Transformer life twin evaluation system and method based on carbon driving

Abstract

The invention provides a transformer life twinborn evaluation system and method based on carbon driving, and solves the problems that current transformer life prediction has high requirements on transformer operation historical data and machine learning, the prediction and estimation process is complex, the accuracy of predicting the thermal life of a dynamic degradation transformer system by applying an empirical formula is not high and the like. The method comprises the following steps: extracting real-time data information and periodic data information, analyzing to obtain multi-source heterogeneous data, constructing a real-time estimation model of the thermal life of the transformer according to the real-time data with rich data volume and poor error level in the multi-source heterogeneous data, and performing periodic calibration. And performing curve fitting and epitaxy on the aging rate temperature change coefficient estimated by the calibrated real-time estimation model so as to predict the thermal life of the transformer, the system mainly comprises a transformer entity control module and a transformer digital twinning module, and predicts and estimates the thermal life of the transformer by coordinating and detecting related data through the modules. The method has the advantages of high prediction accuracy and convenient use.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer detection, and particularly to a carbon-driven transformer life twin evaluation system and method. Background Art

[0002] Power transformers are core equipment in substations. Especially oil-immersed power transformers have become one of the most widely used power transformation equipment in the world due to their simple structure, reliable operation and other advantages. The aging state of the internal insulation of transformers is very complex, which can be roughly divided into thermal aging, electrical aging, mechanical aging and environmental aging. Overload, local overheating caused by excessive fluctuating loads, electromagnetic physical fields, and temperature changes in transformers, especially the phenomenon of thermal life loss, not only accelerates insulation aging, but may even induce inter-turn short circuits in severe cases, causing a large amount of economic losses.

[0003] At present, many scholars have carried out in-depth research on the application of digital twin technology in power transformation equipment represented by oil-immersed power transformers. Literature analysis shows that the construction of digital twin models is mainly achieved through two types of methods: First, the model-driven method, which mainly uses empirical formulas, thermal circuit equivalent methods in the GB / T 1094.7-2008 and IEEE Standard C57.91-2011 standards adopted in engineering, and mathematical equation methods based on finite elements, etc. The main problems are: it is difficult to construct a comprehensive physical model that can fit its complexity and practicality for the transformer entity system. According to the calculation of the relative aging rate of the winding heat-modified insulating paper based on the hot spot temperature in the national standard, the consideration is not comprehensive enough, and the result obtained may be greater than the actual life, which poses a safety hazard; Second, the data-driven method, which mainly uses machine learning methods such as backpropagation neural network algorithm, extreme learning machine, and support vector machine. The main problems are: the process of repeated training and iteration in machine learning can only be based on fitting the long-term historical data of the transformer. Although the more training times, the higher the identification accuracy of the output result will be undoubtedly, considering the error levels and credibility of multi-source heterogeneous data, and the dynamically degraded power transformer, the accuracy of functions such as predicting and evaluating the thermal life of power transformers is not high. Therefore, how to improve the accuracy of predicting the thermal life of power transformers is particularly important.

[0004] "A transformer service life prediction system" disclosed in Chinese patent literature, with the publication number CN105717382A, provides a system for estimating the service life of a transformer. By analyzing the data of the transformer on-line monitoring device, a method for obtaining the equivalent service life of the transformer is obtained. According to the hottest spot temperature of the transformer, the acceleration factor FAA of the thermal aging service life and the acceleration factor MAAF of the electro-thermal aging life are calculated. By comparing these two acceleration factors with the service life of the transformer in an ideal environment, the actual service life of the transformer under different operating conditions is obtained, providing a basis for formulating a more reasonable transformer maintenance plan. However, the accuracy of applying the empirical formula to the thermal life prediction of the dynamic degradation transformer system still needs to be improved. Summary of the Invention

[0005] The present invention aims to solve the problems that the current transformer life prediction has high requirements for the transformer operation history data and machine learning, the prediction and estimation process is relatively complex, and the accuracy of applying the empirical formula to predict the thermal life of the dynamic degradation transformer system is not high.

