A distributed winding deformation online monitoring system and method

Through the synchronous compensation technology of distributed winding deformation online monitoring system and BP neural network algorithm, the problem of centralized system monitoring multiple transformers and long-distance wiring is solved, high-precision winding deformation monitoring is achieved, and the system reliability and Data analysis accuracy.

CN118671666BActive Publication Date: 2024-11-05CHENGDU GONGBAILI AUTOMATION EQUIP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411165994.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-05
Estimated Expiration
2044-08-23

AI Technical Summary

Technical Problem

The existing centralized transformer winding monitoring system cannot monitor multiple transformers at the same time, which is troublesome to long-distance wiring. A single acquisition channel failure will affect the operation of the entire system, and the delay in synchronous signal transmission leads to data synchronization errors.

Method used

The distributed winding deformation online monitoring system is adopted, and the BP neural network algorithm is used for synchronous compensation to realize high-precision synchronous sampling of each distributed acquisition unit. The three-phase winding data is collected through transformers, and the host analyzes and extracts characteristic parameters, combining horizontal and vertical directions. Comparative analysis and monitoring winding deformation.

Benefits of technology

It improves the accuracy of data analysis, enhances the reliability of system operation, facilitates the expansion of interfaces, reduces the failure rate, and improves the convenience of the system and the convenience of use.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118671666B_ABST
    Figure CN118671666B_ABST
Patent Text Reader

Abstract

The invention discloses a distributed winding deformation online monitoring system and method, and relates to the technical field of transformer winding monitoring. A host performs synchronous compensation between distributed acquisition units based on a BP neural network model, and modulates synchronous sampling signals; the distributed acquisition units acquire voltage and current data of a three-phase winding of a transformer through a mutual inductor; the host receives and stores sampling data from each distributed acquisition unit; the host inputs the sampling data into a corresponding transformer model, and extracts characteristic parameters that can represent the transformer winding deformation through analysis and calculation of the model; a horizontal and vertical comparative analysis is performed on the characteristic parameters, and a comprehensive judgment is made on the transformer winding deformation in combination with other monitoring data; the invention can improve the synchronization of distributed unit sampling and enhance the accuracy of data analysis; the invention can improve the reliability of system operation and reduce the failure rate; the invention is convenient for expanding interfaces and enhances the convenience of use.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of transformer winding monitoring, and in particular to a distributed winding deformation online monitoring system and method. Background Art

[0002] In the power system, the transformer plays an important role in power conversion, and its operating status affects the safety and stability of the entire power system. According to statistics, the winding failure rate accounts for a high proportion of transformer failures and is on the rise. Winding deformation includes axial and radial dimensional changes, winding displacement, distortion, bulging, etc. Winding deformation is usually caused by short-circuit shock or long-term operation aging and overload. In severe cases, it will cause transformer failure, affect the normal operation of the system, and may even endanger personal safety. Therefore, it is of great significance to conduct online monitoring of transformer winding deformation.

[0003] Disadvantages of the existing centralized system:

[0004] 1. Multiple transformers cannot be monitored at the same time;

[0005] 2. The centralized structure makes long-distance wiring more troublesome for those collection points that are far away;

[0006] 3. Failure of a single acquisition channel may affect the normal operation of the entire acquisition function.

[0007] Advantages of distributed:

[0008] 1. Distributed acquisition units can easily expand interfaces and support access to data from multiple transformers;

[0009] 2. Distributed acquisition units can be easily installed nearby. You only need to consider installing the communication transmission line from the distributed acquisition unit to the host.

[0010] 3. Distributed acquisition units can improve the reliability of device operation. Each distributed acquisition unit does not interfere with each other and only communicates with the host through transmission lines. At the same time, the distributed acquisition units can be maintained independently, and failure will not affect other units.

[0011] The core of distributed data collection lies in synchronous data collection. In the actual synchronous data collection process, due to the distance between distributed data collection units, there will be a delay problem when the synchronous signal is transmitted through the transmission line, so a certain synchronization error compensation is required.

