Heat supply pipe network operation fault diagnosis method based on digital twinning

Abstract

The invention discloses a heat supply pipe network operation fault diagnosis method based on digital twinning, and the method comprises the following steps: S1, constructing a digital twinning architecture of a heat supply system, and constructing a complete digital twinning system based on a physical entity, a virtual model, a data synchronization layer and a decision control layer; s2, multi-source data fusion and state mapping; s3, a fault prediction and diagnosis mechanism is adopted, an SVM-LSTM joint model is adopted to carry out time sequence anomaly detection and fault prediction, an XGBoost classifier is optimized in combination with an improved sparrow optimization algorithm, and a classification result is output; and S4, performing self-healing control and cooperative regulation and control. According to the invention, a complete digital twin architecture is constructed, through a five-layer system of a physical entity, a sensing and execution network, a data synchronization layer, a virtual model layer and a decision control layer, a full-link closed loop from data acquisition, model driving to intelligent decision is realized, and systematic support is provided for fault diagnosis and self-healing control.

Description

Technical Field

[0001] This application belongs to the field of heating network fault diagnosis technology, specifically involving a method for diagnosing operational faults in heating networks based on digital twins. Background Technology

[0002] As a key component of urban infrastructure, the safe, stable, and efficient operation of heating networks directly impacts people's livelihoods and energy efficiency. With the continuous expansion and increasing complexity of urban heating systems, traditional fault diagnosis methods, relying primarily on manual inspections, threshold alarms, and experience-based judgment, suffer from problems such as delayed response, inaccurate location, high false alarm rates, and a lack of predictive capabilities. These methods are insufficient to meet the real-time, accuracy, and self-healing requirements of modern intelligent heating systems.

[0003] In recent years, digital twin technology has shown great potential in industrial system monitoring and operation and maintenance due to its ability to achieve deep integration and real-time interaction between physical entities and virtual models. However, existing fault diagnosis schemes for heating pipe networks based on digital twins still have the following shortcomings: First, the data fusion mechanism is imperfect, failing to effectively combine mechanistic models with real-time sensor data, resulting in limited accuracy in state estimation; second, fault diagnosis models are mostly based on single algorithms, lacking the ability to identify complex temporal anomalies and multiple types of faults, and lacking early warning and accurate classification capabilities; third, control strategies are mostly statically preset, lacking dynamic collaborative control mechanisms based on multi-objective optimization and virtual simulation, making it difficult to quickly achieve system self-healing and operational optimization after a fault occurs.

[0004] Therefore, there is an urgent need for a method for diagnosing operational faults in heating networks that can deeply integrate multi-source data, achieve accurate prediction and diagnosis, and possess intelligent self-healing and control capabilities, in order to improve system reliability, reduce operation and maintenance costs, and ensure heating quality. Summary of the Invention

[0005] This application provides a method for diagnosing operational faults in heating pipeline networks based on digital twins, aiming to address the problems of imperfect data fusion mechanisms in existing technologies and insufficient ability to identify complex temporal anomalies and multiple types of faults.

[0006] A method for diagnosing operational faults in heating pipe networks based on digital twins includes the following steps:

[0007] S1: Construct a digital twin architecture for the heating system, building a complete digital twin system based on physical entities, virtual models, data synchronization layers, and decision control layers;

[0008] S2: Multi-source data fusion and state mapping, collecting multi-dimensional operating parameters, combining external factors, and performing data cleaning and feature extraction through extended Kalman filtering and principal component analysis to achieve high-frequency synchronization between the physical system and the virtual model;

[0009] S3: Fault prediction and diagnosis mechanism, which uses the SVM-LSTM joint model for time series anomaly detection and fault prediction, and combines the improved sparrow optimization algorithm to optimize the XGBoost classifier and output the classification results;

[0010] S4: Self-healing control and coordinated regulation, triggering preset control strategies based on fault diagnosis results.

[0011] Optionally, constructing a digital twin architecture for the heating system includes:

[0012] The physical entity layer includes all critical equipment and structures in the heating network;

[0013] The sensing and execution network layer is responsible for data acquisition and command execution. The sensor network includes: temperature sensors, pressure sensors, flow sensors, vibration sensors, and acoustic / ultrasonic sensors.

[0014] The execution network includes: frequency converters and electric actuators;

[0015] The data synchronization layer is responsible for the reliable and efficient transmission and fusion of data.