[0006] The above technical problems are solved by the following technical solutions: A carbon-driven transformer life twin evaluation method, including: S1. Monitor the real-time status of various sensors in the transformer to obtain the real-time data information of the transformer, and perform periodic detection on various sensors to obtain the periodic data information of the transformer; S2. Extract the real-time data information and periodic data information and parse them to obtain multi-source heterogeneous data. Use the real-time data in the multi-source heterogeneous data to construct a real-time estimation model for the thermal life of the transformer, and use the periodic data in the multi-source heterogeneous data to perform periodic calibration on the real-time estimation model; S3. Use the calibrated real-time estimation model to estimate the aging rate temperature change coefficient of the transformer, perform curve fitting and extrapolation on the estimated data, and predict the thermal life of the transformer.

[0007] Extract the real-time data information and periodic data information through the corresponding detection unit, perform data analysis to obtain multi-source heterogeneous data, construct a real-time estimation model for the thermal life of the transformer according to the real-time data with rich data volume and slightly poor error level in the multi-source heterogeneous data, and use the periodic data with less data volume and higher error level in the multi-source heterogeneous data to perform periodic calibration on the real-time estimation model. Curve fitting and extrapolation are performed on the aging rate temperature change coefficient estimated by the calibrated real-time estimation model, so as to predict the thermal life of the transformer. Greatly improve the accuracy of applying the empirical formula in the relevant national standards to the thermal life prediction of the dynamic degradation transformer system.

[0008] Preferably, in step S2, the extraction of real-time data information and periodic data information and the parsing to obtain multi-source heterogeneous data include: performing real-time periodic processing on the real-time data information, calculating the weighted error level and credibility of the real-time data information to obtain real-time normalized sensor data; performing weighted processing on the periodic data information to obtain periodic sensing data; combining the real-time normalized sensor data and the periodic sensing data to form multi-source heterogeneous data. Among the obtained multi-source heterogeneous data, an estimation model established based on real-time data with rich data volume and slightly poor error level has strong real-time performance and high accuracy.

[0009] Preferably, the calculation method of the real-time normalized sensor data is S real (t) = Data real (p i )·ε j (p i , p i+1 )·Π i∈T(i) η i (p i ) / z i , where S real (t) is the real-time normalized sensor data at the transformer estimation time t; Data real (p i ) is the sensor data after real-time periodic processing at the sampling time i at the transformer position p; ε i (p i , p i+1 ) is the transfer error from the current sampling time i to the next sampling time i + 1; Π i∈T(i) η i (p i ) is the combined credibility, indicating the cumulative credibility at all sampling times within the detection period T real (t); Z i is the normalization constant, depending on the reference calibration value of the sensor.

[0010] Preferably, the periodic sensing data is S period (t + j·T period ) = Σ j∈N(j) k j ·Data period (q j ), where S period (t + j·T period ) is the periodic sensing data after t + j offline periods T period for the transformer estimation time; Data period (q j ) is the offline period Tperiod The original data obtained by N in - device detection devices; k j is the weighting coefficient corresponding to N detection devices, Σ j∈N(j) k j = 1.

[0011] Preferably, in step S2, using the periodic data in the multi - source heterogeneous data to perform periodic calibration on the real - time estimation model includes: traversing the periodic data in the multi - source heterogeneous data, periodically calibrating the real - time estimation model to obtain an empirical formula for the calibrated aging rate temperature change coefficient, and iterating cyclically until the empirical formula for the calibrated aging rate temperature change coefficient is within a preset range, thereby obtaining the calibrated real - time estimation model.

[0012] Preferably, in step S3, predicting the thermal life of the transformer specifically includes: using the calibrated real - time estimation model to estimate a new aging rate temperature change coefficient, performing curve fitting and extrapolation on the new aging rate temperature change coefficient using the non - linear Boltzmann function, so that the sum of the squared deviations between the experimental values and the fitted values at each time node is minimized, and completing the prediction of the transformer's thermal life.

