Digital twin synchronization method supporting equipment state evolution modeling

Abstract

The invention relates to the technical field of digital twinning, and discloses a digital twinning synchronization method for supporting equipment state evolution modeling, which comprises the following steps of: S1, acquiring and preprocessing multi-source heterogeneous data, and deploying a multi-modal sensor array; s2, establishing a physical-virtual state synchronization mechanism, and establishing a clock synchronization network between the equipment end and the virtual model end based on an IEEE1588 protocol; s3, performing equipment state evolution modeling, and constructing a hybrid model fusing a convolutional neural network and a recurrent neural network to extract equipment degradation features; s4, guaranteeing the consistency of virtual and real states, and designing a dual verification mechanism; through the multi-mode sensor array, various key parameters of the concentrator and the special transformer terminal equipment can be comprehensively and accurately collected, the key parameters comprise a three-phase current harmonic frequency spectrum, temperature field distribution, a vibration acceleration spectrum, an insulation resistance parameter and the like, the comparability of data and the precision of subsequent processing are improved, and a data basis is provided for accurate evaluation of the equipment state.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twins, and specifically provides a digital twin synchronization method supporting device state evolution modeling. Background Art

[0002] With the booming development of Industry 4.0 and Internet of Things technologies, the degree of intelligence of various devices has been increasing day by day. In fields such as complex industrial systems, intelligent production scenarios, and equipment full life cycle management, unprecedented requirements have been put forward for the accurate monitoring, modeling, and analysis of device states. As a key means of deeply integrating physical devices with virtual models, digital twin technology has received extensive attention in recent years.

[0003] After retrieval, the patent with the Chinese patent number CN112365167B discloses a method and device for constructing a digital twin of power grid equipment based on a dynamic ontology model. The method includes: loading the logical model and ontology model of power grid equipment; performing internal data alignment and synchronization processing on the ontology model and the logical model to generate the current logical model of power grid equipment; collecting source business system data, and updating the internal logical data of the current logical model in combination with the source business system data to generate an initial digital twin model of power grid equipment; analyzing and judging the real-time state changes of the initial digital twin model based on complex event processing technology, and driving the internal ontology model to be updated through the triggering of relevant events to generate a digital twin model of power grid equipment. Through the introduction of the digital modeling prototype of power grid equipment and the combination of relevant characteristics and related technologies of the ontology in the embodiments of the present invention, the construction of the digital twin model can be realized, the full life cycle of power grid equipment can be monitored in real time, and decision-making assistance support can be provided.

[0004] In the above-mentioned technology, there are certain difficulties in constructing digital twins of concentrators and dedicated transformer terminal devices and achieving high-frequency synchronization of operating states and physical states. This may cause the state difference between the virtual model and the physical entity to gradually increase over time, and thus it is impossible to reflect the actual operating conditions of the device in a timely and accurate manner; in terms of operation and maintenance visibility, the above technology provides a relatively simple and one-sided visualization display of concentrators and dedicated transformer terminal devices, only providing basic operating parameters and status indicators of the device, lacking the excavation and display of deep-level state information of the device, such as the interaction relationship between internal components of the device, the distribution of potential fault hazards, and the overall health status trend of the device. Based on this, the present invention designs a digital twin synchronization method supporting device state evolution modeling to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a digital twin synchronization method supporting device state evolution modeling, which solves the problem of difficult high-frequency synchronization in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: A digital twin synchronization method for supporting device state evolution modeling, comprising: Step S1, multi-source heterogeneous data collection and preprocessing, deploying a multi-modal sensor array to collect the three-phase current harmonic spectrum, temperature field distribution, vibration acceleration spectrum, and insulation resistance parameters of the concentrator and dedicated transformer terminal equipment; Step S2, physical-virtual state synchronization mechanism, establishing a clock synchronization network between the device end and the virtual model end based on the IEEE1588 protocol, and using a dynamic adaptive Kalman filtering algorithm to eliminate timing deviations, with a synchronization period of 10 ms to 100 ms; Step S3, device state evolution modeling, constructing a hybrid model that combines a convolutional neural network and a recurrent neural network to extract device degradation features, and the input features include device degradation features, operating environment parameters, and maintenance history. In the hybrid model, the convolutional neural network extracts the spatial features of the device operation data, and the recurrent neural network captures the time series features; Step S4, ensuring the consistency of virtual and physical states, designing a double-check mechanism, where the double-check mechanism includes data integrity check and data consistency check. The data integrity check is used to check whether there are missing or incorrect collected data, and the data consistency check is used to compare whether the data on the device end and the virtual model end are consistent.

