Box transformer on-site protection and measurement integration method and device based on edge digital twinning

Abstract

An integrated method and device for on-site protection and measurement and control of box transformer based on edge digital twinning, relating to the field of electric power measurement and control. The method comprises: an edge computing board constructs and updates a digital twin of the box transformer in the local memory according to multi-source operating parameters, and deduces the state response trend of the box transformer equipment under future working conditions based on the updated digital twin; the deduced state value of the digital twin at the current time is compared with the multi-source operating parameters measured at the same time, and a deviation vector is calculated; the reference value or the on-site early warning signal is returned to the protection and measurement and control CPU board through the backplane bus, the protection and measurement and control CPU board corrects the current voltage regulation determination logic and health early warning threshold according to the reference value, and generates and issues a control instruction combined with the preset electrical protection logic. The implementation of the above technical scheme can accurately monitor the operation state of the box transformer and timely discover potential abnormalities.

Description

Technical Field

[0001] This application relates to the field of power measurement and control, and in particular to an integrated method and equipment for local protection measurement and control of transformer substations based on edge digital twins. Background Technology

[0002] With the continuous development of power systems, prefabricated substations, as important transformer equipment, play a crucial role in power distribution and conversion. Their stable and reliable operation is essential for ensuring the quality and continuity of power supply.

[0003] In traditional transformer substation protection and control technology, separate protection and control devices are typically used to handle the electrical protection and operational parameter monitoring of the substation, respectively. The protection device primarily relies on pre-set electrical protection logic to trigger protective actions when abnormal electrical quantities are detected, ensuring equipment safety. The control device, on the other hand, is responsible for collecting the substation's operational parameters, such as current, voltage, and temperature, and uploading the data to the monitoring center. Furthermore, for predicting future loads, simple statistical analysis methods are often used to make approximate estimates based on the average changes in historical loads.

[0004] However, the traditional method of separating protection and measurement and control has drawbacks such as untimely data exchange and poor coordination, making it difficult to comprehensively and accurately reflect the actual operating status of the transformer substation. Simple statistical analysis methods have low accuracy in predicting future load change trends and cannot adapt to complex and ever-changing electricity demands. Summary of the Invention

[0005] This application provides an integrated method and equipment for local protection and control of transformer substations based on edge digital twins, which can accurately monitor the operating status of transformer substations and promptly detect potential anomalies.

[0006] In a first aspect, this application provides an integrated method for local protection and control of prefabricated substations based on edge digital twins, applied to a prefabricated substation protection and control device. The device includes a protection and control CPU board and an edge computing board that communicate via a backplane bus. The method includes: The edge computing board receives multi-source operating parameters collected by the protection and control CPU board, and constructs and updates the digital twin of the transformer substation in local memory in real time based on the multi-source operating parameters. The multi-source operating parameters include electrical parameters, temperature parameters and vibration parameters. The digital twin includes an electrical image, a thermal image and a mechanical image. The edge computing board, based on the updated digital twin, takes the current operating state as the initial condition, and infers the future load change trend based on the historical load sequence through a time-series prediction model. It then applies the corresponding future operating condition change amount to the digital twin and predicts the state response trend of the transformer substation under future operating conditions. The edge computing board compares the inferred state value of the digital twin at the current moment with the measured multi-source operating parameters at the same moment, and calculates the deviation vector. When the deviation values ​​of each dimension in the deviation vector are all less than the preset deviation tolerance threshold, the state of the digital twin is determined to be reliable, and the inferred state value of the digital twin is used as the benchmark reference value of the current operating state of the transformer. When the deviation value of any dimension in the deviation vector is greater than or equal to the preset deviation tolerance threshold, it is determined that there is a potential anomaly in the transformer or that the sensor data is inaccurate, and an on-site early warning signal is triggered. The edge computing board transmits the reference value or the local early warning signal back to the protection and control CPU board through the backplane bus. The protection and control CPU board corrects the current voltage regulation judgment logic and health early warning threshold according to the reference value, and generates and issues control commands in combination with the preset electrical protection logic.

[0007] By employing the above technical solution, a multi-dimensional digital twin integrating electrical, thermal, and mechanical dimensions is constructed. A two-way comparison and verification mechanism between projected and measured values ​​is introduced, enabling self-assessment of the digital twin's reliability. This not only accurately identifies inaccurate sensor data or potential equipment anomalies (triggering local early warnings), but also utilizes the projected values ​​as a benchmark when the twin is reliable. This effectively avoids misjudgments caused by noise interference or sensor drift from single measured data, significantly improving the accuracy and anti-interference capability of assessing the health status of transformer substations. Deeply integrating time-series prediction models with digital twin projections enables the device to proactively predict future state response trends of transformer substations, fundamentally resolving the issues of gear oscillation and threshold mismatch caused by delayed responses in traditional solutions.

[0008] In some embodiments, the step of constructing and updating the digital twin of the transformer substation in local memory in real time based on the multi-source operating parameters includes: The edge computing board records the trajectory of the deviation vector over time and uses the recursive least squares method to determine the current optimal estimates of the winding impedance parameters in the electrical mirror, the equivalent thermal resistance and thermal capacity parameters in the thermal mirror, and the contact wear coefficient in the mechanical mirror. The current optimal estimate is compared with the corresponding fixed parameters in the digital twin. When the relative deviation of any parameter exceeds the preset mismatch warning threshold, the model is determined to be mismatched. When a model mismatch is determined, the corresponding fixed parameters in the digital twin are replaced with the current optimal estimate, and the state variables of the digital twin are reinitialized with the replaced parameters. The changes in parameters before and after the update are encoded into a model state change log, which is then sent to the field operation and maintenance platform via the backplane bus and the protection and control CPU board.

[0009] Using the above technical solution, key physical parameters of electrical, thermal, and mechanical components are identified online using the recursive least squares method. These parameters are then compared with the fixed parameters of the digital twin. When a mismatch is detected, the online identified values ​​replace the fixed parameters, and the state variables are reinitialized. This enables the digital twin to automatically track the slow drift of physical characteristics of the transformer substation equipment caused by long-term aging and changes in operating conditions, ensuring that the twin maintains a high-fidelity mirror image of the physical entity throughout its entire lifecycle. The changes in parameters before and after updates are encoded into a model state change log and uploaded to the site operation and maintenance platform. This transforms the previously invisible adaptive adjustment behavior within the twin into transparent and traceable engineering records, providing complete data support for subsequent fault tracing analysis.

[0010] In some embodiments, the inference of future load change trends based on historical load sequences using a time-series prediction model includes: Historical active power time series data are input into the first prediction model constructed based on a temporal convolutional network to deduce the ultra-short-term load change trend for the first preset time in the future. The predicted values ​​of light intensity and / or wind speed sent by the meteorological terminal, along with the historical load data at the corresponding time, are input into the second prediction model constructed based on a long short-term memory network to deduce the short-term load change trend for the second preset time in the future. The first preset time is shorter than the second preset time. The confidence interval widths corresponding to each predicted value in the load change trend are calculated using Bayesian approximation inference. The ultra-short-term load change trend and the short-term load change trend are weighted and superimposed in the same time domain to generate a comprehensive load change trend. The width of each confidence interval is compared with the preset confidence threshold in real time. If it is less than the confidence threshold, the corresponding prediction value is determined to be reliable. The corresponding prediction value is applied to the digital twin as the future operating condition change amount. If it is greater than or equal to the confidence threshold, the corresponding prediction value is discarded. The zero change amount of the load maintaining the current state is used as the future operating condition change amount. The cumulative number of consecutive abandonments is recorded. When the abandonment frequency threshold is exceeded within a preset time window, the fusion weight of the first prediction model is increased and the fusion weight of the second prediction model is decreased simultaneously. A voltage regulation strategy conservative instruction is sent to the protection and control CPU board to temporarily tighten the load change slope threshold to half of the preset load change slope threshold until the abandonment frequency returns to normal.

[0011] By employing the aforementioned technical solution, a dual-model parallel prediction architecture is constructed, and Bayesian approximate inference is introduced to calculate the confidence interval width of the predicted values, enabling a quantitative assessment of the reliability of each prediction result. A selective mechanism of "weighted fusion applied to the twin if reliable, discarding if unreliable and replacing with zero change" ensures that the future operating condition changes input to the digital twin are always based on high-confidence data. This eliminates the pollution of twin inference caused by prediction distortions due to sudden weather changes or abnormal loads, significantly improving the accuracy and robustness of the prediction input. In extreme scenarios of drastic weather changes or prediction system inaccuracies, the device can autonomously avoid the risk of incorrect voltage regulation caused by erroneous predictions, completely eradicating the gear oscillation risk caused by prediction failures in severe weather conditions, a problem inherent in traditional solutions.

[0012] In some embodiments, the method further includes mutual inspection of the operating status and dynamic fault tolerance between the protection and control CPU board and the edge computing board, specifically including: The protection and control CPU board and the edge computing board send heartbeat signals and operation status self-test summaries to each other at fixed intervals through the backplane bus. The self-test summaries include CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature. The protection and control CPU board determines in real time whether the edge computing board is in an abnormal state based on the self-test summary of the edge computing board. When the edge computing board is determined to be in an abnormal state, the protection and control CPU board sends a load reduction command to the edge computing board. In response to the load reduction command, the edge computing board immediately stops the digital twin virtual simulation and prediction model inference, and switches the simulation operation to a conservative simulation mode based on a preset fixed rate of change. When the protection and control CPU board fails to receive a response from the edge computing board for several consecutive heartbeat cycles, it is determined that the edge computing board has completely failed. The protection and control CPU board switches to a backup voltage regulation mode based on a single-dimensional criterion of instantaneous load rate and a backup early warning mode based on a fixed threshold of a single temperature parameter, and sends a failure alarm through the communication board until the edge computing board recovers and the operation and maintenance platform confirms that it has switched back to normal mode.

[0013] By adopting the above technical solution, a two-way heartbeat mutual inspection mechanism is established between the protection and control CPU board and the edge computing board, along with a two-stage hierarchical fault-tolerance strategy. This ensures that the basic protection, voltage regulation, and early warning functions of the transformer substation remain uninterrupted under any circumstances, including edge computing board failure, crashes, or restarts. This completely eliminates the potential risk of single-point failures brought about by intelligent upgrades, guaranteeing that the absolute safety baseline of the equipment remains intact. The crude logic of traditional embedded watchdog timeouts followed by immediate resets is upgraded to a state-aware hierarchical fault-tolerance mechanism, enabling the device to proactively perceive and report its own operational health.