[0012] Therefore, a distributed winding deformation online monitoring system and method has become an urgent problem to be solved. Summary of the invention

[0013] The purpose of the present invention is to propose a distributed winding deformation online monitoring system and method, which performs synchronous compensation based on the BP neural network algorithm to achieve high-precision synchronous sampling of each distributed acquisition unit, which is beneficial to improving the accuracy of data analysis; at the same time, the distributed structure is easy to use and expand, thereby improving the reliability of system operation.

[0014] To achieve the above object, the present invention provides a technical solution: a distributed winding deformation online monitoring method, comprising the following steps:

[0015] S1. The host performs synchronous compensation for the transmission delay between the distributed acquisition units based on the BP neural network model and modulates the synchronous sampling signal;

[0016] S2, the distributed acquisition unit collects the voltage and current data of the three-phase winding of the transformer through the mutual inductor;

[0017] S3, the host receives and stores the sampled data from each distributed acquisition unit;

[0018] S4, the host inputs the sampled data into the corresponding transformer model, and extracts characteristic parameters that can represent the deformation of the transformer winding through analysis and calculation of the model;

[0019] S5. Conduct horizontal and vertical comparative analysis on the characteristic parameters, and make a comprehensive judgment on the transformer winding deformation in combination with other monitoring data;

[0020] S6. Generate alarm prompts based on the analysis results of the host and provide reference opinions for operation and maintenance.

[0021] Furthermore, the method for synchronously compensating the transmission delay in S1 is as follows:

[0022] a) In the initialization phase, the host sends a pre-synchronization command to each distributed acquisition unit at t0;

[0023] b) After receiving the pre-synchronization command from the host, each distributed acquisition unit generates a special time series pulse and sends it to the host;

[0024] c) The time taken from the host sending the pre-synchronization command to each distributed acquisition unit receiving the pre-synchronization command t MX The time t taken from each distributed acquisition unit sending a pulse signal to the host receiving the pulse signal XM are the same, and the time for each distributed acquisition unit FPGA to process the signal is the same, where X represents a distributed acquisition unit;

[0025] That is: MA =t AM ,t MB =t BM ,tMC =t CM ;

[0026] t MA The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit A to receive the pre-synchronization command;

[0027] t AM It is the time taken from the distributed acquisition unit A sending out a pulse signal to the host receiving the pulse signal;

[0028] t MB The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit B to receive the pre-synchronization command;

[0029] t BM The time taken for the distributed acquisition unit B to send out a pulse signal and for the host to receive the pulse signal;

[0030] t MC The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit C to receive the pre-synchronization command;

[0031] t CM The time taken for the distributed acquisition unit C to send out a pulse signal and for the host to receive the pulse signal;

[0032] d) The time when the host receives the special pulse signal sent by each distributed acquisition unit is t X , get the transmission delay between the host and each distributed acquisition unit

[0033] e) Repeat steps a)-d) n times to obtain n groups of t MX Sample test data;

[0034] f) Group n into t MX The sample test data is input into the trained BP neural network model, and the output is the predicted next transmission delay t MX ;

[0035] g) The host sends clock synchronization commands to each distributed acquisition unit at the same time;

[0036] h) After receiving the clock synchronization command, each distributed acquisition unit performs the transmission delay t obtained in step f) MX The sampling clock is synchronized with the maximum value of

[0037] Furthermore, the method for establishing the BP neural network model in S1 is as follows:

[0038] The BP neural network model includes input layer, hidden layer and output layer; among them,

[0039] Input layer: the obtained n groups tMX Sample data, i.e. the transmission delay data t obtained from the first n tests MX ; In order to improve the training speed of BP neural network, it is necessary to normalize the sample data, that is, select the first sample data as the reference value, and normalize all inputs to the range of [-1, 1] as the input of the input layer;