[0007] Preferably, in step S1, the original signal needs to be normalized, and the calculation formula is: x_norm = (x - u) / σ, where μ is the signal mean and σ is the standard deviation, and the normalized data range is limited to the interval [-1, 1]; The preprocessing stage further includes: step S1.1, using a sliding window technique with a window length of 512 sampling points and an overlap rate of 50%; step S1.2, combining the Daubechies wavelet basis function to perform noise reduction processing on the original signal to remove high-frequency noise components; step S1.3, using linear interpolation to complete the missing data, and setting the interpolation threshold to continuous missing points ≤ 5.

[0008] As can be seen from the above technical solutions: during signal preprocessing, normalization is first performed to limit the signal data within the range of [-1, 1] through the formula x_norm = (x - μ) / σ, eliminating the influence of dimensional and numerical differences. Subsequently, the sliding window technique is adopted, with a window of 512 sampling points and an overlap rate of 50%, to locally observe the signal changes and achieve smooth transition to avoid information loss. Then, the time-frequency localization characteristics of the Daubechies wavelet basis are utilized to remove high-frequency noise and retain low-frequency effective information. Finally, for data with the number of consecutive missing points ≤ 5, based on the assumption of signal continuity, linear interpolation is used to complete the missing data, making the signal complete and conforming to the original change trend.

[0009] Preferably, the method further includes: step S5, deep state visualization, generating a dynamic association topology map of the internal components of the device, and calculating the edge weights using the Pearson correlation coefficient to reflect the interaction strength between components in real time.

[0010] Preferably, the multimodal sensor array includes a current sensor, a temperature sensor, an acceleration sensor, and an insulation resistance tester. The sampling frequencies of each sensor are optimized according to the operating characteristics of the device, and the specific configuration is as follows: for the current sensor, the sampling frequency is 10 kHz; for the temperature sensor, an infrared thermal imager is used with a resolution of 640×480 pixels and a temperature measurement range of -20°C to 1500°C; for the triaxial vibration sensor, the sensitivity is 100 mV / g, the range is ±50 g, and the frequency response range is 10 Hz - 5 kHz; for the insulation resistance tester, the output voltage is 5000 V, the test range is 0 - 10 GΩ, and the dynamic sampling interval is 0.1 Hz.

[0011] As can be seen from the above technical solutions: the multimodal sensor array realizes multi-dimensional state perception of the device through collaborative monitoring. The current sensor (HCT228B, 10 kHz) is based on the principle of electromagnetic induction to collect the instantaneous changes of the working current of the device in real time, providing basic information on electrical performance. The infrared thermal imager generates a thermal imaging map of 640×480 pixels by sensing the thermal radiation differences of objects, covering a temperature measurement range of -20°C to 1500°C, to achieve non-contact temperature field monitoring. The triaxial vibration sensor uses the piezoelectric effect to convert mechanical vibration into an electrical signal. The sensitivity of 100 mV / g and the range of ±50 g can capture the dynamic vibration characteristics within the frequency response range of 10 Hz - 5 kHz. The insulation resistance tester applies a voltage of 5000 V and uses a dynamic sampling of 0.1 Hz to monitor the weak leakage current and infer the change of the insulation resistance value in the range of 0 - 10 GΩ, realizing the quantitative evaluation of electrical insulation performance. Each sensor optimizes the sampling frequency according to the physical characteristics of the device to form a comprehensive monitoring network for electrical, thermal, mechanical, and insulation performance.

[0012] Preferably, in step S2, the physical-virtual state synchronization mechanism further includes: when the packet loss rate exceeds 1%, starting the redundant data retransmission mechanism with a maximum retransmission count of 3 times; dynamically adjusting the synchronization period, extending it to 100 ms during the peak device load period and shortening it to 10 ms during the non-peak period.