[0013] Preferably, in step S1, the real - time data information includes the ambient temperature, top - layer oil temperature, load change, core grounding current, and discharge information of various sensors. The discharge information includes the discharge quantity, discharge times, and discharge potential. The periodic data information includes the grease polymerization degree, water content, and dissolved gas content of various sensors.

[0014] Preferably, after predicting the thermal life of the transformer in step S3, it further includes: maintaining data communication with the outside world and logging the evaluation result of the transformer's thermal life. Uploading the evaluation result of the transformer's thermal life is convenient for subsequent analysis and data management.

[0015] The present invention also provides a system for transformer life twin evaluation and adaptation using the above - mentioned method, including: a transformer entity control module, including an online monitoring unit and an offline detection unit. The online monitoring unit and the offline detection unit are respectively connected to various sensors; a transformer digital twin module, including a multi - source heterogeneous data processing unit and a thermal life loss prediction unit. The online monitoring unit and the offline detection unit are respectively connected to the multi - source heterogeneous data processing unit, and the multi - source heterogeneous data processing unit is connected to the thermal life loss prediction unit.

[0016] The online monitoring unit is used to real - time monitor the ambient temperature, top - layer oil temperature, load change, core grounding current, discharge quantity, discharge times, and discharge potential of various sensors in the transformer to obtain the real - time data information of the transformer; the offline detection unit is used to periodically detect the grease polymerization degree, water content, and dissolved gas content of various sensors in the transformer to obtain the periodic data information of the transformer.

[0017] The multi-source heterogeneous data processing unit is used to extract real-time data information and periodic data information, parse them to obtain multi-source heterogeneous data, construct a real-time estimation model of the transformer thermal life using the real-time data in the multi-source heterogeneous data, and perform periodic calibration on the real-time estimation model using the periodic data in the multi-source heterogeneous data. The multi-source heterogeneous data processing unit can perform real-time periodic processing on the real-time data information, and perform weighted calculations on the error level and credibility of the real-time data information to obtain real-time normalized sensor data; perform weighted processing on the periodic data information to obtain periodic sensing data; the real-time normalized sensor data and the periodic sensing data constitute multi-source heterogeneous data. The thermal life prediction module is used to perform curve fitting and extrapolation on the aging rate temperature change coefficient estimated by the calibrated real-time estimation model, so as to predict the thermal life of the transformer.

[0018] Preferably, the transformer entity control module further includes a fault diagnosis unit. The fault diagnosis unit is provided with a communication device. The thermal life loss prediction unit is connected to the communication device, and the communication device is connected to the evaluation data processing unit. The fault diagnosis unit is mainly used for communicating with the outside world and recording the log of the transformer thermal life evaluation result.

[0019] The beneficial effects of the present invention are as follows: extract real-time data information and periodic data information, parse them to obtain multi-source heterogeneous data, construct a real-time estimation model of the transformer thermal life according to the real-time data with rich data volume and slightly poor error level in the multi-source heterogeneous data, and perform periodic calibration. Perform curve fitting and extrapolation on the aging rate temperature change coefficient estimated by the calibrated real-time estimation model, so as to predict the thermal life of the transformer, greatly improving the accuracy of the thermal life prediction when the empirical formula in the relevant national standard is applied to the dynamic degradation transformer system, and achieving the purpose of energy conservation and emission reduction in the substation. Description of the Drawings

[0020] Figure 1 is the flowchart of the method of the present invention.

[0021] Figure 2 is the structural schematic diagram of the present invention.

[0022] Figure 3 is the environmental temperature sensing data diagram of the present invention.

[0023] Figure 4 is the load factor sensing data diagram of the present invention.

[0024] Figure 5 is the top oil temperature sensing data diagram of the present invention.

[0025] Figure 6 is the aging rate temperature change coefficient curve diagram of the present invention.

