes mapping the measuring point information in the time series feature vector to the measuring point node, and the second mapping relationship includes mapping the maintenance record information in the semantic feature vector to the equipment node; Building a data association between the first mapping relationship and the second mapping relationship based on the pipe network relationship edge, wherein the data association includes associating the measurement point node to the pipe segment node through the spatial relationship edge, and associating the equipment node to the pipe segment node through the functional relationship edge; Based on the first mapping relationship, the second mapping relationship and the data association, the real-time monitoring data and the historical operation and maintenance data are semantically associated and integrated in the domain knowledge graph to form a knowledge-enhanced pipeline network dataset with knowledge association.

3. The method according to claim 1, characterized in that Extracting the pipe network performance attenuation characteristics respectively by the forward feature extraction unit and the reverse feature extraction unit; The performance degradation characteristics of the pipeline network are deeply integrated with the expert experience rules in the domain knowledge graph to construct a performance degradation analysis model including: The forward feature extraction unit extracts the forward feature of the pipeline network performance attenuation. The extraction process of the forward feature of the pipeline network performance attenuation includes: extracting the time series change feature based on the multi-layer perceptron network, using the attention mechanism to identify the performance inflection point to obtain the key time node feature, calculating the performance deviation feature based on residual learning, and combining the time series change feature, the key time node feature and the performance deviation feature to form the forward feature of the pipeline network performance attenuation; The reverse feature extraction unit extracts the reverse feature of the pipeline network performance attenuation. The extraction process of the reverse feature of the pipeline network performance attenuation includes: establishing a fault-symptom association model based on a Bayesian network to obtain a propagation path feature, using causal inference to obtain a performance attenuation root cause feature, and obtaining a performance degradation evolution feature based on a time series reverse analysis. The propagation path feature, the performance attenuation root cause feature, and the performance degradation evolution feature are combined to form the reverse feature of the pipeline network performance attenuation; The forward feature of the pipeline network performance attenuation and the reverse feature of the pipeline network performance attenuation are feature-fused to obtain the pipeline network performance attenuation feature, wherein the feature fusion process includes: constructing a feature association matrix to calculate feature relevance, using an autoencoder to reduce the dimension of highly correlated features, and using a graph convolutional network to combine features; Expert experience rules are extracted from the domain knowledge graph, and the expert experience rules are converted into calculation rules based on the rule formal language. The calculation rules are quantified through the rule reasoning mechanism to obtain rule features. The neural symbolic fusion network is used to deeply fuse the pipeline network performance attenuation features with the rule features, and a knowledge-guided attention mechanism is constructed based on the rule features to optimize the pipeline network performance attenuation features to obtain fusion features. A performance attenuation analysis model is constructed based on the fusion features.

4. The method according to claim 1, characterized in that: Calculate the performance decay curve and remaining service life of the pipe network equipment based on the performance decay analysis model; Identify key performance inflection points based on the performance degradation curve and predict potential failure risks including: Inputting the output characteristics of the performance decay analysis model into a performance status quantitative index system, wherein the performance status quantitative index system includes an overall equipment performance evaluation index, a performance loss rate index, a reliability evaluation index, and an efficiency utilization rate index, and constructing a multidimensional performance evaluation matrix based on the performance status quantitative index system; Based on the multidimensional performance evaluation matrix, a local weighted regression algorithm is used to process performance data fluctuations, a continuous performance curve is constructed in combination with piecewise polynomial fitting, and the continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance attenuation curve; Establishing a deep learning prediction network, inputting the performance decay curve into the deep learning prediction network, fusing physical model constraints to extract multi-scale life characteristics, the multi-scale life characteristics including short-term fluctuation characteristics, medium-term trend characteristics and long-term evolution characteristics, and predicting the remaining service life of the equipment based on the multi-scale life characteristics; Calculating the first-order derivative of the performance attenuation curve to obtain the change rate inflection point, analyzing the second-order derivative to obtain the acceleration change point, extracting the local extreme point to obtain the mutation feature point, and combining the change rate inflection point, the acceleration change point and the mutation feature point to form a key performance inflection point feature; A density clustering algorithm is used to classify the key performance inflection point features, and the common features of inflection points of different categories are extracted to establish an inflection point evolution law model, and a risk propagation network is constructed based on the inflection point evolution law model; The risk propagation network is deeply integrated with historical failure data, expert experience rules and environmental influencing factors to construct a multi-source information fusion failure risk prediction model, and the failure probability and risk level are calculated based on the failure risk prediction model.