[0013] Preferably, the device state evolution modeling further includes a multi-scale attention mechanism that weights the feature contribution degrees of different sensor channels through spatial attention, with a convolution kernel size of 7×7; the temporal attention module analyzes the device degradation trend based on the temporal features with a sliding window length of 10.

[0014] Preferably, the sliding window length in the virtual-real state consistency guarantee is automatically adjusted according to the dynamic characteristics of the device operating state, and the state reconstruction algorithm reconstructs the device state by combining the historical operation data and physical model of the device.

[0015] Preferably, the deep state visualization further includes constructing a three-dimensional visualization model and developing an interactive visualization interface. The three-dimensional visualization model is created based on the CAD model and actual size ratio of the device, and combines the operation data and state information of the device to visually display the device.

[0016] Preferably, the method further includes step S6, abnormal state warning and alarm. Based on the device state evolution model and the virtual-real state consistency guarantee mechanism, an abnormal state detection algorithm is established, which combines statistical methods and machine learning methods to monitor and analyze the device operation data in real time. Step S6 includes the following steps: Step S6.1, the statistical method uses the 3σ criterion, and a primary alarm is triggered when 3 consecutive data points exceed the range of μ±3σ; Step S6.2, the machine learning method uses the isolation forest algorithm, sets the abnormal score threshold to 0.65, and the model training data set contains 500 groups of historical fault samples; Step S6.3, a multi-level alarm mechanism, the primary alarm is pushed to the local monitoring terminal, and the high-level alarm synchronously triggers SMS notifications and cloud work order generation with a response delay ≤200 ms.

[0017] It can be seen from the above technical solutions that: Step S6 realizes device abnormal warning by integrating statistical and machine learning methods. Statistical monitoring judges data anomalies according to the 3σ criterion, and a primary alarm is triggered when 3 consecutive points exceed the threshold; machine learning uses the trained isolation forest algorithm to identify complex abnormal patterns. The multi-level alarm mechanism pushes local prompts or SMS notifications according to the degree of abnormality and quickly generates cloud work orders with a response delay controlled within 200 ms.

[0018] Preferably, the method further includes step S7, remote monitoring and operation and maintenance management. A remote monitoring platform is built and connected to the device end through the Internet to achieve remote real-time monitoring and data analysis of the device. The platform has functions such as user management, permission management, data storage, and data backup, and supports multi-user simultaneous access and collaborative work. The platform architecture adopts the browser-server mode. The backend service is developed based on the Java language, and the database uses a MySQL relational database cluster, which supports database sharding, table sharding, and read-write separation operations. The data storage module is configured to store real-time monitoring data in the in-memory database Redis for 30 days, and the data sampling interval is ≤1 second. Historical data is stored in the distributed file system HDFS in the compressed storage format of Parquet, and the storage period is ≥10 years. The platform meets the third-level security protection capabilities of GB / T 22239-2019 "Information Security Technology - Basic Requirements for Network Security Level Protection", specifically including: data transmission is encrypted using the TLS 1.3 protocol, the key exchange algorithm is ECDHE-RSA, and the encryption suite is AES_256_GCM; user identity authentication supports two-factor verification, including dynamic passwords and digital certificates.

[0019] It can be seen from the above technical solutions that the remote monitoring and operation and maintenance management platform is based on the browser-server architecture, with the backend developed in Java and data stored in a MySQL cluster, supporting database sharding, table sharding, and read-write separation. Real-time monitoring data is stored in Redis with a sampling interval of ≤1 second and retained for 30 days; historical data is compressed and stored in HDFS in the Parquet format with a storage period of ≥10 years. The platform meets the requirements of the third-level information security protection. Data transmission is encrypted using TLS 1.3, and two-factor authentication is supported.

[0020] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: 1. In the present invention, through the multi-modal sensor array, various key parameters of the concentrator and the dedicated transformer terminal device can be comprehensively and accurately collected, including three-phase current harmonic spectra, temperature field distributions, vibration acceleration spectra, insulation resistance parameters, etc., improving the comparability of data and the accuracy of subsequent processing, and providing a data basis for the accurate assessment of the device status.