[0040] Hidden layer: According to the empirical method, m=2n+1, where m is the number of nodes in the hidden layer and n is the number of nodes in the input; therefore, the number of nodes in the hidden layer is set to m;

[0041] Output layer: predicted next transmission delay;

[0042] Training samples: Pre-test to obtain s groups of sample data, use the mean absolute error (MAD) as the model evaluation standard, and continuously optimize the weights and thresholds of the model training function and the number of nodes in the hidden layer to achieve the best training effect;

[0043] Let y i (i=1,2…s) is the predicted transmission delay value, x i (i=1,2…s) is the actual test value corresponding to each time, then the prediction accuracy sequence e i =y i -x i (i=1,2…s)

[0044]

[0045] The BP neural network model is used for calculation, and the latest test t is used in each calculation. MX The sample data replaces the oldest t MX Sample data forms a rolling time series and continuously updates the test samples to obtain the latest predicted transmission delay.

[0046] Furthermore, the method of modulating the synchronous sampling signal in S1 is as follows:

[0047] 1) The host sends a synchronous acquisition command to each distributed acquisition unit at the same time at time t0;

[0048] 2) After receiving the synchronous acquisition command, each distributed acquisition unit calculates the predicted transmission delay t according to the output of the BP neural network model. MX The maximum value of the delay calculation is carried out, and the delay is Δt X After that, a sampling control command is sent to the ADC to start sampling;

[0049] 3) Delay calculation: Assume that the predicted transmission delay of the distributed acquisition unit C is t MC maximum,

[0050] Then the delay corresponding to the distributed acquisition unit A is Δt A =t MC -t MA ;

[0051] The delay corresponding to the distributed acquisition unit B is Δt B =t MC -t MB ;

[0052] The delay corresponding to the distributed acquisition unit C is Δt C =0;

[0053] 4) After each distributed acquisition unit completes a sampling process according to the sampling time and sampling accuracy of the host, it caches the sampled data in the FPGA and sends a sampling completion signal to the host;

[0054] 5) After receiving the sampling completion signal, the host reads the cache data of each distributed acquisition unit and reads the data according to the sampling start time t0+t MX Timestamps are added for later analysis.

[0055] The present invention also provides a distributed winding deformation online monitoring system, comprising a distributed acquisition unit and a host, wherein the distributed acquisition unit is arranged at different positions of the transformer winding and communicates with the host through optical fiber;

[0056] The distributed acquisition unit acquires voltage and current data of the three-phase winding of the transformer through a mutual inductor;

[0057] The host performs synchronous compensation between the distributed acquisition units based on the BP neural network and modulates the synchronous acquisition signal; receives data from the distributed acquisition units, analyzes the winding deformation of the current transformer through the model, and provides reference data for operation and maintenance.

[0058] The advantages of the present invention compared with the prior art are:

[0059] 1. The present invention can improve the synchronization of distributed unit sampling and enhance the accuracy of data analysis;

[0060] 2. The present invention utilizes distributed acquisition units to improve the reliability of system operation and reduce the failure rate;

[0061] 3. The present invention facilitates the expansion of interfaces and enhances the convenience of use. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a system block diagram of a distributed winding deformation online monitoring system.

[0063] Figure 2 It is a flow chart of the transmission delay synchronization compensation method.

[0064] Figure 3 is a flow chart of the synchronous sampling method. DETAILED DESCRIPTION

[0065] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by technicians in the technical field of the present disclosure; the terms used in the specification of the application in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure; the terms "including" and "having" and any variations thereof in the specification and claims of the present disclosure and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present disclosure or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0066] Reference to "embodiments" in the present invention means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present disclosure. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in the present invention may be combined with other embodiments.

