A control method, device and equipment of a transformer

Abstract

This invention discloses a transformer control method, apparatus, and equipment, relating to the field of transformer control technology. The method includes: obtaining a multi-dimensional heterogeneous fused state vector from digital monitoring data collected using preset sensors; using a digital twin model of the transformer to obtain the current health status assessment result and the predicted trajectory of key state parameters within a prediction time window; determining the dynamic safety domain boundary within the prediction time window based on the current health status assessment result, the predicted trajectory of key state parameters, and grid fluctuation tolerance information; determining the transformer's operating status based on the multi-dimensional heterogeneous fused state vector, the predicted trajectory of key state parameters, and the dynamic safety domain boundary; and executing corresponding protection actions when the operating status is the target state. This invention improves the accuracy of transformer control, promptly identifies and warns of latent faults, reduces protection malfunctions, and ensures the safe and reliable operation of the power system.

Description

Technical Field

[0001] This invention relates to the field of transformer control technology, and in particular to a transformer control method, device, and equipment. Background Technology

[0002] With the continued deepening of the global energy transition and the diversification and intelligentization of electricity consumption behavior on the user side, the operating conditions of distribution transformers, as key nodes connecting the power grid and end users, have become extremely complex. The power flow direction of the lines they are on can frequently reverse bidirectionally within milliseconds to seconds, and the voltage level also fluctuates drastically. This poses unprecedented challenges to the protection and control of transformers.

[0003] In related technologies, transformer protection schemes mainly rely on real-time monitoring of transformer electrical quantities (such as voltage and current) and fault identification and isolation based on preset fixed thresholds or offline-tuned characteristic curves. This mode is prone to protection malfunctions when facing short-term inrush currents or power backflows caused by new energy sources and flexible loads, which may exceed the set value but are normal operating conditions. This can unnecessarily disconnect normally operating transformers and damage power supply reliability. Conversely, for some latent faults such as insulation overheating caused by the slow accumulation of minor abnormalities, the response may be slow, resulting in the failure to identify the fault in time and its expansion, leading to serious equipment damage accidents.

[0004] Therefore, improving the accuracy of transformer control, timely identifying and providing early warning of latent faults, and reducing the triggering of protection malfunctions are urgent problems that need to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a transformer control method, device, and equipment to improve the accuracy of transformer control, promptly identify and provide early warning protection for latent faults, reduce the triggering of protection malfunctions, and thus ensure the safe and reliable stable operation of the power system.

[0006] To solve the above-mentioned technical problems, the present invention provides a transformer control method, comprising: Based on the digital monitoring data collected by preset sensors, a multidimensional heterogeneous fusion state vector is obtained at the current moment; wherein, the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data; Based on the multidimensional heterogeneous fused state vector, the transformer digital twin model is used to obtain the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window; wherein, the prediction time window includes the current moment and a future preset time period; the predicted trajectory of key state parameters includes at least one of the predicted values ​​of the transformer winding hottest spot temperature, insulation aging rate, and dissolved gas generation rate in oil. Based on the current health status assessment results, the predicted trajectory of the key status parameters, and the power grid fluctuation tolerance information, the dynamic safety domain boundary within the prediction time window is determined; Based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of the key state parameters, and the dynamic safety domain boundary, the operating status of the transformer is determined, and when the operating status is the target state, corresponding protection actions are executed; wherein, the target state includes abnormal state and state with risk.

[0007] On the other hand, the step of obtaining the multidimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by preset sensors includes: Based on the timestamp information in the digital monitoring data, the digital monitoring data is time-aligned to obtain time-aligned digital monitoring data. Electrical feature extraction is performed on the electrical quantity data in the time-aligned digital monitoring data to obtain electrical features; wherein, the electrical feature extraction operation includes at least one of effective value calculation, harmonic component extraction and phase calculation; A state feature extraction operation is performed on the non-electrical quantity data in the time-aligned digital monitoring data to obtain state features; wherein, the state feature extraction operation includes rate of change calculation and / or trend analysis; The electrical features and the state features are combined and encapsulated to obtain the multidimensional heterogeneous fused state vector at the current moment.

[0008] On the other hand, the preset sensor includes electrical quantity sensors and non-electrical quantity sensors. The electrical quantity sensors include voltage sensors and / or current sensors, and the non-electrical quantity sensors include at least one of dissolved gas in oil sensors, top oil temperature sensors, and / or oil level sensors.

[0009] On the other hand, the step of obtaining the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window based on the multidimensional heterogeneous fused state vector and using the transformer digital twin model includes: The model input driving quantity is parsed from the multidimensional heterogeneous fused state vector; wherein, the model input driving quantity includes electrical quantity data components and non-electrical quantity data components; Using the physical calculation kernel of the transformer digital twin model, the input driving quantity of the model is forward extrapolated to obtain the predicted trajectory of key state parameters corresponding to the extrapolation results at each discrete time point within the prediction time window; wherein, the physical calculation kernel is a set of mechanism equations established based on the physical laws of heat conduction, the aging reaction kinetics of insulating materials, and the dissolution and release mechanism of gas in oil.

[0010] On the other hand, the physical calculation kernel of the transformer digital twin model is used to perform forward extrapolation calculations on the input driving quantity of the model to obtain the predicted trajectory of key state parameters corresponding to the extrapolation results at each discrete time point within the prediction time window, including: Using the physical computing kernel, the input driving quantity of the model is forward extrapolated and calculated, and the extrapolation result is calibrated using the data-driven correction module of the transformer digital twin model, so as to obtain the predicted trajectory of key state parameters corresponding to the calibrated extrapolation result at each discrete time point within the prediction time window.

[0011] On the other hand, the step of obtaining the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window based on the multidimensional heterogeneous fused state vector and using the transformer digital twin model includes: The current health status assessment result of the transformer is determined based on each feature vector in the multidimensional heterogeneous fusion state vector and its corresponding preset health benchmark quantity.

[0012] On the other hand, the predicted trajectory of the key state parameters includes the predicted temperature of the hottest spot in the transformer winding. Determining the dynamic safety domain boundary within the prediction time window based on the current health status assessment results, the predicted trajectory of the key state parameters, and grid fluctuation tolerance information includes: Based on the predicted temperature of the hottest spot in the transformer winding and the current health status assessment result, a dynamic current action threshold curve is determined within the prediction time window. The dynamic current action threshold curve is the curve corresponding to the maximum allowable load current value at each discrete time point within the prediction time window, provided that the predicted temperature of the hottest spot in the transformer winding does not exceed its respective safety upper limit. The safety upper limit is the predicted temperature of the hottest spot in the transformer winding at the corresponding discrete time point or a corrected absolute safety upper limit, wherein the corrected absolute safety upper limit is a value determined based on the current health status assessment result and a preset absolute safety upper limit. Based on the dynamic current action threshold curve and the power grid fluctuation tolerance information, the dynamic current action threshold curve is adjusted to obtain an adjusted dynamic current action threshold curve; wherein, the dynamic safety domain boundary includes the adjusted dynamic current action threshold curve.

[0013] On the other hand, determining the operating status of the transformer based on the multidimensional heterogeneous fused state vector, the predicted trajectory of the key state parameters, and the dynamic safety domain boundary includes: Each feature vector in the multidimensional heterogeneous fusion state vector is compared with the threshold value at the current moment on the corresponding curve in the dynamic security domain boundary. If there is a feature vector greater than the corresponding threshold, then the operating condition is determined to be the abnormal state; If there is no feature vector greater than the corresponding threshold, the predicted trajectory of the key state parameter is compared with the boundary of the dynamic safety domain to determine whether there is a predicted value greater than the corresponding threshold in the predicted trajectory of the key state parameter within the future preset time period. If there is a predicted value greater than the corresponding threshold, then the operating condition is determined to be a state of risk. If there is no predicted value greater than the corresponding threshold, then the operating condition is determined to be normal.

[0014] The present invention also provides a control device for a transformer, comprising: The data acquisition module is used to obtain a multi-dimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by the preset sensors; wherein, the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data; The trajectory prediction module is used to obtain the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window based on the multidimensional heterogeneous fused state vector and the transformer digital twin model; wherein, the prediction time window includes the current moment and a future preset time period; the predicted trajectory of key state parameters includes at least one of the predicted values ​​of the transformer winding hotspot temperature, insulation aging rate, and dissolved gas generation rate in oil. The boundary determination module is used to determine the dynamic safety domain boundary within the prediction time window based on the current health status assessment results, the predicted trajectory of the key status parameters, and the power grid fluctuation tolerance information. The operation protection module is used to determine the operating status of the transformer based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of the key state parameters, and the dynamic safety domain boundary, and to execute corresponding protection actions when the operating status is the target state; wherein, the target state includes abnormal state and state with risk.

[0015] Furthermore, the present invention also provides a control device for a transformer, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the transformer control method described above when executing the computer program.

[0016] The present invention provides a transformer control method, comprising: acquiring a multi-dimensional heterogeneous fusion state vector at the current moment based on digital monitoring data collected by preset sensors; wherein the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data; acquiring the current health status assessment result of the transformer and the predicted trajectory of key state parameters within a prediction time window using a digital twin model of the transformer based on the multi-dimensional heterogeneous fusion state vector; wherein the prediction time window includes the current moment and a preset future time period; the predicted trajectory of key state parameters includes at least one of the predicted values ​​of the transformer winding hotspot temperature, insulation aging rate, and dissolved gas generation rate in the oil; determining the dynamic safety domain boundary within the prediction time window based on the current health status assessment result, the predicted trajectory of key state parameters, and grid fluctuation tolerance information; determining the operating status of the transformer based on the multi-dimensional heterogeneous fusion state vector, the predicted trajectory of key state parameters, and the dynamic safety domain boundary, and executing corresponding protection actions when the operating status is a target state; wherein the target state includes abnormal states and states with risks.