[0021] 2. In the present invention, through the clock synchronization network established based on the IEEE1588 protocol and combined with the dynamic adaptive Kalman filtering algorithm, timing deviations can be eliminated, and real-time synchronization between the device end and the virtual model end can be achieved. The synchronization period can be automatically adjusted according to the device load and network conditions, and a redundant data retransmission mechanism is started when the packet loss rate exceeds the standard, ensuring the integrity and reliability of data transmission.

[0022] 3. In the present invention, by constructing a hybrid model that integrates a convolutional neural network and a recurrent neural network, the spatial feature extraction ability of the convolutional neural network for device operation data and the time series feature capture ability of the recurrent neural network are fully utilized. At the same time, a multi-scale attention mechanism is introduced. The spatial attention weights the feature contribution degrees of different sensor channels, and the time attention module analyzes the device degradation trend based on the sliding window length, so as to achieve accurate evolution modeling of the device state. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is the system architecture diagram of the present invention; Figure 2 is the flowchart of the physical-virtual state synchronization mechanism of the present invention; Figure 3 is the structure diagram of the hybrid model for device state evolution of the present invention; Figure 4 is the logic diagram for ensuring the consistency of virtual and real states of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] Embodiment 1; Please refer to Figures 1 - 4 , a digital twin synchronization method for supporting device state evolution modeling, including: Step S1, multi-source heterogeneous data collection and preprocessing, deploying a multi-modal sensor array to collect the three-phase current harmonic spectrum, temperature field distribution, vibration acceleration spectrum, and insulation resistance parameters of the concentrator and special transformer terminal equipment; Step S2, physical-virtual state synchronization mechanism, establishing a clock synchronization network between the device end and the virtual model end based on the IEEE1588 protocol, using a dynamic adaptive Kalman filtering algorithm to eliminate the timing deviation, and the synchronization period is 10 ms to 100 ms; Step S3, device state evolution modeling, constructing a hybrid model that integrates a convolutional neural network and a recurrent neural network to extract device degradation features, and the input features include device degradation features, operating environment parameters, and maintenance history. In the hybrid model, the convolutional neural network extracts the spatial features of device operation data, and the recurrent neural network captures the time series features; Step S4, ensuring the consistency of virtual and real states, designing a double-check mechanism, and the double-check mechanism includes data integrity check and data consistency check. The data integrity check is used to check whether there are missing or incorrect collected data, and the data consistency check is used to compare whether the data between the device end and the virtual model end is consistent.

[0026] In step S1, the original signal needs to be normalized, and the calculation formula is: x_norm = (x - μ) / σ, where μ is the signal mean and σ is the standard deviation, and the normalized data range is limited within the interval [-1, 1]; The preprocessing stage further includes: step S1.1, adopting the sliding window technique with a window length of 512 sampling points and an overlap rate of 50%; step S1.2, combining the Daubechies wavelet basis function to perform noise reduction processing on the original signal to remove high-frequency noise components; step S1.3, using the linear interpolation method to complete the missing data, and the interpolation threshold is set as the number of consecutive missing points ≤ 5. The method also includes: step S5, deep state visualization, generating a dynamic association topology map of the internal components of the device, and calculating the edge weights using the Pearson correlation coefficient to reflect the interaction strength between components in real time. The multi-modal sensor array includes a current sensor, a temperature sensor, an acceleration sensor, and an insulation resistance tester. The sampling frequencies of each sensor are optimized according to the operating characteristics of the device, and the specific configuration is: for the current sensor, the sampling frequency is 10 kHz; for the temperature sensor, an infrared thermal imager is used with a resolution of 640 × 480 pixels and a temperature measurement range of -20°C to 1500°C; the sensitivity of the three-axis vibration sensor is 100 mV / g, the range is ±50 g, and the frequency response range is 10 Hz - 5 kHz; the output voltage of the insulation resistance tester is 5000 V, the test range is 0 - 10 GΩ, and the dynamic sampling interval is 0.1 Hz. Step S2, the physical-virtual state synchronization mechanism also includes: when the packet loss rate exceeds 1%, start the redundant data retransmission mechanism, and the maximum number of retransmissions is 3 times; dynamically adjust the synchronization period, extend it to 100 ms during the peak load period of the device, and shorten it to 10 ms during the non-peak period.