[0026] Figure 7It is the estimated curve graph of the aging rate temperature change coefficient after calibration of the present invention. Detailed implementation manners

[0027] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application and should not be construed as a limitation to the present application.

[0028] Embodiment 1: This embodiment provides a carbon-driven twin evaluation method for transformer life, as Figure 1 shown, including: S1. Monitor the real-time status of various sensors in the transformer to obtain the real-time data information of the transformer, and perform periodic detection on various sensors to obtain the periodic data information of the transformer; S2. Extract the real-time data information and periodic data information and analyze them to obtain multi-source heterogeneous data. Use the real-time data in the multi-source heterogeneous data to construct a real-time estimation model for the thermal life of the transformer, and use the periodic data in the multi-source heterogeneous data to perform periodic calibration on the real-time estimation model; S3. Use the calibrated real-time estimation model to estimate the aging rate temperature change coefficient of the transformer, perform curve fitting and extrapolation on the estimated data, and predict the thermal life of the transformer.

[0029] In S1, the real-time data information includes the ambient temperature, top oil temperature, load change, core grounding current, and discharge information of various sensors. The discharge information includes the discharge amount, discharge times, and discharge potential. The periodic data information includes the grease polymerization degree, water content, and dissolved gas content of various sensors.

[0030] In S2, extracting the real-time data information and periodic data information and analyzing them to obtain multi-source heterogeneous data includes: performing real-time periodic processing on the real-time data information, performing weighted calculation on the error level and credibility of the real-time data information to obtain real-time normalized sensor data, and the calculation method of the real-time normalized sensor data is S real (t) = Data real (p i )·ε i (p i , p i+1 )·Π i∈T(i) η i (p i ) / z i , where, S real (t) is the real-time normalized sensor data of the transformer at the estimated time t; Data real(p i ) is the sensor data after real-time periodic processing at the sampling moment i located at the transformer position p; ε i (p i , p i+1 ) is the transfer error from the current sampling moment i to the next sampling moment i + 1; Πi ∈T(i) η i (p i ) is the combined credibility, representing the cumulative credibility at all sampling moments within the detection period T real (t); Z i is a normalization constant, depending on the reference calibration value of the sensor.

[0031] Perform weighted processing on the periodic data information to obtain periodic sensing data; the periodic sensing data is S period (t + j·T period ) = Σ j∈N(j) k j ·Data period (q j ), where S period (t + j·T period ) is the periodic sensing data after t + j offline periods T period for the transformer estimation moment; Data period (q j ) is the raw data obtained by N detection devices within the offline period T period ; k j is the weighting coefficient corresponding to N detection devices, and Σ j∈N(j) k j = 1. Combine the real-time normalized sensor data and the periodic sensing data to form multi-source heterogeneous data.

[0032] Performing periodic calibration of the real-time estimation model according to the periodic data in the multi-source heterogeneous data includes: traversing the periodic data in the multi-source heterogeneous data, periodically calibrating the real-time estimation model to obtain an empirical formula for the calibrated aging rate temperature change coefficient, and iterating in a loop until the empirical formula for the calibrated aging rate temperature change coefficient is within a preset range to obtain the calibrated real-time estimation model.

[0033] In S3, predicting the thermal life of the transformer specifically includes: using the calibrated real-time estimation model to estimate a new aging rate temperature change coefficient, performing curve fitting and extrapolation on the new aging rate temperature change coefficient using the non-linear Boltzmann function, and achieving the minimum sum of squared deviations between the experimental values and the fitted values at each time node to complete the prediction of the thermal life of the transformer.

[0034] After completing the prediction of the transformer's thermal life, the prediction and evaluation results are interacted with the outside world through the communication device of the fault diagnosis unit, and the log of the transformer's thermal life evaluation results is recorded for subsequent analysis and data management.

