Low-voltage transformer online health monitoring method and system based on end side AI

Abstract

The invention discloses a low-voltage transformer online health monitoring method and system based on end side AI, and the method comprises the steps: carrying out the side entry of a processing unit at a secondary side embedded section of a low-voltage transformer, carrying out the multi-cycle electric signal collection, end side AI reasoning, sliding window anti-shake post-processing, fault output, and the like. The traditional low-voltage mutual inductor is enabled to have independent health state sensing and evaluating capability; according to the invention, real-time, continuous and online autonomous monitoring of the operation state of the mutual inductor is realized, continuous intervention of an external upper computer is not needed, the speed and accuracy of fault discovery are significantly improved, and the labor cost of operation and maintenance is reduced.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, and more specifically, to a method and system for online health monitoring of low-voltage instrument transformers based on end-side AI. Background Technology

[0002] Low-voltage instrument transformers are core sensing devices in power distribution systems for metering, load monitoring, and fault protection, and are widely used in critical scenarios such as metering boxes, intelligent measuring switches, and low-voltage distribution circuits. Traditional low-voltage instrument transformers can only perform basic electrical signal acquisition and transmission, acting as the "eyes" of the power distribution system but lacking the ability to "self-sensitize" and self-diagnose their own operating status. During long-term operation, instrument transformers are prone to various faults such as inter-turn short circuits and open circuits, core saturation and aging, loose wiring and poor contact, zero-point drift and abnormal linearity, harmonic distortion, and metering inaccuracies. These faults directly lead to distorted power distribution metering data, malfunctioning protection devices, and invalid monitoring data, seriously affecting the stable and reliable operation of low-voltage power distribution systems.

[0003] Currently, the industry mainly relies on manual periodic inspections and offline verification to monitor the status of current transformers. This not only consumes a lot of manpower and resources, but also fails to achieve online, real-time, and end-side intelligent diagnosis of the operating status of current transformers. Faults are often only discovered after they occur, resulting in a significant monitoring lag. Summary of the Invention

[0004] To address at least one of the aforementioned technical problems, the present invention aims to provide a method and system for online health monitoring of low-voltage instrument transformers based on end-side AI, which can significantly improve the speed and accuracy of fault detection and reduce maintenance manpower costs.

[0005] The first aspect of this invention provides a method for online health monitoring of low-voltage transformers based on end-side AI, comprising: Based on a preset time period, the electrical signal output from the secondary side of the current transformer for N power frequency cycles is periodically collected, and M sampling points are collected in each cycle to form the original sampling matrix. The original sampling matrix is ​​sent to a preset terminal for processing to determine the health index and fault confidence. If the health index is less than the preset first health index threshold or the fault confidence is greater than the preset first fault confidence threshold, it is determined that there is a health anomaly at the current moment, and the collected data of the current period is marked as a preliminary anomaly. Based on a preset sliding time window, extract the number of preliminary anomaly markers within the corresponding sliding time window; If the number of initial anomaly markers within the sliding time window exceeds the preset anti-shake threshold, a final anomaly marker is generated, and a health warning message is triggered.

[0006] In this scheme, the step of sending the original sampling matrix to a preset terminal for processing to determine the health index and fault confidence specifically includes: Each period of the original sampling matrix is ​​normalized and aligned with the waveform to obtain a normalized waveform matrix. The feature parameters of each period are then extracted, including the effective value and harmonic amplitude value. Based on the harmonic amplitude, determine the maximum value of the waveform distortion rate; extract the odd harmonic amplitude from the harmonic amplitude, and determine the maximum value of the proportion of odd harmonic amplitude based on the odd harmonic amplitude; The waveform quality index is determined based on the maximum percentage of odd harmonic amplitude and the maximum waveform distortion rate. Its formula is: ,in , These represent the maximum percentage of odd harmonic amplitude and the maximum waveform distortion rate, respectively. , These represent the health baseline values ​​for the proportion of odd harmonic amplitude and waveform distortion rate, respectively. This represents the corresponding weighting coefficient; Extract the harmonics of the current cycle and the harmonics of the adjacent previous cycle; By comparing and analyzing the harmonics of the current cycle with the harmonics of the adjacent previous cycle, the waveform similarity value of the current cycle is determined. After traversing the harmonics of all cycles, the waveform similarity value set is determined. The periodic stability index of the current harmonic is determined based on the effective value and the set of waveform similarity values. Based on a preset algorithm, the health index and fault confidence level are determined according to the periodic stability index and waveform quality index of the current harmonics.