[0027] The working principle of the embodiment of the present invention is: collect the original data such as the three-phase current harmonic spectrum, temperature field distribution, vibration acceleration spectrum, and insulation resistance parameters of the device through the multi-modal sensor array, and perform normalization processing on the original signal to eliminate the influence of different dimensions and dimension units, and improve the comparability of the data and the accuracy of subsequent processing. Subsequently, based on the IEEE1588 protocol, establish a clock synchronization network between the device end and the virtual model end, and use the dynamic adaptive Kalman filtering algorithm to eliminate the timing deviation to ensure the real-time synchronization between the device end and the virtual model end. The synchronized data is input into the device state evolution modeling module, and a hybrid model integrating a convolutional neural network and a recurrent neural network is constructed to extract the device degradation characteristics and realize the evolution modeling of the device state. At the same time, the virtual-real state consistency guarantee module ensures the consistency between the virtual model and the physical entity state through a double verification mechanism. Finally, the deep state visualization module generates a dynamic association topology map of the internal components of the device and performs visual display, enabling the operation and maintenance personnel to intuitively understand the association relationship between the internal components of the device and their dynamic changes.

[0028] Example 2; Please refer to Figures 1 - 4 , in the embodiment of the present invention, the device state evolution modeling further includes a multi-scale attention mechanism, which weights the feature contribution degrees of different sensor channels through spatial attention, and the convolution kernel size is 7×7; the temporal attention module analyzes the device degradation trend based on the temporal features with a sliding window length of 10. The sliding window length in the virtual-real state consistency guarantee is automatically adjusted according to the dynamic characteristics of the device operation state, and the state reconstruction algorithm reconstructs the device state by combining the historical operation data and physical model of the device. The deep state visualization further includes constructing a three-dimensional visualization model and developing an interactive visualization interface. The three-dimensional visualization model is created based on the CAD model and actual size ratio of the device, and combines the operation data and state information of the device to perform visual display on the device.

[0029] The method further includes step S6, abnormal state warning and alarm. Based on the device state evolution model and the virtual-real state consistency guarantee mechanism, an abnormal state detection algorithm is established, which combines a statistical method and a machine learning method to perform real-time monitoring and analysis on the device operation data. Step S6 includes the following steps: Step S6.1, the statistical method adopts the 3σ criterion, and a primary alarm is triggered when three consecutive data points exceed the range of μ±3σ; Step S6.2, the machine learning method adopts the isolation forest algorithm, and the abnormal score threshold is set to 0.65. The model training data set contains 500 groups of historical fault samples; Step S6.3, a multi-level alarm mechanism, the primary alarm is pushed to the local monitoring terminal, and the high-level alarm synchronously triggers a text message notification and cloud work order generation, and the response delay ≤ 200ms. The method further includes step S7, remote monitoring and operation and maintenance management. A remote monitoring platform is built, which is connected to the device end through the Internet to achieve remote real-time monitoring and data analysis of the device. The platform has functions such as user management, permission management, data storage, and data backup, and supports multi-user simultaneous access and collaborative work; the platform architecture adopts the browser-server mode, the backend service is developed based on the Java language, and the database adopts the MySQL relational database cluster, which supports database sharding, table sharding, and read-write separation operations; the data storage module is configured such that the real-time monitoring data is stored in the in-memory database Redis for 30 days, and the data sampling interval ≤ 1 second; the historical data is stored in the distributed file system HDFS, and the compressed storage format is Parquet, and the storage period ≥ 10 years; the platform meets the third-level security protection capabilities of GB / T 22239-2019 "Information Security Technology - Basic Requirements for Network Security Level Protection", specifically including: data transmission uses the TLS1.3 protocol for encryption, the key exchange algorithm is ECDHE-RSA, and the encryption suite is AES_256_GCM; user identity authentication supports two-factor verification, including dynamic passwords and digital certificates.

