Generator digital twinning diagnosis method based on temperature field online simulation

Abstract

The invention relates to the technical field of equipment intelligent diagnosis, in particular to a generator digital twinning diagnosis method based on temperature field online simulation, which comprises the following steps: respectively constructing a three-dimensional virtual model, a temperature field simulation model and an intelligent agent diagnosis model; the digital twin is in data connection with the physical entity; data alignment is achieved through a unified space-time coordinate system, a mapping relation is established between a vertex of the three-dimensional virtual model and a grid node in the temperature field simulation model, and a fault probability matrix output by the intelligent agent diagnosis model is matched to a three-dimensional grid of the three-dimensional virtual model through spatial interpolation; according to a fault probability output by the intelligent agent diagnosis model, correcting a boundary condition of the temperature field simulation model, triggering local grid adaptive encryption of the temperature field simulation model, and marking a fault position in real time through a three-dimensional virtual model; and the simulation result of the temperature field simulation model verifies the fault probability output by the intelligent agent diagnosis model. Through the method, the fault identification accuracy can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent equipment diagnosis, and in particular to a generator digital twin diagnosis method based on online temperature field simulation. Background Art

[0002] Multi-physics online simulation is a crucial foundation for the development of digital twins. Current multi-physics simulation, based on the fusion of artificial intelligence models, enables high-efficiency, high-precision, and highly generalizable online real-time evaluation and prediction of three-dimensional physical fields. By building a three-dimensional digital twin of a device, integrated with physical state information, and based on a three-dimensional virtual volume and the dynamic properties of multiple physical fields, this technology enables comprehensive perception of device status, intelligent assessment, diagnosis, and early warning.

[0003] For example, in the prior art, a Chinese invention patent document with publication number CN118746134A and publication date October 8, 2024, discloses the following technical solution: a gas equipment temperature monitoring method and system based on temperature field and digital twin, wherein the method comprises the following steps: constructing a three-dimensional simulation model and a temperature field simulation model of the gas equipment; performing data dimension reduction on a number of historical gas equipment monitoring data according to preset gas equipment monitoring parameters; performing temperature field simulation simulation based on a number of reduced-dimensional historical gas equipment monitoring data using a temperature field simulation model; using a deep learning algorithm to construct a corresponding gas equipment temperature anomaly detection model; constructing a gas equipment temperature monitoring digital twin model; based on a set of key monitoring indicators, collecting corresponding real-time gas equipment monitoring data, and using the gas equipment temperature monitoring digital twin model to perform gas equipment temperature monitoring.

[0004] The purpose of the simulation in the above technical solution is to provide richer samples for the equipment temperature anomaly detection model, and the interaction between the models is simple. The accuracy, interpretability and interactivity of the anomaly detection and recognition results need to be improved. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a generator digital twin diagnosis method based on online temperature field simulation, which can effectively improve the accuracy of fault identification and enhance interpretability and interactivity.

[0006] The present invention is achieved by adopting the following technical solutions: A generator digital twin diagnosis method based on online temperature field simulation includes the following steps: Constructing a digital twin, which includes a three-dimensional virtual model, a temperature field simulation model, and an intelligent agent diagnosis model; Real-time access to generator operating data through sensors and IoT technology enables operational data connection between the digital twin and the physical entity; The data intercommunication between the 3D virtual model, the temperature field simulation model and the intelligent agent diagnosis model is achieved through a unified spatiotemporal benchmark, and the dynamic mutual verification between the intelligent agent diagnosis model and the temperature field simulation model is achieved through a two-way feedback mechanism. Specifically, data interoperability involves aligning data through a unified spatiotemporal coordinate system, establishing a mapping relationship between the vertices of the 3D virtual model and the grid nodes in the temperature field simulation model, and matching the fault probability matrix output by the agent-based diagnostic model to the 3D grid of the 3D virtual model through spatial interpolation. The two-way feedback mechanism specifically includes: establishing dynamic data interaction verification middleware, correcting the boundary conditions of the temperature field simulation model according to the fault probability output by the intelligent agent diagnosis model and triggering the local grid adaptive encryption of the temperature field simulation model, and marking the fault location in real time through the three-dimensional virtual model; the simulation results of the temperature field simulation model verify the fault probability output by the intelligent agent diagnosis model.