[0067] In order to enable those skilled in the art to better understand the solution of the present disclosure, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0068] Combined with Figure 1-3 The specific implementation process of a distributed winding deformation online monitoring system and method of the present invention is as follows:

[0069] A distributed winding deformation online monitoring method comprises the following steps:

[0070] S1. The host performs synchronous compensation for the transmission delay between the distributed acquisition units based on the BP neural network model and modulates the synchronous sampling signal;

[0071] S2, the distributed data collection unit collects the voltage and current data of the three-phase winding of the transformer through the mutual inductor; it also has an interface for collecting the grounding current of the core and the clamp, the busbar current, etc.;

[0072] S3, the host receives and stores the sampled data from each distributed acquisition unit;

[0073] S4, the host inputs the sampled data into the corresponding transformer model, and extracts characteristic parameters that can represent the deformation of the transformer winding through analysis and calculation of the model;

[0074] S5. Conduct horizontal and vertical comparative analysis on the characteristic parameters, and make a comprehensive judgment on the transformer winding deformation in combination with other monitoring data;

[0075] S6. Generate alarm prompts based on the analysis results of the host and provide reference opinions for operation and maintenance.

[0076] Preferably, the method for synchronously compensating the transmission delay in S1 is as follows:

[0077] a) In the initialization phase, the host sends a pre-synchronization command to each distributed acquisition unit at t0;

[0078] b) After receiving the pre-synchronization command from the host, each distributed acquisition unit generates a special time series pulse and sends it to the host;

[0079] c) The time taken from the host sending the pre-synchronization command to each distributed acquisition unit receiving the pre-synchronization command t MX The time t taken from each distributed acquisition unit sending a pulse signal to the host receiving the pulse signal XM are the same, and the time for each distributed acquisition unit FPGA to process the signal is the same, where X represents a distributed acquisition unit;

[0080] That is: MA =t AM ,t MB =t BM ,t MC =t CM ;

[0081] t MA The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit A to receive the pre-synchronization command;

[0082] t AM It is the time taken from the distributed acquisition unit A sending out a pulse signal to the host receiving the pulse signal;

[0083] t MB The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit B to receive the pre-synchronization command;

[0084] t BM The time taken for the distributed acquisition unit B to send out a pulse signal and for the host to receive the pulse signal;

[0085] t MC The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit C to receive the pre-synchronization command;

[0086] t CM The time taken for the distributed acquisition unit C to send out a pulse signal and for the host to receive the pulse signal;

[0087] d) The time when the host receives the special pulse signal sent by each distributed acquisition unit is t X , get the transmission delay between the host and each distributed acquisition unit

[0088] e) Repeat steps a)-d) 10 times to obtain 10 groups of t MX Sample test data;

[0089] f) Put 10 groups of t MX The sample test data is input into the trained BP neural network model, and the output is the predicted next transmission delay t MX ;

[0090] g) The host sends clock synchronization commands to each distributed acquisition unit at the same time;

[0091] h) After receiving the clock synchronization command, each distributed acquisition unit performs the transmission delay t obtained in step f) MX The sampling clock is synchronized with the maximum value of

[0092] Preferably, the method for establishing the BP neural network model in S1 is as follows:

[0093] The BP neural network model includes input layer, hidden layer and output layer; among them,

[0094] Input layer: The 10 groups of t MX Sample data, i.e. the transmission delay data obtained from the first 10 tests t MX ; In order to improve the training speed of BP neural network, it is necessary to normalize the sample data, that is, select the first sample data as the reference value, and normalize all inputs to the range of [-1, 1] as the input of the input layer;

[0095] Hidden layer: According to the empirical method, m=2n+1, where m is the number of nodes in the hidden layer and n is the number of nodes in the input; therefore, the number of nodes in the hidden layer is set to 21;

[0096] Output layer: predicted next transmission delay;

[0097] Training samples: 100 sets of sample data were obtained through pre-testing. The mean absolute error (MAD) was used as the model evaluation standard. The weights and thresholds of the model training function and the number of nodes in the hidden layer were continuously optimized to achieve the best training effect.

