A transformer operation and maintenance method based on digital twin technology

Abstract

The present invention discloses a transformer operation and maintenance method based on digital twin technology, comprising the following steps: constructing a digital twin model of the transformer and separately constructing fault sample vectors for multiple components; periodically obtaining the actual values of various operation status indicators of the transformer and comparing the actual values with the normal values; establishing an abnormal status vector according to the results of the above comparison step; matching the abnormal status vector with all the fault sample vectors one by one; if there is a matching fault sample vector, a component fault prompt is issued, and if there is no matching fault sample vector, an indicator abnormality prompt is issued. When different components have problems, they will have different impacts on various operation status indicators of the transformer. By comparing the various operation status indicators of the transformer with the fault sample vectors of the components, it is determined which components cannot work properly, and then a reminder is sent to the maintenance personnel, improving the maintenance efficiency of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of power transformers, and particularly relates to a transformer operation and maintenance method based on digital twin technology. Background Art

[0002] A transformer is a commonly used device in the power system and is one of the important hubs for power grid transmission. Since faults are likely to occur during the use of the transformer, it is necessary to perform operation and maintenance as well as regular inspections on it. With the gradual maturity of various types of sensor technologies, it has become very convenient to monitor the various index parameters of the transformer during operation in real time. Therefore, the real-time monitoring and condition analysis of the transformer have been studied more and more deeply. The current monitoring methods for the operation status of the transformer mainly include the characteristic gas method and the three-ratio method, both of which analyze the components of the gas generated by the transformer oil in the transformer to determine whether there are faults or abnormalities in the transformer. However, through the analysis of these two methods, only the fault types of the transformer can be judged, such as arc discharge, partial discharge, spark discharge, high-temperature overheating, etc., and it is impossible to accurately give which specific component has a fault. When the maintenance personnel find that the transformer status is abnormal, they still need to check each component one by one, which brings inconvenience to the maintenance work. Summary of the Invention

[0003] In order to solve the technical problem that the existing transformer monitoring technology cannot accurately judge the working status of each component, the present application provides a transformer operation and maintenance method based on digital twin technology, including the following steps:

[0004] S1. Construct a digital twin model of the transformer, and construct fault sample vectors for multiple components respectively;

[0005] S2. Periodically obtain the actual values of the various operation status indicators of the transformer, and compare the actual values with the normal values corresponding to the operation status indicators;

[0006] S3. Establish an abnormal state vector according to the results of the above comparison step;

[0007] S4. Match the abnormal state vector with all the fault sample vectors one by one;

[0008] S5. If there is a matching fault sample vector, issue a component fault prompt; if there is no matching fault sample vector, issue an index abnormality prompt.

[0009] When different components have problems, they will have different impacts on the various operation status indicators of the transformer. Therefore, by analyzing the differences between the various operation status indicators and the normal values, it is possible to accurately determine which components are in an abnormal working state, and then remind the maintenance personnel to check and maintain the transformer in time, effectively improving the maintenance efficiency of the transformer.

[0010] Specifically, the fault sample vector is obtained by the following method:

[0011] S11. For a component, obtain multiple historical fault data, and construct a corresponding historical fault vector according to each piece of historical fault data. Each component of the historical fault vector corresponds to a state value of an operating state index when the transformer fails;

[0012] S12. Calculate the weights of the operating state indicators according to the number of abnormalities of the operating state indicators in all the historical fault vectors;

[0013] S13. Calculate the weighted average vector of the multiple historical fault vectors;

[0014] S14. Modify the weighted average vector to obtain the fault sample vector.

[0015] Considering that different degrees of damage to a component will cause different changes in the operating state indicators of the transformer, so when the same component fails each time, it is not always the same few operating state indicators that are abnormal. By calculating the proportion of the number of abnormalities of each operating state indicator in the fault, the weights of each operating state indicator are changed to make the judgment result more accurate.

[0016] Further, step S14 further includes:

[0017] S141. Obtain new historical fault vectors, and use a neural network to fit all the historical fault vectors to obtain a historical fault fitting vector;

[0018] S142. Construct a loss function according to the historical fault fitting vector and the weighted average vector;

[0019] S143. Modify the weighted average vector according to the loss function to obtain the fault sample vector.