[0030] The working principle of the embodiments of the present invention is as follows: The device state evolution modeling module further introduces a multi-scale attention mechanism, which weights the feature contribution degrees of different sensor channels through spatial attention, and the convolution kernel size is 7×7; the temporal attention module analyzes the device degradation trend based on the temporal features with a sliding window length of 10, enhancing the model's prediction ability for the device degradation trend. The sliding window length in the virtual-real state consistency guarantee module can be automatically adjusted according to the dynamic characteristics of the device operation state, ensuring that the state reconstruction algorithm efficiently reconstructs the device state. The deep state visualization module not only dynamically associates topological graphs but also constructs a three-dimensional visualization model and develops an interactive visualization interface. In addition, the present invention also covers an abnormal state warning and alarm mechanism as well as remote monitoring and operation and maintenance management functions. The device operation data is monitored and analyzed in real time through an abnormal state detection algorithm. Once an abnormal device state is detected, the operation and maintenance personnel are immediately notified in multiple ways, and alarm information is provided.

[0031] Embodiment 3; Please refer to Figures 1 - 4 , which provides an embodiment of a digital twin synchronization method for power equipment based on dynamic feature fusion. In this embodiment, a 110kV oil-immersed power transformer is used as the implementation object, and the specific process and parameters are as follows: Multi-source heterogeneous data collection and preprocessing. Deploy a multi-modal sensor array, including: a current sensor with a sampling frequency of 10kHz, collecting the three-phase current harmonic spectrum, covering harmonic components from 0 to 50 times; an infrared thermal imager with a resolution of 640 pixels × 480 pixels, a temperature measurement range of -20 degrees Celsius to 1500 degrees Celsius, generating a temperature field distribution map, and a spatial resolution of 0.1 degrees Celsius; a three-axis vibration sensor with a sensitivity of 100 millivolts per gravitational acceleration and a range of ±50 gravitational acceleration, recording the vibration acceleration spectrum, with a frequency band covering 10 Hz to 5000 Hz; an insulation resistance tester with an output voltage of 5000 volts and an accuracy of ±2%, dynamically monitoring the winding insulation parameters. Where μ is the signal mean and σ is the standard deviation, and the normalized data range is limited to the interval from -1 to 1. In the preprocessing stage, a sliding window technique is adopted, with a window length of 512 sampling points and an overlap rate of 50%, and noise reduction is carried out in combination with the Daubechies wavelet basis function.

[0032] Physical-virtual state synchronization mechanism, constructs a clock synchronization network based on the IEEE1588v2 protocol, and the master clock node and 12 slave nodes form a ring topology. The dynamic adaptive Kalman filtering algorithm is adopted, the process noise covariance Q is set to 0.01, the observation noise covariance R is set to 0.1, and the timing deviation compensation accuracy reaches ±1 microsecond. The dynamic adjustment rule of the synchronization period is as follows: during the peak load period from 08:00 to 18:00 every day, the synchronization period is 100 milliseconds, and the data packet size limit is 1 megabyte; during the non-peak period, the synchronization period is shortened to 10 milliseconds, and the LZ77 data compression algorithm is enabled, and the compression rate is not less than 60%. When the data packet loss rate exceeds 1%, the redundant retransmission mechanism is started, the maximum number of retransmissions is 3 times, and the ACK timeout threshold is 200 milliseconds.

[0033] Device state evolution modeling, constructs a hybrid model of a convolutional neural network and a bidirectional long short-term memory network: the spatial feature extraction layer uses a 7×7 convolutional kernel to extract the spatial distribution features of the temperature field; the timing analysis layer configures 128 hidden node bidirectional LSTM units to process the vibration spectrum time series; the multi-scale attention mechanism fuses the 7×7 spatial attention weight matrix and the time sliding window features of length 10. The input features include the current harmonic distortion rate, the hot spot temperature gradient, the vibration main frequency offset, and the maintenance records of the past three years, forming a 32-dimensional feature vector. The model is trained using the Adam optimizer, with a learning rate of 0.001, a batch size of 64, and achieves a prediction accuracy of 98.7% on 500 sets of historical fault data sets.

[0034] Ensuring the consistency of virtual and real states, implements a double-check mechanism: for data integrity verification, CRC32 check codes are compared, and abnormal data automatically triggers the re-acquisition process; for consistency verification, a dynamic association topology graph is constructed, which contains no less than 50 nodes, and the edge weight is calculated using the Pearson correlation coefficient. When the deviation between the virtual model and the entity data exceeds the temperature threshold of ±3 degrees Celsius or the vibration acceleration threshold of ±0.5 gravitational acceleration, the state reconstruction algorithm is started. The verification results are pushed to the operation and maintenance platform through the MQTT protocol, and the end-to-end response delay does not exceed 50 milliseconds.