[0017] As can be seen, this invention achieves a comprehensive and three-dimensional perception of the transformer's operating status by fusing electrical and non-electrical quantity data into a multi-dimensional heterogeneous fused state vector, overcoming the shortcomings of traditional protection schemes that rely solely on the perception of a single electrical quantity. By utilizing a digital twin model of the transformer, it simultaneously obtains the current health status assessment results and the predicted trajectory of key state parameters within the prediction time window, enabling control decisions to be based on both the actual health level of the equipment and the ability to predict future operational risks. Furthermore, by combining the current health status assessment results, the predicted trajectory of key state parameters, and grid fluctuation tolerance information, a dynamic safety domain boundary is determined. This allows the protection boundary to adaptively tighten as the equipment's health deteriorates and reasonably widen as the equipment's health improves, while also considering the grid's fluctuation tolerance requirements. This maximizes the equipment's transmission capacity while ensuring equipment safety and effectively avoids protection malfunctions caused by fixed thresholds. Finally, based on the multi-dimensional heterogeneous fused state vector, the predicted trajectory of key state parameters, and the dynamic safety domain boundary, the operating status is comprehensively judged, and corresponding protection actions are executed under abnormal and risky conditions. This achieves graded and precise protection for timely handling of anomalies and early warning and prevention of potential risks, thereby ensuring the safe and reliable stable operation of the power system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart of a transformer control method provided in an embodiment of the present invention; Figure 2 This is a flowchart of obtaining a multidimensional heterogeneous fusion state vector provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a transformer control device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a transformer control device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please refer to Figure 1 , Figure 1 A flowchart illustrating a transformer control method provided in an embodiment of the present invention. The method may include: Step 101: Obtain the multidimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by the preset sensors.

[0022] Among them, the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data.

[0023] It is understood that the preset sensor in this embodiment can be a sensor pre-installed on and around the transformer (such as a distribution transformer) to monitor the transformer's operation, such as a high-precision intelligent sensing unit. The digital monitoring data in this embodiment can be data collected by the preset sensor.

[0024] Correspondingly, the specific number and type of preset sensors in this step, i.e., the specific number and type of digital monitoring data, can be set by the designer according to the practical scenario and user needs. For example, the preset sensors may include electrical quantity sensors for collecting electrical quantity data and non-electrical quantity sensors for collecting non-electrical quantity data; electrical quantity sensors may include voltage sensors and / or current sensors; non-electrical quantity sensors may include at least one of dissolved gas sensors in oil, top oil temperature sensors, and / or oil level sensors; wherein, voltage sensors and current sensors can be respectively set on the high-voltage side and / or low-voltage side of the transformer to collect electrical quantity data such as voltage and current data on each side of the transformer; dissolved gas sensors in oil are used to collect concentration data of characteristic gases such as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), and acetylene (C2H2) in transformer oil; top oil temperature sensors can be used to collect oil temperature data at the top of the transformer tank; and oil level sensors are used to collect oil level data in the transformer tank. In other embodiments, the preset sensor may also include other electrical and non-electrical sensors, such as ambient temperature sensors, core grounding current sensors, and vibration sensors, etc., and the embodiments of this application do not impose any limitations on this.

[0025] Accordingly, the method provided in this embodiment may also include the step of receiving digital monitoring data collected by preset sensors; for example, the transformer control equipment can synchronously collect digital monitoring data from preset sensors through at least two different types of communication interfaces. For instance, after the system is powered on, the processor (or main control unit, such as a programmable logic controller PLC) of the transformer control equipment can send data requests or listen to data reports to each preset sensor in parallel through its integrated first type of communication interface and second type of communication interface; wherein, the first type of communication interface is used to connect voltage sensors and current sensors that monitor the operating status of the high-voltage side and low-voltage side of the transformer to obtain electrical quantity data characterizing the electrical operating status; the second type of communication interface is used to connect oil dissolved gas sensors, top oil temperature sensors, and oil level sensors that monitor the physical status of the transformer body to obtain non-electrical quantity data characterizing the physical status of the transformer body. The first type of communication interface can use a high-speed, real-time industrial Ethernet interface, supporting Modbus TCP (an industrial communication protocol) or IEC61850MMS (an industrial communication protocol) to meet the requirements of electrical quantity data for high sampling rate and low transmission delay. The sampling period of voltage and current sensors can be set to a time value between 1ms and 10ms to ensure that detailed changes and transient characteristics within the power frequency cycle can be captured. The second type of communication interface can use an RS-485 (a serial communication standard) bus interface, supporting the Modbus RTU (a serial communication protocol). These non-electrical parameters change relatively slowly. The sampling period of the dissolved gas sensor, top oil temperature sensor, and oil level sensor can be set to a time value between 100ms and 1s. While ensuring real-time status, this effectively reduces the bus communication load and the processing pressure of the control equipment. This design, which configures the communication interface and sampling period differently according to data characteristics, achieves a balance between resource optimization and performance assurance.

[0026] It should be noted that the specific method for obtaining the multidimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by the preset sensors in this step can be set by the designer according to the practical scenario and user needs. For example, electrical feature extraction can be performed on the electrical quantity data in the collected digital monitoring data to obtain electrical features, and state feature extraction can be performed on the non-electrical quantity data to obtain state features. The electrical features and state features are then combined and encapsulated to obtain the multidimensional heterogeneous fusion state vector at the current moment. Among them, the electrical feature extraction operation includes at least one of effective value calculation, harmonic component extraction and phase calculation; the state feature extraction operation includes rate of change calculation and / or trend analysis.

[0027] Correspondingly, since each preset sensor converts the acquired analog signals into digital signals locally, and then continuously transmits the digital monitoring data to the control device through its respective communication interface with its own preset sampling period and communication protocol, the digital monitoring data received by the control device from the communication interface (such as the first type of communication interface and the second type of communication interface mentioned above) is asynchronous in time sequence due to the possible differences in the start time and period of data transmission from each preset sensor. This asynchronicity reflects the true physical basis of multi-source heterogeneous sensor data fusion. The synchronous acquisition intention of the control device refers to logically initiating acquisition tasks concurrently, while the original data stream of the received digital monitoring data must have timestamp interleaving. Therefore, this step can include a process of time alignment processing of the digital monitoring data to ensure that subsequent multi-dimensional data can be accurately fused and analyzed at the same time segment, avoiding feature distortion and protection misjudgment caused by inconsistent data delays. Figure 2 As shown, step 101 may include: Step 1011: Based on the timestamp information in the digital monitoring data, perform time alignment processing on the digital monitoring data to obtain time-aligned digital monitoring data.

[0028] In this embodiment, the control device receives digital monitoring data from various preset sensors via a communication interface. It performs time alignment based on the timestamp information carried in the data frames (or the reception time of the main control device's internal clock), unifying all digital monitoring data to the same reference time point. The core of time alignment is establishing a unified time reference axis. Asynchronous data (i.e., digital monitoring data) with their own timestamp information can be interpolated or assigned to the nearest, regular time point on this axis (e.g., aligned to integer milliseconds per second). Time alignment can preferentially use the high-precision time stamp (i.e., timestamp information) built into the data frames; if not, it relies on the internal clock mark added by the main control device at the moment the data is received. This step organizes the originally chaotic and asynchronous raw data stream into a series of data snapshots at strictly aligned time points, providing a stable and consistent input for subsequent feature extraction.

[0029] Step 1012: Perform electrical feature extraction on the electrical quantity data in the time-aligned digital monitoring data to obtain electrical features; wherein, the electrical feature extraction operation includes at least one of effective value calculation, harmonic component extraction and phase calculation.

[0030] Understandably, this step can perform electrical feature extraction operations on time-aligned electrical quantity data (such as voltage and current data) to obtain the corresponding electrical features. For example, electrical feature extraction operations include RMS value calculation, harmonic component extraction, and phase calculation.

[0031] For example, the above-mentioned effective value calculation process may include: for a sequence of instantaneous sampled values ​​of a certain electrical quantity (such as voltage or current) within a complete power frequency cycle after time alignment, firstly, the square of each sampled value in the sequence is calculated to obtain a sequence of squared values; then, the arithmetic mean of all elements in the sequence of squared values ​​is calculated; finally, the square root of the arithmetic mean is taken, and the result is used as the effective value characteristic of the electrical quantity data at the current moment. A complete power frequency cycle can be 20ms (corresponding to a 50Hz system) or 16.67ms (corresponding to a 60Hz system); a power frequency cycle tracking module can be set in the control equipment to dynamically adjust the calculation window, ensuring that calculations can still be performed based on accurate data from one cycle even when the power grid frequency fluctuates, thereby obtaining an effective value that truly reflects the work capacity and eliminating errors caused by non-integer cycle calculations.

[0032] Correspondingly, the process of extracting harmonic components can include: applying Fast Fourier Transform (FFT) to the instantaneous sampled value sequence of a time-aligned electrical quantity (such as voltage or current) for digital signal processing, transforming the time-domain sequence (i.e., the instantaneous sampled value sequence) into a frequency-domain sequence; in the frequency-domain sequence, determining the amplitude component corresponding to the fundamental frequency as the fundamental amplitude, and determining the amplitude components corresponding to other frequency points that are integer multiples of the power frequency as the harmonic amplitude characteristics of the corresponding order. The number of points in the Fast Fourier Transform can be set according to accuracy and speed requirements, such as 1024 points or 2048 points; during calculation, a Hanning window is added to the input sequence (i.e., the instantaneous sampled value sequence) to reduce spectral leakage; the extracted harmonic orders typically include the 2nd to 13th orders, with a focus on low-order harmonics (such as the 3rd, 5th, and 7th orders) that may cause additional losses to the transformer or trigger local overheating, as well as high-order harmonics that indicate specific faults; these harmonic amplitude characteristics are important bases for assessing power quality and identifying fault types.