[0020] During the use of the transformer, new faults will also occur. According to the new historical fault vectors, the original fault sample vector is corrected by machine learning to further improve the accuracy of the judgment result.

[0021] Further, the normal value is obtained by the following method:

[0022] S21. Obtain the operating load data and external environment data of the transformer in real time;

[0023] S22. Substitute the operating load data and the external environment data into the digital twin model to obtain the predicted values of the operating state indicators of the transformer;

[0024] S23. Modify the predicted value to obtain the normal value.

[0025] The operating state indicators of the transformer are determined by the operating load data and external environment data of the transformer. Therefore, based on the geometric model of the transformer and combined with mechanics, electromagnetics, thermodynamics, etc., the digital twin model of the transformer can calculate the predicted values of various operating state indicators according to the operating load data and external environment data. This predicted value can be used as the basis for judging whether the operating state indicators of the transformer are normal.

[0026] Furthermore, step S23 further includes:

[0027] S231. Calculate the fitted value of the operating state indicator using a neural convolutional network according to the historical data of the operating state indicator.

[0028] S232. Construct a loss function based on the fitted value and the predicted value.

[0029] S233. Modify the predicted value according to the loss function to obtain the normal value.

[0030] Considering that the digital twin model is an idealized model of the transformer, there will be a certain deviation between the predicted value obtained from it and the true value. Therefore, through machine learning, the predicted value is modified according to the historical data of the operating state indicator to make the obtained normal value more accurate.

[0031] Furthermore, the abnormal state vector is obtained through the following steps:

[0032] S31. Construct an empty vector. The dimension of the empty vector is the same as that of the fault sample vector, and the operating state indicator corresponding to each component of the empty vector is the same as that of the fault sample vector.

[0033] S32. For the actual value of each obtained operating state indicator, calculate the absolute difference between the actual value and the normal value, and compare the absolute difference with a first threshold.

[0034] S33. If the absolute difference is greater than the first threshold, mark the component corresponding to the actual value in the empty vector as abnormal. If the absolute difference is less than the first threshold, mark the component corresponding to the actual value in the empty vector as normal.

[0035] Specifically, in step S4, the process of matching the abnormal state vector with all the fault sample vectors one by one includes:

[0036] S41. Calculate the norm of the abnormal state vector and the norm of the fault sample vector, and calculate the absolute difference between the norms of the two vectors.

[0037] S42. Calculate the angle between the abnormal state vector and the fault sample vector.

[0038] S43. When the absolute difference between the norms of the two vectors is less than the second threshold and the angle between the two vectors is less than the third threshold, it is determined that the abnormal state vector matches the fault sample vector.

[0039] Further, the steps of issuing an indicator anomaly prompt in step S5 include:

[0040] S51. Construct a visualization model of the transformer according to the digital twin model and the geometric model of the transformer.

[0041] S52. Issue a component fault prompt through the visualization model.

[0042] The digital twin model can conveniently visualize the working state of the transformer, enabling maintenance personnel to more directly observe the state of the transformer.

[0043] Further, the steps of issuing an indicator anomaly prompt in step S5 include:

[0044] S53. Issue an indicator anomaly prompt through the visualization model.

[0045] Further, the steps of issuing an indicator anomaly prompt in step S5 include:

[0046] S54. Calculate the percentage of the ratio of the actual value to the normal value.

[0047] S55. Issue an indicator anomaly prompt through the visualization model and display the percentage.

[0048] In some special cases, when only one operating state indicator is abnormal and it is impossible to accurately determine the specific faulty component, it is necessary to remind the maintenance personnel for further confirmation.

[0049] The technical effects and advantages of the present invention:

[0050] 1. By establishing a digital twin model of the transformer, calculating the normal values of the various operating state indicators of the transformer according to the load state of the transformer, comparing the actual values of the various operating state indicators of the measured transformer with the normal values, determining which operating state indicator of the transformer is abnormal, and judging the faulty components of the transformer according to the abnormal operating state indicators, so as to remind the maintenance personnel for precise maintenance.