[0035] Deep state visualization and early warning, develops a three-dimensional visualization system, implemented based on the Unity3D engine: the dynamic topology displays the real-time rendered winding temperature field distribution, and the color mapping range is from 30 degrees Celsius to 120 degrees Celsius; the interactive diagnosis supports axial and radial profile analysis, the slice resolution is 0.1 millimeter, and the historical data backtracking time axis accuracy is 1 second; for abnormal early warning, the insulation resistance drop rate threshold is set at 5% per hour and the third harmonic content threshold is 8%, triggering a multi-level alarm mechanism, synchronously generating a maintenance work order and recommending a spare part list, with a recommended accuracy of 92%.

[0036] Remote operation and maintenance management, building a browser-server architecture monitoring platform, using the Java Spring Boot framework and MySQL cluster: The device profile aggregates more than 150 operating parameters in 12 categories, and calculates the health index hourly; Predictive maintenance calculates the remaining life based on the Weibull distribution model, with an error range of ±72 hours; The knowledge base is linked to automatically match historical similar cases, with a cosine similarity threshold of 0.85, and recommended disposal solutions are provided. The platform supports 200 concurrent accesses, and the data storage period is 10 years, meeting the requirements of the third-level information security protection of GB / T 22239-2019 Information Security Technology.

[0037] Working principle: First, the multi-modal sensor array collects the original data of the concentrator and the special transformer terminal equipment, such as the three-phase current harmonic spectrum, temperature field distribution, vibration acceleration spectrum, and insulation resistance parameters, and performs normalization processing to eliminate the influence of dimensions and improve data accuracy. Then, a clock synchronization network is established based on the IEEE 1588 protocol, and the dynamic adaptive Kalman filtering algorithm is used to eliminate the timing deviation to ensure the accurate real-time synchronization between the device end and the virtual model end. The synchronization period can be automatically adjusted, and when the packet loss rate exceeds the standard, the redundant data retransmission mechanism is started. The synchronized data is input into the device state evolution modeling module, which integrates the convolutional neural network and the recurrent neural network to extract the device degradation characteristics. At the same time, a multi-scale attention mechanism is introduced. Through the spatial attention to weight the feature contribution degree, and the time attention module analyzes the degradation trend to realize the device state evolution modeling. The virtual-real state consistency guarantee module ensures the consistency between the virtual model and the physical entity state through a dual verification mechanism. When the data deviation exceeds the standard, the state reconstruction algorithm based on the sliding window combines historical data and physical models to reconstruct the device state. The deep state visualization module generates a dynamic association topology map and visualizes it. The edge weight calculation uses the Pearson correlation coefficient. At the same time, a three-dimensional visualization model is constructed and an interactive visualization interface is developed to help the operation and maintenance personnel intuitively understand the device state. In addition, the abnormal state warning and alarm mechanism monitors and analyzes the device operation data in real time, and notifies the operation and maintenance personnel in time when an abnormality is detected. The remote monitoring and operation and maintenance management function improves the efficiency and flexibility of device operation and maintenance management through the remote monitoring platform.

[0038] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A digital twin synchronization method for supporting device state evolution modeling, characterized in that, Including: Step S1, multi-source heterogeneous data acquisition and preprocessing, deploying a multi-modal sensor array to collect the three-phase current harmonic spectrum, temperature field distribution, vibration acceleration spectrum, and insulation resistance parameters of the concentrator and dedicated transformer terminal equipment; Step S2, physical-virtual state synchronization mechanism, establishing a clock synchronization network between the device end and the virtual model end based on the IEEE1588 protocol, using the dynamic adaptive Kalman filtering algorithm to eliminate the timing deviation, and the synchronization period is 10ms to 100ms; Step S3, device state evolution modeling, constructing a hybrid model that combines a convolutional neural network and a recurrent neural network to extract device degradation features. The input features include device degradation features, operating environment parameters, and maintenance history. In the hybrid model, the convolutional neural network extracts the spatial features of the device operating data, and the recurrent neural network captures the time series features; Step S4, ensuring the consistency of virtual and real states, designing a dual verification mechanism, which includes data integrity verification and data consistency verification. Data integrity verification is used to check whether there are missing or incorrect collected data, and data consistency verification is used to compare whether the data between the device end and the virtual model end is consistent.