5. The method according to claim 4, characterized in that Based on the multi-dimensional performance evaluation matrix, a local weighted regression algorithm is used to process performance data fluctuations, and a continuous performance curve is constructed by combining piecewise polynomial fitting. The continuous performance curve is smoothed by a kernel function smoothing technique to obtain a performance attenuation curve, including: Receive performance data in the multidimensional performance evaluation matrix, calculate bandwidth parameters based on the performance data using an adaptive bandwidth selection strategy, dynamically adjust the bandwidth parameters with data density, construct a three-core weight function based on the bandwidth parameters, introduce time series correlation weights to form a dynamic weight calculation model; use a local weighted regression algorithm to perform fluctuation processing on the performance data, input the dynamic weight calculation model into a local weighted least squares objective function, use QR decomposition to solve the weighted normal equations, improve fitting accuracy through an iterative reweighting mechanism, and determine the optimal polynomial order in combination with cross-validation to obtain denoised performance data; Based on the denoised performance data, the data curvature change is calculated, the segmentation point position is optimized by a dynamic programming algorithm, a criterion for determining the number of segmentation points is constructed, and the optimal segmentation point set is determined by local adjustment of the segmentation points; in combination with the piecewise polynomial fitting method, a cubic Hermite interpolation polynomial is constructed based on the optimal segmentation point set, a piecewise function continuity constraint condition and an endpoint derivative matching condition are established, and a monotonicity preservation strategy is introduced to form a piecewise function construction model; the piecewise function construction model is input into the overall smoothness objective function, a piecewise function parameter joint optimization algorithm is designed, a constraint condition processing mechanism is constructed, and a continuous performance curve is obtained through iterative optimization and solution; The continuous performance curve is processed by kernel function smoothing technology, a variable bandwidth Gaussian kernel function is constructed, a bandwidth adaptive adjustment strategy is designed, a directional weight adjustment mechanism is introduced, and a kernel function shape optimization method is established to obtain an adaptive kernel function; a multi-scale decomposition framework is constructed based on the adaptive kernel function, a scale weight adaptive allocation strategy is designed, and a feature transfer mechanism between scales is established to achieve multi-scale result fusion; the boundary effect of the multi-scale result fusion is processed, a boundary correction kernel function and a boundary extrapolation strategy are designed, boundary constraints are constructed, and the final performance attenuation curve is obtained through a gradual transition processing mechanism.

6. The method according to claim 1, characterized in that When the decay rate of the performance decay curve exceeds a preset threshold, a fault diagnosis analysis is performed in combination with the domain knowledge graph to generate equipment maintenance suggestions including: A benchmark threshold matrix is ​​constructed by performing probability density estimation on the performance attenuation curve, a seasonal adjustment factor is introduced to dynamically modify the benchmark threshold matrix, and a personalized threshold adjustment coefficient is constructed according to the equipment operating years and load characteristics to obtain a dynamic threshold criterion; The local decay rate of the performance decay curve is calculated by using a sliding window technique, the local decay rate is input into an exponential smoothing processing model to eliminate short-term fluctuations, and the future decay trend is predicted based on time series analysis to obtain a device state prediction result; the device state prediction result is compared with the dynamic threshold criterion, and when the device state prediction result exceeds the dynamic threshold criterion, a fault diagnosis process is triggered; Construct an equipment ontology knowledge base including equipment structure and function parameters, establish a fault case knowledge base including fault features and solutions, formalize the expert diagnosis rules to form an expert experience knowledge base, and construct a probability association network between features and fault causes through association rule mining based on the equipment ontology knowledge base, the fault case knowledge base and the expert experience knowledge base; The equipment state prediction result is input into the probability association network, a multi-path reasoning algorithm is designed, the uncertainty in the diagnosis process is processed by evidence theory, a diagnosis credibility evaluation model is constructed by comprehensively considering the knowledge support and rule matching, and the fault causes are sorted based on the multi-criteria decision-making method to obtain the fault diagnosis result; Based on the fault diagnosis results, a hierarchical maintenance solution system including emergency treatment and planned maintenance is established, the economic feasibility of different maintenance solutions is evaluated, the impact of the remaining life of the equipment on the maintenance strategy is considered, and a maintenance condition evaluation model is constructed; the maintenance condition evaluation model is input into the maintenance resource availability and time window constraint analysis module, the maintenance priority is determined in combination with the fault risk and the importance of the equipment, and a combination optimization model of the solution for multi-equipment collaborative maintenance is established to obtain the maintenance solution combination result; The maintenance scheme combination results are displayed in grades according to urgency and importance, a maintenance step flow chart is generated, and maintenance effect feedback data is input into the fault case knowledge base and the expert experience knowledge base for updating to form closed-loop optimized equipment maintenance recommendations.

7. A digital twin-based park pipe network monitoring and early warning system, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to construct a knowledge graph of the park pipe network domain, wherein the domain knowledge graph includes pipe network equipment entity nodes and pipe network relationship edges; Collect real-time monitoring data of the park pipe network, the real-time monitoring data includes pressure data, flow data, and temperature data; obtain historical operation and maintenance data of the park pipe network, the historical operation and maintenance data includes maintenance record data and replacement record data; perform data association mapping between the real-time monitoring data and the historical operation and maintenance data and the domain knowledge graph to form a knowledge-enhanced pipe network data set; The second unit is used to segment the knowledge-enhanced pipe network data set in time sequence to construct a sliding time window sequence; input the sliding time window sequence into a bidirectional gated recurrent neural network, the bidirectional gated recurrent neural network includes a forward feature extraction unit and a reverse feature extraction unit; extract the pipe network performance attenuation characteristics through the forward feature extraction unit and the reverse feature extraction unit respectively; deeply integrate the pipe network performance attenuation characteristics with the expert experience rules in the domain knowledge graph to construct a performance attenuation analysis model; The third unit is used to input the real-time collected park pipe network performance parameters into the performance decay analysis model; Calculate the performance decay curve and remaining service life of the pipe network equipment based on the performance decay analysis model; Identify key performance inflection points based on the performance decay curve and predict potential failure risks; when the decay rate of the performance decay curve exceeds a preset threshold, perform fault diagnosis analysis in combination with the domain knowledge graph and generate equipment maintenance recommendations; The remaining service life, potential failure risks, and equipment maintenance recommendations are integrated into an intelligent assessment report and sent to the pipeline network operation and maintenance terminal.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images