[0033] Accordingly, the aforementioned phase calculation process may include: extracting complex results at the fundamental frequency from the frequency domain sequences corresponding to the voltage and current data, calculating the angle difference between these two complex results, and using this angle difference as the voltage and current fundamental phase difference feature at the current moment. The voltage and current fundamental phase difference feature directly reflects the transformer's power factor angle at the current moment and is one of the key dynamic features for determining the reactive power flow direction and identifying the difference between inrush current and internal fault current; its calculation relies on the aforementioned synchronously aligned voltage and current data, ensuring the consistency of phase information.

[0034] Step 1013: Perform state feature extraction on the non-electrical quantity data in the time-aligned digital monitoring data to obtain state features; wherein, the state feature extraction operation includes rate of change calculation and / or trend analysis.

[0035] Understandably, this step can perform state feature extraction operations on time-aligned non-electrical quantity data (such as characteristic gas concentration data, oil temperature data, and oil level data) to obtain the corresponding state features. For example, electrical feature extraction operations include rate of change calculation and trend analysis.

[0036] For example, the process of calculating the rate of change mentioned above may include: for a non-electrical data sequence (such as continuous oil temperature values) that has been time-aligned and arranged in chronological order, calculating the difference between the latest data value and the Nth data value preceding it, and then dividing this difference by the time interval corresponding to these two data values. The quotient is taken as the rate of change feature of the non-electrical data at the current moment. Here, N is a preset positive integer; the value of N is related to the physical change inertia and sampling period of the corresponding non-electrical parameter; for example, for the rate of change of oil temperature, N can be the data index difference corresponding to a time interval of 1 minute to 5 minutes; for the rate of change of hydrogen (H2) concentration in oil, since it may rise rapidly in the early stage of a fault, N can be the data index difference corresponding to a time interval of 30 seconds to 2 minutes; this rate of change feature can sensitively capture the acceleration or deceleration of the state and provide early warning of abnormal trends than absolute thresholds.

[0037] Correspondingly, the aforementioned trend analysis process may include: for a non-electrical quantity data sequence within a preset time window after time alignment, applying a linear regression algorithm with time as the independent variable and the data value of the non-electrical quantity data as the dependent variable, to calculate a best-fit line; the slope of this best-fit line is used as the trend characteristic reflecting the recent change direction of the non-electrical quantity data. The length of the preset time window can be set according to the monitoring purpose; for example, for monitoring long-term aging trends, the window length can be several hours; for monitoring the fault development process, the window length can be several minutes to tens of minutes. A positive slope value indicates an upward trend for the parameter, while a negative slope indicates a downward trend; the absolute value reflects the strength of the trend. This trend characteristic, combined with the aforementioned instantaneous rate of change characteristic, provides a complete profile of state changes from short-term drastic changes to long-term slow evolution, greatly enhancing the ability to identify latent and progressive faults.

[0038] In this embodiment, the specific order of steps 1012 and 1013 is not limited. As shown in this embodiment, step 1012 can be performed first and then step 1013, or step 1013 can be performed first and then step 1012, or both steps can be performed simultaneously.

[0039] Step 1014: Combine and encapsulate the electrical features and state features to obtain the multidimensional heterogeneous fused state vector at the current moment.

[0040] In this step, all extracted electrical and state features are combined and encapsulated according to a predefined vector structure (i.e., a preset vector structure template) to generate a multidimensional heterogeneous fusion state vector that contains multiple feature dimensions and strictly corresponds to a single moment (i.e., the current moment).

[0041] Correspondingly, the predefined vector structure mentioned above is an ordered data structure with fixed dimensions. The specific structural settings of the predefined vector structure, i.e., the specific structure and content of the multi-dimensional heterogeneous fusion state vector, can be set by the designer according to practical scenarios and user needs. For example, the multi-dimensional heterogeneous fusion state vector can sequentially include: the effective values ​​of phase A voltage, phase B voltage, and phase C voltage on the high-voltage side, as well as the effective values ​​of the three-phase current, total harmonic distortion rate, third harmonic content, fifth harmonic content, phase A power factor angle, top oil temperature, oil temperature change rate, oil temperature change trend slope, hydrogen concentration, acetylene concentration, and hydrogen concentration change rate, etc. This vector encapsulates the panoramic fusion information of the transformer at the same moment (i.e., the current moment), serving as a unified and regular input object for subsequent analysis, completely solving the problem that multi-source heterogeneous data is difficult to directly use in advanced algorithms.

[0042] Step 102: Based on the multidimensional heterogeneous fusion state vector, use the transformer digital twin model to obtain the current health status assessment results of the transformer and the predicted trajectory of key state parameters within the prediction time window.

[0043] The prediction time window includes the current moment and a preset future time period; the prediction trajectory of key state parameters includes at least one of the following: the predicted value of the hottest spot temperature of the transformer winding, the predicted value of the insulation aging rate, and the predicted value of the dissolved gas generation rate in the oil.

[0044] It is understood that the transformer digital twin model in this embodiment can be a virtual model constructed in digital space that maps to the physical transformer entity. It takes a multi-dimensional heterogeneous fused state vector as input and can mirror the current state of the transformer in real time and predict its future state evolution. The prediction time window can include the current moment and a preset future time period. The length of the preset future time period can be set according to application requirements, such as 30 minutes, 2 hours, or 24 hours. The prediction time window can be divided into several discrete time points according to a preset step size (e.g., 1 minute, 5 minutes). The key state parameter prediction trajectory can be a time series composed of the predicted values ​​of key state parameters at each discrete time point within the prediction time window, and can include at least one of the following: the predicted value of the transformer winding's hottest spot temperature, the predicted value of the insulation aging rate, and the predicted value of the dissolved gas generation rate in the oil.

[0045] Correspondingly, the specific method for obtaining the current health status assessment results of the transformer and the prediction trajectory of key state parameters within the prediction time window based on the multidimensional heterogeneous fusion state vector and the transformer digital twin model in this step, i.e., the specific model content of the transformer digital twin model, can be set by the designer according to the usage scenario and user needs. For example, the transformer digital twin model can include a physical calculation kernel, which can be a set of mechanism equations based on the physical laws of heat conduction, the aging reaction kinetics of insulating materials, and the dissolution and release mechanism of gases in oil, used to describe the hot spot temperature of transformer windings, the cumulative effect of insulation aging, and the change law of characteristic gas concentration in oil. For example, the creation process of the mechanism equations based on the physical laws of heat conduction in the physical calculation kernel can include: based on the equivalent thermal circuit model of the transformer, the winding and core are regarded as heat sources, and the insulating oil and oil tank are regarded as thermal resistance and thermal capacity. A set of difference equations is established to associate the square of the effective value of the winding current, the ambient temperature, the temperature of the hottest spot of the winding, and the temperature of the top oil layer, to describe the relationship between temperature and time. The equivalent thermal circuit model can be simulated using a multi-order thermal resistance-thermal capacity network. For example, a topology model containing multiple layers of thermal resistance and thermal capacity, such as winding to oil and oil to environment, can be established. Its parameters (such as thermal resistance and thermal capacity values) are obtained by analyzing the transformer design drawings, material properties, and temperature rise test data. The established difference equations can simulate the dynamic process of the temperature of the hottest spot of the winding rising or falling with time lag after load changes. This is the physical basis for realizing ultra-short-term thermal stress prediction.

[0046] Correspondingly, the creation process of the mechanism equation based on the aging reaction kinetics of insulation materials in the physical calculation kernel can include: establishing an exponential function with winding hot spot temperature as the independent variable based on the Arrhenius reaction rate theory to describe the aging reaction rate of the insulation material, and integrating this rate over time to characterize the cumulative effect of insulation aging; the Arrhenius reaction rate theory shows that the aging rate of insulating paper accelerates exponentially with increasing temperature; the established function includes activation energy parameters related to the properties of the insulation material; by integrating the aging reaction rate over operating time, the equivalent cumulative aging amount or percentage of life loss can be calculated, thereby quantitatively assessing the health status and remaining life trend of transformer insulation, providing a basis for preventive maintenance.

[0047] Correspondingly, the creation process of the mechanistic equations based on the dissolution and release mechanism of gases in oil in the physical calculation kernel can include: establishing a set of differential equations describing the relationship between the concentration changes of each characteristic gas in the oil and the type, energy magnitude, and duration of internal electrothermal faults in the transformer, according to Henry's law and the principle of gas diffusion; the characteristic gases can include at least hydrogen, methane, ethane, ethylene, and acetylene; according to fault diagnosis theories such as the Duval trigonometric method, different gas components and proportions correspond to different fault types (such as partial discharge, low-temperature overheating, high-temperature overheating, and arc discharge); the established set of differential equations can describe the gas generation rate of various gases, their dissolution equilibrium in the oil, and the dynamic process of escaping from the oil to the gas relay under specific fault energies, providing a mechanistic model for predicting fault development through gas data.

[0048] In other words, step 102 may include parsing the model input driving quantity from the multidimensional heterogeneous fused state vector; wherein the model input driving quantity includes electrical quantity data components and non-electrical quantity data components; using the physical calculation kernel of the transformer digital twin model, the model input driving quantity is forward deduced and calculated to obtain the predicted trajectory of key state parameters corresponding to the deduction results at each discrete time point within the prediction time window.