[0007] The following steps are involved in achieving operational data connection between the digital twin and the physical entity: Step S 11 Establish a multi-level data transmission architecture that includes sensors, intelligent edge nodes, high-speed transmission networks, and intelligent analysis platforms; Step S 12 Data collection through sensors; Step S 13 Introducing lossless data compression algorithms, leveraging high-speed transmission networks to achieve real-time data transmission, and implementing data preprocessing and multi-dimensional data quality assessment at intelligent edge nodes; Step S 14 . Implement multi-source sensor data fusion on the intelligent analysis platform and transmit the fused data to the digital twin in real time.

[0008] The data preprocessing includes outlier removal and missing data completion.

[0009] The missing data completion specifically refers to: reconstructing using the spatiotemporal Kriging interpolation method and using the variation function model to determine the weight λi.

[0010] The multi-dimensional data quality assessment is used to evaluate the quality of pre-processed data from the perspectives of data completeness, accuracy, consistency, timeliness, uniqueness and credibility.

[0011] The multi-source sensor data fusion specifically includes: realizing data layer fusion through adaptive weighted averaging and adaptive Kalman filtering; using principal component analysis to extract key features and combining convolutional neural network and long short-term memory network hybrid model to perform deep learning analysis on the extracted key features to realize feature layer fusion; and realizing decision layer fusion by combining evidence theory with fault tree analysis.

[0012] The method for constructing a three-dimensional virtual model includes: establishing a hierarchical three-dimensional geometric model from the macroscopic whole to the key features, and performing lightweight rendering on the three-dimensional geometric model; preprocessing the time series data of multiple sensors through time alignment and spatial normalization, and deep mapping and fusing material parameters, operating conditions, and equipment status data with the lightweight rendered three-dimensional geometric model to obtain a three-dimensional virtual model.

[0013] The construction method of the temperature field simulation model includes: establishing a dynamic temperature field simulation model based on a neural network algorithm, and using Kalman filtering to fuse real-time sensor data to correct the simulation results; completing the temperature field cloud map reconstruction through vertex shading and texture baking methods.

[0014] The intelligent agent diagnosis model includes a spatiotemporal feature extraction model based on a deep learning algorithm and an operating status anomaly recognition model; the spatiotemporal feature extraction model is used to construct a multidimensional feature vector for the time series data of multiple sensors and the simulation data of the temperature field simulation model, and extract the spatial and temporal features of the multidimensional feature vector; the operating status anomaly recognition model adopts a generative adversarial network architecture to realize status anomaly recognition by calculating the anomaly score or probability based on the features extracted by the spatiotemporal feature extraction model.

[0015] The intelligent agent diagnosis model also includes a time series data analysis and trend prediction model, including feature attention for completing the weight allocation of each sensor and time attention for completing the focus on key time steps; the time series data analysis and trend prediction model is used to realize state prediction based on multivariate historical sequence data; wherein, the multivariate historical sequence data includes the time series data of the sensor and the historical result data output by the intelligent agent diagnosis model.

[0016] The intelligent agent diagnosis model also includes a fault tracing model based on a dynamic Bayesian network; the fault tracing model constructs the equipment fault propagation path through hierarchical time series nodes and uses the expectation maximization algorithm to learn state transition probabilities from historical data; during online monitoring, it fuses multi-source sensor data based on the particle filter algorithm, iteratively updates the posterior probability of each node, and locates the maximum likelihood fault chain.

[0017] The fault propagation path is displayed by combining the probabilistic root cause analysis provided by the fault tracing model, the comparative heat map generated by the operating status anomaly identification model, and the three-dimensional visual interactive interface.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention constructs a complete closed-loop system from data acquisition to simulation calculation to diagnostic feedback through the deep collaboration of three-dimensional virtual models, temperature field simulation models and intelligent diagnostic models. The temperature field simulation model can dynamically adjust the local grid accuracy according to the diagnostic results, and the three-dimensional virtual model marks the fault location in real time, forming an intelligent cycle of "monitoring-analysis-correction-verification", improving the accuracy of fault identification, and is superior to the traditional single-model diagnostic method. Among them, monitoring refers to the operation data connection, analysis refers to the simulation analysis of the temperature field simulation model and the intelligent diagnosis of the intelligent diagnostic model, correction refers to the correction of the temperature field simulation model by the intelligent diagnostic conclusion of the intelligent diagnostic model, and verification refers to the mutual verification between the simulation analysis results of the temperature field simulation model and the intelligent diagnostic conclusion of the intelligent diagnostic model.