[0014] In some embodiments, the protection and control CPU board determines in real time whether the edge computing board is in an abnormal state based on the self-test summary of the edge computing board, specifically including: The protection and control CPU board establishes a sliding time window for each operating status indicator of the edge computing board, records the historical change trajectory of each operating status indicator within the window, and uses least squares linear fitting to calculate the change rate of each operating status indicator. The operating status indicators include CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature. The protection and control CPU board normalizes and weights the current instantaneous value of each of the operating status indicators and the corresponding calculated change rate to generate an independent anomaly score for each of the operating status indicators, and then weights and sums the independent anomaly scores of all the operating status indicators to obtain the comprehensive degradation index of the edge computing board. The protection and control CPU board compares the comprehensive degradation index with preset mild degradation threshold, moderate degradation threshold and severe degradation threshold in sequence to determine the current degradation state, and performs a graded response according to the degradation state.

[0015] By adopting the above technical solution, a sliding time window is established for each operational status indicator, and the rate of change is calculated using least-squares linear fitting. This upgrades the evaluation dimension from the traditional single snapshot mode of whether the current instantaneous value exceeds the limit to a dual-dimensional weighted fusion evaluation mode of "current absolute value + historical trend". This can accurately distinguish between instantaneous load spikes and continuous performance degradation, effectively avoiding false triggering of load reduction due to accidental fluctuations, and achieving trend-level early warning of the health status of edge computing boards, significantly improving the accuracy and lead time of anomaly detection.

[0016] In some embodiments, the method further includes: The edge computing board uses the reference value as the current operating base point to calculate the expected changes in bus voltage, winding temperature and mechanical wear of on-load tap changer after this voltage regulation action, and adds the expected changes to the reference value to obtain the predicted values ​​of each operating parameter after this voltage regulation action. The predicted values ​​of each operating parameter are compared with the preset safe operating boundary range in each dimension. When the predicted value of any dimension exceeds the safe operating boundary range, it is determined that there is a risk of exceeding the limit in this voltage regulation action. In response to the over-limit risk determination, the edge computing board generates a voltage regulation suppression command containing an over-limit dimension identifier and an over-limit amplitude, and sends it to the protection and control CPU board through the backplane bus. The protection and control CPU board intercepts the issuance of this gear switching command according to the voltage regulation suppression command, and uploads alarm information carrying the over-limit dimension identifier to the site operation and maintenance platform. After intercepting the voltage regulation command, the protection and control CPU board starts a preset suppression and blocking timer. During the suppression and blocking timer, it continuously receives the updated benchmark reference value from the edge computing board. When the updated benchmark reference value shows that the predicted values ​​of each operating parameter have returned to the safe operating boundary range and the suppression and blocking timer ends, the voltage regulation suppression state is released and the normal issuance permission of the voltage regulation command is restored.

[0017] By adopting the above technical solution, using a reliable benchmark value verified by twins as a base, the three-dimensional expected impact of this voltage regulation on bus voltage, winding temperature, and tap changer mechanical wear is pre-calculated and compared dimension by dimension with the preset safe operating boundary. When any dimension's predicted value exceeds the limit, the voltage regulation is proactively intercepted. This upgrades the voltage regulation decision from a single-dimensional judgment of whether it meets grid requirements to a dual verification of whether it simultaneously meets grid requirements and equipment safety boundaries. This eliminates the potential for voltage regulation itself to become a cause of equipment damage, achieving proactive self-protection of the transformer substation.

[0018] In some embodiments, the transformer substation protection and control device is connected to the industrial ring network of the substation via a communication board, forming a distributed edge computing cluster with the edge computing boards of other transformer substations within the substation. The method further includes distributed collaborative voltage regulation of multiple transformer substations, specifically including: The edge computing board of the current transformer substation receives the local predicted load change trend and current voltage regulation action intention broadcast by the edge computing boards of adjacent transformer substations through the industrial ring network, and synchronously broadcasts the local comprehensive load change trend to the adjacent transformer substations. Each transformer's edge computing board selects one transformer as the master node within a preset collaborative time window based on the load forecast trends and voltage regulation intentions received from all adjacent transformers. The election weight is dynamically determined based on the proximity of each transformer's current load rate to the light load threshold. The master control node summarizes the load forecast data and current voltage status of all transformer substations. With the optimization goal of minimizing the total number of voltage regulation actions of all transformer substations and avoiding the time interval between voltage regulation actions of any two transformer substations being less than the preset safety interval, the master control node uniformly arranges the voltage regulation action sequence of each transformer substation within the collaborative time window and sends the arranged voltage regulation action sequence instructions to each subordinate transformer substation. Each subordinate transformer substation executes the voltage regulation action sequentially according to the voltage regulation action timing instructions. After the current transformer substation's edge computing board completes the local voltage regulation action, it feeds back the actual voltage regulation effect inferred by the digital twin to the master control node through the industrial ring network. The master control node dynamically adjusts the scheduling strategy of the next collaborative time window based on the feedback effect.

[0019] By adopting the above technical solution, a distributed edge computing cluster is constructed through an industrial ring network. Each transformer substation broadcasts local load forecasts and voltage regulation intentions, and dynamically elects a master control node based on the proximity of the load rate to the light load threshold. This fundamentally solves the problem of bus voltage superposition and grid oscillation caused by multiple transformer substations making independent judgments and potentially acting simultaneously in traditional solutions. At the same time, centralized scheduling transforms the dispersed and disordered voltage regulation behavior into an ordered queue, effectively smoothing the substation-level voltage regulation curve, improving the overall grid-connected power quality, and reducing the overall cumulative wear of the mechanical structures of multiple tap changers.

[0020] In a second aspect, embodiments of this application provide a computer system including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the steps of the method described in any possible implementation of the first aspect.

[0021] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method described in any possible implementation of the first aspect.

[0022] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method described in any possible implementation of the first aspect.

[0023] It is understood that the computer system provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided in this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. Construct a three-dimensional digital twin (electrical, thermal, and mechanical) at the edge, and fuse the multi-source heterogeneous parameters of the transformer substation in local memory in real time to form a synchronous multi-physics mapping of the physical transformer substation. Simultaneously, apply future load trends derived from time-series prediction models as input to the twin, enabling forward-looking projection of the transformer substation's future state response trends. This upgrades operational decisions from relying on historical data statistics to prediction based on physical models, providing more time for voltage regulation and protection strategies to respond in advance. 2. By comparing the inferred state value of the twin with the measured value at the same time dimension by dimension and calculating the deviation vector, when the deviation exceeds the tolerance threshold, the sensor is directly judged to be inaccurate or the equipment is potentially abnormal and an on-site warning is triggered. This enables the system to have the ability to evaluate the credibility of its own sensing data, effectively identify data quality problems such as sensor drift, zero-point offset, and sudden interference, avoid AI decision traps, and upgrade the traditional single-point over-threshold alarm to the prediction and actual measurement deviation alarm, significantly improving the sensitivity and lead time of anomaly detection. 3. Only when all dimensions of the deviation vector are less than the tolerance threshold will the twin-inferred value be sent back to the protection and control CPU board as a benchmark reference value for actively correcting the voltage regulation judgment logic and health warning threshold. This means that voltage regulation and warning no longer rely on the unverified original sensor instantaneous value, but use the twin-inferred value as the decision basis, realizing the fusion decision of physical model and measured data, which greatly improves the accuracy and reliability of voltage regulation action and health judgment. 4. By leveraging the physical separation and backplane bus interaction between the protection and control CPU board and the edge computing board, the edge computing board handles the computationally intensive twin simulation and predictive reasoning, while the protection and control CPU board focuses on executing voltage regulation decisions, protection logic, and command issuance optimized by twin verification. Neither board competes for resources, and the AI ​​algorithm's performance is optimized to the extreme without sacrificing the real-time performance of protection. Simultaneously, all calculations are completed locally at the edge, and all outputs originate from a trusted twin reference value that has undergone bidirectional verification, fundamentally eliminating dependence on cloud computing power and the site network. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the integrated method for local protection and control of transformer substations based on edge digital twins in this application embodiment; Figure 2 This is a flowchart illustrating the mutual inspection and dynamic fault tolerance of the operating status between the protection and control CPU board and the edge computing board in this embodiment of the application. Figure 3 This is a schematic diagram of an exemplary hardware structure of a computer system in an embodiment of this application. Detailed Implementation

[0026] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0027] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0028] The following is combined Figure 1 The method of the embodiments of this application will be described below.

[0029] Please see Figure 1 This is a flowchart illustrating the integrated method for on-site protection and control of transformer substations based on edge digital twins, as described in this application. Figure 1 As shown, the method includes the following steps: S101. The edge computing board receives multi-source operating parameters collected by the protection and control CPU board, and constructs and updates the digital twin of the transformer substation in local memory in real time based on the multi-source operating parameters. The multi-source operating parameters include electrical parameters, temperature parameters and vibration parameters. The digital twin includes an electrical image, a thermal image and a mechanical image. S102. The edge computing board, based on the updated digital twin, takes the current operating state as the initial condition, and infers the future load change trend based on the historical load sequence through a time-series prediction model. It then applies the corresponding future operating condition change amount to the digital twin and deduces the state response trend of the transformer substation under future operating conditions. S103. The edge computing board compares the inferred state value of the digital twin at the current moment with the measured multi-source operating parameters at the same moment, and calculates the deviation vector. S104. When the deviation values ​​of each dimension in the deviation vector are all less than the preset deviation tolerance threshold, the state of the digital twin is determined to be reliable, and the inferred state value of the digital twin is used as the benchmark reference value of the current operating state of the transformer. When the deviation value of any dimension in the deviation vector is greater than or equal to the preset deviation tolerance threshold, it is determined that there is a potential abnormality in the transformer or the sensor data is inaccurate, and an on-site early warning signal is triggered. S105. The edge computing board transmits the reference value or the local early warning signal back to the protection and control CPU board through the backplane bus. The protection and control CPU board corrects the current voltage regulation judgment logic and health early warning threshold according to the reference value, and generates and issues control commands in combination with the preset electrical protection logic.