[0035] This embodiment also provides a carbon-driven transformer life twin evaluation system adapted to the above method, as Figure 2 shown, which mainly includes a transformer entity control module and a transformer digital twin module. The transformer entity control module includes an on-line monitoring unit and an off-line detection unit, and the on-line monitoring unit and the off-line detection unit are respectively connected to various sensors; the transformer digital twin module includes a multi-source heterogeneous data processing unit and a thermal life loss prediction unit, and the on-line monitoring unit and the off-line detection unit are respectively connected to the multi-source heterogeneous data processing unit, and the multi-source heterogeneous data processing unit is connected to the thermal life loss prediction unit.

[0036] In addition, the transformer entity control module further includes a fault diagnosis unit. The fault diagnosis unit is provided with a communication device, the thermal life loss prediction unit is connected to the communication device, and the communication device is connected to the evaluation data processing unit.

[0037] Embodiment 2: This embodiment provides an actual scenario for detecting the transformer life by applying its adapted system based on the above evaluation method, and adopts the following technical means: The oil-immersed direct-cooled power transformer in the substation is monitored in real time for a time span of 2 years, and the sampling interval is 1 hour. The ambient temperature sensing data is as Figure 3 shown, denoted as S real (t, p1), and its value is calculated from the data of the AD590 crystal temperature sensor monitored in real time, denoted as Data real (p1). Among them, the accuracy of the temperature sensor is 1°C, the combined credibility Π i∈T(i) η i (p i ) takes a value of 99%, the transfer error ε i (p i , p i+1 ) takes a value of 0.99, and the normalization constant takes a value of 1. The data S real (t, p1) shows that it significantly changes seasonally, with an average minimum temperature of about -8°C and an average maximum temperature of about 38°C.

[0038] The real-time transformer load factor K(t) at the corresponding time is as Figure 4As shown, by using the conversion model between the load rate and the hot spot temperature, the calculation results of heat sources with different load rates can be imported into the digital twin component of the transformer to estimate the hot spot temperature under different load rates. IEC 60076-7-2005 Power transformers - Part 7: Loading guide for oil-immersed power transformers gives the calculation method for the temperature rise of each part of the oil-immersed transformer under rated load, and points out that the load factor K(t, p2), the ambient temperature θ a , the top oil temperature rise Δθ 顶层 , the temperature rise Δθ 绕组 of the transformer winding, etc. have a great influence on the winding hot spot temperature θ h . Specifically, θ h = θ a + Δθ 顶层 + Δθ 绕组 , and Δθ 顶层 , Δθ 绕组 can be calculated by Equation (1):

[0039] where Δθ or is the top oil temperature rise; R is the ratio of the load loss to the air loss under rated current; H is the hot spot factor; x is the oil index; y is the winding index; g r is the gradient of the average temperature of the winding to the average temperature of the transformer oil under rated current. The above parameters can be determined according to the cooling method. For example, for an ONAN type transformer, Δθ or = 52, R = 6, H = 1.3, x = 0.8, y = 1.3, g r = 14.5. Substituting the S Figure 3 (t, p1) data in real into the calculation, the winding hot spot temperature distribution during the operation period of the transformer can be obtained, denoted as S real (t, p2), as shown in Figure 5 . Among them, a relatively high point temperature of about 122 °C appears around May 15th in the first year.

[0040] According to the national standard, the relationship between the thermal aging rate V at the sampling time n and the winding hot spot temperature θ h can be calculated by Equation (2):

[0041] where S is the service life set when the transformer leaves the factory, T is the remaining life of the transformer, and Δt is the duration of the transformer working at the current hot spot temperature. When the transformer winding temperature is in the range of 80 - 130 °C, the aging rate temperature change coefficient P is a constant in the Montsinger formula. The corresponding calculation in the national standard makes P = 6, which is also called the 6 °C rule, that is, for every 6 °C increase in temperature, the insulation life of the transformer is reduced by half. From this, the thermal life loss curve of the transformer can be deduced, as shown inFigure 6 As shown, denoted as

[0042] When the hot-spot temperature of the transformer is below 98 °C, its relative thermal aging rate is 1. When the hot-spot temperature of the winding exceeds 98 °C, the relative thermal aging rate rises sharply. As Figure 5 shown, when the hot-spot temperature exceeds 122 °C, the relative thermal aging rate is as high as 16. If the duration is 1 hour, the loss of the thermal life of the transformer can be as high as 16 hours.