[0016] The virtual model layer is a high-fidelity virtual image composed of multidisciplinary models, including geometric models, physical mechanism models, heat transfer models, equipment performance models, data-driven models, and ISSA-XGBoost fault diagnosis models.

[0017] The decision-making and control layer is responsible for analysis, decision-making, and feedback control.

[0018] Optionally, the decision control layer includes:

[0019] Fault diagnosis and location engine: Receives abnormal signals and diagnostic results output from the virtual model layer, combines them with the knowledge base, performs root cause analysis of the fault, and highlights and locates the fault point in the 3D model;

[0020] The self-healing control strategy library includes: leakage fault strategy; blockage fault strategy; equipment failure strategy.

[0021] Optionally, S2 specifically includes:

[0022] S2.1: Data acquisition and preprocessing, collecting internal operating parameters of the heating network, obtaining external environmental parameters, and load demand data;

[0023] S2.2: Dynamic data fusion and state estimation based on extended Kalman filter, which dynamically fuses the mechanism model with real-time observation data to optimally estimate the data that cannot be directly measured in the system;

[0024] S2.3: Static feature extraction and dimensionality reduction based on principal component analysis: extract essential features from high-dimensional relevant operating parameters for fault diagnosis;

[0025] S2.4: State Mapping and High-Frequency Synchronization: The formation of the fusion feature vector combines the optimal state estimation sequence output by the extended Kalman filter with the principal component scores extracted by principal component analysis to form a fusion feature vector, which drives the virtual model. This fusion feature vector is used as input to drive the high-fidelity mechanism model in the virtual model layer in real time, forming a digital mirror.

[0026] Optionally, S3 includes:

[0027] S3.1: Construct an SVM-LSTM joint prediction model to achieve early anomaly detection and trend warning; the outputs of SVM and LSTM are fused with decision logic to generate early warning signals, and feature data corresponding to the time period of the early warning signals are automatically extracted and sent to the fault sample library.

[0028] S3.2: Construct an ISSA-XGBoost diagnostic model to achieve accurate fault identification and classification, and finally output the globally optimal hyperparameter combination.

[0029] Optionally, S4 is equipped with a hierarchical control strategy library, which is based on a set of "IF-THEN" rules pre-defined by expert knowledge, historical data and simulation, and is divided into three levels according to the control range and objectives.

[0030] Level 1 control: rapid local isolation; Level 2 control: restoration of regional hydraulic balance; Level 3 control: system-level heat source and load scheduling.

[0031] Optionally, S4 includes:

[0032] S4.1: Based on the multi-objective optimization collaborative control mechanism, a multi-objective optimization algorithm is used to select the globally optimal execution scheme. According to the diagnostic results, all feasible control strategy combinations are retrieved from the strategy library. In the digital twin virtual model, the system state after executing each strategy combination is quickly simulated. Based on the pre-simulation results, the objective function value of each strategy is calculated, and the strategy combination that minimizes the objective function is selected as the final execution scheme.

[0033] S4.2: Control command issuance and closed-loop verification, decompose the final execution plan into specific equipment operation command sequences, and perform a final security check in the virtual model;

[0034] Command execution and feedback: Through the data synchronization layer, commands are sent to the PLC, frequency converter and electric actuator in the field in seconds, and the sensor network monitors the system status after control in real time;

[0035] Closed-loop adjustment: The actual system state after control is compared with the predicted state of the digital twin. If the recovery effect does not meet expectations, the control strategy is automatically fine-tuned until the system is fully restored to stability.

[0036] Optionally, the objective function can be optimized as follows:

[0037]

[0038] in:

[0039] Key user room temperature deviation, Changes in total system voltage drop Total system energy consumption : Load shedding cost; α, β, γ, δ: Weighting coefficients, dynamically adjusted according to the severity of the fault and the operating mode.

[0040] Compared with the prior art, this application has at least the following beneficial effects:

[0041] This application constructs a complete digital twin architecture, which realizes a closed-loop system from data acquisition and model-driven to intelligent decision-making through a five-layer system of physical entities, sensing and execution networks, data synchronization layer, virtual model layer and decision control layer, providing systematic support for fault diagnosis and self-healing control.