[0051] 2. Continuously correct the normal values of various operating status indicators according to the historical operating data of the transformer through machine learning to improve the judgment accuracy of component failures. Brief Description of the Drawings

[0052] Figure 1 It is a schematic diagram of the overall operation and maintenance process of the transformer of the present invention;

[0053] Figure 2 It is a schematic diagram of the process for obtaining the normal values of various operating status indicators of the transformer of the present invention;

[0054] Figure 3 It is a schematic diagram of the process for obtaining the fault sample vector of any component of the transformer of the present invention;

[0055] Figure 4 It is a schematic diagram of the process for correcting the fault sample vector of any component of the transformer of the present invention;

[0056] Figure 5 It is a schematic diagram of the process for correcting the predicted values of various operating status indicators of the present invention. Detailed Embodiment

[0057] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0058] Embodiment 1

[0059] Refer to Figure 1 , Embodiment 1 of the present invention provides a transformer operation and maintenance method based on digital twin technology, including the following steps:

[0060] S1. Build a digital twin model of the transformer and build fault sample vectors for multiple components respectively;

[0061] S2. Periodically and real-time obtain the actual values of various operating status indicators of the transformer and compare the actual values with the normal values corresponding to the operating status indicators;

[0062] S3. Establish an abnormal state vector according to the results of the above comparison step;

[0063] S4. Match the abnormal state vector with all the fault sample vectors one by one;

[0064] S5. If there is a matching fault sample vector, issue a component fault prompt; if there is no matching fault sample vector, issue an indicator abnormality prompt.

[0065] The digital twin model is a mathematical model of the transformer established based on the geometric model of the transformer, combined with mechanics, electromagnetics, thermodynamics, etc., so as to accurately calculate the operating state indicators of the transformer according to the load data of the transformer. The operating state indicators of the transformer can include transformer oil temperature, transformer oil level, core temperature, winding temperature, core grounding current, etc., and the load data of the transformer includes high-voltage side voltage, high-voltage side current, low-voltage side voltage, low-voltage side current, ambient temperature, and ambient humidity, etc.

[0066] When each component of the transformer is working normally, the operating state indicators of the transformer under a specific load condition can be accurately predicted through the digital twin model. When a certain component of the transformer has a problem, it will cause several specific operating state indicators to deviate from the normal values. Different components having problems will have different impacts on the operating state indicators of the transformer. Therefore, by analyzing the differences between the operating state indicators and the normal values, it is possible to accurately determine which components are in abnormal working states.

[0067] To sum up, as Figure 2 shown, the normal value can be obtained through the following method:

[0068] S21. Obtain the operating load data and external environment data of the transformer in real time;

[0069] S22. Substitute the operating load data and external environment data into the digital twin model to obtain the predicted values of the operating state indicators of the transformer;

[0070] S23. Correct the predicted values to obtain the said normal value.

[0071] Specifically, as Figure 3 shown, the fault sample vector is obtained through the following method:

[0072] S11. For a component, obtain multiple historical fault data, and construct corresponding historical fault vectors according to each historical fault data. Each component of the historical fault vector corresponds to a state value of an operating state indicator when the transformer fails;

[0073] S12. Calculate the weights of the operating state indicators according to the number of abnormalities of the operating state indicators in all historical fault vectors;

[0074] S13. Calculate the weighted average vector of multiple historical fault vectors;

[0075] S14. Correct the weighted average vector to obtain the fault sample vector.

[0076] When problems occur in different components, they will have different impacts on the various operating state indicators of the transformer. Therefore, by analyzing the differences between the various operating state indicators and the normal values, it is possible to accurately determine which components are in abnormal working conditions, and then remind the maintenance personnel to check and maintain the transformer in a timely manner, effectively improving the maintenance efficiency of the transformer.

[0077] The state values of the various operating state indicators can be either actual numerical values, such as current values, voltage values, temperature, etc., or judgment values of whether they are abnormal. For example, when abnormal, it is 1, and when normal, it is 0. For example, the detected operating state indicators of the transformer include the transformer oil temperature 、the transformer oil level 、the core temperature 、the winding temperature 、the core grounding current . When there is a problem with the sealing parts of the transformer, it will cause the transformer oil to leak, resulting in the transformer oil level deviating from the normal value. And since the amount of transformer oil leakage is small and the transformer oil temperature does not show any abnormality, the historical fault vector corresponding to the problem with this sealing part is:

[0078]

[0079] When there are historical fault vectors of multiple sealing parts, due to the different degrees of severity of each sealing part failure each time, resulting in different abnormal operating state indicators, it is necessary to calculate the weighted average vector to improve the accuracy of fault diagnosis.