[0007] In this solution, the steps of determining the maximum value of waveform distortion rate / determining the maximum value of odd harmonic amplitude proportion specifically include: The harmonic amplitude is divided according to the period to determine the harmonic amplitude of different periods; If the waveform distortion rate is set as THD, then the waveform distortion rate corresponding to the i-th period is THD(i), and its formula is: ,in The fundamental amplitude, Let h be the amplitude of the h-th harmonic within the i-th period. This represents the harmonic fluctuation weighting coefficient for the i-th period; Extract the odd harmonic amplitude from the harmonic amplitude values ​​of the same period; Set the amplitude ratio of odd harmonics to Then the proportion of the odd harmonic amplitude corresponding to the i-th period is Its formula is ; After traversing all cycles, we obtain the waveform distortion rate set and the odd harmonic amplitude ratio set; Extract the maximum values ​​from the waveform distortion rate set and the odd harmonic amplitude percentage set, which correspond to the maximum value of the waveform distortion rate and the maximum value of the odd harmonic amplitude percentage.

[0008] In this scheme, the steps for obtaining the harmonic fluctuation weighting coefficient specifically include: Extract the amplitude sequence of characteristic harmonic orders in each period of the original sampling matrix; The amplitudes of each characteristic harmonic order in the current cycle are compared and analyzed with those in the previous cycle to determine the cross-correlation vector; If the cross-correlation vector is greater than the preset first similarity threshold, it is determined that the harmonic fluctuation of the current period originates from the background harmonic change on the system side, and the preset first value is set as the corresponding harmonic fluctuation weight. If the cross-correlation vector is less than or equal to the preset first similarity threshold, it is determined that the harmonic fluctuation of the current cycle originates from the change of the characteristics of the transformer itself, and the preset second value is set as the corresponding harmonic fluctuation weight. The preset first value is less than the preset second value.

[0009] In this scheme, after obtaining the waveform distortion rate set and the odd harmonic amplitude ratio set, it further includes: Within each power frequency cycle, the DC component of the original sampled signal is extracted to obtain the DC component value; Extract the even harmonic amplitude from the harmonic amplitude values ​​of the same period; Based on the even harmonic amplitude in the same period, determine the proportion of even harmonic amplitude in the corresponding period. If the absolute value of the DC component is greater than the preset DC component threshold, and the amplitude ratio of the even harmonic is greater than the preset even harmonic ratio threshold and the amplitude of the second harmonic is greater than the amplitude of the third harmonic, then the current period is determined to be a DC bias period. The waveform distortion rate and the proportion of odd harmonic amplitude in the DC bias period are deleted.

[0010] This plan also includes: Within each power frequency cycle, extract the phase relationship before and after the current zero-crossing point; Based on a preset voltage phase reference signal, the direction of the current in this cycle is determined to be either positive or negative. Based on the current direction, a positive direction feature buffer and a negative direction feature buffer are constructed respectively, and the positive direction feature buffer and the negative direction feature buffer independently store the corresponding feature parameters; The periodic stability index and waveform quality index are calculated based on the characteristic data of the buffer corresponding to the direction of the current period.

[0011] This plan also includes: Within a preset stabilization period, if the health index is greater than a preset second health index threshold and the fault confidence is less than a preset second fault confidence threshold for n consecutive periods, then a baseline value buffer update is triggered, and the update formula is as follows: ,in This represents the updated baseline value. This indicates the baseline values ​​before the update, which include a healthy baseline value for the proportion of odd harmonic amplitude and a healthy baseline value for waveform distortion rate. For the corresponding adjustment coefficient, These are the feature parameters of the j-th cycle out of the current n cycles; The preset second health index threshold is greater than the preset first health index threshold, and the preset second fault confidence threshold is less than the preset first fault confidence threshold.

[0012] This plan also includes: The mean and variance of the waveform quality index are monitored in real time within a preset first number of window periods, which are set as the first mean and the first variance, respectively. If the first mean exceeds the preset first threshold and the first variance is less than the preset stability threshold, then the current environment is determined to be in a continuous high harmonic state, and information on the length of the extended sliding time window is generated. The length of the sliding time window is adjusted based on a preset delay value to obtain the adjusted sliding time window. And adjust the image stabilization threshold based on the corrected sliding time window. Its formula is: ,in This indicates the corrected sliding time window length, L represents the preset sliding time window length, and K represents the preset anti-shake threshold.