[0042] This application also combines Extended Kalman Filter (EKF) with Principal Component Analysis (PCA), which not only improves the reliability of state estimation by utilizing mechanistic models, but also improves the efficiency and accuracy of subsequent diagnostic models through feature extraction and dimensionality reduction, enabling the virtual model to maintain second-level synchronization with the physical system and form a high-fidelity digital mirror.

[0043] This application proposes a dual mechanism of SVM-LSTM joint prediction and ISSA-XGBoost diagnosis, which takes into account both instantaneous anomaly detection and trend early warning. By improving the Sparrow Search Algorithm (ISSA) to optimize the hyperparameters of XGBoost, the accuracy and generalization ability of fault classification are significantly improved, enabling early detection and accurate identification of various typical faults such as leakage, blockage, and equipment failure. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating a method for diagnosing operational faults in a heating network based on digital twins, as provided in one embodiment of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0046] This application provides a method for diagnosing operational faults in heating pipe networks based on digital twins, comprising the following steps:

[0047] S1: Construct a digital twin architecture for the heating system, based on physical entities (pipelines, pumping stations, heat exchangers, etc.), virtual models (fluid dynamics models, thermodynamic models), a data synchronization layer (sensor networks, 5G communication), and a decision control layer (fault diagnosis and self-healing control) to build a complete digital twin system;

[0048] Specifically, building a digital twin architecture for a heating system includes:

[0049] The physical entity layer is the actual object that the digital twin serves, including all critical equipment and structures in the heating network.

[0050] Pipeline circuit: Water supply and return pipelines of primary and secondary networks.

[0051] Power equipment: circulating water pump, makeup water pump.

[0052] Heat exchange equipment: Plate heat exchangers in the heat exchange station.

[0053] Regulation and actuation mechanisms: electric regulating valves, shut-off valves.

[0054] User terminals: radiators, underfloor heating coils, and user-controlled intelligent temperature valves within the building.

[0055] Ancillary facilities: expansion tank, sludge separator

[0056] The sensing and execution network layer is responsible for data acquisition and command execution. Sensor networks include:

[0057] Temperature sensors: deployed at key pipeline nodes (heat source outlet, heat exchange station inlet and outlet, typical heat inlet, end user) to monitor water temperature;

[0058] Pressure sensors: Deployed at pump station inlets and outlets, and at high and low points in the pipeline network, to monitor system pressure and diagnose leaks and air locks;

[0059] Flow sensors: deployed in each branch and heat exchange station to monitor the circulating water volume and are used for hydraulic balance analysis and fault diagnosis;

[0060] Vibration sensors: installed in the bearing housings of rotating equipment such as water pumps and motors, used for early warning of mechanical failures;

[0061] Acoustic / ultrasonic sensors: used to detect specific frequency sound wave signals generated by leaks inside pipes;

[0062] The actuator network includes: a frequency converter: controlling the pump speed to achieve variable flow rate regulation; and an electric actuator: driving the valve opening to regulate flow and pressure.

[0063] The data synchronization layer is responsible for the reliable and efficient transmission and fusion of data.

[0064] The virtual model layer consists of a high-fidelity virtual image composed of multidisciplinary models, including:

[0065] Geometric model: A three-dimensional visualization model of the pipeline network built based on GIS (Geographic Information System) and BIM (Building Information Modeling), which accurately reflects the spatial layout, diameter, burial depth and relationship with the surrounding environment of the pipeline;

[0066] Physical mechanism model: Fluid dynamics model: Based on the pipeline network topology and fluid dynamics equations, real-time simulation of flow distribution, pressure distribution and hydraulic conditions within the system;

[0067] Heat transfer model: Based on the laws of thermodynamics, it simulates the transport and loss of heat in pipes and the transfer process in heat exchangers, and predicts the temperature field changes of the entire network;

[0068] Equipment performance model: embedding performance curves and efficiency graphs of equipment such as water pumps and heat exchangers to simulate their operating status under different working conditions;

[0069] Data-driven model: SVM-LSTM joint prediction model: used to predict trends and detect early anomalies in key system parameters (such as terminal pressure and heat exchange efficiency) based on historical and real-time data;

[0070] ISSA-XGBoost Fault Diagnosis Model: As the core classifier, it receives outputs from the mechanism model and the prediction model to accurately identify and classify specific fault types;

[0071] The decision-making and control layer is responsible for analysis, decision-making, and feedback control, and includes:

[0072] Fault diagnosis and location engine: Receives abnormal signals and diagnostic results from the virtual model layer, combines them with the knowledge base (stores historical fault cases and maintenance records), performs root cause analysis of faults, and highlights and locates fault points in the 3D model;

[0073] Self-healing control strategy library: Leakage fault strategy: Automatically shut off the valves at both ends of the leaking pipe section, start the backup circuit, and adjust the pump station output;

[0074] Blockage fault strategy: Adjust the opening of relevant valves, perform hydraulic rebalancing, ensure basic heating at the end, and issue a blockage clearing and maintenance alarm;

[0075] Equipment failure strategy: Activate backup equipment and generate an equipment maintenance work order;

[0076] Command issuance and closed-loop verification: Control strategies (such as valve opening commands and pump frequency setpoints) are issued to physical actuators through the data synchronization layer. The control system verifies the control effect in real time in the virtual model based on the new data fed back by the sensors, forming a closed-loop control of "perception-diagnosis-decision-execution-verification".

[0077] S2: Multi-source data fusion and state mapping, collecting multi-dimensional operating parameters such as temperature, pressure, flow rate, and vibration, and combining external factors such as ambient temperature and load changes, using extended Kalman filter (EKF) and principal component analysis (PCA) to perform data cleaning and feature extraction, achieving high-frequency synchronization between the physical system and the virtual model;

[0078] S2 specifically includes:

[0079] S2.1: Data Acquisition and Preprocessing, Data Source:

[0080] Internal operating parameters: Thermal parameters: Supply / return water temperature ( , ), pressure (P), flow rate (F); hydraulic parameters: pump inlet and outlet pressure difference, valve opening; equipment status parameters: pump / motor vibration (V), current (I), speed (RPM);

[0081] External environmental parameters: meteorological data: ambient temperature ( ), wind speed, solar radiation intensity (used to correct the heat loss model of the pipe network); load demand data: predicted heat load obtained from user-side smart meters or building energy consumption models ( );

[0082] Data preprocessing and cleaning: The sliding window method and interquartile range (IQR) rule are used to automatically identify and remove pulse-type abnormal values ​​caused by instantaneous sensor failures;

[0083] Alignment: Apply a uniform timestamp to all data streams to resolve timing misalignment issues caused by different sensor sampling frequencies and communication delays;

[0084] Normalization: Normalize parameters with different dimensions (such as temperature and pressure) to the [0,1] interval to eliminate the influence of magnitude and prepare for subsequent fusion analysis;

[0085] S2.2: Dynamic data fusion and state estimation based on extended Kalman filter (EKF) dynamically fuses the mechanism model with real-time observation data to optimally estimate the key state variables in the system that cannot be directly measured or have high measurement noise.

[0086] It includes the following steps: State variable definition (X):

[0087]

[0088] in, (Effective heat transfer coefficient of pipeline) is a virtual health status variable that cannot be directly measured but reflects the scaling and insulation performance of pipeline.

[0089] State-space model establishment:

[0090] State equations (based on mechanism):

[0091] Where f() is the simplified discrete equation for heat transfer and fluid dynamics in the pipe network. For control inputs (such as water pump frequency). This is process noise.

[0092] Observation equations (based on sensors):

[0093] Where h() is the observation model, representing the relationship between state variables and sensor readings (such as temperature and pressure). To observe noise;

[0094] EKF recursive fusion process:

[0095] Prediction: Based on the state estimate and mechanistic model f() from the previous time step, predict the system state at the current time step. And error covariance;

[0096] Update: Predicted values Compared with the actual sensor observations at the current moment Compare;

[0097] Calculate the Kalman gain: Based on the uncertainty of the prediction and the observation noise, decide whether to trust the model prediction or the sensor data more.

[0098] Fusion correction: The model predictions and actual observations are weighted and fused using Kalman gain to obtain the optimal state estimate at the current time step. ;

[0099] Output: After EKF fusion, we obtain not the original noisy data, but a smoother and more accurate system state sequence constrained and calibrated by the mechanistic model, especially gaining insights into the pipeline's health status. Quantitative assessment;

[0100] S2.3: Static feature extraction and dimensionality reduction based on principal component analysis (PCA) extracts a few irrelevant essential features that can characterize the main operating modes of the system from high-dimensional relevant operating parameters for fault diagnosis.