[0080] Specifically, the abnormal state vector can be obtained through the following steps:

[0081] S31. Construct an empty vector. The dimension of the empty vector is the same as that of the fault sample vector, and the operating state indicators corresponding to each component of the empty vector are the same as those of the fault sample vector;

[0082] S32. For the actual values of the obtained operating state indicators, calculate the absolute difference between the actual value and the normal value, and compare the absolute difference with the first threshold;

[0083] S33. If the absolute difference is greater than the first threshold, mark the component corresponding to the actual value in the empty vector as abnormal. If the absolute difference is less than the first threshold, mark the component corresponding to the actual value in the empty vector as normal.

[0084] For example, each component of the abnormal state vector corresponds one by one to the components of the above-mentioned fault sample vector. When only the core grounding current deviates from the normal value, the abnormal state vector is ;

[0085] The components of the abnormal state vector can also be operating state parameter values, such as current values, voltage values, temperature, etc.

[0086] Further, in step S4, the process of matching the abnormal state vector with all the fault sample vectors one by one includes:

[0087] S41. Calculate the modulus of the abnormal state vector and the modulus of the fault sample vector, and calculate the absolute difference between the moduli of the two vectors;

[0088] S42. Calculate the included angle between the abnormal state vector and the fault sample vector;

[0089] S43. When the absolute difference between the moduli of the two vectors is less than the second threshold and the included angle between the two vectors is less than the third threshold, it is determined that the abnormal state vector matches the fault sample vector.

[0090] The basis for judging whether two vectors are equal is whether the moduli of the two vectors are the same and whether the included angle is zero. Since the fault sample vector is obtained through weighted calculation, the abnormal state vector cannot be exactly equal to the fault sample vector. A certain deviation should be allowed in the judgment. When the deviation is less than a certain value, it can be considered that the two match.

[0091] One of the major advantages of the digital twin model lies in visualization. Through the digital twin model, the state of the transformer can be conveniently displayed in the form of an animation, enabling maintenance personnel to more intuitively obtain the operating state of the transformer.

[0092] For example, in step S5, component fault prompts can be issued through the visualization model, and index anomaly prompts can also be issued through the visualization model. Further, the percentage of the ratio of the actual value to the normal value of the abnormal operating state index can also be calculated to prompt the maintenance personnel of the deviation degree of this index from the normal value.

[0093] Embodiment 2

[0094] When calculating the fault sample vector, considering that new faults may occur during the use of the transformer, it is necessary to continuously correct the fault sample vector according to new data to improve the accuracy of the judgment result. Machine learning can very accurately fit the data and continuously correct the fitted data according to the difference between the predicted value and the actual value.

[0095] Specifically, as Figure 4 shown, step S14 further includes:

[0096] S141. Obtain a new historical fault vector, and use a neural network to fit all the historical fault vectors to obtain a historical fault fitted vector;

[0097] S142. Construct a loss function based on the historical fault fitting vector and the weighted average vector;

[0098] S143. Modify the weighted average vector according to the loss function to obtain a fault sample vector.

[0099] Similarly, when calculating the normal value, machine learning can also be used to correct the predicted value to improve the accuracy of the normal value.

[0100] Specifically, as Figure 5 shown, step S23 further includes:

[0101] S231. Calculate the fitting value of the operating state index using a neural convolutional network based on the historical data of the operating state index;

[0102] S232. Construct a loss function based on the fitting value and the predicted value;

[0103] S233. Modify the predicted value according to the loss function to obtain the normal value.

[0104] Considering that the digital twin model is an idealized model of the transformer, there will be a certain deviation between the predicted value obtained from it and the true value. Therefore, through machine learning, the predicted value is corrected according to the historical data of the operating state index to make the obtained normal value more accurate.