[0019] 2. This invention utilizes lightweight rendering when constructing 3D virtual models, reducing loading times and improving interactive fluency and development efficiency. The temperature field simulation model utilizes an innovative combination of parametric geometric modeling and neural network accelerated simulation, significantly improving computational efficiency while ensuring accuracy.

[0020] 3. In the present invention, the fault tracing model based on the dynamic Bayesian network can provide probabilistic root cause analysis, and the operating status abnormality identification model can generate a comparative heat map. Combined with the three-dimensional visual interactive interface, it can intuitively display the fault propagation path, shortening the decision-making time of operation and maintenance personnel and greatly improving the credibility and operability of the diagnosis results.

[0021] 4. In the present invention, through multi-level data transmission architecture, data quality processing and multi-source sensor data fusion, it is possible to ensure the establishment of high-precision, low-latency data acquisition and transmission channels, ensure data synchronization between digital twins and physical entities, and improve data reliability and enhance state perception capabilities.

[0022] 5. The three-dimensional virtual model established by this method takes into account both lightweight structure and refined properties, and can ensure the smoothness of interactive operations and the integrity of information display.

[0023] 6. The temperature field simulation model established by this method can meet the requirements of calculation accuracy and real-time performance, and can quickly respond to changes in operating conditions.

[0024] 7. Through this intelligent diagnostic model, it can complete operation status assessment, risk identification and warning, trend prediction and fault cause tracing, and has high accuracy and generalization capabilities, and can cope with complex operating environments and diverse fault types. DETAILED DESCRIPTION

[0025] Example 1 As a basic embodiment of the present invention, the present invention includes a generator digital twin diagnosis method based on online temperature field simulation, comprising the following steps: A digital twin is constructed, consisting of a 3D virtual model, a temperature field simulation model, and an agent-based diagnostic model. The high-fidelity 3D virtual model forms the spatial basis of the digital twin. Using 3D modeling and visual rendering techniques, the generator's geometry, material properties, and component spatial relationships are constructed, ensuring that the digital twin aligns with the physical form of the physical entity, providing a spatial framework for subsequent fault location and visual interaction. The temperature field simulation model is used to online simulate the temperature distribution and trends within the generator. Based on heat transfer principles and finite element fast simulation methods, combined with real-time operating data, virtual sensors are constructed to dynamically calculate the internal temperature field of the generator, providing a basis for thermal status assessment and fault diagnosis. The agent-based diagnostic model analyzes the generator's operating status and conducts fault diagnosis based on both physical sensor data and simulated virtual sensor data. Using machine learning or deep learning algorithms, combined with historical and real-time data, the agent-based diagnostic model is trained and constructed to intelligently assess the generator's health, proactively identify potential risks, and provide maintenance recommendations.

[0026] Through sensors and IoT technology, real-time access to generator operating data connects the digital twin to the physical entity. This generator operating data is used for data association display in the 3D virtual model, as boundary conditions and correction parameters for the temperature field simulation model, and as input parameters for the intelligent diagnostic model.

[0027] The data intercommunication of the three-dimensional virtual model, the temperature field simulation model and the intelligent agent diagnosis model is achieved through a unified time-space benchmark, and the dynamic mutual verification of the intelligent agent diagnosis model and the temperature field simulation model is achieved through a two-way feedback mechanism. Among them, data intercommunication specifically refers to: achieving data alignment through a unified time-space coordinate system, establishing a mapping relationship between the vertices of the three-dimensional virtual model and the grid nodes in the temperature field simulation model, and matching the fault probability matrix output by the intelligent agent diagnosis model to the three-dimensional grid of the three-dimensional virtual model through spatial interpolation. The two-way feedback mechanism specifically includes: establishing a dynamic data interaction verification middleware, correcting the boundary conditions of the temperature field simulation model according to the fault probability output by the intelligent agent diagnosis model and triggering the local grid adaptive encryption of the temperature field simulation model, and marking the fault location in real time through the three-dimensional virtual model; the simulation results of the temperature field simulation model verify the fault probability output by the intelligent agent diagnosis model.