[0101] It includes the following steps:

[0102] Construct a feature matrix by forming a data matrix from all running parameters (including data processed by EKF) within a time window;

[0103] Principal components are calculated by linear transformation, which transforms the original correlated variables into a series of new, orthogonal variables (principal components, PCs). The first principal component (PC1) carries the largest variance information in the original data, which usually corresponds to the main load change pattern of the system; the second principal component (PC2) carries the second largest variance, which may correspond to the hydraulic imbalance pattern; subsequent components may contain noise or specific fault information.

[0104] Feature selection: The top k principal components are selected as the baseline feature space representing the healthy operating state of the system, and the residual (the subspace composed of the remaining principal components) is used as a sensitive indicator for monitoring minor anomalies and early failures.

[0105] S2.4: State Mapping and High-Frequency Synchronization:

[0106] The formation of the fusion feature vector involves combining the optimal state estimation sequence output by EKF with the principal component scores (PCScores) extracted by PCA to form a comprehensive, low-dimensional, and highly information-condensed fusion feature vector.

[0107] Drive the virtual model by taking this fused feature vector as input to drive the high-fidelity mechanism model in the virtual model layer in real time.

[0108] Achieving high-frequency synchronization: The entire process of "data acquisition → EKF fusion → PCA feature extraction → model-driven" is completed within a fixed short period (e.g., 10 seconds);

[0109] This allows the simulated state in the virtual model (such as the pressure distribution and temperature field of the entire network) to be synchronized with the actual state of the physical entity at the second or even sub-second level, forming a dynamic and constantly self-correcting digital mirror.

[0110] S3: Fault prediction and diagnosis mechanism. It adopts the SVM-LSTM joint model for time-series anomaly detection and fault prediction, and combines the improved Sparrow Optimization Algorithm (ISSA) to optimize the XGBoost classifier and output classification results to achieve accurate identification and classification of typical faults such as leakage, blockage, and equipment failure.

[0111] S3.1: Construct an SVM-LSTM joint prediction model to achieve early anomaly detection and trend warning;

[0112] Input feature vector: The fused feature vector processed by EKF and PCA includes: [temperature gradient] Standard deviation of pressure fluctuation Flow deviation Vibration energy value Pipeline health index Ambient temperature ,...];

[0113] The SVM transient anomaly detector is responsible for point anomalies. It uses a radial basis function kernel to map nonlinear problems to a high-dimensional space for linear segmentation. It uses a large amount of normal operating data for unsupervised learning (One-Class SVM) or supervised learning on normal / anomaly labeled data to establish a decision boundary of a "normal operating region". If a real-time data point falls outside the decision boundary, a transient anomaly flag is immediately triggered. For example, a sudden drop in pressure is a strong feature of pipeline leakage.

[0114] LSTM trend predictor, responsible for line anomalies: takes a sequence of feature vectors from a sliding time window (e.g., the past 30 minutes) as input;

[0115] Using multi-layer LSTM units, its forget gate, input gate, and output gate mechanisms can effectively learn long-term dependencies;

[0116] Predict key indicators such as supply and return water pressure difference and heat exchange efficiency for a future period of time (e.g., the next 10 minutes);

[0117] Trend Judgment: If the predicted value continues to deviate from the normal range (e.g., the predicted differential pressure will continue to increase), a trend anomaly indicator is triggered. For example, a continuously slow decline in predicted heat exchange efficiency may indicate fouling and blockage of the heat exchanger;

[0118] Joint decision-making and early warning: The outputs of SVM and LSTM (instantaneous anomaly indicators, trend anomaly indicators, and predicted values) are fed into a fusion decision logic to generate early warning signals and automatically extract feature data for that time period, which is then sent to the fault sample library to prepare for subsequent diagnosis.

[0119] S3.2: Construct an ISSA-XGBoost diagnostic model to achieve accurate fault identification and classification, which includes:

[0120] Construct a fault sample library: It includes historical fault data, simulation-generated fault data, and early abnormal samples captured by the SVM-LSTM module. Each sample is labeled with a clear fault type label, such as: 1: Pipeline leakage, 2: Branch blockage, 3: Pump cavitation, 4: Heat exchanger scaling, 0: Normal.

[0121] The sparrow search algorithm is improved by optimizing XGBoost. A control parameter that decreases nonlinearly with the number of iterations is introduced to balance global exploration and local exploitation capabilities. Lévy flight is incorporated into the sparrow's position update to generate a random step size, which effectively escapes local optima. Cauchy mutation is performed on the current optimal solution to increase population diversity and prevent premature convergence.