[0013] This plan also includes: After using the corrected sliding time window, extract the average value of the waveform quality index within a preset second number of window periods and set it as the second average value. If the second mean is less than the preset second threshold, then the length of the next sliding time window will be restored to the preset sliding time window length. The preset second quantity is greater than the preset first quantity, and the preset second threshold is less than the preset first threshold.

[0014] The second aspect of the present invention provides an online health monitoring system for low-voltage transformers based on end-side AI, which further includes: a low-voltage current / voltage transformer, a signal conditioning circuit, an ADC acquisition unit, an MCU main control unit, a storage unit, and a communication unit; The MCU main control unit is connected to the ADC acquisition unit, the storage unit, and the communication unit respectively. The output terminal of the signal conditioning circuit is connected to the ADC acquisition unit, and the output terminal of the low-voltage transformer is connected to the signal conditioning circuit. It is used to execute the low-voltage transformer online health monitoring method based on end-side AI as described in any one of the above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: 1. By calculating the periodic stability index and waveform quality index, the health index HI and fault confidence FC are obtained. The entire process does not rely on any pre-trained machine learning model. Compared with traditional AI methods based on supervised learning, this invention completely eliminates the dependence on a large amount of labeled fault data. It can complete health assessment with only normal operating condition data, solving the pain point of extremely scarce and difficult-to-obtain fault samples in actual engineering. It has good engineering practicality and generalization ability. 2. This application determines whether the current period is a DC bias period by using the even harmonic amplitude, and then filters the determined waveform distortion rate and odd harmonic amplitude to improve data accuracy; 3. This application raises the threshold for abnormal triggering by automatically expanding the sliding window length and synchronously adjusting the anti-shake threshold; when the harmonic level drops, the original window parameters are automatically restored; thus achieving a dynamic balance between environmental adaptation and monitoring sensitivity. In summary, by embedding a segment-side processing unit on the secondary side of the low-voltage instrument transformer, and performing steps such as multi-cycle electrical signal acquisition, end-side AI inference, sliding window anti-jitter post-processing, and fault output, the traditional low-voltage instrument transformer is endowed with independent health status perception and assessment capabilities. This invention achieves real-time, continuous, and online autonomous monitoring of the instrument transformer's operating status without the need for continuous intervention from an external host computer, significantly improving the speed and accuracy of fault detection and reducing maintenance manpower costs. Attached Figure Description

[0016] Figure 1 A flowchart of an online health monitoring method for low-voltage transformers based on end-side AI according to the present invention is shown; Figure 2 A schematic diagram of the normal current waveform of the present invention is shown; Figure 3 A schematic diagram of the core saturation fault waveform of the present invention is shown; Figure 4 A schematic diagram of a loose wiring fault waveform is shown in this invention; Figure 5 A schematic diagram of the DC offset fault waveform of the present invention is shown; Figure 6A block diagram of an online health monitoring system for low-voltage transformers based on end-side AI is shown. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] Figure 1 A flowchart of an online health monitoring method for low-voltage transformers based on end-side AI according to the present invention is shown.

[0020] like Figure 1 As shown, this invention discloses an online health monitoring method for low-voltage current transformers based on end-side AI, comprising: S101, based on a preset time period, periodically collects the electrical signal output from the secondary side of the current transformer for N power frequency cycles, and collects M sampling points in each cycle to form the original sampling matrix; S102, the original sampling matrix is ​​sent to the preset terminal for processing to determine the health index and fault confidence; S103, if the health index is less than the preset first health index threshold or the fault confidence is greater than the preset first fault confidence threshold, it is determined that there is a health abnormality at the current moment, and the collected data of the current period is marked as a preliminary abnormality. S104, Based on a preset sliding time window, extract the number of preliminary anomaly markers within the corresponding sliding time window; S105, If the number of preliminary abnormal markers within the sliding time window exceeds the preset anti-shake threshold, then the final abnormal marker is generated and a health warning message is triggered.