[0030] This application is based on the hardware architecture of a transformer substation protection and control device. The device adopts a dual-board architecture with a protection and control CPU board and an edge computing board, which communicate with each other at high speed via a chassis backplane bus. The protection and control CPU board is equipped with a domestically produced industrial-grade main control chip, responsible for the overall aggregation of raw data collected by the entire device. The edge computing board is independently configured with a domestically produced industrial-grade chip, designed specifically for AI algorithms and data analysis. The protection and control CPU board acquires voltage and current signals from the high-voltage and low-voltage sides of the transformer substation in real time through an analog acquisition board, calculating electrical parameters such as voltage, current, active power, and reactive power. Simultaneously, the protection and control CPU board acquires multi-source data in real time through a non-electrical quantity acquisition expansion board, including transformer winding temperature, cable joint temperature (temperature parameter), enclosure vibration signal (vibration parameter), partial discharge signal, and ambient temperature and humidity. The non-electrical quantity acquisition expansion board natively integrates interface circuits and timing synchronization circuits for multiple types of sensors such as temperature, partial discharge, vibration, and ambient temperature and humidity, achieving native acquisition of multi-source data and hardware-level timing synchronization without the need for an external independent acquisition terminal. All collected electrical and non-electrical parameters are synchronously transmitted from the protection and control CPU board to the edge computing board via the backplane bus. Upon receiving the multi-source operating parameters, the edge computing board performs local data cleaning and anomaly / glitch removal (such as removing invalid data from instantaneous lightning strikes or electromagnetic interference) to ensure the validity of the input samples. Subsequently, the edge computing board constructs a lightweight digital twin of the transformer substation in its local memory. The digital twin includes three sub-models: electrical, thermal, and mechanical. The electrical twin maps the real-time distribution of voltage, current, and active / reactive power in the transformer substation windings, reflecting its electrical operating characteristics. The thermal twin maps the temperature field distribution and dynamic process of heat accumulation / dissipation within the transformer substation windings and enclosure, reflecting its thermodynamic behavior. The mechanical twin maps the vibration characteristics and mechanical wear accumulation of the on-load tap changer and enclosure structure, reflecting its mechanical dynamics. The edge computing board injects multi-source operating parameters into the digital twin in real time at a preset update cycle (which can be set to seconds or minutes), driving each sub-model to update its state variables synchronously, so that the digital twin and the physical transformer substation maintain a real-time state mirror image.

[0031] The edge computing board, based on historical active power data sequences (such as load data from the past 30 minutes) collected and forwarded by the protection and control CPU board, uses a lightweight time-series prediction model (such as a prediction model built based on a Temporal Convolutional Network (TCN) or a Long Short-Term Memory (LSTM) network) deployed on the edge computing board to infer the load change trend for future periods (such as the next 5-30 minutes). The prediction model has been trimmed and lightweighted for deployment on the edge computing board, adapting to low-computing-power edge hardware environments. The edge computing board quantifies the future load change trend inferred from the time-series prediction model into the corresponding future operating condition change, i.e., the expected value of load increment / decrease at future times relative to the current time. Then, using the current state of the digital twin as the initial condition, the edge computing board applies this future operating condition change as an external stimulus input to the three sub-models (electrical mirror, thermal mirror, and mechanical mirror) of the digital twin. The digital twin has a built-in simplified simulation engine based on physical mechanisms. Upon receiving excitations of future operating condition changes, the twin, based on the current state and physical model (including electrical equations, thermodynamic equations, and mechanical dynamics equations), extrapolates and calculates the state response trend of the prefabricated substation equipment under future operating conditions. Specifically: the electrical mirror extrapolates future voltage and current change trends; the thermal mirror extrapolates future winding temperature and enclosure temperature change trends based on the thermal resistance-thermal capacitance network; and the mechanical mirror extrapolates future vibration characteristics and cumulative mechanical wear trends. This extrapolation process is entirely completed in the local memory of the edge computing board, without relying on cloud computing power or the site network.

[0032] Because the multi-source operating parameters (electrical quantities, temperature, vibration) come from different acquisition channels, there is an inherent inconsistency in sampling times (e.g., electrical quantities at millisecond levels, temperature at second levels, vibration at microsecond levels). After receiving multi-source data, the edge computing board performs time-series alignment of the data from each channel using a unified time base. Specifically, this can be achieved by using a time-series synchronization circuit on the non-electrical quantity acquisition expansion board to assign a unified timestamp to all sensor data, or by using an interpolation alignment algorithm after unified time synchronization by the edge computing board. Time-series alignment ensures that the inferred state values ​​and measured values ​​participating in the comparison strictly correspond in time. The edge computing board extracts the inferred state values ​​of each dimension at the current moment from the digital twin, including the voltage, current, and power values ​​in the electrical mirror simulation, the temperature value in the thermal mirror simulation, and the vibration amplitude and cumulative mechanical wear value in the mechanical mirror simulation. The edge computing board compares the projected state value of the digital twin at the current moment with the multi-source operating parameters actually collected from the physical transformer substation at the same moment (i.e., the measured values ​​obtained in real time by the protection and control CPU board through various acquisition boards) dimension by dimension, calculates the deviation value of each dimension, and forms a deviation vector. Each component of the deviation vector corresponds to the difference between the projected value and the measured value of an operating parameter dimension (such as voltage deviation, temperature deviation, vibration deviation, etc.).

[0033] The deviation tolerance threshold is a pre-set upper limit for allowable deviations in each dimension. It can be determined based on the statistical distribution of historical normal operation data of the transformer substation or calculated in real time based on the uncertainty of the digital twin model. Different tolerance thresholds can be set for different dimensions (e.g., the temperature deviation tolerance threshold and voltage deviation tolerance threshold can be set independently). When the deviation values ​​in each dimension of the deviation vector are all less than the preset deviation tolerance threshold, it indicates that the inferred state value of the digital twin and the measured value of the physical entity are highly consistent in each dimension, that is, the mirror accuracy of the digital twin to the physical transformer substation meets the requirements. At this time, the edge computing board determines that the state of the digital twin is reliable and uses the inferred state value of the digital twin as the benchmark reference value of the current operating state of the transformer substation. The benchmark reference value integrates the dual information of the physical model mechanism and the measured data, and has higher reliability and robustness than a single measured value. When the deviation value in any dimension of the deviation vector is greater than or equal to the preset deviation tolerance threshold, it indicates that the inferred state of the digital twin and the measured state of the physical entity have deviated significantly in a certain dimension. This deviation may stem from two reasons: (1) the physical transformer itself has potential anomalies (such as local overheating of windings, insulation deterioration, loose mechanical structure, etc.), causing the actual operating state of the equipment to deviate from the normal trajectory based on the physical model; (2) the sensor data of the corresponding dimension is inaccurate (such as sensor drift, zero-point offset, sudden interference, etc.), resulting in unreliable measured values. Regardless of the reason, the edge computing board determines that there is an anomaly and triggers a local early warning signal. The early warning signal can be generated and stored locally on the edge computing board, and simultaneously transmitted back to the protection and control CPU board via the backplane bus for subsequent processing.

[0034] The edge computing board transmits the obtained reference value or local warning signal back to the protection and control CPU board via the chassis backplane bus (all boards are plugged into the chassis backplane slots, and the boards exchange data and signals through the internal backplane bus). When the protection and control CPU board receives the reference value (not the warning signal), it indicates that the digital twin's status is reliable. The protection and control CPU board uses this reference value to correct the current voltage regulation judgment logic. Specific correction methods include: replacing the original sensor instantaneous values ​​with the load rate and voltage value in the reference value as inputs to the voltage regulation criteria; for example, replacing the real-time load rate in the criterion that the real-time load rate is lower than the light load threshold with the reference reference load rate verified by the digital twin; using the reference value as a pre-verification basis for voltage regulation actions, allowing the voltage regulation command to be issued only when the reference value indicates that the voltage regulation conditions are met. Simultaneously, the protection and control CPU board uses the reference value to correct the health warning threshold. Since the benchmark reference value has been verified for reliability through twin-measurement bidirectional verification, the dynamic threshold calculated or corrected based on it has higher accuracy, effectively avoiding false alarms caused by instantaneous sensor fluctuations and missed alarms caused by sensor drift. The protection and control CPU board integrates basic logic operation units and conventional protection logic units (such as three-stage overcurrent protection, two-stage zero-sequence overcurrent protection, overload protection, two-stage undervoltage protection, two-stage overvoltage protection, etc.). After completing the correction of the voltage regulation judgment logic and health warning threshold, the protection and control CPU board, in conjunction with the preset electrical protection logic, comprehensively judges whether the voltage regulation condition or protection condition is met, and finally generates control commands. Control commands include, but are not limited to: tap-switching commands issued to the on-load tap changer (performing voltage regulation through the tap changer drive circuit of the switch output board); fault isolation commands issued to the trip circuit (performing fault protection tripping through the trip output circuit of the switch output board); and local alarm signals output to the alarm contacts. The control commands issued by the protection and control CPU board are entirely based on local calculation results (reference values ​​or local early warning signals), without relying on cloud computing power or field network. The device only uses data uploading and remote parameter configuration as auxiliary communication functions. Even in the event of a network outage, the device can still independently complete the entire process of voltage regulation determination, health assessment, and control command issuance.

[0035] In some embodiments, the step of constructing and updating the digital twin of the transformer substation in local memory in real time based on the multi-source operating parameters includes: The edge computing board records the trajectory of the deviation vector over time and uses the recursive least squares method to determine the current optimal estimates of the winding impedance parameters in the electrical mirror, the equivalent thermal resistance and thermal capacity parameters in the thermal mirror, and the contact wear coefficient in the mechanical mirror. The current optimal estimate is compared with the corresponding fixed parameters in the digital twin. When the relative deviation of any parameter exceeds the preset mismatch warning threshold, the model is determined to be mismatched. When a model mismatch is determined, the corresponding fixed parameters in the digital twin are replaced with the current optimal estimate, and the state variables of the digital twin are reinitialized with the replaced parameters. The changes in parameters before and after the update are encoded into a model state change log, which is then sent to the field operation and maintenance platform via the backplane bus and the protection and control CPU board.