[0028] Example 2 As a preferred embodiment of the present invention, the present invention includes a generator digital twin diagnosis method based on online temperature field simulation, comprising the following steps: A digital twin is constructed, comprising a 3D virtual model, a temperature field simulation model, and an intelligent diagnostic model. The 3D virtual model is used to restore the generator's geometry, material properties, and component spatial relationships, providing a spatial framework for fault location. The temperature field simulation model is used to simulate internal temperature distribution and changes online. The intelligent diagnostic model combines real-time and historical data to assess the generator's health and predict risks.

[0029] Through real-time access to generator operation data through sensors and IoT technology, the operation data connection between the digital twin and the physical entity is realized. Specifically, the following steps are included: Step S 11 . Establish a multi-level data transmission architecture that includes sensors, intelligent edge nodes, high-speed transmission networks, and intelligent analysis platforms.

[0030] Step S 12 . Use sensors to collect real-time operating data of the physical entity, namely the generator, such as temperature, current, etc.

[0031] Step S 13 A lossless data compression algorithm is introduced, high-speed transmission networks are utilized to achieve real-time data transmission, and data preprocessing and multi-dimensional data quality assessment are implemented on intelligent edge nodes. This data preprocessing includes outlier removal and missing data completion. The multi-dimensional data quality assessment evaluates the quality of the preprocessed data based on its completeness, accuracy, consistency, timeliness, uniqueness, and credibility.

[0032] Step S 14 Multi-source sensor data fusion is implemented on the intelligent analysis platform and transmitted to the digital twin over a high-speed transmission network, ensuring data synchronization between the digital twin and the physical entity. Specifically, this multi-source sensor data fusion includes: data-level fusion through adaptive weighted averaging and adaptive Kalman filtering; feature-level fusion through principal component analysis to extract key features and deep learning analysis using a hybrid model of convolutional neural networks and long-short-term memory networks; and decision-level fusion through a combination of evidence theory and fault tree analysis.

[0033] The data intercommunication of the three-dimensional virtual model, the temperature field simulation model and the intelligent agent diagnosis model is achieved through a unified time-space benchmark, and the dynamic mutual verification of the intelligent agent diagnosis model and the temperature field simulation model is achieved through a two-way feedback mechanism. Among them, data intercommunication specifically refers to: achieving data alignment through a unified time-space coordinate system, establishing a mapping relationship between the vertices of the three-dimensional virtual model and the grid nodes in the temperature field simulation model, and matching the fault probability matrix output by the intelligent agent diagnosis model to the three-dimensional grid of the three-dimensional virtual model through spatial interpolation. The two-way feedback mechanism specifically includes: establishing a dynamic data interaction verification middleware, correcting the boundary conditions of the temperature field simulation model according to the fault probability output by the intelligent agent diagnosis model and triggering the local grid adaptive encryption of the temperature field simulation model, and marking the fault location in real time through the three-dimensional virtual model; the simulation results of the temperature field simulation model verify the fault probability output by the intelligent agent diagnosis model.

[0034] Example 3 As another preferred embodiment of the present invention, the present invention includes a generator digital twin diagnosis method based on online temperature field simulation, comprising the following steps: A digital twin is constructed, which includes a three-dimensional virtual model, a temperature field simulation model and an intelligent body diagnosis model.

[0035] Among them, the method for constructing the three-dimensional virtual model includes: establishing a hierarchical three-dimensional geometric model from the macroscopic whole to the key features, and performing lightweight rendering on the three-dimensional geometric model; preprocessing the time series data of multiple sensors through time alignment and spatial normalization, and deep mapping and fusing the material parameters, operating conditions, and equipment status data with the lightweight rendered three-dimensional geometric model to obtain a three-dimensional virtual model.