[0021] According to an embodiment of the present invention, a preset time period is set as A. If the first data acquisition time is t, then the second data acquisition time is t+A, the third data acquisition time is t+2A, and so on. The preset end side includes an MCU (Microcontroller Unit) main control unit, which is used to process the acquired electrical signals and determine the health index and fault confidence. For example, the preset first health index threshold is 0.8, the preset first fault confidence threshold is 0.5, the preset sliding time window L=3, and the preset anti-shake threshold is 2. Then, when more than two of the three consecutive periods within the preset sliding time window are marked as abnormal, the final abnormal mark is output.

[0022] According to an embodiment of the present invention, the step of sending the original sampling matrix to a preset terminal for processing to determine the health index and fault confidence specifically includes: Each period of the original sampling matrix is ​​normalized and aligned with the waveform to obtain a normalized waveform matrix. The feature parameters of each period are then extracted, including the effective value and harmonic amplitude value. Based on the harmonic amplitude, determine the maximum value of the waveform distortion rate; extract the odd harmonic amplitude from the harmonic amplitude, and determine the maximum value of the proportion of odd harmonic amplitude based on the odd harmonic amplitude; The waveform quality index is determined based on the maximum percentage of odd harmonic amplitude and the maximum waveform distortion rate. Its formula is: ,in , These represent the maximum percentage of odd harmonic amplitude and the maximum waveform distortion rate, respectively. , These represent the health baseline values ​​for the proportion of odd harmonic amplitude and waveform distortion rate, respectively. This represents the corresponding weighting coefficient; Extract the harmonics of the current cycle and the harmonics of the adjacent previous cycle; By comparing and analyzing the harmonics of the current cycle with the harmonics of the adjacent previous cycle, the waveform similarity value of the current cycle is determined. After traversing the harmonics of all cycles, the waveform similarity value set is determined. The periodic stability index of the current harmonic is determined based on the effective value and the set of waveform similarity values. Based on a preset algorithm, the health index and fault confidence level are determined according to the periodic stability index and waveform quality index of the current harmonics.

[0023] It should be noted that the periodic stability index of the current harmonic is set to... The formula is: ,in This represents the effective value of the i-th period. The mean of the effective values ​​over N periods. This represents the waveform similarity value for the i-th cycle; the preset algorithm includes an algorithm for calculating the health index, specifically: HI represents the health index. , These are the corresponding sensitivity coefficients; the preset algorithm includes an algorithm for calculating the fault confidence index, which is specifically as follows: ,in , These are the warning thresholds for the periodic stability index and the waveform quality index, respectively.

[0024] According to an embodiment of the present invention, the step of determining the maximum value of waveform distortion rate / determining the maximum value of odd harmonic amplitude ratio specifically includes: The harmonic amplitude is divided according to the period to determine the harmonic amplitude of different periods; If the waveform distortion rate is set as THD, then the waveform distortion rate corresponding to the i-th period is THD(i), and its formula is: ,in The fundamental amplitude, Let h be the amplitude of the h-th harmonic within the i-th period. This represents the harmonic fluctuation weighting coefficient for the i-th period; Extract the odd harmonic amplitude from the harmonic amplitude values ​​of the same period; Set the amplitude ratio of odd harmonics to Then the proportion of the odd harmonic amplitude corresponding to the i-th period is Its formula is ; After traversing all cycles, we obtain the waveform distortion rate set and the odd harmonic amplitude ratio set; Extract the maximum values ​​from the waveform distortion rate set and the odd harmonic amplitude percentage set, which correspond to the maximum value of the waveform distortion rate and the maximum value of the odd harmonic amplitude percentage.

[0025] According to an embodiment of the present invention, the step of obtaining the harmonic fluctuation weighting coefficient specifically includes: Extract the amplitude sequence of characteristic harmonic orders in each period of the original sampling matrix; The amplitudes of each characteristic harmonic order in the current cycle are compared and analyzed with those in the previous cycle to determine the cross-correlation vector; If the cross-correlation vector is greater than the preset first similarity threshold, it is determined that the harmonic fluctuation of the current period originates from the background harmonic change on the system side, and the preset first value is set as the corresponding harmonic fluctuation weight. If the cross-correlation vector is less than or equal to the preset first similarity threshold, it is determined that the harmonic fluctuation of the current cycle originates from the change of the characteristics of the transformer itself, and the preset second value is set as the corresponding harmonic fluctuation weight. The preset first value is less than the preset second value.