[0105] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A transformer operation and maintenance method based on digital twin technology, characterized in that: The following steps are involved: S1. According to the geometric model of the transformer, a digital twin model of the transformer is constructed by combining mechanics, electromagnetism and thermodynamics, and fault sample vectors are constructed for multiple components respectively; S2. Periodically obtain the actual value of each operating status indicator of the transformer, and compare the actual value with the normal value corresponding to the operating status indicator; S3, establishing an abnormal state vector according to the result of the above comparison step; S4, matching the abnormal state vector with all the fault sample vectors one by one; S5. If there is a matching fault sample vector, a component fault prompt is issued; if there is no matching fault sample vector, an indicator abnormality prompt is issued; The normal value is obtained by the following method: S21, acquiring the operating load data and external environment data of the transformer in real time; S22, bringing the operating load data and the external environment data into the digital twin model to obtain predicted values ​​of various operating status indicators of the transformer; S231, calculating the fitting value of the operating status indicator using a neural convolution network according to the historical data of the operating status indicator; S232, constructing a loss function according to the fitted value and the predicted value; S233. Correct the predicted value according to the loss function to obtain the normal value.

2. The transformer operation and maintenance method based on digital twin technology according to claim 1 is characterized in that: The fault sample vector is obtained by the following method: S11. For a component, multiple pieces of historical fault data are obtained, and a corresponding historical fault vector is constructed according to each piece of historical fault data, wherein each component of the historical fault vector corresponds to a state value of an operating state indicator of the transformer when the transformer fails; S12, calculating the weight of each of the operating status indicators according to the number of abnormalities of each of the operating status indicators in all the historical fault vectors; S13, calculating a weighted average vector of a plurality of the historical fault vectors; S14. Correct the weighted average vector to obtain the fault sample vector.

3. The transformer operation and maintenance method based on digital twin technology according to claim 2 is characterized in that: Step S14 also includes: S141, obtaining new historical fault vectors, and fitting all the historical fault vectors using a neural network to obtain historical fault fitting vectors; S142, constructing a loss function according to the historical fault fitting vector and the weighted average vector; S143. Correct the weighted average vector according to the loss function to obtain the fault sample vector.

4. The transformer operation and maintenance method based on digital twin technology according to claim 1 is characterized in that: The abnormal state vector is obtained by the following steps: S31, constructing an empty vector, wherein the dimension of the empty vector is the same as the dimension of the fault sample vector, and the operating status indicator corresponding to each component of the empty vector is the same as that of the fault sample vector; S32, for the actual value of each of the operating status indicators obtained, calculating the absolute difference between the actual value and the normal value, and comparing the absolute difference with a first threshold; S33. If the absolute difference is greater than the first threshold, the component in the empty vector corresponding to the actual value is recorded as abnormal; if the absolute difference is less than the first threshold, the component in the empty vector corresponding to the actual value is recorded as normal.

5. The transformer operation and maintenance method based on digital twin technology according to claim 1 is characterized in that: In step S4, the process of matching the abnormal state vector with all the fault sample vectors one by one includes: S41, calculating the modulus of the abnormal state vector and the modulus of the fault sample vector, and calculating the absolute difference between the moduli of the two vectors; S42, calculating the angle between the abnormal state vector and the fault sample vector; S43: When the absolute difference between the modules of the two vectors is smaller than the second threshold, and the angle between the two vectors is smaller than the third threshold, it is determined that the abnormal state vector matches the fault sample vector.

6. The transformer operation and maintenance method based on digital twin technology according to claim 1 is characterized in that: The step of issuing an indicator abnormality prompt in step S5 includes: S51. Constructing a visualization model of the transformer according to the digital twin model and the geometric model of the transformer; S52: issuing a component failure prompt through the visualization model.

7. The transformer operation and maintenance method based on digital twin technology according to claim 6 is characterized in that: The step of issuing an indicator abnormality prompt in step S5 includes: S53: issuing an indicator abnormality prompt through the visualization model.

8. The transformer operation and maintenance method based on digital twin technology according to claim 7 is characterized in that: The step of issuing an indicator abnormality prompt in step S5 includes: S54, calculating the percentage of the ratio of the actual value to the normal value; S55. Issue an indicator abnormality prompt through the visualization model and display the percentage.

Images