[0036] The construction method of the temperature field simulation model includes: establishing a dynamic temperature field simulation model based on a neural network algorithm, and using Kalman filtering to fuse real-time sensor data to correct the simulation results; completing the temperature field cloud map reconstruction through vertex shading and texture baking methods.

[0037] The intelligent agent diagnosis model includes a spatiotemporal feature extraction model based on a deep learning algorithm and an operating status anomaly recognition model; the spatiotemporal feature extraction model is used to construct a multidimensional feature vector for the time series data of multiple sensors and the simulation data of the temperature field simulation model, and extract the spatial and temporal features of the multidimensional feature vector; the operating status anomaly recognition model adopts a generative adversarial network architecture to realize status anomaly recognition by calculating the anomaly score or probability based on the features extracted by the spatiotemporal feature extraction model.

[0038] Through sensors and Internet of Things technology, real-time access to generator operation data is achieved to realize the operation data connection between the digital twin and the physical entity.

[0039] The data intercommunication of the three-dimensional virtual model, the temperature field simulation model and the intelligent agent diagnosis model is achieved through a unified time-space benchmark, and the dynamic mutual verification of the intelligent agent diagnosis model and the temperature field simulation model is achieved through a two-way feedback mechanism. Among them, data intercommunication specifically refers to: achieving data alignment through a unified time-space coordinate system, establishing a mapping relationship between the vertices of the three-dimensional virtual model and the grid nodes in the temperature field simulation model, and matching the fault probability matrix output by the intelligent agent diagnosis model to the three-dimensional grid of the three-dimensional virtual model through spatial interpolation. The two-way feedback mechanism specifically includes: establishing a dynamic data interaction verification middleware, correcting the boundary conditions of the temperature field simulation model according to the fault probability output by the intelligent agent diagnosis model and triggering the local grid adaptive encryption of the temperature field simulation model, and marking the fault location in real time through the three-dimensional virtual model; the simulation results of the temperature field simulation model verify the fault probability output by the intelligent agent diagnosis model.

[0040] Example 4 As the best embodiment of the present invention, the present invention includes a generator digital twin diagnosis method based on online temperature field simulation. The method dynamically monitors the internal heating state of the generator through virtual temperature sensors to achieve online risk warning and fault diagnosis. Specifically, the method includes the following steps: Construct a digital twin, which includes a three-dimensional virtual model, a temperature field simulation model, and an intelligent diagnostic model. The three-dimensional virtual model is a high-fidelity, parameterized, and computable model that takes into account both lightweight model and refined attributes to ensure smooth interactive operations and complete information display. It is used to restore the generator's geometric structure, material properties, and component spatial relationships, providing a spatial framework for fault location. The method for constructing the three-dimensional virtual model includes: Multi-scale parametric modeling: Establish hierarchical 3D geometric models ranging from macroscopic wholes to key features, and use methods such as mesh simplification, instanced rendering, and texture compression for lightweight rendering to reduce loading time and improve interactive fluency and development efficiency.

[0041] Spatiotemporal attribute mapping: Dynamic data, or time-series data from multiple sensors, is preprocessed through time alignment and spatial normalization. Material parameters, operating conditions, and equipment status data (temperature, stress, etc.) are then deeply mapped and integrated with the lightweight rendered 3D geometric model, allowing for dynamic display and interaction to create a 3D virtual model. Dynamic data refers to data that is updated over time.

[0042] The temperature field simulation model is used to simulate the internal temperature distribution and changes online. Its construction method includes: based on the high-fidelity finite element simulation data set, establishing a dynamic temperature field simulation model based on the neural network algorithm, introducing real-time sensor data, and using Kalman filter fusion to dynamically adjust the Kalman gain. K k, weighted fusion predictions and measured data are corrected in real time. Temperature field cloud maps are reconstructed through vertex shading and texture baking, enabling rapid simulation and high-fidelity interactive visualization of temperature fields on a million-level grid. Specifically, vertex shading is used to map temperature values ​​to vertex colors, enabling interactive applications such as displaying the current location's real-time temperature value when the mouse is hovering. Texture baking is used to map the temperature field to a UV unwrapped texture for high-precision display, allowing for control of historical data playback via a time slider.