[0026] It should be noted that since harmonic variations are caused by inverter load adjustments, the harmonic amplitude changes in adjacent cycles tend to follow the same trend. If the cross-correlation vector is greater than a preset first similarity threshold, it is determined that background harmonics dominate, and the THD and [other values] of that cycle are [calculated]. The harmonic fluctuation weight, determined by the maximum value of waveform distortion rate / the maximum value of odd harmonic amplitude proportion, is reduced to a preset first value, for example, setting the preset first value to 0.2 and the preset second value to 1; thereby making It is almost unaffected, thus avoiding false alarms; the cross-correlation vector can be replaced by the similarity value of the amplitude of each characteristic harmonic number of the corresponding period and the previous period.

[0027] According to an embodiment of the present invention, after obtaining the waveform distortion rate set and the odd harmonic amplitude ratio set, the method further includes: Within each power frequency cycle, the DC component of the original sampled signal is extracted to obtain the DC component value; Extract the even harmonic amplitude from the harmonic amplitude values ​​of the same period; Based on the even harmonic amplitude in the same period, determine the proportion of even harmonic amplitude in the corresponding period. If the absolute value of the DC component is greater than the preset DC component threshold, and the amplitude ratio of the even harmonic is greater than the preset even harmonic ratio threshold and the amplitude of the second harmonic is greater than the amplitude of the third harmonic, then the current period is determined to be a DC bias period. The waveform distortion rate and the proportion of odd harmonic amplitude in the DC bias period are deleted.

[0028] It should be noted that the amplitude proportion of even harmonics is set as Its formula is: The preset DC component threshold value is 0.5% to 2% of the rated current of the transformer, and the preset even harmonic proportion threshold value ranges from 1% to 3%.

[0029] According to an embodiment of the present invention, it further includes: Within each power frequency cycle, extract the phase relationship before and after the current zero-crossing point; Based on a preset voltage phase reference signal, the direction of the current in this cycle is determined to be either positive or negative. Based on the current direction, a positive direction feature buffer and a negative direction feature buffer are constructed respectively, and the positive direction feature buffer and the negative direction feature buffer independently store the corresponding feature parameters; The periodic stability index and waveform quality index are calculated based on the characteristic data of the buffer corresponding to the direction of the current period.

[0030] It should be noted that at the distributed power grid connection point, power may flow bidirectionally. When the load exceeds the photovoltaic power generation, current flows from the grid to the load; when photovoltaic power generation is in surplus, current flows from the user side to the grid. The phase and amplitude characteristics of the current on the secondary side of the transformer may differ significantly in different directions. Mixing bidirectional data for statistical analysis would lead to distortion of the effective value mean and deviations in the stability index calculation. Therefore, by establishing separate forward and reverse characteristic buffers, health assessments are based on electrical signal data from the same direction, ensuring the consistency of statistical characteristics.

[0031] According to an embodiment of the present invention, it further includes: Within a preset stabilization period, if the health index is greater than a preset second health index threshold and the fault confidence is less than a preset second fault confidence threshold for n consecutive periods, then a baseline value buffer update is triggered, and the update formula is as follows: ,in This represents the updated baseline value. This indicates the baseline values ​​before the update, which include a healthy baseline value for the proportion of odd harmonic amplitude and a healthy baseline value for waveform distortion rate. For the corresponding adjustment coefficient, These are the feature parameters of the j-th cycle out of the current n cycles; The preset second health index threshold is greater than the preset first health index threshold, and the preset second fault confidence threshold is less than the preset first fault confidence threshold.

[0032] It should be noted that during long-term operation, the electrical characteristics of current transformers may slowly drift due to factors such as insulation aging, temperature changes, and minor wear. If the reference value remains constant, this slow drift may be misinterpreted as an anomaly. Conversely, if the reference value can slowly track changes in the normal state, the true fault can be accurately identified. For example, setting the preset second health index threshold to 0.95 and the preset second fault confidence threshold to 0.3... Set it to 0.98.

[0033] According to an embodiment of the present invention, it further includes: The mean and variance of the waveform quality index are monitored in real time within a preset first number of window periods, which are set as the first mean and the first variance, respectively. If the first mean exceeds the preset first threshold and the first variance is less than the preset stability threshold, then the current environment is determined to be in a continuous high harmonic state, and information on the length of the extended sliding time window is generated. The length of the sliding time window is adjusted based on a preset delay value to obtain the adjusted sliding time window. And adjust the image stabilization threshold based on the corrected sliding time window. Its formula is: ,in This indicates the corrected sliding time window length, L represents the preset sliding time window length, and K represents the preset anti-shake threshold.