[0036] The components of the deviation vector reflect the dynamic evolution of the difference between the digital twin's projected values ​​and the physical transformer's measured values. For example, a continuous increase in temperature deviation may indicate that the thermal mirror parameters deviate from the actual thermal characteristics, while a continuous change in voltage deviation may reflect changes in winding impedance parameters due to heating. Recording the change trajectory aims to provide a time-series data basis for parameter identification, enabling the system to track the historical evolution of deviations in each dimension, rather than relying solely on instantaneous deviation values ​​at a single moment. The edge computing board employs recursive least squares, using the historical trajectory of the deviation vector as observation data, to perform online optimal estimation of the core physical parameters of each sub-model in the digital twin. Recursive least squares is an adaptive filtering algorithm that recursively updates parameter estimates using the latest observation data at each moment. Its core idea is to minimize the cumulative sum of the squared errors between the model output and the actual observations, thereby obtaining the optimal estimate of the parameters under the current operating conditions. Compared to traditional least squares methods, which require processing all historical data at once, recursive least squares methods only need to store the parameter estimates and covariance matrix from the previous time step. They can then recursively update parameter estimates based on new observations at the current time step, resulting in extremely low computational and storage overhead, making them suitable for the limited memory and computing resources of edge computing boards. In the electrical mirror, winding impedance parameters determine the voltage drop and power loss characteristics of the transformer windings. As the operating temperature of the transformer changes and the winding insulation ages, the winding impedance parameters will slowly drift. Based on the output voltage, current, and power data from the electrical mirror, the recursive least squares method uses winding impedance as the parameter to be identified, aiming to minimize the deviation between the model-derived voltage and the measured voltage, and identifies the optimal estimate of the winding impedance under the current operating conditions online. In the thermal mirror, equivalent thermal resistance and thermal capacity are the core parameters determining the temperature field distribution and dynamic process of heat accumulation / dissipation within the transformer windings and enclosure. Equivalent thermal resistance reflects the resistance to heat transfer from the heat source (winding) to the environment, while equivalent thermal capacity reflects the ability of each component of the transformer to store heat. As equipment ages, insulation materials degrade, and heat dissipation conditions change (such as dust accumulation on heat sinks), the equivalent thermal resistance and thermal capacity parameters will change. The recursive least squares method, based on the deviation sequence between the temperature extrapolation value from the thermal image and the measured temperature value, identifies the optimal estimated values ​​of the equivalent thermal resistance and thermal capacity parameters under the current operating conditions online. In the mechanical image, the contact wear coefficient is a core characteristic parameter of the mechanical life of the on-load tap changer. It determines the cumulative damage to the switch contacts from each tap change operation, thus affecting the prediction accuracy of the remaining number of tap changer operations. As the actual number of tap changer operations accumulates, the contact wear coefficient will also change due to factors such as contact surface erosion and spring fatigue. The recursive least squares method, based on the deviation sequence between the vibration characteristic extrapolation value from the mechanical image and the measured vibration signal, identifies the optimal estimated value of the contact wear coefficient under the current operating conditions online.

[0037] When the digital twin is initially constructed, the core physical parameters (winding impedance, equivalent thermal resistance and thermal capacity, contact wear coefficient, etc.) within each sub-model are set based on the factory design values ​​or initial calibration values ​​of the transformer substation. These parameters are stored as fixed parameters. Fixed parameters represent the physical characteristic benchmark of the equipment in a brand-new state or initial operating conditions. As the transformer substation operates for a long time, factors such as equipment aging, component wear, and environmental changes will cause the actual physical parameters to gradually deviate from the fixed parameter values. The edge computing board compares the current optimal estimate of each parameter obtained by recursive least squares identification with the corresponding fixed parameters stored in the digital twin parameter by parameter. The comparison method is to calculate the relative deviation, that is, the absolute value of "(current optimal estimate - fixed parameter) / fixed parameter", which is expressed as a percentage to represent the degree of deviation of the current parameter from the initial benchmark value. The edge computing board pre-sets corresponding mismatch warning thresholds for each type of parameter (such as winding impedance mismatch warning threshold, equivalent thermal resistance mismatch warning threshold, equivalent thermal capacity mismatch warning threshold, contact wear coefficient mismatch warning threshold). When the relative deviation of any parameter exceeds its corresponding mismatch warning threshold, it indicates that the actual physical value of that parameter has significantly deviated from the parameter value currently used by the digital twin, and the digital twin's ability to represent the physical characteristics of that dimension has been severely distorted, i.e., model mismatch. Causes of model mismatch include, but are not limited to: slow aging due to long-term operation of the transformer substation, sudden parameter mutations caused by sudden faults (such as a drastic impedance change due to a partial short circuit in the winding), and thermal parameter shifts caused by drastic environmental changes. When the edge computing board determines model mismatch, it will trigger a parameter adaptive update mechanism. Specifically, it replaces the corresponding mismatched fixed parameter in the digital twin with the optimal estimate obtained at the current moment using the recursive least squares method. For example, when the relative deviation of the equivalent thermal resistance parameter exceeds the mismatch warning threshold, the latest optimal estimate of the equivalent thermal resistance obtained will be written into the thermal mirror sub-model of the digital twin, replacing the original fixed equivalent thermal resistance value. The replacement operation only applies to the parameter dimension determined to be mismatched; parameters that are not mismatched remain unchanged, avoiding unnecessary parameter disturbances. After parameter replacement, the underlying physical model of the digital twin has changed (e.g., a change in thermal resistance parameters will alter the steady-state distribution of the temperature field under the same heat source). At this point, the edge computing board reinitializes all state variables of the digital twin with the replaced parameter values. This means using the latest measured data as initial conditions, substituting it into the updated physical model to recalculate the initial values ​​of each state variable, ensuring the consistency and continuity of the twin's state after the parameter update. After reinitialization, the digital twin continues operating with the updated physical parameters and the current measured state, avoiding jumps in projected values ​​due to parameter mutations. The edge computing board encodes the changes before and after the parameter update (i.e., the difference between the updated optimal estimate and the previously fixed parameters, and the relative percentage deviation) into a model state change log according to a preset log format.The log content includes, but is not limited to: change timestamp, mismatch parameter name, fixed parameter value before update, optimal estimated value after update, relative deviation percentage, deviation threshold triggering mismatch judgment, and action number triggering the change. The purpose of log encoding is to transform the invisible adaptive adjustment behavior within the digital twin into structured, parsable, and storable data records. The edge computing board sends the encoded model state change log to the protection and control CPU board via the chassis backplane bus. The protection and control CPU board then uploads the log to the site maintenance platform or remote control center via the communication board (industrial ring network module, multi-protocol communication interface). After the model state change log is uploaded, site maintenance personnel can remotely view the parameter adaptation history of each transformer's digital twin. When the same parameter in the log undergoes frequent and significant updates, it can serve as an early indication of accelerated aging of the corresponding component (such as windings, cooling systems, tap changer contacts). For example, a continuous and gradual increase in the equivalent thermal resistance parameter may indicate that dust accumulation on the heat sink or aging of the insulation material is causing a decrease in heat dissipation performance; a significant increase in the contact wear coefficient may indicate accelerated tap changer contact ablation. Based on this log data, maintenance personnel can develop maintenance plans and prepare spare parts in advance.

[0038] In some embodiments, the inference of future load change trends based on historical load sequences using a time-series prediction model includes: Historical active power time series data are input into the first prediction model constructed based on a temporal convolutional network to deduce the ultra-short-term load change trend for the first preset time in the future. The predicted values ​​of light intensity and / or wind speed sent by the meteorological terminal, along with the historical load data at the corresponding time, are input into the second prediction model constructed based on a long short-term memory network to deduce the short-term load change trend for the second preset time in the future. The first preset time is shorter than the second preset time. The confidence interval widths corresponding to each predicted value in the load change trend are calculated using Bayesian approximation inference. The ultra-short-term load change trend and the short-term load change trend are weighted and superimposed in the same time domain to generate a comprehensive load change trend. The width of each confidence interval is compared with the preset confidence threshold in real time. If it is less than the confidence threshold, the corresponding prediction value is determined to be reliable. The corresponding prediction value is applied to the digital twin as the future operating condition change amount. If it is greater than or equal to the confidence threshold, the corresponding prediction value is discarded. The zero change amount of the load maintaining the current state is used as the future operating condition change amount. The cumulative number of consecutive abandonments is recorded. When the abandonment frequency threshold is exceeded within a preset time window, the fusion weight of the first prediction model is increased and the fusion weight of the second prediction model is decreased simultaneously. A voltage regulation strategy conservative instruction is sent to the protection and control CPU board to temporarily tighten the load change slope threshold to half of the preset load change slope threshold until the abandonment frequency returns to normal.

[0039] Temporal convolutional networks (TCNNs) are convolutional neural network architectures specifically designed for time-series data processing. They achieve parallel computing capabilities through causal convolution and dilated convolution, resulting in fast inference speeds and low computational resource consumption. In this application, the edge computing board obtains historical active power time-series data from the protection and control CPU board and inputs it into a first prediction model (built based on a temporal convolutional network) deployed on the edge computing board to deduce the ultra-short-term load change trend for the first preset time in the future (e.g., 5-15 minutes). This model relies only on the transformer's own historical load data and does not depend on external meteorological information, resulting in a fast response speed and suitability for ultra-short-term load prediction requirements for real-time voltage regulation decisions. Long Short-Term Memory (LSTM) networks are a variant of recurrent neural networks that excels at capturing long-term dependencies in time-series data, making them suitable for processing medium- to long-term time-series forecasts that include periodic patterns and external influencing factors. Photovoltaic power plants' meteorological terminals continuously send predicted solar irradiance values, and wind farms' meteorological terminals continuously send predicted wind speed values; these meteorological data are key external factors determining the output of new energy sources. The edge computing board inputs meteorological forecasts from the meteorological terminal along with historical load data for the corresponding time into the second prediction model to extrapolate short-term load change trends for a second preset time period (e.g., 30 minutes to 4 hours in the future). This model integrates historical load patterns and external meteorological factors, enabling it to capture load trend reversals caused by changes in sunlight / wind speed, but it involves relatively large computational loads and long inference times. The first preset time period is shorter than the second preset time period, meaning that ultra-short-term forecasts cover a smaller near-future period, while short-term forecasts cover a larger far-future period. Bayesian approximation inference is a method for approximating the posterior probability distribution under limited computing resources. The edge computing board uses Bayesian approximation inference to calculate the confidence interval width for each predicted load value output by the first and second prediction models. The confidence interval width reflects the degree of uncertainty of the predicted value: a narrower width indicates higher reliability of the model's prediction for that time (more concentrated prediction results), while a wider width indicates greater prediction uncertainty (more dispersed prediction results). For example, under stable, sunny conditions, the prediction confidence interval of the second prediction model is narrow; while under cloudy conditions with rapidly moving clouds, the drastic fluctuations in sunlight lead to a significant widening of the prediction confidence interval.