[0043] The intelligent agent diagnostic model includes a spatiotemporal feature extraction model based on deep learning algorithms, an operational anomaly identification model, a time series data analysis and trend prediction model, and a fault tracing model based on a dynamic Bayesian network. Leveraging machine learning or deep learning algorithm training, combined with real-time sensor data and temperature field simulation data, it performs operational status assessment, risk identification and early warning, trend prediction, and fault cause tracing.

[0044] Specifically, the spatiotemporal feature extraction model based on the deep learning algorithm is the CNN-GRU spatiotemporal feature extraction model: for multi-sensor time series data (temperature, vibration, current, etc.) and three-dimensional temperature field simulation data, a multi-dimensional feature vector is established, spatial features are extracted through CNN, and GRU captures temporal dependencies, while realizing local abnormal hotspot (space) and degradation trend (time) analysis.

[0045] The operational status anomaly recognition model can be a generative adversarial network (GAN)-based model. Using a GAN architecture, the generator learns the distribution characteristics of normal operating data, while the discriminator distinguishes between real and generated data. During the training phase, only healthy data is used, and network parameters are optimized by minimizing the Wasserstein distance. During online monitoring, real-time sensor data is input to the generator to obtain a reconstructed output. The reconstruction error (e.g., mean squared error) is calculated as an anomaly score, which is then comprehensively evaluated in combination with the anomaly probability output by the discriminator. A dynamic threshold mechanism is used, with alarm thresholds set based on extreme value theory. An alert is triggered when the anomaly score exceeds μ + 3σ.

[0046] The time series data analysis and trend prediction model is a long short-term memory (LSTM) time series state prediction model. For multivariate historical sequence data, a dual attention mechanism is established: feature attention is used to assign weights to each sensor; temporal attention is used to focus on key time steps, thereby achieving state prediction. The historical sequence data includes sensor time series data and historical results output by the intelligent agent diagnostic model.

[0047] The fault tracing model based on a dynamic Bayesian network constructs a device fault propagation path using hierarchical time-series nodes (root cause layer, physical layer, and phenomenon layer). It then uses the expectation-maximization (EM) algorithm to learn state transition probabilities from historical data. During online monitoring, it uses a particle filter algorithm to fuse multi-source sensor data, iteratively update the posterior probability of each node, and locate the maximum likelihood fault chain.

[0048] Combining the probabilistic root cause analysis provided by the fault tracing model, the comparative heat map generated by the operating status anomaly identification model, and the three-dimensional visual interactive interface, it can display the fault propagation path, shorten the decision-making time of operation and maintenance personnel, and greatly improve the credibility and operability of the diagnostic results.

[0049] Through sensors and IoT technology, real-time access to generator operation data is achieved to connect the operation data between the digital twin and the physical entity. Specifically, the following steps are included: Step S 11 Establish a multi-level data transmission architecture that includes sensors, intelligent edge nodes, high-speed transmission networks, and intelligent analysis platforms. Use high-speed transmission protocols to collect and transmit measurement point data. At the same time, introduce lossless data compression algorithms (preferably Delta+Huffman) to reduce bandwidth usage and further improve transmission efficiency.

[0050] Step S 12 . Data collection is achieved through sensors.

[0051] Step S 13 . Introducing lossless data compression algorithms, utilizing high-speed transmission networks to achieve real-time data transmission, and implementing data preprocessing and multi-dimensional data quality assessment on intelligent edge nodes.

[0052] Data preprocessing includes outlier removal and missing data completion. Specifically, a sliding window-based Grubbs test is established to remove outliers, and missing data is reconstructed using spatiotemporal kriging interpolation. The weights λi are determined by the variogram model to ensure data integrity. The multi-dimensional data quality assessment is used to evaluate the quality of preprocessed data based on data integrity, accuracy, consistency, timeliness, uniqueness, and credibility. The assessment methods may include: data missing check, outlier check, sensor validity check, and repeatability check.

[0053] Step S 14 Implement multi-source sensor data fusion on the intelligent analysis platform. Establish a high-precision, low-latency transmission channel to ensure data synchronization between the digital twin and the physical entity, providing data support for subsequent temperature field simulation and intelligent diagnosis.