[0034] It should be noted that, for example, if the preset delay value is 2 and the sliding time window length is 3, then the corrected sliding time window length is 2+3=5, where ceil is the rounding operation. For example, the preset first threshold can be set to 1.5 times the base value.

[0035] According to an embodiment of the present invention, it further includes: After using the corrected sliding time window, extract the average value of the waveform quality index within a preset second number of window periods and set it as the second average value. If the second mean is less than the preset second threshold, then the length of the next sliding time window will be restored to the preset sliding time window length. The preset second quantity is greater than the preset first quantity, and the preset second threshold is less than the preset first threshold.

[0036] It should be noted that, for example, if the preset second quantity is set to 5 and the preset second threshold is 1.2 times the benchmark value, then when the average value of the waveform quality index is less than 1.2 times the benchmark value within 5 consecutive window periods, the original window parameters are restored, for example, to 3.

[0037] Figure 2 A schematic diagram of the normal current waveform of the present invention is shown.

[0038] like Figure 2 As shown, Figure 2 This is a schematic diagram of a normal current waveform. The waveform is a standard power frequency sine wave with a uniform period, complete symmetry between the positive and negative half-cycles, no distortion, and no DC offset. It is a typical waveform characteristic of a current transformer in normal operation.

[0039] Figure 3 A schematic diagram of the core saturation fault waveform of the present invention is shown.

[0040] like Figure 3 As shown, Figure 3The diagram shows a waveform of a core saturation fault. The tops of both the positive and negative half-cycles of the waveform are horizontally clipped, exhibiting a distinct flat-top waveform characteristic. The harmonic content is significantly increased, which is a typical electrical signal manifestation of a core saturation fault.

[0041] Figure 4 A schematic diagram of a loose wiring fault waveform is shown in this invention.

[0042] like Figure 4 As shown, Figure 4 This is a waveform diagram of a loose wiring fault. The waveform shows obvious discontinuities, accompanied by random amplitude fluctuations, momentary drops, and jumps. The period is also unstable, accurately reflecting the fault characteristics of loose wiring / poor contact.

[0043] Figure 5 A schematic diagram of the DC offset fault waveform of the present invention is shown.

[0044] like Figure 5 As shown, Figure 5 The diagram shows a DC offset fault waveform. The entire sine wave shows an upward or downward overall offset, the positive and negative half-waves lose their symmetrical characteristics, and there is a significant DC component. This is a typical electrical signal manifestation of a DC offset fault.

[0045] Figure 6 A block diagram of an online health monitoring system for low-voltage transformers based on end-side AI is shown.

[0046] like Figure 6 As shown, the second aspect of the present invention provides an online health monitoring system for low-voltage transformers based on end-side AI, which further includes: a low-voltage current / voltage transformer, a signal conditioning circuit, an ADC acquisition unit, an MCU main control unit, a storage unit, and a communication unit; The MCU main control unit is connected to the ADC acquisition unit, the storage unit, and the communication unit respectively. The output terminal of the signal conditioning circuit is connected to the ADC acquisition unit, and the output terminal of the low-voltage transformer is connected to the signal conditioning circuit. It is used to execute the low-voltage transformer online health monitoring method based on end-side AI as described in any one of the above.

[0047] According to an embodiment of the present invention, the ADC acquisition unit is used to acquire the secondary current signal of the low-voltage current / voltage transformer. For example, the ADC acquisition unit continuously acquires the secondary current signal of the low-voltage transformer at a sampling rate of 12.8kHz, acquiring 256 points per power frequency cycle (20ms), and continuously acquiring N=10 cycles to form a signal. The original sampling matrix; the signal conditioning circuit is used to convert the weak and easily interfered signals output by the transformer into standardized electrical signals that can be effectively processed by subsequent circuits; the MCU active unit includes an end-side AI inference module and a fault diagnosis and early warning model, wherein the end-side AI inference module has an algorithm and steps for calculating the health index and fault confidence based on the electrical signals, and the fault diagnosis and early warning model is set with the judgment thresholds for the health index and fault confidence and a health warning device; when a health warning information is triggered, the preset management terminal is notified through the communication unit, and the storage unit is used to store the data in the entire online health monitoring process for traceability and retrieval.