[0043] Periodically detect the degree of polymerization of grease, water content, and dissolved gas content of various sensors in the transformer to obtain the periodic data information of the transformer. The specific implementation is as follows: the content of dissolved gases, such as carbon monoxide, carbon dioxide, hydrogen, etc., and the content of furfural and its derivatives, etc. The degree of polymerization (DP) of oil-paper of an oil-immersed direct-cooled power transformer in a certain substation was detected at an interval of 6 months, that is, T period = 180 days. The relationship between the decrease in the degree of polymerization of the oil-paper insulation of the transformer and the aging rate k1 is 1 / DP t - 1 / DP0 = k1t, where DP0 is the initial degree of insulation polymerization and DP t is the degree of insulation polymerization at time t. However, the quantitative analysis steps of DP are complicated, and the specific steps are carried out with reference to GB1548-2006, which seriously affects the efficiency of the digital twin modeling of the transformer. Therefore, in actual engineering, the relatively high linear correlation between CO2 and DP in the dissolved gas (DAG) detection information is usually used, or the DP value of the transformer is predicted from the furfural content, and then the remaining operating life of the transformer is calculated from the k1 value and the DP threshold. As shown in Equation (3), a method for evaluating the life of a transformer with DP as a parameter, where η is the characteristic life, F is the lower limit value of DP, C0 is the initial water content in the oil-paper, T is the temperature, and A and B are undetermined constants related to the transformer to be measured, which can be determined by experience:

[0044] The off-line detection of DGA / furfural of the transformer is carried out at the following times: T0 (November 15th of the first year), T1 (May 15th of the second year), T2 (November 15th of the second year), and T3 (May 15th of the third year). According to the foregoing research results, the life loss of the transformer to be measured at the above time nodes is calculated in this embodiment. The life loss at T0 days; the life loss at T1 days; the life loss at T2 days; the life loss at T3 days.

[0045] Extract real-time data information and periodic data information for parsing to obtain multi-source heterogeneous data. Use the real-time data in the multi-source heterogeneous data to construct a real-time estimation model for the thermal life of the transformer, and use the periodic data in the multi-source heterogeneous data to perform periodic calibration on the real-time estimation model. The specific implementation is as follows: The life loss values obtained based on chemical characteristic parameters at times T0 to T3 are used to segmentally calibrate the life loss of the original hot-spot temperature rise obtained from formula (2). Analysis shows that when a series of values of P range from 5.63 to 8.24, it can more accurately correspond to different insulation aging types and better predict the thermal life of the transformer. Figure 7 The value of the aging rate temperature change coefficient after periodic calibration and the thermal life estimation shown by the solid line part in. The life loss prediction obtained at times T0 to T3 is denoted as and Among them, for the real-time estimation model before time T0, at this time, the aging rate temperature change coefficient in formula (2) is corrected, that is, P1 = 8.24; for the real-time estimation model within the period from T0 to T1, at this time, P2 = 7.55; for the real-time estimation model within the period from T1 to T2, at this time, P3 = 5.63; for the real-time estimation model within the period from T2 to T3, at this time, P4 = 7.26. Generally speaking, the calibrated real-time estimation model reflects a more severe aging trend of the transformer. The direct reason is that the calibrated real-time estimation model takes into account more aging factors and characteristics, and the processing of multi-source heterogeneity improves the accuracy of the real-time estimation model.

[0046] Curve fit and extend the aging rate temperature change coefficient estimated by the calibrated real-time estimation model to predict the thermal life loss of the transformer.