[0040] Ultra-short-term forecasts (the first forecasting model) have high accuracy but short coverage periods, while short-term forecasts (the second forecasting model) have long coverage periods but are more susceptible to meteorological uncertainties. The edge computing board weights and superimposes both models within the same time domain—that is, within the shared time interval, the ultra-short-term forecast is given a higher weight, and the short-term forecast a lower weight. Beyond the coverage of the ultra-short-term forecast, the short-term forecast takes over completely, generating a comprehensive load change trend covering the entire future period. The edge computing board compares the confidence interval width corresponding to each forecast value in the comprehensive load change trend with a preset confidence threshold in real time. When the confidence interval width is less than the confidence threshold, it indicates that the confidence level of the forecast value at that moment meets the requirements. The system then applies this forecast value as a future load condition change to the digital twin, driving the twin to extrapolate future state responses. For example, under stable, sunny conditions, forecast values ​​with narrower confidence intervals are continuously adopted, and the system drives the twin extrapolation entirely according to the forecast trend. When the confidence interval width is greater than or equal to the confidence threshold, it indicates that the uncertainty of the predicted value at that moment is too high (such as severe fluctuations in sunlight due to cloud cover, or inaccurate short-term weather forecasts). If the predicted value is continued to be adopted, it may cause the digital twin to extrapolate based on incorrect future operating conditions, thereby triggering incorrect voltage regulation. At this time, the edge computing board actively discards the predicted value and applies the zero change in the load to maintain the current state as the future operating condition change to the digital twin. That is, no forward adjustment is made to the current state, and the existing voltage regulation level remains unchanged until a high-confidence predicted value is obtained again at the next moment before making a decision. Multiple consecutive discards indicate that the meteorological conditions are fluctuating drastically in the current period, and the prediction results of the second prediction model (which depends on meteorological data) are continuously unreliable. The edge computing board records in real time the cumulative number of times the predicted value is continuously discarded within a preset time window (such as within the past 30 minutes). When the cumulative number exceeds the preset discard frequency threshold (such as more than 15 discards within 30 minutes), the system determines that the current operating condition is a severe weather condition with drastic changes.

[0041] At this point, the edge computing board performs dual proactive fault tolerance actions. First, it increases the fusion weight of the first prediction model (ultra-short-term prediction based on TCN, independent of meteorological data) while simultaneously decreasing the fusion weight of the second prediction model (short-term prediction based on LSTM, dependent on meteorological data). This ensures that the system prioritizes the more reliable ultra-short-term predictions during weighted summation, reducing the impact of unreliable short-term predictions on the overall trend. Second, it sends a conservative voltage regulation strategy command to the protection and control CPU board via the backplane bus, temporarily tightening the load change slope threshold to half of the preset value (i.e., from 1.2% / min to 0.6% / min). This means that under severe operating conditions with excessive abandonment frequency, even if the three-dimensional criteria for voltage regulation are met, the load change slope needs to be gentler than under normal operating conditions (less than the stricter threshold) to trigger voltage regulation, effectively avoiding unnecessary gear switching during periods of unreliable predictions due to drastic meteorological changes. When the frequency of abandonment returns to normal (e.g., the proportion of qualified predicted values ​​within the built-in signal interval increases for a continuous period), the system automatically restores the original fusion weight ratio and load change slope threshold, exits the conservative strategy, and re-enters the normal prediction-deduction-decision mode. This mechanism enables online adaptive adjustment of the prediction model weights and adaptive switching of the voltage regulation strategy across different scenarios. Under stable meteorological conditions, it fully leverages the forward-looking advantages of dual-model fusion prediction, while proactively adopting a conservative approach under severe meteorological conditions, thus balancing the foresight and safety of predictive voltage regulation.

[0042] In some embodiments, the method further includes mutual inspection of the operating status and dynamic fault tolerance between the protection and control CPU board and the edge computing board, specifically including: The protection and control CPU board and the edge computing board send heartbeat signals and operation status self-test summaries to each other at fixed intervals through the backplane bus. The self-test summaries include CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature. The protection and control CPU board determines in real time whether the edge computing board is in an abnormal state based on the self-test summary of the edge computing board. When the edge computing board is determined to be in an abnormal state, the protection and control CPU board sends a load reduction command to the edge computing board. In response to the load reduction command, the edge computing board immediately stops the digital twin virtual simulation and prediction model inference, and switches the simulation operation to a conservative simulation mode based on a preset fixed rate of change. When the protection and control CPU board fails to receive a response from the edge computing board for several consecutive heartbeat cycles, it is determined that the edge computing board has completely failed. The protection and control CPU board switches to a backup voltage regulation mode based on a single-dimensional criterion of instantaneous load rate and a backup early warning mode based on a fixed threshold of a single temperature parameter, and sends a failure alarm through the communication board until the edge computing board recovers and the operation and maintenance platform confirms that it has switched back to normal mode.

[0043] Both the protection and control CPU board and the edge computing board are independent standard rectangular plug-in boards, plugged into slots on the chassis backplane. A high-speed data exchange link is established between the two boards via the internal bus of the backplane. To achieve mutual operational status checks, the two boards send heartbeat signals and operational status self-check summaries to each other via the backplane bus at fixed intervals (e.g., every 100 milliseconds or every second). The heartbeat signal confirms that the other board is still operating normally; the operational status self-check summary is a quantitative assessment report of each board's own operational health status. The self-check summary specifically includes the following indicators: CPU utilization (current processor load percentage), remaining memory capacity (available memory size), critical task execution timeout count (e.g., whether AI inference tasks and data acquisition tasks timed out, and the number of timeouts), and board temperature (current board surface temperature). The critical task execution timeout count is implemented through the task scheduler built into the edge computing board. Each critical function APP is timestamped at the start of execution and its completion time is recorded after execution. When the task execution time exceeds the preset maximum allowed execution time, the timeout counter automatically increments. Board temperature is collected in real time by temperature sensors deployed on each board. CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature are the four key metrics that form the complete data foundation for judging the health status of the edge computing board. CPU utilization and remaining memory capacity reflect the board's load pressure, critical task execution timeout count reflects the software task's running status, and board temperature reflects the hardware operating environment. After receiving the self-test summary sent by the edge computing board, the protection and control CPU board performs real-time evaluation of various operating status metrics to comprehensively determine whether the edge computing board is in an abnormal state. The specific judgment logic includes: when CPU utilization is consistently too high (e.g., exceeding 90% for multiple cycles), remaining memory capacity is below the safety threshold (e.g., less than 10%), critical task execution timeout count increases rapidly within a unit of time, or board temperature exceeds the upper limit of safe operating temperature (e.g., exceeding 85℃), the protection and control CPU board determines that the edge computing board has entered an abnormal state. The determination of abnormal states adopts a comprehensive evaluation of multiple indicators, which can effectively avoid misjudgment caused by the instantaneous fluctuation of a single indicator. For example, when the peak CPU load exceeds 90% for a short period of time but immediately drops back, the abnormal state determination will not be triggered; the determination will only be triggered when the indicators continue to deviate and multiple indicators deteriorate at the same time.

[0044] After the protection and control CPU board determines that the edge computing board has entered an abnormal state, it immediately sends a load reduction command to the edge computing board via the backplane bus. The purpose of the load reduction command is to actively unload the current high-load computing tasks of the edge computing board, reduce its CPU utilization and heat generation, and avoid overheating and crashing or hardware damage caused by continuous high-load operation. Upon receiving the load reduction command, the edge computing board immediately stops running two types of high-computation-consuming tasks: digital twin virtual simulation and time-series prediction model inference, releasing the occupied CPU and memory resources. At the same time, the edge computing board switches the simulation operation to a conservative simulation mode based on a preset fixed rate of change. That is, it replaces the inference results of the time-series prediction model with a pre-set fixed rate of change (such as the load changing at a fixed rate of 0.5% / minute) as the input for the future operating conditions of the digital twin. The conservative simulation mode does not rely on the calculation of the time-series prediction model, and the computing power consumption is extremely low, but it sacrifices prediction accuracy and foresight. It is only used as a transitional strategy during the load reduction period to ensure that the basic simulation function of the twin is not interrupted. During the switching process, the historical projections already completed by the edge computing board remain unchanged; only the input source for future operating conditions is switched to ensure a smooth transition. When the edge computing board completely stops responding due to software deadlock, hardware failure, system crash, or unexpected power outage, its heartbeat signal is synchronously interrupted. The protection and control CPU board continuously monitors the heartbeat signal from the edge computing board. When no response is received from the edge computing board for several consecutive heartbeat cycles (e.g., 3 consecutive cycles, each 100 milliseconds, or 300 milliseconds; or 3 consecutive cycles, each 1 second, or 3 seconds), the protection and control CPU board determines that the edge computing board has completely failed. At this time, the protection and control CPU board performs a two-level backup switch: the first level of switch involves voltage regulation mode. The protection and control CPU board abandons all advanced voltage regulation logic (three-dimensional timing criteria, predictive voltage regulation, etc.) that relies on the edge computing board, and switches to a backup voltage regulation mode based on a single-dimensional criterion of instantaneous load rate. This involves a simple comparison between the current instantaneous load rate and a fixed light load threshold. When the instantaneous load rate is below the threshold, a derating / voltage reduction operation is performed; when it is above the cut-off threshold, a boosting / voltage increase operation is performed. This mode is the voltage regulation logic of traditional basic protection and control devices, with lower accuracy but simple implementation and reliable execution. The second-level switch involves an early warning mode. The protection and control CPU board abandons all advanced health assessment logic (digital twin simulation, multi-source data fusion, dynamic thresholds, etc.) that relies on the edge computing board, and switches to a backup early warning mode based on a fixed threshold for a single temperature parameter—that is, using a fixed threshold for the winding temperature as a single parameter for over-limit judgment. When the temperature exceeds the fixed threshold, an overheat alarm is triggered. Although this mode cannot comprehensively assess multi-dimensional health status, it ensures uninterrupted basic temperature safety monitoring.After the backup mode switch is completed, the protection and control CPU board sends an edge computing board failure alarm to the site maintenance platform via the communication board (industrial ring network module). The alarm content includes information such as the failure time, the number of consecutive unresponsive heartbeat cycles, and the type of backup mode switched, so that maintenance personnel can be remotely informed and respond promptly. When the edge computing board restarts or recovers after troubleshooting, the protection and control CPU board receives its heartbeat signal and self-test summary again. At this time, the two boards will not immediately and automatically switch back to normal mode. The system is designed so that the maintenance platform confirms and then manually issues a switchback command to prevent a hasty switchback that could lead to secondary failures if the edge computing board has restarted but is not yet fully ready (e.g., model loading is not complete, internal status is not synchronized). After the maintenance personnel confirm that the edge computing board is running stably, they issue a switchback command through the maintenance platform. The protection and control CPU board resumes the normal dual-board collaborative working mode, and the edge computing board restarts advanced functions such as digital twin construction and predictive model inference.