[0098] Let y i (i=1,2…100) is the predicted transmission delay value, x i (i=1,2…100) is the actual test value corresponding to each time, then the prediction accuracy sequence e i =yi -x i (i=1,2…s(s=100))

[0099]

[0100] The BP neural network model is used for calculation, and the latest test t is used in each calculation. MX The sample data replaces the oldest t MX Sample data forms a rolling time series and continuously updates the test samples to obtain the latest predicted transmission delay.

[0101] Preferably, the method for modulating the synchronous sampling signal in S1 is as follows:

[0102] 1) The host sends a synchronous acquisition command to each distributed acquisition unit at the same time at time t0;

[0103] 2) After receiving the synchronous acquisition command, each distributed acquisition unit calculates the predicted transmission delay t according to the output of the BP neural network model. MX The maximum value of the delay calculation is carried out, and the delay is Δt X After that, a sampling control command is sent to the ADC to start sampling;

[0104] 3) Delay calculation: Assume that the predicted transmission delay of the distributed acquisition unit C is t MC maximum,

[0105] Then the delay corresponding to the distributed acquisition unit A is Δt A =t MC -t MA ;

[0106] The delay corresponding to the distributed acquisition unit B is Δt B =t MC -t MB ;

[0107] The delay corresponding to the distributed acquisition unit C is Δt C =0;

[0108] 4) After each distributed acquisition unit completes a sampling process according to the sampling time and sampling accuracy of the host, it caches the sampled data in the FPGA and sends a sampling completion signal to the host;

[0109] 5) After receiving the sampling completion signal, the host reads the cache data of each distributed acquisition unit and reads the data according to the sampling start time t0+t MX Timestamps are added for later analysis.

[0110] Based on the same concept as the above-mentioned distributed winding deformation online monitoring method, this embodiment also provides a distributed winding deformation online monitoring system, which can be used to execute the above-mentioned distributed winding deformation online monitoring method.

[0111] A distributed winding deformation online monitoring system includes a distributed acquisition unit and a host, wherein the distributed acquisition unit is arranged at different positions of the transformer winding and communicates with the host through optical fiber;

[0112] The distributed acquisition unit collects the voltage and current data of the three-phase winding of the transformer through the mutual inductor; the distributed acquisition unit has an interface for collecting the grounding current of the core and the clamp, and the busbar current; the distributed acquisition unit realizes the synchronous acquisition of different distributed acquisition units based on the synchronous acquisition signal of the host.

[0113] The host performs synchronous compensation between the distributed acquisition units based on the BP neural network and modulates the synchronous acquisition signal; receives data from each distributed acquisition unit, analyzes the winding deformation of the current transformer through the model, and provides reference data for operation and maintenance.