[0049] Furthermore, to improve the accuracy of predicted trajectories for key state parameters, the transformer digital twin model can also include a data-driven correction module for calibrating the extrapolation results. For example, the data-driven correction module can be a time-series prediction algorithm trained based on historical operating data of the transformer under various known operating conditions, which learns and compensates for deviations in the physical calculation kernel under specific application scenarios. For instance, the training process of the data-driven correction module (i.e., the time-varying electrical load, oil temperature, and gas concentration sequences in the oil recorded in historical operating data can be used as input features. The sequence of winding hotspot temperature changes or accelerated insulation aging measured by high-precision sensors within the same time period can be used as target labels. Supervised training of a recurrent neural network or long short-term memory network model is then performed, enabling the network to learn the complex nonlinear mapping relationship from input features to actual physical state changes. The trained network constitutes the data-driven correction module. The structure of the recurrent neural network or long short-term memory network model can be designed according to the amount and complexity of data, for example, containing 1 to 3 hidden layers; the training process adopts the error backpropagation algorithm with the goal of minimizing the mean square error between the network prediction value and the actual measurement label; the role of the data-driven correction module is to learn and compensate for factors that are not covered by the physical model or whose parameters are inaccurate, such as the actual decay of radiator efficiency, the impact of oil aging on thermal characteristics, and the heat dissipation conditions of a specific installation environment, thereby significantly improving the prediction accuracy of the digital twin model on specific transformers.

[0050] Correspondingly, the physical calculation kernel and the data-driven correction module can be coupled and integrated into the transformer digital twin model. For example, the transformer digital twin model can first use the physical calculation kernel to perform a forward deduction based on the input quantity to obtain a first intermediate result; then, the same input quantity and the first intermediate result are input together to the data-driven correction module, which outputs a compensation quantity; finally, the first intermediate result and the compensation quantity are added to obtain the final output value of the transformer digital twin model. This serial coupling integration method fully leverages the advantages of the physical model's strong interpretability and good extrapolation, as well as the advantages of the data-driven model's strong fitting ability and ability to correct system errors. The physical calculation kernel ensures the basic physical rationality of the prediction results, while the data-driven correction module performs personalized calibration. Finally, what is solidified and loaded into the main control unit (such as a PLC) is an executable calculation program package that integrates all parameters and logic, ensuring that the digital twin model can run efficiently and reliably even on resource-constrained edge devices.

[0051] Correspondingly, the process described above, which utilizes the physical calculation kernel of the transformer digital twin model to perform forward extrapolation calculations on the model input driving quantities and obtain the predicted trajectories of key state parameters corresponding to the extrapolation results at each discrete time point within the prediction time window, can be specifically described as follows: The physical calculation kernel performs forward extrapolation calculations on the model input driving quantities, and the data-driven correction module of the transformer digital twin model is used to calibrate the extrapolation results, obtaining the predicted trajectories of key state parameters corresponding to the calibrated extrapolation results at each discrete time point within the prediction time window. For example, after receiving the multidimensional heterogeneous fused state vector, the transformer digital twin model first parses from the vector the electrical quantity data components, non-electrical quantity data components (including information such as oil temperature and gas concentration in the oil), and the current operating condition information reflected by the multidimensional heterogeneous fused state vector, which serve as the model input driving quantities.

[0052] Then, the physical calculation kernel of the transformer digital twin model performs forward extrapolation calculations using these components as boundary conditions. At the same time, its data-driven correction module performs real-time calibration of the physical extrapolation results based on the correlations learned from the historical operating data trained during the construction phase. Together, they complete an ultra-short-term operating trajectory prediction calculation, thereby outputting a continuous change trajectory containing the predicted values ​​of the transformer winding hotspot temperature, insulation aging rate, and dissolved gas generation rate at each discrete time point in the future. This trajectory is the key state parameter prediction trajectory.

[0053] The process of jointly completing an ultra-short-term trajectory prediction calculation can be specifically described as follows: within a future preset time period of the prediction time window, the simulation is iterated forward step by step with fixed and tiny time steps; at each time step, the physical calculation kernel calculates the first predicted value of the current step state value based on the state of the previous time step and the current input driving quantity; then, the data-driven correction module outputs a correction value for the current step based on the same input driving quantity and the first predicted value; the first predicted value and the correction value are superimposed to obtain the final state prediction value of the current time step, and this value is used as one of the initial states for the next time step iteration; this process is repeated until the entire future preset time period of the prediction time window is covered, thereby generating a trajectory formed by connecting the final state prediction values ​​of each time step; Correspondingly, the preset time period can be set according to the protection and control needs, such as 30 seconds to 5 minutes in the future, to balance the real-time requirements of prediction and the ability to cover short-term risks; the fixed, small time step is usually set to 1 to 10 seconds, which matches the dynamic characteristics of the model and the computing power of the main control unit; this iterative rolling prediction method can dynamically integrate the latest measured data, so that the prediction trajectory is always based on the current latest state, thereby significantly improving the ability to track and predict sudden changes in working conditions.

[0054] It should be noted that this step may include the process of determining the current health status assessment result of the transformer using a digital twin model of the transformer. For example, the current health status assessment result of the transformer can be determined based on each feature vector in the multidimensional heterogeneous fused state vector and its corresponding preset health benchmark quantity. For instance, while performing ultra-short-term operating trajectory prediction calculations, the transformer digital twin model can also calculate a quantified health status index through a built-in health assessment function based on the values ​​of each feature vector in the input multidimensional heterogeneous fused state vector and their deviation from the predefined health benchmark quantity (i.e., the preset health benchmark quantity) that characterizes the transformer's health benchmark state. This index is the real-time health status assessment result of the transformer (i.e., the current health status assessment result).

[0055] Correspondingly, the process of calculating the health status index using a built-in health assessment function can include: calculating the relative deviation between the value of each feature vector in the multidimensional heterogeneous fusion state vector and the benchmark value of the corresponding component in the predefined health benchmark state feature vector (i.e., the preset health benchmark quantity); then, assigning a preset weight coefficient to the relative deviation of each feature vector, which reflects the importance of the feature vector to the overall health status of the transformer; finally, summing the weighted relative deviations of all feature vectors and processing them through a preset normalization function, mapping the processing result to a value within a set range, which is the quantified health status index (i.e., the current health status assessment result). The predefined health baseline characteristic vector contains typical or allowable upper limits for each characteristic vector of the transformer in its new and healthy state (e.g., winding temperature at rated current, normal background oil gas content). The weighting coefficients can be assigned based on expert experience or statistical analysis of historical fault data. For example, acetylene (C2H2) concentration and winding hot spot temperature can be given higher weights because they directly indicate serious faults; total hydrocarbon content and load current can be given the next highest weights. The normalization function typically maps the comprehensive cumulative value to a range of 0 to 100, where 100 represents complete health and 0 represents serious faults. For example, a health status index above 90 is considered "excellent," 80-90 is "good," 70-80 is "attention," below 70 is "warning," and below 50 is "serious." This quantified health status index provides maintenance personnel with an intuitive status assessment and can be directly used to trigger different levels of early warning or preventative control strategies.

[0056] Accordingly, the method provided in this embodiment may also include the process of acquiring a transformer digital twin model, such as directly receiving and storing the transformer digital twin model; or creating a transformer digital twin model based on historical operating data. This embodiment does not impose any limitations on this.

[0057] Step 103: Based on the current health status assessment results, the predicted trajectory of key status parameters, and the grid fluctuation tolerance information, determine the dynamic security domain boundary within the prediction time window.

[0058] Understandably, the dynamic safety domain boundary can be used to define a set of protection action thresholds and characteristic curves that vary over time and are adapted to the actual operating conditions of the transformer and the power grid environment. In other words, it is a set of thresholds that change over time within the prediction time window, defining the safe operating range of the transformer. These are no longer fixed thresholds in the traditional sense, but rather a set of boundary curves that dynamically adjust with the health status of the equipment, the evolution trend of its thermal state, and the operational needs of the power grid. Power grid fluctuation tolerance information is used to characterize the power grid's ability to withstand disturbances. This information can be obtained from the distribution automation master station or statistically derived from historical operating data.

[0059] Correspondingly, in this step, the dynamic safety domain boundary within the prediction time window can be determined based on the current health status assessment results and key state parameter prediction trajectories output by the transformer digital twin model, combined with grid fluctuation tolerance information. For example, based on the predicted values ​​of the transformer winding hotspot temperature, insulation aging rate, and dissolved gas generation rate in the key state parameter prediction trajectory, the overall equipment health level reflected by the current health status assessment results, and the grid side's ability to withstand disturbances as represented by grid fluctuation tolerance information, a set of boundary conditions (i.e., dynamic safety domain boundaries) that evolve over time can be comprehensively determined.

[0060] Correspondingly, the dynamic safety domain boundary may include: a dynamic current action threshold curve determined based on thermal stress constraints, an overload tolerance time curve determined based on insulation life loss constraints, and / or an oil gas concentration warning threshold curve determined based on fault gas development constraints. The essence of the dynamic safety domain boundary is to transform traditional fixed and isolated protection settings into a series of interrelated protection surfaces or protection envelopes that evolve in real time with the transformer's own state and the power grid environment; it is not a single numerical value, but a comprehensive set of rules defining the interface between the safe operating region and the risky action region in a multi-dimensional state space. This leap from point thresholds to a dynamic domain is the core of achieving intelligent adaptive protection in this embodiment.

[0061] It should be noted that the specific method for determining the dynamic safety domain boundary within the prediction time window based on the current health status assessment results, the predicted trajectory of key state parameters, and the grid fluctuation tolerance information in this step can be set by the designer according to the practical scenario and user needs. For example, the calculation and determination process of the dynamic current action threshold curve determined based on thermal stress constraints can include: dynamically deriving the maximum load current value allowed to flow at the current moment under the premise that the temperature of the hottest spot of the transformer winding does not exceed its safety limit at any future moment, based on the predicted value of the hottest spot temperature of the transformer winding in the predicted trajectory of key state parameters, and the overall health level of the equipment reflected by the current health status assessment results (which can be converted into a reduction factor for the insulation thermal tolerance capability), and then dynamically deriving the maximum load current value allowed to flow at the current moment. The curve formed by the maximum load current value at the current moment and at each future moment constitutes the dynamic current action threshold curve.