[0054] The multi-source sensor data fusion method specifically involves: achieving data-level fusion through adaptive weighted averaging and adaptive Kalman filtering; extracting key features using principal component analysis and performing deep learning analysis on the extracted key features using a hybrid model of convolutional neural networks and long-short-term memory networks to achieve feature-level fusion; and achieving decision-level fusion through a combination of evidence theory and fault tree analysis. This multi-level fusion method can improve data reliability and enhance state perception capabilities.

[0055] Data interoperability among the 3D virtual model, temperature field simulation model, and agent-based diagnostic model is achieved through a unified spatiotemporal benchmark. Dynamic mutual verification between the agent-based diagnostic model and the temperature field simulation model is achieved through a two-way feedback mechanism. Specifically, the 3D virtual model, temperature field simulation model, and agent-based diagnostic model are encapsulated as independent modules, and data alignment is achieved through a unified spatiotemporal coordinate system. A mapping relationship is established between the vertices of the 3D virtual model and the grid nodes in the temperature field simulation model. The fault probability matrix output by the agent-based diagnostic model is then spatially interpolated and matched to the 3D mesh of the 3D virtual model.

[0056] The two-way feedback mechanism specifically includes: establishing dynamic data interaction verification middleware, correcting the boundary conditions of the temperature field simulation model according to the fault probability output by the intelligent agent diagnosis model and triggering the local grid adaptive encryption of the temperature field simulation model, and marking the fault location in real time through the three-dimensional virtual model; the simulation results of the temperature field simulation model verify the fault probability output by the intelligent agent diagnosis model.

[0057] By integrating real-time data, simulation results, and diagnostic algorithms through the digital twin, we can build an online diagnostic decision support system that features real-time monitoring, dynamic simulation, intelligent evaluation, and three-dimensional updates. The system generates a diagnostic report every 5 seconds, including key parameter trends and operation and maintenance recommendations. It dynamically evaluates the generator status and displays it visually, providing operation and maintenance decision support, thereby improving operational reliability and maintenance convenience and reducing unplanned downtime.

[0058] In summary, after reading the present invention document, ordinary technicians in this field can make various other corresponding transformation schemes based on the technical solutions and technical concepts of the present invention without creative mental work, which all fall within the scope of protection of the present invention.

Claims

1. A generator digital twin diagnosis method based on online temperature field simulation, characterized by: The following steps are involved: Constructing a digital twin, which includes a three-dimensional virtual model, a temperature field simulation model, and an intelligent agent diagnosis model; Real-time access to generator operating data through sensors and IoT technology enables operational data connection between the digital twin and the physical entity; The data intercommunication between the 3D virtual model, the temperature field simulation model and the intelligent agent diagnosis model is achieved through a unified spatiotemporal benchmark, and the dynamic mutual verification between the intelligent agent diagnosis model and the temperature field simulation model is achieved through a two-way feedback mechanism. Specifically, data interoperability involves aligning data through a unified spatiotemporal coordinate system, establishing a mapping relationship between the vertices of the 3D virtual model and the grid nodes in the temperature field simulation model, and matching the fault probability matrix output by the agent-based diagnostic model to the 3D grid of the 3D virtual model through spatial interpolation. The two-way feedback mechanism specifically includes: establishing dynamic data interaction verification middleware, correcting the boundary conditions of the temperature field simulation model according to the fault probability output by the intelligent agent diagnosis model and triggering the local grid adaptive encryption of the temperature field simulation model, and marking the fault location in real time through the three-dimensional virtual model; the simulation results of the temperature field simulation model verify the fault probability output by the intelligent agent diagnosis model.

2. The generator digital twin diagnosis method based on online temperature field simulation according to claim 1 is characterized by: The following steps are involved in achieving operational data connection between the digital twin and the physical entity: Step S 11 Establish a multi-level data transmission architecture that includes sensors, intelligent edge nodes, high-speed transmission networks, and intelligent analysis platforms; Step S 12 Data collection through sensors; Step S 13 Introducing lossless data compression algorithms, leveraging high-speed transmission networks to achieve real-time data transmission, and implementing data preprocessing and multi-dimensional data quality assessment at intelligent edge nodes; Step S 14 . Implement multi-source sensor data fusion on the intelligent analysis platform and transmit the fused data to the digital twin in real time.