[0048] This invention discloses an online health monitoring method and system for low-voltage instrument transformers based on end-side AI. By embedding an end-side processing unit on the secondary side of the low-voltage instrument transformer, it performs steps such as multi-cycle electrical signal acquisition, end-side AI inference, sliding window anti-jitter post-processing, and fault output, enabling traditional low-voltage instrument transformers to have independent health status perception and assessment capabilities. This invention achieves real-time, continuous, and online autonomous monitoring of the instrument transformer's operating status without the need for continuous intervention from an external host computer, significantly improving the speed and accuracy of fault detection and reducing maintenance manpower costs.

[0049] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0050] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0051] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0052] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A method for online health monitoring of low-voltage instrument transformers based on end-side AI, characterized in that, include: Based on a preset time period, the electrical signal output from the secondary side of the current transformer for N power frequency cycles is periodically collected, and M sampling points are collected in each cycle to form the original sampling matrix. The original sampling matrix is ​​sent to a preset terminal for processing to determine the health index and fault confidence. If the health index is less than the preset first health index threshold or the fault confidence is greater than the preset first fault confidence threshold, it is determined that there is a health anomaly at the current moment, and the collected data of the current period is marked as a preliminary anomaly. Based on a preset sliding time window, extract the number of preliminary anomaly markers within the corresponding sliding time window; If the number of initial anomaly markers within the sliding time window exceeds the preset anti-shake threshold, a final anomaly marker is generated, and a health warning message is triggered.

2. The method for online health monitoring of low-voltage current transformers based on end-side AI according to claim 1, characterized in that, The step of sending the original sampling matrix to a preset terminal for processing to determine the health index and fault confidence specifically includes: Each period of the original sampling matrix is ​​normalized and aligned with the waveform to obtain a normalized waveform matrix. The feature parameters of each period are then extracted, including the effective value and harmonic amplitude value. Based on the harmonic amplitude, determine the maximum value of the waveform distortion rate; extract the odd harmonic amplitude from the harmonic amplitude, and determine the maximum value of the proportion of odd harmonic amplitude based on the odd harmonic amplitude; The waveform quality index is determined based on the maximum percentage of odd harmonic amplitude and the maximum waveform distortion rate. Its formula is: ,in , These represent the maximum percentage of odd harmonic amplitude and the maximum waveform distortion rate, respectively. , These represent the health baseline values ​​for the proportion of odd harmonic amplitude and waveform distortion rate, respectively. This represents the corresponding weighting coefficient; Extract the harmonics of the current cycle and the harmonics of the adjacent previous cycle; By comparing and analyzing the harmonics of the current cycle with the harmonics of the adjacent previous cycle, the waveform similarity value of the current cycle is determined. After traversing the harmonics of all cycles, the waveform similarity value set is determined. The periodic stability index of the current harmonic is determined based on the effective value and the set of waveform similarity values. Based on a preset algorithm, the health index and fault confidence level are determined according to the periodic stability index and waveform quality index of the current harmonics.

3. The method for online health monitoring of low-voltage transformers based on end-side AI according to claim 2, characterized in that, The steps of determining the maximum value of waveform distortion rate / determining the maximum value of odd harmonic amplitude proportion specifically include: The harmonic amplitude is divided according to the period to determine the harmonic amplitude of different periods; If the waveform distortion rate is set as THD, then the waveform distortion rate corresponding to the i-th period is THD(i), and its formula is: ,in The fundamental amplitude, Let h be the amplitude of the h-th harmonic within the i-th period. This represents the harmonic fluctuation weighting coefficient for the i-th period; Extract the odd harmonic amplitude from the harmonic amplitude values ​​of the same period; Set the amplitude ratio of odd harmonics to Then the proportion of the odd harmonic amplitude corresponding to the i-th period is Its formula is ; After traversing all cycles, we obtain the waveform distortion rate set and the odd harmonic amplitude ratio set; Extract the maximum values ​​from the waveform distortion rate set and the odd harmonic amplitude percentage set, which correspond to the maximum value of the waveform distortion rate and the maximum value of the odd harmonic amplitude percentage.