[0062] The specific process of dynamically deriving the maximum load current value can be as follows: Select the predicted value of the hottest point temperature of the winding at a specific future moment in the predicted trajectory, compare this predicted value with the absolute safe upper limit value of the winding hot spot temperature, and select the smaller value as the selected value; the absolute safe upper limit value of the winding hot spot temperature is usually set according to the heat resistance grade of the transformer insulation material. For example, for oil-immersed transformers with Class A insulation, this value is often set to 140°C; at the same time, obtain the insulation heat resistance reduction coefficient calculated from the current health status assessment results. This coefficient is a value less than or equal to 1, used to correct the absolute safe upper limit value to obtain the actually usable dynamic upper limit temperature; the reduction coefficient can be dynamically mapped according to the health status index. For example, the coefficient is 1.0 when the health index is greater than 90, 0.95 when it is 80-90, 0.9 when it is 70-80, and so on. This reflects the concept that equipment "operating with defects" needs more conservative protection. Based on the relationship between winding heating and the square of current described in the equivalent thermal circuit model of the transformer, a reverse calculation is performed: the difference between the upper limit of dynamic temperature and the current actual oil temperature (or ambient temperature) is calculated, and then based on this difference, the thermal resistance parameter of the transformer, and a preset time interval from the current moment to a specific future moment, a square value of current is deduced; the thermal resistance parameter of the transformer is a known parameter directly obtained from the physical calculation kernel of the digital twin model.

[0063] The derivation time interval can be equal to the protection action delay, for example, 0.5 to 2 seconds. This ensures that the calculated current threshold can effectively prevent temperature overruns before the protection takes effect. Finally, the square root of this current square value is taken to obtain the maximum load current value allowed to flow at the current moment, provided that the temperature does not exceed the limit at a specific future moment. This calculation is repeated for multiple future moments on the predicted trajectory to obtain a series of maximum load current values. Connecting these values ​​forms the dynamic current action threshold curve. Graphically, this curve is usually represented as a curve that slowly decreases over time, intuitively showing the time-varying characteristics of the transformer load capacity under the current thermal state. Compared with the fixed overcurrent setting, it is more lenient in the early stage of slow load increase, but more stringent when approaching the thermal limit, achieving precise matching between protection and equipment capacity.

[0064] Correspondingly, the calculation and determination process of the tolerable overload time curve based on the insulation life loss constraint can include: based on the predicted insulation aging rate value in the predicted trajectory of key state parameters, performing cumulative integration over a future preset time period to obtain the predicted insulation life loss within that future preset time period; and then, combined with the remaining life margin reflected in the current health status assessment results, calculating the overload time that the transformer can continuously withstand under the current operating state, provided that the insulation life loss does not exceed the preset threshold. This time relationship constitutes the tolerable overload time curve.

[0065] The specific process of obtaining the insulation life loss by cumulative integration is as follows: within a future preset time period, the predicted insulation aging rate at each tiny time step of the predicted trajectory is multiplied by the length of that time step to obtain the tiny life loss within that step; then, the tiny life loss of all tiny time steps within the entire time window is accumulated and summed, and the sum obtained is the predicted insulation life loss within the future preset time period.

[0066] The specific operational process for calculating the tolerable overload time is as follows: A preset threshold value for insulation life loss is set; this threshold value is typically set as an acceptable short-term life consumption ratio, such as 0.1% (one-thousandth) of the rated life; the percentage or absolute value reflecting the remaining life margin of the equipment is extracted from the current health status assessment results; the preset threshold value is compared with the remaining life margin, and the smaller of the two is taken as the maximum allowable life loss; under the assumption that the current predicted insulation aging rate (or a set higher overload aging rate) remains unchanged, the time required for the cumulative life loss to reach its maximum is calculated. The calculated time to reach the maximum allowable lifespan loss is the overload time that the transformer can withstand under the current operating conditions. By changing the assumptions of the current operating conditions (such as changing the load current) and repeating this calculation, we can obtain the tolerable overload time curve describing the relationship between different overload levels and the duration of overload. This curve is essentially an inverse time characteristic curve, but its baseline and slope are determined by the real-time aging status of the equipment and the predicted future aging rate, rather than fixed values. For example, a new transformer with ample remaining lifespan will have a longer allowable short-term overload time than an older transformer, reflecting the equipment differences in protection strategies.

[0067] Accordingly, the calculation and determination process of the oil gas concentration warning threshold curve based on the constraints of fault gas development can include: directly obtaining or deducing the predicted values ​​of the concentrations of key fault gases such as hydrogen and acetylene in the oil within a future preset time period based on the predicted values ​​of dissolved gas production rate in the oil in the predicted trajectory of key state parameters; dynamically calculating the concentration warning thresholds of various fault gases at the current moment based on preset gas concentration attention values ​​and warning values, and the set of these thresholds constitutes the oil gas concentration warning threshold curve.

[0068] The specific operation process for extrapolating the predicted values ​​of gas concentrations in critical faults is as follows: starting from the actual gas concentration in the oil monitored at the current moment, the predicted value of the future dissolved gas production rate in the oil in the prediction trajectory is accumulated (integrated) at each corresponding small time step, thereby extrapolating the predicted values ​​of gas concentrations at each future moment.

[0069] The specific operational process for dynamically calculating the concentration warning threshold is as follows: Two concentration reference values ​​are preset for each type of critical fault gas (such as hydrogen and acetylene): a lower attention value and a higher warning value. Taking hydrogen as an example, its attention value can be set to 150 μL / L, and its warning value to 500 μL / L; the attention value for acetylene can be set to 5 μL / L, and its warning value to 30 μL / L. These values ​​are based on industry standards and adjusted according to the specific transformer voltage level. The predicted concentration values ​​of the critical fault gases obtained through deduction are observed, and the future time points corresponding to the first occurrence of the predicted values ​​reaching the attention and warning values ​​are identified. Based on the principle that "warnings need a certain lead time," the time point before the first occurrence of the warning value is advanced... The previous fixed time interval (e.g., 5 minutes) is determined as the "current moment" at which an alert needs to be issued. Then, the slightly lower predicted concentration value corresponding to this "current moment" is dynamically calculated as the concentration alert threshold for that type of gas at the current moment. For example, if the predicted hydrogen concentration will reach the alert value of 500 μL / L in 8 minutes, the system will use the predicted concentration value corresponding to the current moment (8 minutes in advance) (which may only be 200 μL / L) as the dynamic alert threshold. Once the real-time monitored hydrogen concentration exceeds this dynamic threshold, an alert is triggered, achieving "prevention before the event". This calculation is performed separately for each type of gas to obtain a set of thresholds that change over time, forming the gas concentration alert threshold curve in the oil.

[0070] Finally, grid-side constraints are integrated; for example, the dynamic current action threshold curve, the overload tolerance time curve, and the oil gas concentration warning threshold curve calculated above are comprehensively checked with the grid fluctuation tolerance information obtained from the distribution automation master station; the grid fluctuation tolerance information is used to dynamically relax or tighten the instantaneous current setpoint in the dynamic current action threshold curve to ensure that the transformer protection action is coordinated with the transient stability requirements of the grid, and finally generate the checked dynamic safety domain boundary as the final result; The specific operational process for comprehensive verification and adjustment is as follows: Analyze the grid fluctuation tolerance information, which contains the quantitative level or percentage information of the current grid stability margin; establish adjustment rules: when the signal indicates a high grid stability margin, the instantaneous overcurrent setting used to determine instantaneous faults (such as short circuits) in the dynamic current action threshold curve is widened upwards by a preset ratio to reduce protection maloperation caused by normal grid fluctuations such as new energy switching; for example, when the grid stability margin is "high," the instantaneous overcurrent setting can be increased by 20%; when the signal indicates a low grid stability margin, the instantaneous overcurrent setting is tightened downwards by a preset ratio to improve protection sensitivity and ensure rapid fault isolation when the grid is weak, preventing the accident from escalating; for example, when the grid stability margin is "low," the instantaneous overcurrent setting can be reduced by 15%; combine the dynamic current action threshold curve adjusted according to these rules with the unadjusted overload tolerance time curve and the oil gas concentration warning threshold curve to form the verified dynamic safety domain boundary.

[0071] This integration step is crucial because it unifies the protection requirements of the equipment itself with the safe operation requirements of the power system within a single decision-making framework, resolving potential conflicts between equipment protection and system stability. For example, during the vulnerable period following a grid fault, even if the transformer's thermal stress still has a margin, the system requires its protection to be more sensitive to quickly clear the fault. This mechanism achieves this system-level requirement by tightening the fast-break setting value, reflecting the collaborative concept of local protection serving overall safety.

[0072] In other words, this step determines the dynamic current action threshold curve within the prediction time window based on the predicted temperature of the hottest spot in the transformer winding and the current health status assessment results. The dynamic current action threshold curve is the curve corresponding to the maximum allowable load current value at each discrete time point within the prediction time window, provided that the predicted temperature of the hottest spot in the transformer winding does not exceed its respective safety upper limit. The safety upper limit is the predicted temperature of the hottest spot in the transformer winding at the corresponding discrete time point or the corrected absolute safety upper limit. The corrected absolute safety upper limit is a value determined based on the current health status assessment results and the preset absolute safety upper limit. The dynamic current action threshold curve is adjusted based on the dynamic current action threshold curve and grid fluctuation tolerance information to obtain the adjusted dynamic current action threshold curve. The dynamic safety domain boundary includes the adjusted dynamic current action threshold curve. Correspondingly, the adjusted dynamic current action threshold curve can be combined with the overload tolerance time curve and the oil gas concentration warning threshold curve to obtain the dynamic safety domain boundary.

[0073] Step 104: Determine the transformer's operating status based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of key state parameters, and the dynamic safety domain boundary, and execute corresponding protection actions when the operating status is the target state; wherein, the target state includes abnormal state and state with risk.