[0045] In some embodiments, the protection and control CPU board determines in real time whether the edge computing board is in an abnormal state based on the self-test summary of the edge computing board, specifically including: The protection and control CPU board establishes a sliding time window for each operating status indicator of the edge computing board, records the historical change trajectory of each operating status indicator within the window, and uses least squares linear fitting to calculate the change rate of each operating status indicator. The operating status indicators include CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature. The protection and control CPU board normalizes and weights the current instantaneous value of each of the operating status indicators and the corresponding calculated change rate to generate an independent anomaly score for each of the operating status indicators, and then weights and sums the independent anomaly scores of all the operating status indicators to obtain the comprehensive degradation index of the edge computing board. The protection and control CPU board compares the comprehensive degradation index with preset mild degradation threshold, moderate degradation threshold and severe degradation threshold in sequence to determine the current degradation state, and performs a graded response according to the degradation state.

[0046] For four indicators—CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature—the protection and control CPU board maintains a fixed-length time window for each indicator in its local memory (e.g., the window length is set to the most recent 60 sampling periods, with each sampling period being 1 second, thus covering the data of the most recent 60 seconds). The window uses a first-in, first-out (FIFO) update rule—each time a new set of self-test summary data is received, the new data is added to the end of the window, while the oldest historical data within the window is removed, ensuring that the window always focuses on the recent operational status changes of the edge computing board and avoids interference from outdated historical data. When a sufficient number of data samples have accumulated within the window (e.g., the window is full or the number of samples exceeds the preset minimum fitting sample size), the protection and control CPU board uses the least squares linear fitting method to perform linear regression on the historical data points of each indicator within the window, fitting an optimal straight line. The slope of this line represents the rate of change of that indicator in the current time period. For example, if the CPU utilization rate gradually increases from 60% to 85% within the window, the fitted change rate is positive and relatively large; if the remaining memory capacity continuously decreases within the window, the fitted change rate is negative; if the board temperature remains stable within the window, the fitted change rate is close to zero; and if the critical task execution timeout count continuously accumulates within the window, the fitted change rate remains positive. Because the dimensions and numerical ranges of various operational status indicators are different, they cannot be directly weighted and fused. The protection and control CPU board first normalizes the current instantaneous value and change rate separately, mapping them to a unified scoring range. Normalization uses preset upper and lower boundaries for mapping. For example, the normal range for CPU utilization is 0%~80%, exceeding 80% enters the danger zone, so normalization maps the current instantaneous CPU utilization rate to a score of 0~100; the change rate is mapped to the same scoring range based on the preset maximum allowable change rate. After normalization, the instantaneous value score and change rate score of each indicator fall within the same numerical range, possessing comparability and additivity.

[0047] For each operational status indicator, the normalized instantaneous value score and the normalized rate of change score are weighted and fused to generate an independent anomaly score for that indicator. The instantaneous value score reflects the absolute severity of the current state; for example, a CPU utilization rate of 95% indicates an extremely heavy load. The rate of change score reflects the rate of deterioration of the state; for example, although the CPU utilization rate is only 60%, it has been rising at a rate of 20 percentage points per minute over the past 60 seconds, indicating that it is about to reach the danger zone. After weighted fusion, the independent anomaly score can simultaneously reflect the current absolute level and the trend of change, achieving a dual assessment of the current situation and trend. The specific fusion method is: Independent Anomaly Score = α × Instantaneous Value Score + β × Rate of Change Score, where α and β are preset weighting coefficients that can be adjusted according to the characteristics of the indicator (e.g., the weight of the rate of change can be appropriately increased for indicators sensitive to temperature changes). The comprehensive degradation index of the edge computing board is obtained by weighted summing of the independent anomaly scores of all operational status indicators. The weights of CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature can be set separately according to the degree of impact of each indicator on the overall health of the edge computing board, with a total sum of 1. The comprehensive degradation index quantitatively characterizes the overall degradation degree of the edge computing board; a higher index indicates a worse overall health status of the board, while a lower index indicates a healthier operational status.

[0048] The protection and control CPU board is pre-configured with three progressively increasing threshold values ​​during factory or remote parameter configuration: a mild degradation threshold (e.g., comprehensive degradation index ≥ 30), a moderate degradation threshold (e.g., comprehensive degradation index ≥ 60), and a severe degradation threshold (e.g., comprehensive degradation index ≥ 85). These three thresholds classify the health status of the edge computing board into four levels: normal (comprehensive degradation index < mild degradation threshold), mild degradation (mild degradation threshold ≤ comprehensive degradation index < moderate degradation threshold), moderate degradation (moderate degradation threshold ≤ comprehensive degradation index < severe degradation threshold), and severe degradation (comprehensive degradation index ≥ severe degradation threshold). The protection and control CPU board executes the corresponding graded response strategy based on the determined degradation status. Normal status (comprehensive degradation index below the mild degradation threshold): This indicates that all operating status indicators of the edge computing board are within the normal range, requiring no intervention; the two boards continue to maintain normal collaborative operation. The protection and control CPU board continuously monitors the self-test summary and maintains a periodic status assessment cycle. Mild degradation state (overall degradation index between the mild and moderate degradation thresholds): This indicates that the edge computing board has experienced some performance degradation, but it is not yet severe. The protection and control CPU board records the degradation status information in its local log and sends it to the maintenance platform via the communication board. At this time, the system does not actively trigger load reduction operations to avoid unnecessary operating mode switching affecting normal functions. Moderate degradation state (overall degradation index between the moderate and severe degradation thresholds): This indicates that the edge computing board's performance continues to deteriorate and is approaching the danger zone. The protection and control CPU board performs active load reduction, sending a load reduction command to the edge computing board, requiring it to stop high-computing-power-consuming digital twin virtual simulation and prediction model inference, and switch to a conservative simulation mode to reduce CPU load and board temperature, thus slowing down the degradation trend. Severe degradation status (overall degradation index reaches or exceeds the severe degradation threshold): This indicates that the overall condition of the edge computing board is extremely poor. Continuing to maintain the current operating mode may lead to hardware damage or sudden complete failure of the board. The protection and control CPU board directly determines that the edge computing board is about to fail or has already failed completely, and switches to a backup voltage regulation mode based on a single-dimensional criterion of instantaneous load rate and a backup early warning mode based on a fixed threshold of a single temperature parameter. It also sends a severe failure alarm to the site operation and maintenance platform via the communication board. At this time, regardless of whether the heartbeat signal still exists, as long as the overall degradation index remains high, the protection and control CPU board will continue to operate in backup mode until the operation and maintenance personnel complete the on-site troubleshooting and repair and manually confirm the recovery.

[0049] In some embodiments, the method further includes: The edge computing board uses the reference value as the current operating base point to calculate the expected changes in bus voltage, winding temperature and mechanical wear of on-load tap changer after this voltage regulation action, and adds the expected changes to the reference value to obtain the predicted values ​​of each operating parameter after this voltage regulation action. The predicted values ​​of each operating parameter are compared with the preset safe operating boundary range in each dimension. When the predicted value of any dimension exceeds the safe operating boundary range, it is determined that there is a risk of exceeding the limit in this voltage regulation action. In response to the over-limit risk determination, the edge computing board generates a voltage regulation suppression command containing an over-limit dimension identifier and an over-limit amplitude, and sends it to the protection and control CPU board through the backplane bus. The protection and control CPU board intercepts the issuance of this gear switching command according to the voltage regulation suppression command, and uploads alarm information carrying the over-limit dimension identifier to the site operation and maintenance platform. After intercepting the voltage regulation command, the protection and control CPU board starts a preset suppression and blocking timer. During the suppression and blocking timer, it continuously receives the updated benchmark reference value from the edge computing board. When the updated benchmark reference value shows that the predicted values ​​of each operating parameter have returned to the safe operating boundary range and the suppression and blocking timer ends, the voltage regulation suppression state is released and the normal issuance permission of the voltage regulation command is restored.

[0050] Switching the tap position of an on-load tap changer alters the transformer's turns ratio, thus changing the low-voltage bus voltage. The edge computing board, based on the transformer turns ratio parameters in the electrical mirror sub-model and the current bus voltage reference value, calculates the voltage change step size corresponding to this tap position switch (up or down), i.e., the expected increase or decrease in bus voltage. After the voltage regulation action changes the bus voltage, the transformer's output power and winding current may change accordingly, affecting winding heat generation. The edge computing board, based on the thermal resistance-capacitance network parameters in the thermal mirror sub-model and the current winding temperature reference value, calculates the expected steady-state temperature change relative to the current temperature after voltage regulation. Each tap position switch operation causes cumulative wear on the tap changer's mechanical contacts. The edge computing board, based on the contact wear coefficient stored in the mechanical mirror sub-model and the current accumulated mechanical wear reference value, calculates the additional mechanical wear caused by this voltage regulation action. By superimposing the expected increments in the three dimensions onto the corresponding dimensions of the reference values, the predicted values ​​of each operating parameter after this voltage regulation action can be obtained. The safe operating boundary range refers to the safe operating range of various operating parameters of the transformer substation equipment. It is preset in the factory configuration or remote parameter configuration and is based on equipment design parameters, industry standards, and operation and maintenance procedures. Specifically, it includes: the allowable upper and lower limits of bus voltage (e.g., within ±5% or ±7% of the rated voltage), the upper limit of winding temperature (e.g., maximum allowable temperature of 105℃ for Class A insulation, 130℃ for Class B insulation, etc., determined according to the insulation class of the transformer substation), and the maximum allowable value of cumulative mechanical wear of the on-load tap changer (converted from the allowable number of operations throughout the switch's lifespan given by the switch manufacturer). The edge computing board compares the predicted values ​​of each operating parameter obtained in the previous step with the corresponding safe operating boundary ranges dimension by dimension: whether the predicted voltage value is between the upper and lower voltage limits; whether the predicted temperature value is lower than the upper temperature limit; and whether the predicted value of cumulative mechanical wear is lower than the maximum allowable wear. When all three predicted values ​​are within their respective safe boundary ranges, it indicates that the voltage regulation action is safe and can be executed. When the predicted value of any dimension exceeds the safety boundary range, such as the predicted voltage value after voltage regulation exceeding the upper limit of voltage, or the predicted temperature value exceeding the allowable temperature, or the predicted value of cumulative wear exceeding the maximum allowable value, the edge computing board determines that there is a risk of exceeding the limit for this voltage regulation action. That is, it has predicted that the voltage regulation will cause the equipment operating parameters to exceed the safety boundary even before the voltage regulation is actually executed.