4. The method for online health monitoring of low-voltage transformers based on end-side AI according to claim 3, characterized in that, The steps for obtaining the harmonic fluctuation weighting coefficient specifically include: Extract the amplitude sequence of characteristic harmonic orders in each period of the original sampling matrix; The amplitudes of each characteristic harmonic order in the current cycle are compared and analyzed with those in the previous cycle to determine the cross-correlation vector; If the cross-correlation vector is greater than the preset first similarity threshold, it is determined that the harmonic fluctuation of the current period originates from the background harmonic change on the system side, and the preset first value is set as the corresponding harmonic fluctuation weight. If the cross-correlation vector is less than or equal to the preset first similarity threshold, it is determined that the harmonic fluctuation of the current cycle originates from the change of the characteristics of the transformer itself, and the preset second value is set as the corresponding harmonic fluctuation weight. The preset first value is less than the preset second value.

5. The method for online health monitoring of low-voltage transformers based on end-side AI according to claim 3, characterized in that, After obtaining the waveform distortion rate set and the odd harmonic amplitude ratio set, the method further includes: Within each power frequency cycle, the DC component of the original sampled signal is extracted to obtain the DC component value; Extract the even harmonic amplitude from the harmonic amplitude values ​​of the same period; Based on the even harmonic amplitude in the same period, determine the proportion of even harmonic amplitude in the corresponding period. If the absolute value of the DC component is greater than the preset DC component threshold, and the amplitude ratio of the even harmonic is greater than the preset even harmonic ratio threshold and the amplitude of the second harmonic is greater than the amplitude of the third harmonic, then the current period is determined to be a DC bias period. The waveform distortion rate and the proportion of odd harmonic amplitude in the DC bias period are deleted.

6. The method for online health monitoring of low-voltage current transformers based on end-side AI according to claim 2, characterized in that, Also includes: Within each power frequency cycle, extract the phase relationship before and after the current zero-crossing point; Based on a preset voltage phase reference signal, the direction of the current in this cycle is determined to be either positive or negative. Based on the current direction, a positive direction feature buffer and a negative direction feature buffer are constructed respectively, and the positive direction feature buffer and the negative direction feature buffer independently store the corresponding feature parameters; The periodic stability index and waveform quality index are calculated based on the characteristic data of the buffer corresponding to the direction of the current period.

7. The method for online health monitoring of low-voltage transformers based on end-side AI according to claim 2, characterized in that, Also includes: Within a preset stabilization period, if the health index is greater than a preset second health index threshold and the fault confidence is less than a preset second fault confidence threshold for n consecutive periods, then a baseline value buffer update is triggered, and the update formula is as follows: ,in This represents the updated baseline value. This indicates the baseline values ​​before the update, which include a healthy baseline value for the proportion of odd harmonic amplitude and a healthy baseline value for waveform distortion rate. For the corresponding adjustment coefficient, These are the feature parameters of the j-th cycle out of the current n cycles; The preset second health index threshold is greater than the preset first health index threshold, and the preset second fault confidence threshold is less than the preset first fault confidence threshold.

8. The method for online health monitoring of low-voltage current transformers based on end-side AI according to claim 1, characterized in that, Also includes: The mean and variance of the waveform quality index are monitored in real time within a preset first number of window periods, which are set as the first mean and the first variance, respectively. If the first mean exceeds the preset first threshold and the first variance is less than the preset stability threshold, then the current environment is determined to be in a continuous high harmonic state, and information on the length of the extended sliding time window is generated. The length of the sliding time window is adjusted based on a preset delay value to obtain the adjusted sliding time window. And adjust the image stabilization threshold based on the corrected sliding time window. Its formula is: ,in This indicates the corrected sliding time window length, L represents the preset sliding time window length, and K represents the preset anti-shake threshold.

9. The method for online health monitoring of low-voltage current transformers based on end-side AI according to claim 8, characterized in that, Also includes: After using the corrected sliding time window, extract the average value of the waveform quality index within a preset second number of window periods and set it as the second average value. If the second mean is less than the preset second threshold, then the length of the next sliding time window will be restored to the preset sliding time window length. The preset second quantity is greater than the preset first quantity, and the preset second threshold is less than the preset first threshold.

10. A low-voltage current transformer online health monitoring system based on end-side AI, characterized in that, Also includes: Low-voltage current / voltage transformer, signal conditioning circuit, ADC acquisition unit, MCU main control unit, storage unit and communication unit; The MCU main control unit is connected to the ADC acquisition unit, the storage unit, and the communication unit respectively. The output terminal of the signal conditioning circuit is connected to the ADC acquisition unit, and the output terminal of the low-voltage transformer is connected to the signal conditioning circuit. It is used to execute the low-voltage transformer online health monitoring method based on end-side AI as described in any one of claims 1-9.

Images