[0074] Understandably, the specific method for determining the transformer's operating status based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of key state parameters, and the dynamic safety domain boundary in this step can be set by the designer. For example, each feature vector in the multidimensional heterogeneous fusion state vector can be compared with the threshold at the current moment on the corresponding curve in the dynamic safety domain boundary. If there is a feature vector greater than the corresponding threshold, the operating status is determined to be abnormal. If there is no feature vector greater than the corresponding threshold, the predicted trajectory of key state parameters is compared with the dynamic safety domain boundary to determine whether there is a predicted value greater than the corresponding threshold in the predicted trajectory of key state parameters within a preset time period in the future. If there is a predicted value greater than the corresponding threshold, the operating status is determined to be at risk. If there is no predicted value greater than the corresponding threshold, the operating status is determined to be normal.

[0075] For example, the dynamic safety domain boundary includes a dynamic current action threshold curve based on thermal stress constraints, an overload tolerance time curve based on insulation life loss constraints, and an oil gas concentration warning threshold curve based on fault gas development constraints. Therefore, the comparison process can specifically include: comparing the effective current value component in the multidimensional heterogeneous fusion state vector with the threshold at the current moment on the dynamic current action threshold curve, and comparing the non-electrical parameter components such as oil temperature, oil gas concentration, and their corresponding rate of change and trend slope in the multidimensional heterogeneous fusion state vector with the threshold at the current moment on the overload tolerance time curve and the oil gas concentration warning threshold curve.

[0076] The specific process of comparing the effective current value component with the threshold at the current moment on the dynamic current action threshold curve can be as follows: The effective current value component representing the transformer load (e.g., the effective value of the three-phase current on the low-voltage side) is read from the multi-dimensional heterogeneous fused state vector. Simultaneously, the maximum allowable current threshold corresponding to the current moment is retrieved from the dynamic current action threshold curve. Then, the read effective current value component is directly compared with the retrieved maximum allowable current threshold to determine whether the effective current value is greater than the threshold. The maximum allowable current threshold corresponding to the current moment is dynamically calculated in step 103 based on the real-time thermal state of the equipment and future thermal stress prediction. It may be higher or lower than the traditional fixed overcurrent setting. For example, during cold start-up of the transformer, this dynamic threshold may allow continuous operation at 1.2 times the rated current, while after long-term high load, the threshold may drop to 1.05 times the rated current, thus achieving precise matching between the protection setting and the actual thermal capacity of the equipment. This dynamic comparison is crucial for preventing maloperation (false tripping when the equipment is capable) and failure to operate (failure to trip when the equipment is overheated). The specific process of comparing the non-electrical parameter components with the threshold at the current moment on the overload tolerance time curve can be as follows: Read the parameter components characterizing the transformer's thermal state (such as top oil temperature and oil temperature change rate) from the multi-dimensional heterogeneous fused state vector, and locate them in the coordinate system defined by the overload tolerance time curve, combined with the current load level; one axis of the coordinate system represents the overload degree (or equivalent thermal stress), and the other axis represents the allowable duration; compare the current operating point reflected by the state vector (with the overload degree and the current duration as coordinates) with the allowable duration-overload degree relationship defined at the current moment on the curve to determine the current operating... Whether the point falls outside the allowable area defined by the curve determines whether the current overload condition has exceeded the allowable duration. The allowable area is a region in the first quadrant of the coordinate system enclosed by the overload time curve. Below the curve indicates safety, and above the curve indicates that the insulation life loss will exceed the limit. For example, for an overload of 1.3 times the rated current, the curve may allow it to last for 20 minutes, while for 1.5 times the rated current, it may only allow it to last for 5 minutes. This comparison transforms the abstract insulation life consumption into an intuitive remaining safe time, providing operators with a clear basis for decision-making, and can automatically trigger alarms or tripping before the remaining safe time is exhausted.

[0077] The specific process of comparing the non-electrical parameter components in the state vector with the current threshold on the oil gas concentration warning threshold curve can be as follows: The current concentration component of each critical fault gas (such as hydrogen and acetylene) is read from the multi-dimensional heterogeneous fused state vector, while the dynamic warning thresholds for each type of gas corresponding to the current moment are retrieved from the oil gas concentration warning threshold curve. Then, the current concentration component of each gas is directly compared with the corresponding dynamic warning threshold to determine whether the gas concentration is greater than its corresponding dynamic warning threshold. The dynamic warning threshold is a threshold with "advance" derived from the gas production rate prediction, and its value is usually lower than the static attention value. For example, even if the current hydrogen concentration is only 200 μL / L (lower than the static attention value of 150 μL / L), if the system predicts that it will rise rapidly in the short term, the dynamic warning threshold may be set to 180 μL / L, thus achieving early warning. This comparison mechanism transforms the traditional "alarm after exceeding the standard" into "early warning upon adverse trend," greatly advancing the fault detection time.

[0078] If the comparison results show that the value of any one or more feature vectors in the multidimensional heterogeneous fusion state vector exceeds the current action threshold of the corresponding curve in the dynamic safety domain boundary, it is determined to be an abnormal state of fault or anomaly, and the corresponding abnormal protection action can be triggered immediately. The abnormal protection action may include outputting a command to drive the circuit breaker to perform a trip, outputting a command to activate the audible and visual alarm device, and generating fault information containing fault type, fault parameters, over-limit time, and action type. This fault information, as part of the state information, is called the protection action record.

[0079] Correspondingly, the specific process for generating protection action records can be as follows: the system records the moment when protection is triggered as the over-limit moment; based on which feature vector(s) exceeded the threshold of which curve, the fault type is determined (e.g., current exceeding the dynamic current action threshold is determined as an overcurrent fault, hydrogen concentration exceeding its dynamic warning threshold is determined as a high hydrogen fault); the specific values ​​of the relevant feature vectors in the multidimensional heterogeneous fusion state vector corresponding to the over-limit moment are recorded as fault parameters; the specific protection action type executed is recorded (e.g., instantaneous tripping, overcurrent tripping, or gas warning); finally, the over-limit moment, fault type, and fault parameters are combined. The four types of information—number, action type, and action quantity—are combined according to a preset format to form a structured protection action record. This record is crucial evidence for accident retrospective analysis and responsibility analysis. Its structured format ensures the integrity and readability of the information, facilitating automatic import into the fault information management system. For example, a record might read: "2023-10-27 14:05:33.450, Overcurrent Fault (Instantaneous Trip), Ia=5120A, Ib=4980A, Ic=5050A, Instantaneous Trip". This detailed record provides a complete data chain for assessing the correctness of protection actions and analyzing the causes of faults.

[0080] Correspondingly, if the comparison results show that none of the feature vectors have exceeded the limit at the current time, but the predicted trajectory of the key state parameters output in step 102 indicates that the predicted value of a feature vector will exceed the dynamic safety domain boundary within a preset time period in the future, then it is determined that there is a risk state and corresponding preventive protection actions can be executed.

[0081] The specific process of determining whether the predicted value of a feature vector will exceed the dynamic safety domain boundary within a preset future time period, based on the key state parameter prediction trajectory output in step 102, can be as follows: Obtain the key state parameter prediction trajectory output in step 102, which contains the predicted values ​​of key state parameters at a series of future time points; simultaneously obtain the boundary conditions of the dynamic safety domain boundary calculated in step 103 at the same future time; within the preset future time period, check sequentially: compare the predicted value of the state parameter at each future time with the corresponding dynamic safety domain boundary; if, at any future time, the predicted value of any state parameter exceeds the corresponding dynamic safety domain boundary, it is determined that a risk state exists where the predicted value of a feature vector will exceed the boundary. This process achieves the "predictive" nature of the protection. For example, the system may predict, even when the current is not currently exceeding the limit, that the winding hotspot temperature will reach 138°C (close to the 140°C limit) in 15 minutes due to increased ambient temperature and load, thus determining a risk state in advance; this provides valuable preparation time for implementing preventative protection actions.

[0082] Correspondingly, preventative protective actions may include sending early warning information to the superior system or local interface, slightly adjusting the internal settings of protective components based on prediction results to improve the sensitivity of subsequent monitoring; at the same time, generating information containing risk prediction content and suggested control measures, which, as another part of the status information, is called an early warning record.

[0083] The specific process of slightly adjusting the internal settings of protection components based on prediction results can be as follows: Identify which state parameter (such as winding hot spot temperature) is predicted to exceed the safety boundary in the future and at what time; based on the urgency of the parameter exceeding the limit (e.g., how close the predicted time of exceeding the limit is to the current moment), temporarily lower the internal settings of the relevant protection components according to a preset adjustment ratio; for example, if the predicted hot spot temperature will exceed the limit in 10 minutes, the operating current setting of the overload protection will be temporarily lowered by 2%; if it is predicted to exceed the limit in 2 minutes, it will be lowered by 5%; this adjustment only takes effect in the background logic to improve monitoring sensitivity, without changing the externally released setting value; this soft adjustment strategy automatically improves the system's defense level when the risk increases, which is equivalent to adding an immune response mechanism to the protection system; it can capture the deterioration trend earlier and avoid the confusion caused to operators by frequent changes in settings.

[0084] The specific process for generating an early warning record can be as follows: the system records the moment the early warning is generated; based on the prediction and comparison results, risk prediction content is generated, describing which parameter, when it is predicted to be exceeded, and what safety boundary it will exceed; based on the implemented preventive control strategy, suggested control measures are generated, such as monitoring the winding temperature and automatically increasing the overload protection sensitivity; finally, the three types of information—early warning time, risk prediction content, and suggested control measures—are combined according to a preset format to form a structured early warning record. Early warning records are important inputs for condition-based maintenance and preventive maintenance; for example, an early warning record might read: "2023-10-27 13:50:00, predicted winding hotspot temperature will exceed 138°C at 14:05, recommended to check the cooling system, overload protection sensitivity has been temporarily increased by 5%"; it provides maintenance personnel with clear inspection targets and handling suggestions, transforming passive maintenance into proactive maintenance.