[0122] Its optimization process:

[0123] Initialization: The position of each sparrow in the sparrow population represents a set of XGBoost hyperparameter combinations;

[0124] Fitness evaluation: The diagnostic accuracy on the validation set is used as the fitness value when the XGBoost model is trained with each hyperparameter combination.

[0125] Location Update: Sparrows (discoverers, followers, and watchers) update their locations based on improved ISSA rules (i.e., searching for better hyperparameters);

[0126] Iterative output: Repeat the process until the globally optimal combination of hyperparameters is output;

[0127] XGBoost Fault Classification: Using the optimal hyperparameters found by ISSA, the XGBoost model is trained on the complete training set. XGBoost performs ensemble learning by building multiple decision trees, which can automatically evaluate the importance of features (such as finding that the combination of "pressure drop" and "abnormal flow" is most important for diagnosing "leak"). For a new data sample, the trained ISSA-XGBoost model not only outputs the fault type, but also gives the confidence level in the form of probability.

[0128] S4: Self-healing control and coordinated regulation. Based on the fault diagnosis results, it triggers preset control strategies (such as valve adjustment, pump speed control, and load switching) to achieve rapid system recovery and minimize heating interruption.

[0129] It has a hierarchical control strategy library, based on expert knowledge, historical data and simulation simulation pre-set "IF-THEN" rule set, and is divided into three levels according to the control range and objectives;

[0130] Level 1 control (local rapid isolation) prevents the fault from escalating and ensures the safety of personnel and equipment;

[0131] Secondary control (regional hydraulic balance restoration) quickly restores the hydraulic balance of the remaining pipe network after fault isolation, ensuring normal heating in unaffected areas;

[0132] Three-level control (system-level heat source and load dispatching) is used to deal with major faults, conduct energy dispatching from the perspective of the entire network, and prioritize the protection of critical loads.

[0133] S4.1: A collaborative control mechanism based on multi-objective optimization is used to select the globally optimal execution scheme using a multi-objective optimization algorithm. Its objective function is:

[0134]

[0135] in:

[0136] For key users with room temperature deviations, priority should be given to ensuring heating quality in key areas. Changes in total system pressure drop, maintaining pipeline stability, and avoiding pressure fluctuations. The total system energy consumption should be considered in conjunction with economic efficiency during the recovery process. : Load shedding cost, minimizing the impact of interruption; α, β, γ, δ: weighting coefficients, which can be dynamically adjusted according to the severity of the fault and the operating mode (e.g., the weight of α increases under extreme weather conditions).

[0137] Its collaborative decision-making process is as follows: Based on the diagnostic results, all feasible control strategy combinations are retrieved from the strategy library. In the digital twin virtual model, the system state (pressure, flow, temperature distribution) after executing each strategy combination is quickly simulated. Based on the simulation results, the objective function value J of each strategy is calculated, and the strategy combination that minimizes the objective function J is selected as the final execution plan.

[0138] S4.2: Control command issuance and closed-loop verification, command generation and security verification: Decompose the final execution plan into specific equipment operation command sequences, and perform a final security verification in the virtual model to prevent misoperation;

[0139] Command execution and feedback: Through the data synchronization layer, commands are sent to the PLC (Programmable Logic Controller), frequency converter and electric actuator in the field in seconds, and the sensor network monitors the system status after control in real time (such as whether the pressure is stable and whether the temperature is rising).

[0140] Closed-loop adjustment: The actual system state after control is compared with the predicted state of the digital twin. If the recovery effect does not meet expectations (such as the pressure at a certain point is still too low), the control strategy is automatically fine-tuned to form a complete closed loop of "perception-diagnosis-decision-execution-verification-optimization" until the system is fully restored to stability.

[0141] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for diagnosing operational faults in a heating network based on digital twins, characterized in that, Includes the following steps: S1: Construct a digital twin architecture for the heating system, building a complete digital twin system based on physical entities, virtual models, data synchronization layers, and decision control layers; S2: Multi-source data fusion and state mapping, collecting multi-dimensional operating parameters, combining external factors, and performing data cleaning and feature extraction through extended Kalman filtering and principal component analysis to achieve high-frequency synchronization between the physical system and the virtual model; S3: Fault prediction and diagnosis mechanism, which uses the SVM-LSTM joint model for time series anomaly detection and fault prediction, and combines the improved sparrow optimization algorithm to optimize the XGBoost classifier and output the classification results; S4: Self-healing control and coordinated regulation, triggering preset control strategies based on fault diagnosis results.

2. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 1, characterized in that, Building a digital twin architecture for a heating system includes: The physical entity layer includes all critical equipment and structures in the heating network; The sensing and execution network layer is responsible for data acquisition and command execution. The sensor network includes: temperature sensors, pressure sensors, flow sensors, vibration sensors, and acoustic / ultrasonic sensors. The execution network includes: frequency converters and electric actuators; The data synchronization layer is responsible for the reliable and efficient transmission and fusion of data. The virtual model layer is a high-fidelity virtual image composed of multidisciplinary models, including geometric models, physical mechanism models, heat transfer models, equipment performance models, data-driven models, and ISSA-XGBoost fault diagnosis models. The decision-making and control layer is responsible for analysis, decision-making, and feedback control.

3. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 2, characterized in that, The decision-making and control layer includes: Fault diagnosis and location engine: Receives abnormal signals and diagnostic results output from the virtual model layer, combines them with the knowledge base, performs root cause analysis of the fault, and highlights and locates the fault point in the 3D model; The self-healing control strategy library includes: leakage fault strategy; blockage fault strategy; equipment failure strategy.

4. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 1, characterized in that, S2 specifically includes: S2.1: Data acquisition and preprocessing, collecting internal operating parameters of the heating network, obtaining external environmental parameters, and load demand data; S2.2: Dynamic data fusion and state estimation based on extended Kalman filter, which dynamically fuses the mechanism model with real-time observation data to optimally estimate the data that cannot be directly measured in the system; S2.3: Static feature extraction and dimensionality reduction based on principal component analysis: extract essential features from high-dimensional relevant operating parameters for fault diagnosis; S2.4: State Mapping and High-Frequency Synchronization: The formation of the fusion feature vector combines the optimal state estimation sequence output by the extended Kalman filter with the principal component scores extracted by principal component analysis to form a fusion feature vector, which drives the virtual model. This fusion feature vector is used as input to drive the high-fidelity mechanism model in the virtual model layer in real time, forming a digital mirror.

5. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 1, characterized in that, S3 include: S3.1: Construct an SVM-LSTM joint prediction model to achieve early anomaly detection and trend warning; The outputs of SVM and LSTM are fused into a decision logic to generate early warning signals, and feature data corresponding to the time period of the early warning signals are automatically extracted and sent to the fault sample library. S3.2: Construct an ISSA-XGBoost diagnostic model to achieve accurate fault identification and classification, and finally output the globally optimal hyperparameter combination.

6. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 1, characterized in that, S4 has a hierarchical control strategy library, which is based on expert knowledge, historical data and simulation simulation of a pre-defined set of "IF-THEN" rules, and is divided into three levels according to the control range and objectives. Level 1 control: rapid local isolation; Level 2 control: restoration of regional hydraulic balance; Level 3 control: system-level heat source and load scheduling.

7. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 1, characterized in that, S4 include: S4.1: Based on the multi-objective optimization collaborative control mechanism, a multi-objective optimization algorithm is used to select the globally optimal execution scheme. According to the diagnostic results, all feasible control strategy combinations are retrieved from the strategy library. In the digital twin virtual model, the system state after executing each strategy combination is quickly simulated. Based on the pre-simulation results, the objective function value of each strategy is calculated, and the strategy combination that minimizes the objective function is selected as the final execution scheme. S4.2: Control command issuance and closed-loop verification, decompose the final execution plan into specific equipment operation command sequences, and perform a final security check in the virtual model; Command execution and feedback: Through the data synchronization layer, commands are sent to the PLC, frequency converter and electric actuator in the field in seconds, and the sensor network monitors the system status after control in real time; Closed-loop adjustment: The actual system state after control is compared with the predicted state of the digital twin. If the recovery effect does not meet expectations, the control strategy is automatically fine-tuned until the system is fully restored to stability.

8. The method for diagnosing operational faults in heating pipe networks based on digital twins according to claim 7, characterized in that, The objective function to be optimized is: ； in: Key user room temperature deviation, Changes in total system voltage drop Total system energy consumption : Load shedding cost; α, β, γ, δ: Weighting coefficients, dynamically adjusted according to the severity of the fault and the operating mode.

Images