[0051] When a voltage regulation action is determined to have an over-limit risk, the edge computing board generates a voltage regulation suppression command. This suppression command contains two key pieces of information: an over-limit dimension identifier (specifically indicating the dimension of the over-limit) and an over-limit amplitude (the predicted value exceeding the safety boundary by a specific value or percentage, facilitating assessment of the severity of the over-limit). The edge computing board sends the voltage regulation suppression command to the protection and control CPU board via the backplane bus. The protection and control CPU board had already generated the current gear switching command based on the voltage regulation judgment logic and was awaiting execution. Upon receiving the suppression command, the protection and control CPU board immediately intercepts the gear switching command, terminating the issuance process of this voltage regulation command, and the on-load tap changer will not perform any action. Simultaneously, the protection and control CPU board uploads the alarm information carrying the over-limit dimension identifier to the site operation and maintenance platform via the communication board, enabling maintenance personnel to remotely obtain the reason for the voltage regulation being intercepted and the over-limit dimension. The voltage regulation suppression in this application is a pre-emptive safety check and proactive interception performed before the actual execution of the voltage regulation action, its function being to prevent voltage regulation actions that would cause equipment damage from being executed. After the protection and control CPU board intercepts the voltage regulation command, it initiates a preset suppression and blocking timer (e.g., 30 or 60 minutes). During this timer, the protection and control CPU board continuously receives updated baseline reference values ​​from the edge computing board. Due to continuous changes in the operating conditions of the transformer substation (e.g., load decrease, temperature drop), the predicted values ​​that initially caused the limit exceedance may gradually return to the safe range. Restoring the normal issuance authority of the voltage regulation command requires two conditions to be met simultaneously: first, the updated baseline reference values ​​show that if the voltage regulation action is re-executed, the predicted values ​​of all operating parameters have returned to the safe operating boundary range, indicating that the current limit exceedance risk has been eliminated; second, the preset suppression and blocking timer has ended, indicating that a sufficient safety observation period has passed to prevent the operating conditions from deteriorating again after a brief return to normal, leading to a renewed limit exceedance after voltage regulation. When both conditions are met, the protection and control CPU board releases the voltage regulation suppression state, restores the normal issuance authority of the voltage regulation command, and the system re-enters the adjustable voltage state. If the predicted values ​​of all operating parameters have not all returned to the safe boundary range by the end of the suppression and blocking timer, the system continues to maintain the suppression state until the conditions are met or maintenance personnel intervene remotely. This mechanism not only prevents harmful voltage regulation from being repeatedly executed when the safety margin is insufficient, but also avoids the voltage regulation function from being locked for a long time due to a single over-limit judgment, ensuring that the grid voltage regulation demand can still be responded to in a timely manner after the safety conditions are met.

[0052] In some embodiments, the transformer substation protection and control device is connected to the industrial ring network of the substation via a communication board, forming a distributed edge computing cluster with the edge computing boards of other transformer substations within the substation. The method further includes distributed collaborative voltage regulation of multiple transformer substations, specifically including: The edge computing board of the current transformer substation receives the local predicted load change trend and current voltage regulation action intention broadcast by the edge computing boards of adjacent transformer substations through the industrial ring network, and synchronously broadcasts the local comprehensive load change trend to the adjacent transformer substations. Each transformer's edge computing board selects one transformer as the master node within a preset collaborative time window based on the load forecast trends and voltage regulation intentions received from all adjacent transformers. The election weight is dynamically determined based on the proximity of each transformer's current load rate to the light load threshold. The master control node summarizes the load forecast data and current voltage status of all transformer substations. With the optimization goal of minimizing the total number of voltage regulation actions of all transformer substations and avoiding the time interval between voltage regulation actions of any two transformer substations being less than the preset safety interval, the master control node uniformly arranges the voltage regulation action sequence of each transformer substation within the collaborative time window and sends the arranged voltage regulation action sequence instructions to each subordinate transformer substation. Each subordinate transformer substation executes the voltage regulation action sequentially according to the voltage regulation action timing instructions. After the current transformer substation's edge computing board completes the local voltage regulation action, it feeds back the actual voltage regulation effect inferred by the digital twin to the master control node through the industrial ring network. The master control node dynamically adjusts the scheduling strategy of the next collaborative time window based on the feedback effect.

[0053] Each transformer substation's protection and control device connects to the station's fiber optic industrial Ethernet ring network via its communication board's industrial ring network module. High-speed data exchange channels (millisecond latency) are established between the edge computing boards of each substation through this ring network, forming a distributed edge computing cluster. This means each substation is both an independent decision-making intelligent terminal and a node in the collaborative network. Information sharing is divided into two directions: receiving and sending. Receiving direction: The current substation's edge computing board continuously monitors the industrial ring network, receiving local information broadcast by the edge computing boards of adjacent substations, including local predicted load change trends and current voltage regulation intentions. Sending direction: The current substation's edge computing board synchronously broadcasts the comprehensive load change trend calculated locally to adjacent substations. Through this two-way broadcasting mechanism, each substation in the cluster can obtain load prediction information and voltage regulation intentions from all other substations, providing a complete information foundation for subsequent collaborative decision-making. After information sharing is completed, each substation initiates a master control node election within a preset collaborative time window. The election process does not rely on a fixed master control device or external server, but is completed through distributed negotiation. Each transformer substation independently performs election calculations based on the same information set and the same election rules. Therefore, the election results from each transformer substation are consistent, eliminating the need for additional voting confirmation. The core election rule dynamically determines the election weight based on the proximity of each transformer substation's current load rate to the light load threshold. Specifically, the transformer substation with the current load rate closest to the light load threshold receives the highest election weight. The logic behind this rule is that the transformer substation with the load rate closest to the light load threshold is most likely to trigger voltage regulation in the future. A load rate slightly below the light load threshold requires capacity / voltage reduction, while a load rate slightly above the light load threshold requires capacity / voltage increase. Electing this most urgent transformer substation as the master control node, which then uniformly orchestrates the voltage regulation timing within the coordination window, ensures that subsequent voltage regulation coordination decisions are led by the transformer substation that best understands the current voltage regulation needs. This avoids decision delays or mismatches caused by electing a transformer substation with no current voltage regulation needs as the master control node. The election process is executed independently within each collaborative time window. When the window ends, the master node's identity automatically expires, and a new election is held in the next window.

[0054] The master control node in this application is dynamically elected by each transformer substation, and its identity is not fixed. Furthermore, the master control node itself is still an edge computing board of a transformer substation, performing its own local measurement and control tasks. The master control function and local functions reuse the same hardware resources. This design avoids the risk of single-point failure, as any transformer substation can be elected as the master control node. The edge computing board of the transformer substation elected as the master control node aggregates the load forecast data and current tap status of all transformer substations. Based on this, it performs voltage regulation action timing arrangement within the coordinated time window. The arrangement is based on the following two optimization objectives: Objective 1 is to minimize the total number of voltage regulation actions of all transformer substations. When multiple transformer substations meet their respective voltage regulation conditions within a similar time period, the master control node assesses whether some transformer substations can temporarily postpone voltage regulation, meeting the global voltage regulation requirements through load transfer or reactive power compensation within the station. Minimizing the total number of actions reduces the cumulative mechanical wear of tap changers throughout the station, extending the overall equipment lifespan. Objective 2 is to avoid the time interval between voltage regulation actions of any two transformer substations being less than the preset safety interval. If multiple transformer substations operate consecutively within a very short time interval (e.g., two substations switch voltages simultaneously or successively within hundreds of milliseconds), the voltage regulation effects of each substation will be superimposed on the bus side, causing cumulative impacts and repeated oscillations in the bus voltage. The master control node ensures that the voltage regulation actions of each substation are evenly staggered on the time axis by setting a preset safety interval as a constraint (e.g., at least 5 or 10 seconds between any two voltage regulation actions). The master control node jointly solves the two optimization objectives, generating a timing arrangement of the voltage regulation actions of each subordinate substation within a preset collaborative time window, thus clarifying when and what voltage regulation action each substation will perform. After the arrangement is completed, the master control node sends the timing instructions one by one to the corresponding subordinate substation edge computing boards through the industrial ring network. After receiving the instructions, each subordinate substation confirms the receipt, and the master control node confirms that all instructions have been delivered, preparing to enter the execution phase.

[0055] After receiving the voltage regulation action sequence command, each subordinate transformer substation executes the voltage regulation action sequentially according to the scheduled time nodes. Since the timing of each substation is staggered, the time interval between any two substations is no less than the preset safety interval, avoiding the impact of multiple voltage regulation actions on the bus voltage. After the current substation's edge computing board completes its local voltage regulation action, it uses the constructed digital twin to deduce the actual effect of this voltage regulation, including the actual change in bus voltage after regulation, the actual response of winding temperature, and the actual contribution of this action to the mechanical wear of the tap changer. This deduced effect is fed back to the master control node through the industrial ring network. After receiving the actual voltage regulation effects from all subordinate transformer substations, the master control node compares and analyzes them with the expected voltage regulation effects during scheduling, and dynamically adjusts the scheduling strategy for the next coordinated time window based on the direction and magnitude of the deviation. Specific correction methods include: if the actual response speed of a transformer substation is slow (the time from command issuance to action completion exceeds expectations), its subsequent action sequence will be appropriately advanced to compensate for the response delay; if the voltage change of a transformer substation after voltage regulation is greater than expected, its single tap-switching step size will be reduced or its priority in participating in coordinated voltage regulation will be lowered; if the temperature overshoot of a transformer substation after execution exceeds expectations, its action sequence will be postponed in the next window scheduling to give it more time for heat dissipation and recovery. Through this closed-loop mechanism of "scheduling-execution-feedback-correction", the coordinated strategy is continuously iterated and optimized according to the actual response characteristics of each transformer substation, load fluctuation patterns, and equipment aging status, achieving adaptive evolution of the voltage regulation order of the entire station.