[0085] It should be noted that the method provided in this embodiment can be applied to transformers or edge devices near transformers (such as intelligent protection devices); it can also be applied to distribution automation master stations or cloud servers that are communicatively connected to transformers or edge devices. This embodiment does not impose any limitations on this.

[0086] Correspondingly, when the method provided in this embodiment is applied to a transformer or edge devices near a transformer, after step 104, protection control information can also be sent to the distribution automation master station to achieve system-level collaborative optimization. The protection control information can include the predicted trajectory of key state parameters, the dynamic safety domain boundary, and the control commands and status information corresponding to the protection action. That is, after executing the protection action corresponding to the target state, the control device can encapsulate the predicted trajectory of key state parameters output in step 102, the dynamic safety domain boundary calculated in step 103, and the control commands and status information corresponding to the protection action generated in step 104, which are cached internally. According to the data model and message format specified by the selected communication protocol (such as the Modbus serial communication protocol or the IEC61850 communication standard), the four types of data—the predicted trajectory of key state parameters, the dynamic safety domain boundary, the control commands, and the status information—are mapped to the data objects or register addresses specified by the protocol, and necessary message headers and device addresses are added. Control information such as address and checksum is assembled into one or more complete communication protocol data units that can be transmitted over the network. For example, when using the IEC61850 protocol, the predicted trajectory of key state parameters and the dynamic safety domain boundary can be mapped to a dataset in the logical node (LN) under the logical device (LD), and actively sent up through the report or log service model. Protection action records and early warning records can be sent up as time-stamped disturbance records (DDR). This standardized mapping ensures that the master station system can unambiguously parse and utilize this high-level information, realizing semantic interoperability between the station-end intelligent device and the master station system.

[0087] Correspondingly, the encapsulated data packets can be sent to the distribution automation master station through the communication interface integrated into the main control device, which supports the Modbus or IEC61850 protocol. The distribution automation master station receives and parses this data to remotely monitor the operating status of transformers and protection actions, analyze faults and early warning events, and perform distribution network-level operation optimization and coordinated control. The master station system utilizes predicted trajectories, dynamic boundaries, and early warning information uploaded from multiple such intelligent protection devices to perform distribution network-level load forecasting, weak link analysis, and collaborative optimization of protection settings. For example, the master station can comprehensively judge the power supply reliability of a certain area and, in emergency situations, issue more stringent grid fluctuation tolerance information to relevant transformers to improve overall protection speed and prevent fault escalation. This forms a two-layer intelligent architecture of single-point intelligent perception-decision and system-wide coordination-optimization, ultimately improving the resilience and operating efficiency of the entire distribution network.

[0088] In this embodiment, the present invention achieves a comprehensive and three-dimensional perception of the transformer's operating status by fusing electrical and non-electrical quantity data into a multi-dimensional heterogeneous fused state vector. This overcomes the shortcomings of traditional protection schemes that rely solely on a single electrical quantity for perception. By utilizing a digital twin model of the transformer, it simultaneously obtains the current health status assessment results and the predicted trajectory of key state parameters within the prediction time window. This allows control decisions to be based on both the actual health level of the equipment and the ability to predict future operational risks. Furthermore, by combining the current health status assessment results, the predicted trajectory of key state parameters, and grid fluctuation tolerance information, a dynamic safety domain boundary is determined. This allows the protection boundary to adaptively tighten as the equipment's health deteriorates and reasonably widen as the equipment's health improves, while also considering the grid's fluctuation tolerance requirements. This maximizes the equipment's transmission capacity while ensuring equipment safety and effectively avoids protection malfunctions caused by fixed thresholds. Finally, based on the multi-dimensional heterogeneous fused state vector, the predicted trajectory of key state parameters, and the dynamic safety domain boundary, the operating status is comprehensively determined, and corresponding protection actions are executed under abnormal and risky conditions. This achieves graded and precise protection for timely handling of anomalies and early warning and prevention of potential risks, thereby ensuring the safe and reliable stable operation of the power system.

[0089] Corresponding to the above method embodiments, this invention also provides a transformer control device. The transformer control device described below and the transformer control method described above can be referred to each other.

[0090] Please refer to Figure 3 , Figure 3 This is a structural block diagram of a transformer control device provided in an embodiment of the present invention. The device may include: The data acquisition module 10 is used to acquire a multi-dimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by the preset sensors; wherein, the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data. The trajectory prediction module 20 is used to obtain the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window based on the multi-dimensional heterogeneous fused state vector and the transformer digital twin model. The prediction time window includes the current time and a preset future time period. The predicted trajectory of key state parameters includes at least one of the predicted values ​​of the transformer winding hottest spot temperature, insulation aging rate, and dissolved gas generation rate in oil. The boundary determination module 30 is used to determine the dynamic safety domain boundary within the prediction time window based on the current health status assessment results, the predicted trajectory of key status parameters, and the power grid fluctuation tolerance information. The operation protection module 40 is used to determine the operating status of the transformer based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of key state parameters and the dynamic safety domain boundary, and to execute corresponding protection actions when the operating status is the target state; wherein, the target state includes abnormal state and state with risk.

[0091] In some embodiments, the acquisition module 10 may include: The alignment submodule is used to perform time alignment processing on the digital monitoring data based on the timestamp information in the digital monitoring data, so as to obtain the time-aligned digital monitoring data. The first extraction submodule is used to perform electrical feature extraction operations on the electrical quantity data in the time-aligned digital monitoring data to obtain electrical features; wherein, the electrical feature extraction operation includes at least one of effective value calculation, harmonic component extraction and phase calculation; The second extraction submodule is used to perform state feature extraction operations on non-electrical quantity data in the time-aligned digital monitoring data to obtain state features; wherein, the state feature extraction operation includes rate of change calculation and / or trend analysis; The combination submodule is used to combine and encapsulate electrical and state features to obtain a multidimensional heterogeneous fused state vector at the current moment.

[0092] In some embodiments, the preset sensor includes an electrical quantity sensor and a non-electrical quantity sensor. The electrical quantity sensor includes a voltage sensor and / or a current sensor, and the non-electrical quantity sensor includes at least one of a dissolved gas sensor, a top oil temperature sensor, and / or an oil level sensor.

[0093] In some embodiments, the trajectory prediction module 20 may include: The parsing submodule is used to parse the model input driving quantities from the multidimensional heterogeneous fused state vector; wherein, the model input driving quantities include electrical quantity data components and non-electrical quantity data components; The deduction submodule is used to perform forward deduction calculations on the input driving quantities of the transformer digital twin model using the physical calculation kernel, and obtain the prediction trajectory of key state parameters corresponding to the deduction results at each discrete time point within the prediction time window; the physical calculation kernel is a set of mechanism equations based on the physical laws of heat conduction, the aging reaction kinetics of insulating materials, and the dissolution and release mechanism of gas in oil.

[0094] In some embodiments, the deduction submodule can be specifically used to perform forward deduction calculations on the model input driving quantities using the physical computing kernel, and to calibrate the deduction results using the data-driven correction module of the transformer digital twin model, so as to obtain the predicted trajectory of key state parameters corresponding to the calibrated deduction results at each discrete time point within the prediction time window.

[0095] In some embodiments, the trajectory prediction module 20 may include: The state assessment submodule is used to determine the current health status assessment result of the transformer based on each feature vector in the multidimensional heterogeneous fused state vector and its corresponding preset health benchmark quantity.

[0096] In some embodiments, the boundary determination module 30 may include: The current action threshold determination submodule is used to determine the dynamic current action threshold curve within the prediction time window based on the predicted temperature of the hottest spot in the transformer winding and the current health status assessment results. The dynamic current action threshold curve is the curve corresponding to the maximum allowable load current value at each discrete time point within the prediction time window, provided that the predicted temperature of the hottest spot in the transformer winding does not exceed its respective safety upper limit. The safety upper limit is the predicted temperature of the hottest spot in the transformer winding at the corresponding discrete time point or the corrected absolute safety upper limit. The corrected absolute safety upper limit is a value determined based on the current health status assessment results and the preset absolute safety upper limit. The fluctuation adjustment submodule is used to adjust the dynamic current action threshold curve based on the dynamic current action threshold curve and the grid fluctuation tolerance information to obtain the adjusted dynamic current action threshold curve; wherein, the dynamic safety domain boundary includes the adjusted dynamic current action threshold curve.

[0097] In some embodiments, the operation protection module 40 may include: The current comparison submodule is used to compare each feature vector in the multidimensional heterogeneous fusion state vector with the threshold at the current moment on the corresponding curve in the dynamic security domain boundary; if there is a feature vector greater than the corresponding threshold, the running condition is determined to be an abnormal state. The prediction comparison submodule is used to compare the predicted trajectory of key state parameters with the boundary of the dynamic safety domain if no feature vector is greater than the corresponding threshold. It determines whether there is a predicted value greater than the corresponding threshold in the predicted trajectory of key state parameters within a preset time period in the future. If there is a predicted value greater than the corresponding threshold, the operation is determined to be in a risky state. If there is no predicted value greater than the corresponding threshold, the operation is determined to be in a normal state.

[0098] Corresponding to the above method embodiments, this invention also provides a transformer control device. The transformer control device described below and the transformer control method described above can be referred to each other.

[0099] Please refer to Figure 4 , Figure 4 This is a schematic diagram of a transformer control device provided in an embodiment of the present invention. The control device may include: Memory D1 is used to store computer programs; The processor D2 is used to execute computer programs to implement the steps of the transformer control method provided in the above method embodiments.

[0100] In this embodiment, the control device for the transformer can specifically be a transformer or an edge device near the transformer, such as an intelligent protection device.

[0101] Corresponding to the above method embodiments, this invention also provides a computer program product. The computer program product described below and the transformer control method described above can be referred to each other.