[0114] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A distributed winding deformation online monitoring method, characterized in that: The following steps are included: S1. The host performs synchronous compensation for the transmission delay between the distributed acquisition units based on the BP neural network model and modulates the synchronous sampling signal; Among them, the method for synchronously compensating the transmission delay is as follows: a) In the initialization phase, the host sends a pre-synchronization command to each distributed acquisition unit at t0; b) After receiving the pre-synchronization command from the host, each distributed acquisition unit generates a special time series pulse and sends it to the host; c) The time taken from the host sending the pre-synchronization command to each distributed acquisition unit receiving the pre-synchronization command t MX The time t taken from each distributed acquisition unit sending a pulse signal to the host receiving the pulse signal XM are the same, and the time for each distributed acquisition unit FPGA to process the signal is the same, where X represents a distributed acquisition unit; That is: tMA=tAM, tMB=tBM, tMc=tcM t MA The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit A to receive the pre-synchronization command; t AM It is the time taken from the distributed acquisition unit A sending out a pulse signal to the host receiving the pulse signal; t MB The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit B to receive the pre-synchronization command; t BM The time taken for the distributed acquisition unit B to send out a pulse signal and for the host to receive the pulse signal; t MC The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit C to receive the pre-synchronization command; t CM The time taken for the distributed acquisition unit C to send out a pulse signal and for the host to receive the pulse signal; d) The time when the host receives the special pulse signal sent by each distributed acquisition unit is t X , get the transmission delay between the host and each distributed acquisition unit e) Repeat steps a)-d) n times to obtain n groups of t MX Sample test data; f) Group n into t MX The sample test data is input into the trained BP neural network model, and the output is the predicted next transmission delay t MX ; g) The host sends clock synchronization commands to each distributed acquisition unit at the same time; h) After receiving the clock synchronization command, each distributed acquisition unit performs the transmission delay t obtained in step f) MX The maximum value of the sampling clock is synchronized; The method of modulating the synchronous sampling signal is as follows: 1) The host sends a synchronous acquisition command to each distributed acquisition unit at the same time at time t0; 2) After receiving the synchronous acquisition command, each distributed acquisition unit calculates the predicted transmission delay t according to the output of the BP neural network model. MX The maximum value of the delay calculation is carried out, and the delay is Δt X After that, a sampling control command is sent to the ADC to start sampling; 3) Delay calculation: Assume that the predicted transmission delay of the distributed acquisition unit C is t MC maximum, Then the delay corresponding to the distributed acquisition unit A is Δt A =t MC -t MA ; The delay corresponding to the distributed acquisition unit B is Δt B =t MC -t MB ; The delay corresponding to the distributed acquisition unit C is Δt C =0; 4) After each distributed acquisition unit completes a sampling process according to the sampling time and sampling accuracy of the host, it caches the sampled data in the FPGA and sends a sampling completion signal to the host; 5) After receiving the sampling completion signal, the host reads the cache data of each distributed acquisition unit and reads the data according to the sampling start time t0+t MX Timestamp for later analysis; S2, the distributed acquisition unit collects the voltage and current data of the three-phase winding of the transformer through the mutual inductor; S3, the host receives and stores the sampled data from each distributed acquisition unit; S4, the host inputs the sampled data into the corresponding transformer model, and extracts characteristic parameters that can represent the deformation of the transformer winding through analysis and calculation of the model; S5. Conduct horizontal and vertical comparative analysis on the characteristic parameters, and make a comprehensive judgment on the transformer winding deformation in combination with other monitoring data; S6. Generate alarm prompts based on the analysis results of the host and provide reference opinions for operation and maintenance.

2. A distributed winding deformation online monitoring method according to claim 1, characterized in that: The mutual inductor also collects the core and clamp grounding current and busbar current.

3. A distributed winding deformation online monitoring method according to claim 2, characterized in that: The method of establishing the BP neural network model in S1 is as follows: The BP neural network model includes input layer, hidden layer and output layer; in, Input layer: the obtained n groups t MX Sample data, i.e. the transmission delay data t obtained from the first n tests MX ; In order to improve the training speed of BP neural network, it is necessary to normalize the sample data, that is, select the first sample data as the reference value, and normalize all inputs to the range of [-1, 1] as the input of the input layer; Hidden layer: According to the empirical method, m=2n+1, where m is the number of nodes in the hidden layer and n is the number of nodes in the input; therefore, the number of nodes in the hidden layer is set to m; Output layer: predicted next transmission delay; Training samples: Pre-test to obtain s groups of sample data, use the mean absolute error (MAD) as the model evaluation standard, and continuously optimize the weights and thresholds of the model training function and the number of nodes in the hidden layer to achieve the best training effect; Let y i (i=1,2…s) is the predicted transmission delay value, x i (i=1,2…s) is the actual test value corresponding to each time, then the prediction accuracy sequence e i =y i -x i (i=1,2…s); The BP neural network model is used for calculation, and the latest test t is used in each calculation. MX The sample data replaces the oldest t MX Sample data forms a rolling time series and continuously updates the test samples to obtain the latest predicted transmission delay.