[0056] The above describes the integrated method for local protection and control of transformer substations based on edge digital twins in the embodiments of this application. The computer system in the embodiments of this application will be described in detail below in conjunction with the above integrated method for local protection and control of transformer substations based on edge digital twins.

[0057] Please see Figure 3 This is a schematic diagram of an exemplary hardware structure of a computer system in an embodiment of this application.

[0058] In some embodiments, the computer system 300 includes a computer device, which may be a terminal device. The computer device includes a processor 301, a memory 302, a sensor module 303, a communication module 304, an input device 305, and an output device 306 connected via a system bus. The processor 301 of the computer device provides computing and control capabilities. The memory 302 of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database is used to store data.

[0059] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0060] In some embodiments of this application, a computer-readable storage medium is provided, including instructions that, when executed on the computer system 300, cause the computer system 300 to execute the integrated method for local protection and control of transformer substation based on edge digital twins as described in this application.

[0061] In some embodiments of this application, a computer program product is also provided, which, when running on a computer system 300, causes the computer system 300 to execute the integrated method for local protection and control of transformer substation based on edge digital twins as described in this application.

[0062] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0063] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0064] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for integrated local protection and control of a transformer substation based on edge digital twin, characterized in that, This is applied to a transformer substation protection and control device, which includes a protection and control CPU board and an edge computing board that communicate via a backplane bus, and includes: The edge computing board receives multi-source operating parameters collected by the protection and control CPU board, and constructs and updates the digital twin of the transformer substation in local memory in real time based on the multi-source operating parameters. The multi-source operating parameters include electrical parameters, temperature parameters and vibration parameters. The digital twin includes an electrical image, a thermal image and a mechanical image. The edge computing board, based on the updated digital twin, takes the current operating state as the initial condition, and infers the future load change trend based on the historical load sequence through a time-series prediction model. It then applies the corresponding future operating condition change amount to the digital twin and predicts the state response trend of the transformer substation under future operating conditions. The edge computing board compares the inferred state value of the digital twin at the current moment with the measured multi-source operating parameters at the same moment, and calculates the deviation vector. When the deviation values ​​of each dimension in the deviation vector are all less than the preset deviation tolerance threshold, the state of the digital twin is determined to be reliable, and the inferred state value of the digital twin is used as the benchmark reference value of the current operating state of the transformer. When the deviation value of any dimension in the deviation vector is greater than or equal to the preset deviation tolerance threshold, it is determined that there is a potential anomaly in the transformer or that the sensor data is inaccurate, and an on-site early warning signal is triggered. The edge computing board transmits the reference value or the local early warning signal back to the protection and control CPU board through the backplane bus. The protection and control CPU board corrects the current voltage regulation judgment logic and health early warning threshold according to the reference value, and generates and issues control commands in combination with the preset electrical protection logic.

2. The method according to claim 1, characterized in that, The digital twin of the transformer substation, constructed and updated in real time in local memory based on the multi-source operating parameters, includes: The edge computing board records the trajectory of the deviation vector over time and uses the recursive least squares method to determine the current optimal estimates of the winding impedance parameters in the electrical mirror, the equivalent thermal resistance and thermal capacity parameters in the thermal mirror, and the contact wear coefficient in the mechanical mirror. The current optimal estimate is compared with the corresponding fixed parameters in the digital twin. When the relative deviation of any parameter exceeds the preset mismatch warning threshold, the model is determined to be mismatched. When a model mismatch is determined, the corresponding fixed parameters in the digital twin are replaced with the current optimal estimate, and the state variables of the digital twin are reinitialized with the replaced parameters. The changes in parameters before and after the update are encoded into a model state change log, which is then sent to the field operation and maintenance platform via the backplane bus and the protection and control CPU board.

3. The method according to claim 1, characterized in that, The future load change trend inferred from historical load sequences using a time-series prediction model includes: Historical active power time series data are input into the first prediction model constructed based on a temporal convolutional network to deduce the ultra-short-term load change trend for the first preset time in the future. The predicted values ​​of light intensity and / or wind speed sent by the meteorological terminal, along with the historical load data at the corresponding time, are input into the second prediction model constructed based on a long short-term memory network to deduce the short-term load change trend for the second preset time in the future. The first preset time is shorter than the second preset time. The confidence interval widths corresponding to each predicted value in the load change trend are calculated using Bayesian approximation inference. The ultra-short-term load change trend and the short-term load change trend are weighted and superimposed in the same time domain to generate a comprehensive load change trend. The width of each confidence interval is compared with the preset confidence threshold in real time. If it is less than the confidence threshold, the corresponding prediction value is determined to be reliable. The corresponding prediction value is applied to the digital twin as the future operating condition change amount. If it is greater than or equal to the confidence threshold, the corresponding prediction value is discarded. The zero change amount of the load maintaining the current state is used as the future operating condition change amount. The cumulative number of consecutive abandonments is recorded. When the abandonment frequency threshold is exceeded within a preset time window, the fusion weight of the first prediction model is increased and the fusion weight of the second prediction model is decreased simultaneously. A voltage regulation strategy conservative instruction is sent to the protection and control CPU board to temporarily tighten the load change slope threshold to half of the preset load change slope threshold until the abandonment frequency returns to normal.

4. The method according to claim 1, characterized in that, The method also includes mutual inspection of the operating status and dynamic fault tolerance between the protection and control CPU board and the edge computing board, specifically including: The protection and control CPU board and the edge computing board send heartbeat signals and operation status self-test summaries to each other at fixed intervals through the backplane bus. The self-test summaries include CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature. The protection and control CPU board determines in real time whether the edge computing board is in an abnormal state based on the self-test summary of the edge computing board. When the edge computing board is determined to be in an abnormal state, the protection and control CPU board sends a load reduction command to the edge computing board. In response to the load reduction command, the edge computing board immediately stops the digital twin virtual simulation and prediction model inference, and switches the simulation operation to a conservative simulation mode based on a preset fixed rate of change. When the protection and control CPU board fails to receive a response from the edge computing board for several consecutive heartbeat cycles, it is determined that the edge computing board has completely failed. The protection and control CPU board switches to a backup voltage regulation mode based on a single-dimensional criterion of instantaneous load rate and a backup early warning mode based on a fixed threshold of a single temperature parameter, and sends a failure alarm through the communication board until the edge computing board recovers and the operation and maintenance platform confirms that it has switched back to normal mode.

5. The method according to claim 4, characterized in that, The protection and control CPU board determines in real time whether the edge computing board is in an abnormal state based on the self-test summary of the edge computing board, specifically including: The protection and control CPU board establishes a sliding time window for each operating status indicator of the edge computing board, records the historical change trajectory of each operating status indicator within the window, and uses least squares linear fitting to calculate the change rate of each operating status indicator. The operating status indicators include CPU utilization, remaining memory capacity, critical task execution timeout count, and board temperature. The protection and control CPU board normalizes and weights the current instantaneous value of each of the operating status indicators and the corresponding calculated change rate to generate an independent anomaly score for each of the operating status indicators, and then weights and sums the independent anomaly scores of all the operating status indicators to obtain the comprehensive degradation index of the edge computing board. The protection and control CPU board compares the comprehensive degradation index with preset mild degradation threshold, moderate degradation threshold and severe degradation threshold in sequence to determine the current degradation state, and performs a graded response according to the degradation state.

6. The method according to claim 1, characterized in that, The method further includes: The edge computing board uses the reference value as the current operating base point to calculate the expected changes in bus voltage, winding temperature and mechanical wear of on-load tap changer after this voltage regulation action, and adds the expected changes to the reference value to obtain the predicted values ​​of each operating parameter after this voltage regulation action. The predicted values ​​of each operating parameter are compared with the preset safe operating boundary range in each dimension. When the predicted value of any dimension exceeds the safe operating boundary range, it is determined that there is a risk of exceeding the limit in this voltage regulation action. In response to the over-limit risk determination, the edge computing board generates a voltage regulation suppression command containing an over-limit dimension identifier and an over-limit amplitude, and sends it to the protection and control CPU board through the backplane bus. The protection and control CPU board intercepts the issuance of this gear switching command according to the voltage regulation suppression command, and uploads alarm information carrying the over-limit dimension identifier to the site operation and maintenance platform. After intercepting the voltage regulation command, the protection and control CPU board starts a preset suppression and blocking timer. During the suppression and blocking timer, it continuously receives the updated benchmark reference value from the edge computing board. When the updated benchmark reference value shows that the predicted values ​​of each operating parameter have returned to the safe operating boundary range and the suppression and blocking timer ends, the voltage regulation suppression state is released and the normal issuance permission of the voltage regulation command is restored.

7. The method according to claim 1, characterized in that, The transformer substation protection and control device is connected to the industrial ring network of the substation via a communication board, forming a distributed edge computing cluster with the edge computing boards of other transformer substations within the substation. The method also includes distributed collaborative voltage regulation of multiple transformer substations, specifically including: The edge computing board of the current transformer substation receives the local predicted load change trend and current voltage regulation action intention broadcast by the edge computing boards of adjacent transformer substations through the industrial ring network, and synchronously broadcasts the local comprehensive load change trend to the adjacent transformer substations. Each transformer's edge computing board selects one transformer as the master node within a preset collaborative time window based on the load forecast trends and voltage regulation intentions received from all adjacent transformers. The election weight is dynamically determined based on the proximity of each transformer's current load rate to the light load threshold. The master control node summarizes the load forecast data and current voltage status of all transformer substations. With the optimization goal of minimizing the total number of voltage regulation actions of all transformer substations and avoiding the time interval between voltage regulation actions of any two transformer substations being less than the preset safety interval, the master control node uniformly arranges the voltage regulation action sequence of each transformer substation within the collaborative time window and sends the arranged voltage regulation action sequence instructions to each subordinate transformer substation. Each subordinate transformer substation executes the voltage regulation action sequentially according to the voltage regulation action timing instructions. After the current transformer substation's edge computing board completes the local voltage regulation action, it feeds back the actual voltage regulation effect inferred by the digital twin to the master control node through the industrial ring network. The master control node dynamically adjusts the scheduling strategy of the next collaborative time window based on the feedback effect.

8. A computer system comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images