2. The digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that In step S1, it is necessary to perform normalization processing on the original signal, and the calculation formula is: x_norm = (x - u) / σ, where μ is the signal mean and σ is the standard deviation, and the normalized data range is limited within the interval [-1, 1]; The multi-source heterogeneous data acquisition and preprocessing further includes: Step S1.1, using the sliding window technique, the window length is 512 sampling points, and the overlap rate is 50%; Step S1.2, combining the Daubechies wavelet basis function to perform noise reduction processing on the original signal to remove high-frequency noise components; Step S1.3, using the linear interpolation method to complete the missing data, and the interpolation threshold is set to the number of consecutive missing points ≤ 5.

3. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that The method further includes: Step S5, deep state visualization, generating a dynamic association topology map of the internal components of the device, and calculating the edge weights using the Pearson correlation coefficient to reflect the interaction strength between components in real time.

4. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that: The multi-modal sensor array includes a current sensor, a temperature sensor, an acceleration sensor, and an insulation resistance tester. The sampling frequency of each sensor is optimized according to the device operating characteristics. The specific configuration is as follows: current sensor, sampling frequency 10kHz; the temperature sensor uses an infrared thermal imager with a resolution of 640×480 pixels and a temperature measurement range of -20°C to 1500°C; the sensitivity of the three-axis vibration sensor is 100mV / g, the range is ±50g, and the frequency response range is 10Hz - 5kHz; the output voltage of the insulation resistance tester is 5000V, the test range is 0 - 10GΩ, and the dynamic sampling interval is 0.1Hz.

5. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that: In step S2, the physical-virtual state synchronization mechanism further includes: when the packet loss rate exceeds 1%, starting the redundant data retransmission mechanism, and the maximum number of retransmissions is 3 times; dynamically adjusting the synchronization period, extending it to 100ms during the peak device load period and shortening it to 10ms during the non-peak period.

6. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that: The device state evolution modeling also includes a multi-scale attention mechanism that weights the feature contribution degrees of different sensor channels through spatial attention, with a convolution kernel size of 7×7; the temporal attention module analyzes the device degradation trend based on the temporal features with a sliding window length of 10.

7. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that: The sliding window length in the virtual-real state consistency guarantee is automatically adjusted according to the dynamic characteristics of the device operation state, and the state reconstruction algorithm reconstructs the device state by combining the historical operation data and physical model of the device.

8. A digital twin synchronization method for supporting device state evolution modeling according to claim 3, characterized in that: The deep state visualization also includes constructing a three-dimensional visualization model and developing an interactive visualization interface. The three-dimensional visualization model is created based on the CAD model and actual size ratio of the device, and combines the operation data and state information of the device to visually display the device.

9. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that: The method also includes step S6, abnormal state warning and alarm. Based on the device state evolution model and the virtual-real state consistency guarantee mechanism, an abnormal state detection algorithm is established, which combines a statistical method and a machine learning method to monitor and analyze the device operation data in real time. Step S6 includes the following steps: Step S6.1, the statistical method uses the 3σ criterion, and a primary alarm is triggered when three consecutive data points exceed the range of μ±3σ. Step S6.2, the machine learning method uses the Isolation Forest algorithm, sets the abnormal score threshold to 0.65, and the model training data set contains 500 groups of historical fault samples. Step S6.3, a multi-level alarm mechanism. The primary alarm is pushed to the local monitoring terminal, and the advanced alarm synchronously triggers SMS notifications and cloud work order generation with a response delay ≤200ms.

10. A digital twin synchronization method for supporting device state evolution modeling according to claim 1, characterized in that: The method also includes step S7, remote monitoring and operation and maintenance management. A remote monitoring platform is built, connected to the device end through the Internet to achieve remote real-time monitoring and data analysis of the device. The platform has functions such as user management, permission management, data storage, and data backup, and supports multi-user simultaneous access and collaborative work.

Images