3. The generator digital twin diagnosis method based on online temperature field simulation according to claim 2 is characterized by: The data preprocessing includes outlier removal and missing data completion.

4. The generator digital twin diagnosis method based on online temperature field simulation according to claim 3 is characterized by: The missing data completion specifically refers to: reconstructing using the spatiotemporal Kriging interpolation method and using the variation function model to determine the weight λi.

5. The generator digital twin diagnosis method based on online temperature field simulation according to claim 2 is characterized by: The multi-dimensional data quality assessment is used to evaluate the quality of pre-processed data from the perspectives of data completeness, accuracy, consistency, timeliness, uniqueness and credibility.

6. The generator digital twin diagnosis method based on online temperature field simulation according to claim 2 is characterized by: The multi-source sensor data fusion specifically includes: realizing data layer fusion through adaptive weighted averaging and adaptive Kalman filtering; using principal component analysis to extract key features and combining convolutional neural network and long short-term memory network hybrid model to perform deep learning analysis on the extracted key features to realize feature layer fusion; and realizing decision layer fusion by combining evidence theory with fault tree analysis.

7. The generator digital twin diagnosis method based on online temperature field simulation according to claim 1 is characterized by: The method for constructing a three-dimensional virtual model includes: establishing a hierarchical three-dimensional geometric model from the macroscopic whole to the key features, and performing lightweight rendering on the three-dimensional geometric model; preprocessing the time series data of multiple sensors through time alignment and spatial normalization, and deep mapping and fusing material parameters, operating conditions, and equipment status data with the lightweight rendered three-dimensional geometric model to obtain a three-dimensional virtual model.

8. The generator digital twin diagnosis method based on online temperature field simulation according to claim 1 is characterized by: The construction method of the temperature field simulation model includes: establishing a dynamic temperature field simulation model based on a neural network algorithm, and using Kalman filtering to fuse real-time sensor data to correct the simulation results; completing the temperature field cloud map reconstruction through vertex shading and texture baking methods.

9. The generator digital twin diagnosis method based on online temperature field simulation according to claim 1 is characterized by: The intelligent agent diagnosis model includes a spatiotemporal feature extraction model based on a deep learning algorithm and an operating status anomaly recognition model; the spatiotemporal feature extraction model is used to construct a multidimensional feature vector for the time series data of multiple sensors and the simulation data of the temperature field simulation model, and extract the spatial and temporal features of the multidimensional feature vector; the operating status anomaly recognition model adopts a generative adversarial network architecture to realize status anomaly recognition by calculating the anomaly score or probability based on the features extracted by the spatiotemporal feature extraction model.

10. The generator digital twin diagnosis method based on online temperature field simulation according to claim 9 is characterized in that: The intelligent agent diagnosis model also includes a time series data analysis and trend prediction model, including feature attention for completing the weight allocation of each sensor and time attention for completing the focus on key time steps; the time series data analysis and trend prediction model is used to realize state prediction based on multivariate historical sequence data; wherein, the multivariate historical sequence data includes the time series data of the sensor and the historical result data output by the intelligent agent diagnosis model.

11. The generator digital twin diagnosis method based on online temperature field simulation according to claim 10, characterized in that: The intelligent agent diagnosis model also includes a fault tracing model based on a dynamic Bayesian network; the fault tracing model constructs a device fault propagation path through hierarchical time series nodes and learns state transition probabilities from historical data using an expectation maximization algorithm; During online monitoring, the particle filter algorithm is used to fuse multi-source sensor data, iteratively update the posterior probability of each node, and locate the maximum likelihood fault chain.

12. The generator digital twin diagnosis method based on online temperature field simulation according to claim 11, characterized in that: The fault propagation path is displayed by combining the probabilistic root cause analysis provided by the fault tracing model, the comparative heat map generated by the operating status anomaly identification model, and the three-dimensional visual interactive interface.