4. A distributed winding deformation online monitoring system, characterized in that: It includes a distributed acquisition unit and a host, wherein the distributed acquisition unit is arranged at different positions of the transformer winding and communicates with the host through optical fiber; The distributed acquisition unit acquires voltage and current data of the three-phase winding of the transformer through a mutual inductor; The host performs synchronous compensation of transmission delay between distributed acquisition units based on BP neural network and modulates synchronous acquisition signals; receives data from each distributed acquisition unit, analyzes the winding deformation of the current transformer through the model, and provides reference data for operation and maintenance; Among them, the method for synchronously compensating the transmission delay is as follows: a) In the initialization phase, the host sends a pre-synchronization command to each distributed acquisition unit at t0; b) After receiving the pre-synchronization command from the host, each distributed acquisition unit generates a special time series pulse and sends it to the host; c) The time taken from the host sending the pre-synchronization command to each distributed acquisition unit receiving the pre-synchronization command t MX The time t taken from each distributed acquisition unit sending a pulse signal to the host receiving the pulse signal XM are the same, and the time for each distributed acquisition unit FPGA to process the signal is the same, where X represents a distributed acquisition unit; That is: t MA = t AM , t MB = t BM , t MC = t CM ; t MA The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit A to receive the pre-synchronization command; t AM It is the time taken from the distributed acquisition unit A sending out a pulse signal to the host receiving the pulse signal; t MB The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit B to receive the pre-synchronization command; t BM The time taken for the distributed acquisition unit B to send out a pulse signal and for the host to receive the pulse signal; t MC The time taken for the host to send out a pre-synchronization command and for the distributed acquisition unit C to receive the pre-synchronization command; t CM The time taken for the distributed acquisition unit C to send out a pulse signal and for the host to receive the pulse signal; d) The time when the host receives the special pulse signal sent by each distributed acquisition unit is t X , get the transmission delay between the host and each distributed acquisition unit e) Repeat steps a)-d) n times to obtain n groups of t MX Sample test data; f) Group n into t MX The sample test data is input into the trained BP neural network model, and the output is the predicted next transmission delay t MX ; g) The host sends clock synchronization commands to each distributed acquisition unit at the same time; h) After receiving the clock synchronization command, each distributed acquisition unit performs the transmission delay t obtained in step f) MX The maximum value of the sampling clock is synchronized; Modulation The method of modulating the synchronous sampling signal is as follows: 1) The host sends a synchronous acquisition command to each distributed acquisition unit at the same time at time t0; 2) After receiving the synchronous acquisition command, each distributed acquisition unit calculates the predicted transmission delay t according to the output of the BP neural network model. MX The maximum value of the delay calculation is carried out, and the delay is Δt X After that, a sampling control command is sent to the ADC to start sampling; 3) Delay calculation: Assume that the predicted transmission delay of the distributed acquisition unit C is t MC maximum, Then the delay corresponding to the distributed acquisition unit A is Δt A =t MC -t MA ; The delay corresponding to the distributed acquisition unit B is Δt B =t MC -t MB ; The delay corresponding to the distributed acquisition unit C is Δt C =0; 4) After each distributed acquisition unit completes a sampling process according to the sampling time and sampling accuracy of the host, it caches the sampled data in the FPGA and sends a sampling completion signal to the host; 5) After receiving the sampling completion signal, the host reads the cache data of each distributed acquisition unit and reads the data according to the sampling start time t0+t MX Timestamps are added for later analysis.

5. A distributed winding deformation online monitoring system according to claim 4, characterized in that: The distributed acquisition unit realizes synchronous acquisition of different distributed acquisition units based on the synchronous acquisition signal of the host.

6. A distributed winding deformation online monitoring system according to claim 5, characterized in that: The distributed data acquisition unit is provided with an interface for collecting the grounding current of the core and the clamp, and the busbar current.