[0047] In this embodiment, the thermal life loss deviation between the reference base point and the calibrated real-time estimation model is compared and analyzed. The selected comparison samples are the sampling values of the above-mentioned Boltzmann function curve fitting and extension at times T1 to T3, that is, and See Figure 7 the dotted line part in, that is, complete the prediction of the life of the transformer after the current time node for medium and short time spans, such as 6 months later.

[0048] The following table shows the Boltzman fitting parameters for the thermal life prediction of the transformer by the calibrated real-time estimation model. In this embodiment, Origin fitting is used, and the date / hour coding format is used as the data grouping identifier. The value of parameter x0 is affected by this:

[0049] Analysis of the obtained data shows that at time T1, the prediction deviation of the classical model is about 30%, and the prediction deviation after calibration is about 31%. The difference between the two is not significant. However, it should be noted that the prediction of the classical model is based on the transformer operation data before time T1, while the prediction after calibration is only based on the data before time T0. The superiority of the two is immediately apparent. The calibrated real-time estimation model only requires about half the number of transformer operation data under the condition of approximate deviation; at time T2, the prediction deviation of the classical model is about 25%, and the prediction deviation after calibration is about 10%. The difference between the two gradually increases; at time T3, the prediction deviation of the classical model is about 27%, and the prediction deviation after calibration is about 5%. The difference between the two is significant. Thus, it can be seen that the calibratable thermal life prediction method is more accurate than the classical model.

[0050] It should be noted that by traversing the periodic data in the multi-source heterogeneous data and performing periodic calibration on the real-time estimation model, an empirical formula for the aging rate temperature change coefficient after calibration is obtained, where P0 is the initial value of the aging rate temperature change coefficient; the value of Δp depends on the working state of the transformer and ranges from (3, 5). Thus, the calibrated empirical formula is iteratively looped and applied to the thermal life prediction of dynamic degradation transformers at different stages to improve accuracy.

[0051] In summary, the above embodiments disclose a calibratable carbon-driven transformer life twin evaluation system and method. By extracting real-time data information and periodic data information for analysis to obtain multi-source heterogeneous data, a real-time estimation model of the transformer thermal life is constructed using the real-time data with rich data volume and slightly worse error level in the multi-source heterogeneous data, and the real-time estimation model is periodically calibrated using the periodic data with less data volume and higher error level in the multi-source heterogeneous data. The aging rate temperature change coefficient estimated by the calibrated real-time estimation model is curve-fitted and extrapolated to predict the transformer thermal life.

[0052] The specific embodiments described in this application are only illustrative of the spirit of this application. Those skilled in the art of this application can make various modifications or supplements to the described specific embodiments or use similar methods for substitution, but will not deviate from the spirit of this application or exceed the scope defined by the appended claims.

[0053] Although this application uses terms such as transformer entity control module, transformer digital twin module, online monitoring unit, offline detection unit, multi-source heterogeneous data processing unit, and thermal life loss prediction unit more frequently, the possibility of using other terms is not excluded. Using these terms is only to describe and understand the essence of this application more conveniently; interpreting them as any additional limitation is contrary to the spirit of this application.

Claims

1. A carbon-driven twin evaluation method for transformer life, characterized in that, Including: S1. Monitor the real-time status of various sensors in the transformer to obtain the real-time data information of the transformer, and perform periodic detection on various sensors to obtain the periodic data information of the transformer; S2. Extract the real-time data information and periodic data information and parse them to obtain multi-source heterogeneous data. Use the real-time data in the multi-source heterogeneous data to construct a real-time estimation model for the thermal life of the transformer, and use the periodic data in the multi-source heterogeneous data to perform periodic calibration on the real-time estimation model; S3. Use the calibrated real-time estimation model to estimate the aging rate temperature change coefficient of the transformer, perform curve fitting and extrapolation on the estimated data, and predict the thermal life of the transformer.

2. The carbon-driven twin evaluation method for transformer life according to claim 1, characterized in that, In step S2, the extraction of the real-time data information and periodic data information and parsing them to obtain multi-source heterogeneous data includes: performing real-time periodic processing on the real-time data information, performing weighted calculation on the error level and credibility of the real-time data information to obtain real-time normalized sensor data; performing weighted processing on the periodic data information to obtain periodic sensing data; combining the real-time normalized sensor data and the periodic sensing data to form multi-source heterogeneous data.