[0102] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the transformer control method provided in the above-described method embodiments.

[0103] Corresponding to the above method embodiments, this invention also provides a computer-readable storage medium. The computer-readable storage medium described below and the transformer control method described above can be referred to in correspondence.

[0104] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the transformer control method described in the above method embodiments.

[0105] The computer-readable storage medium can specifically be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0106] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatuses, devices, computer program products, and computer-readable storage media disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant details can be found in the method section.

[0107] The control method, apparatus, and device for a transformer provided by the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the present invention.

Claims

1. A control method for a transformer, characterized in that, include: Based on the digital monitoring data collected by preset sensors, a multidimensional heterogeneous fusion state vector is obtained at the current moment; wherein, the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data; Based on the multidimensional heterogeneous fused state vector, the transformer digital twin model is used to obtain the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window; wherein, the prediction time window includes the current moment and a future preset time period; the predicted trajectory of key state parameters includes the predicted value of the transformer winding hotspot temperature, the predicted value of insulation aging rate, and the predicted value of dissolved gas generation rate in oil. Based on the current health status assessment results, the predicted trajectories of the key status parameters, and the grid fluctuation tolerance information, the dynamic safety domain boundary within the prediction time window is determined. The dynamic safety domain boundary includes a dynamic current action threshold curve based on thermal stress constraints, an overload tolerance time curve based on insulation life loss constraints, and an oil gas concentration warning threshold curve based on fault gas development constraints. The dynamic current action threshold curve is the curve corresponding to the maximum allowable load current value at each discrete time point within the prediction time window, provided that the predicted temperature of the hottest point of the transformer winding does not exceed its respective safety upper limit. The safety upper limit is the predicted temperature of the hottest point of the transformer winding at the corresponding discrete time point or a corrected absolute safety upper limit. The corrected absolute safety upper limit is determined based on the current... The values ​​determined by the health status assessment results and the preset absolute safety upper limit are: the dynamic current action threshold curve is adjusted according to the dynamic current action threshold curve and the power grid fluctuation tolerance information to obtain the adjusted dynamic current action threshold curve; the dynamic safety domain boundary includes the adjusted dynamic current action threshold curve; the overload tolerance time curve is determined by accumulating the insulation aging rate prediction value in the key state parameter prediction trajectory over a preset time period to obtain the insulation life loss, and combined with the remaining life margin in the current health status assessment results; the oil gas concentration warning threshold curve is derived from the oil dissolved gas production rate prediction value in the key state parameter prediction trajectory to obtain the predicted value of the future key fault gas concentration, and is dynamically determined based on the preset gas concentration attention value and warning value; Based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of the key state parameters, and the dynamic safety domain boundary, the operating status of the transformer is determined, and when the operating status is the target state, corresponding protection actions are executed; wherein, the target state includes abnormal states and states with risks. The step of determining the operating status of the transformer based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of the key state parameters, and the dynamic safety domain boundary includes: Each feature vector in the multidimensional heterogeneous fusion state vector is compared with the threshold value at the current moment on the corresponding curve in the dynamic security domain boundary. If there is a feature vector greater than the corresponding threshold, then the operating condition is determined to be the abnormal state; If there is no feature vector greater than the corresponding threshold, the predicted trajectory of the key state parameter is compared with the boundary of the dynamic safety domain to determine whether there is a predicted value greater than the corresponding threshold in the predicted trajectory of the key state parameter within the future preset time period. If there is a predicted value greater than the corresponding threshold, then the operating condition is determined to be a state of risk. If there is no predicted value greater than the corresponding threshold, then the operating condition is determined to be normal.

2. The transformer control method according to claim 1, characterized in that, The step of obtaining the multi-dimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by preset sensors includes: Based on the timestamp information in the digital monitoring data, the digital monitoring data is time-aligned to obtain time-aligned digital monitoring data. Electrical feature extraction is performed on the electrical quantity data in the time-aligned digital monitoring data to obtain electrical features; wherein, the electrical feature extraction operation includes at least one of effective value calculation, harmonic component extraction and phase calculation; A state feature extraction operation is performed on the non-electrical quantity data in the time-aligned digital monitoring data to obtain state features; wherein, the state feature extraction operation includes rate of change calculation and / or trend analysis; The electrical features and the state features are combined and encapsulated to obtain the multidimensional heterogeneous fused state vector at the current moment.

3. The transformer control method according to claim 1, characterized in that, The preset sensors include electrical quantity sensors and non-electrical quantity sensors. The electrical quantity sensors include voltage sensors and / or current sensors, and the non-electrical quantity sensors include at least one of dissolved gas in oil sensors, top oil temperature sensors, and / or oil level sensors.

4. The transformer control method according to claim 1, characterized in that, The step of obtaining the current health status assessment result and the predicted trajectory of key state parameters within the prediction time window of the transformer based on the multidimensional heterogeneous fused state vector and using the transformer digital twin model includes: The model input driving quantity is parsed from the multidimensional heterogeneous fused state vector; wherein, the model input driving quantity includes electrical quantity data components and non-electrical quantity data components; Using the physical calculation kernel of the transformer digital twin model, the input driving quantity of the model is forward extrapolated to obtain the predicted trajectory of key state parameters corresponding to the extrapolation results at each discrete time point within the prediction time window; wherein, the physical calculation kernel is a set of mechanism equations established based on the physical laws of heat conduction, the aging reaction kinetics of insulating materials, and the dissolution and release mechanism of gas in oil.

5. The transformer control method according to claim 4, characterized in that, The physical calculation kernel of the transformer digital twin model is used to perform forward extrapolation calculations on the input driving quantity of the model, obtaining the predicted trajectory of key state parameters corresponding to the extrapolation results at each discrete time point within the prediction time window, including: Using the physical computing kernel, the input driving quantity of the model is forward extrapolated and calculated, and the extrapolation result is calibrated using the data-driven correction module of the transformer digital twin model, so as to obtain the predicted trajectory of key state parameters corresponding to the calibrated extrapolation result at each discrete time point within the prediction time window.

6. The transformer control method according to claim 1, characterized in that, The step of obtaining the current health status assessment result and the predicted trajectory of key state parameters within the prediction time window of the transformer based on the multidimensional heterogeneous fused state vector and using the transformer digital twin model includes: The current health status assessment result of the transformer is determined based on each feature vector in the multidimensional heterogeneous fusion state vector and its corresponding preset health benchmark quantity.

7. A control device for a transformer, characterized in that, include: The data acquisition module is used to obtain a multi-dimensional heterogeneous fusion state vector at the current moment based on the digital monitoring data collected by the preset sensors; wherein, the preset sensors are used to monitor the transformer, and the digital monitoring data includes electrical quantity data and non-electrical quantity data; The trajectory prediction module is used to obtain the current health status assessment result of the transformer and the predicted trajectory of key state parameters within the prediction time window based on the multidimensional heterogeneous fused state vector and the transformer digital twin model; wherein, the prediction time window includes the current time and a future preset time period; the predicted trajectory of key state parameters includes the predicted value of the transformer winding hotspot temperature, the predicted value of the insulation aging rate, and the predicted value of the dissolved gas generation rate in the oil. The boundary determination module is used to determine the dynamic safety domain boundary within the prediction time window based on the current health status assessment results, the predicted trajectory of the key status parameters, and the grid fluctuation tolerance information. The dynamic safety domain boundary includes a dynamic current action threshold curve based on thermal stress constraints, an overload tolerance time curve based on insulation life loss constraints, and an oil gas concentration warning threshold curve based on fault gas development constraints. The dynamic current action threshold curve is the curve corresponding to the maximum allowable load current value at each discrete time point within the prediction time window, provided that the predicted temperature of the hottest point of the transformer winding does not exceed its respective safety upper limit. The safety upper limit is the predicted temperature of the hottest point of the transformer winding at the corresponding discrete time point or the corrected absolute safety upper limit, where the corrected absolute safety upper limit is the root... The values ​​are determined based on the current health status assessment results and the preset absolute safety upper limit; the dynamic current action threshold curve is adjusted according to the dynamic current action threshold curve and the power grid fluctuation tolerance information to obtain the adjusted dynamic current action threshold curve; the dynamic safety domain boundary includes the adjusted dynamic current action threshold curve; the overload tolerance time curve is determined by accumulating the insulation aging rate prediction value in the key state parameter prediction trajectory over a preset time period to obtain the insulation life loss, and combined with the remaining life margin in the current health status assessment results; the oil gas concentration warning threshold curve is derived from the oil dissolved gas production rate prediction value in the key state parameter prediction trajectory to obtain the predicted value of the future key fault gas concentration, and is dynamically determined based on the preset gas concentration attention value and warning value; The operation protection module is used to determine the operating status of the transformer based on the multidimensional heterogeneous fusion state vector, the predicted trajectory of the key state parameters, and the dynamic safety domain boundary, and to execute corresponding protection actions when the operating status is a target state; wherein, the target state includes abnormal state and state with risk. The operation protection module includes: The current comparison submodule is used to compare each feature vector in the multidimensional heterogeneous fusion state vector with the threshold at the current moment on the corresponding curve in the dynamic security domain boundary; if there is a feature vector greater than the corresponding threshold, the running condition is determined to be an abnormal state. The prediction comparison submodule is used to compare the predicted trajectory of the key state parameters with the boundary of the dynamic safety domain if there is no feature vector greater than the corresponding threshold, and determine whether there is a predicted value greater than the corresponding threshold in the predicted trajectory of the key state parameters within the future preset time period; if there is a predicted value greater than the corresponding threshold, the operating condition is determined to be the state with risk; if there is no predicted value greater than the corresponding threshold, the operating condition is determined to be the normal state.

8. A control device for a transformer, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the transformer control method as described in any one of claims 1 to 6 when executing the computer program.

Images