3. The carbon-driven twin evaluation method for transformer life according to claim 2, characterized in that, The calculation method of the real-time normalized sensor data is S real S(t) = Data real (p i )·ε i (p i ,p i+1 )·Π i∈T(i) η i (p i ) / z i , Among them, S real (t) is the real-time normalized sensor data of the transformer at the estimated time t; Data real (p i ) is the real-time periodic processed sensor data at the sampling time i located at the transformer position p; ε i (p i , p i+1 ) is the transfer error from the current sampling time i to the next sampling time i + 1; Π i∈T(i) η i (p i ) is the joint credibility, indicating the cumulative credibility at all sampling times within the detection period T real (t); Z i is a normalization constant, which depends on the reference calibration value of the sensor.

4. The carbon-driven twin evaluation method for transformer life according to claim 2, characterized in that, The periodic sensing data is S period (t + j·T period ) = ∑ j∈N(j) k j ·Data period (q j ) Among them, S period (t + j·T period ) is the periodic sensing data of the transformer at the estimation time of t + j offline cycles T period later; Data period (q j ) is the original data obtained by N detection devices within the offline cycle T period ; k j is the weighting coefficient corresponding to N detection devices, ∑ j∈N(j) k j = 1.

5. The carbon-driven twin evaluation method for transformer life according to claim 1, characterized in that, In step S2, the use of the periodic data in the multi-source heterogeneous data to perform periodic calibration on the real-time estimation model includes: traversing the periodic data in the multi-source heterogeneous data, periodically calibrating the real-time estimation model to obtain an empirical formula for the calibrated aging rate temperature change coefficient, and performing cyclic iteration until the empirical formula for the calibrated aging rate temperature change coefficient is within a preset range to obtain the calibrated real-time estimation model.

6. The carbon-driven twin evaluation method for transformer life according to claim 5, characterized in that, In step S3, the specific prediction of the thermal life of the transformer includes: using the calibrated real-time estimation model to estimate a new aging rate temperature change coefficient, performing curve fitting and extrapolation on the new aging rate temperature change coefficient using a non-linear Boltzmann function, and achieving the minimum sum of the squared deviations between the experimental values and the fitted values at each time node to complete the prediction of the thermal life of the transformer.

7. The carbon-driven twin evaluation method for transformer life according to claim 1 or 2, characterized in that, In step S1, the real-time data information includes the ambient temperature, top oil temperature, load change, core grounding current, and discharge information of various sensors. The discharge information includes the discharge amount, discharge times, and discharge potential. The periodic data information includes the grease polymerization degree, water content, and dissolved gas content of various sensors.

8. The carbon-driven twin evaluation method for transformer life according to claim 1 or 6, characterized in that, In step S3, after predicting the thermal life of the transformer, it further includes: maintaining data communication with the outside world and recording the log of the evaluation result of the thermal life of the transformer.

9. A carbon-driven twin evaluation system for transformer life, adopting the method in any one of claims 1-8, characterized in that, Including: The transformer entity control module includes an on-line monitoring unit and an off-line detection unit. The on-line monitoring unit and the off-line detection unit are respectively connected to various sensors; The transformer digital twin module includes a multi-source heterogeneous data processing unit and a thermal life loss prediction unit. The on-line monitoring unit and the off-line detection unit are respectively connected to the multi-source heterogeneous data processing unit, and the multi-source heterogeneous data processing unit is connected to the thermal life loss prediction unit.

10. The carbon-driven twin evaluation system for transformer life according to claim 9, characterized in that, The transformer entity control module further includes a fault diagnosis unit. The fault diagnosis unit is provided with a communication device. The thermal life loss prediction unit is connected to the communication device, and the communication device is connected to the evaluation data processing unit.

Images