Fault monitoring device for power line

Abstract

The invention provides a fault monitoring device for a power line, and the device comprises a plurality of probes, a time service module, and a processing unit. The probes are isolated from one another and are respectively connected with the processing unit; the probe at least comprises an electric field probe and a magnetic field probe; the probe is used for detecting electric field and magnetic field data corresponding to a power line; the time service module is used for providing a corresponding clock signal for the processing unit; the processing unit is used for carrying out local time maintenance according to a built-in timer and carrying out long-period calibration according to a received clock signal so as to ensure the accuracy of time service; and the processor is also used for preprocessing the received electric field and magnetic field data, identifying the frequency characteristic of the fault by comparing the change of the high-frequency and low-frequency signals, and obtaining a fault monitoring result. According to the invention, through innovation in two aspects of hardware and algorithm application, fault monitoring on multi-scene power equipment is realized, an edge detection artificial intelligence algorithm is provided, and a detection result is timely and efficiently obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system fault detection, in particular to a fault monitoring device for a power line. BACKGROUND

[0002] Power line fault monitoring is the core link to ensure the stable operation of the power grid. Traditional contact monitoring (such as voltage / current transformers) needs to be directly connected to the line, which has problems such as insulation aging and complex installation. With the development of smart grid, non-contact monitoring technology has become a research hotspot due to its high safety and flexible deployment. The core is to indirectly sense the line state through electromagnetic field, infrared and other physical quantities to avoid electrical contact risk. However, the reliability and accuracy of a single sensor may be affected by external interference, and the detection result is not accurate. The probe sensor is not separated and may be damaged as a whole. The fault sampling rate is fixed, and the adaptive scene is fixed. The signal recognition accuracy is insufficient, and deep learning and algorithm are not added. It does not have edge computing power, and all computing power depends on the server, which is poor in real-time and limited by network conditions. No multi-data fusion analysis is performed, and comprehensive analysis of multiple signals is lacking. The data communication reliability is not processed, resulting in data loss when there is no network. SUMMARY

[0003] The purpose of the present application is to provide a fault monitoring device for a power line, which solves the technical problem of detecting power line faults in a non-contact manner.

[0004] In one aspect, a fault monitoring device for a power line is provided, comprising:

[0005] a plurality of probes, a time-providing module, and a processing unit; each probe is isolated from each other and connected to the processing unit; the probe at least includes an electric field probe and a magnetic field probe;

[0006] The probe is used to detect the electric field and magnetic field data corresponding to the power line;

[0007] The time-providing module is used to provide a corresponding clock signal for the processing unit;

[0008] The processing unit is used to maintain the local time according to the built-in timer, and to calibrate the long period according to the received clock signal to ensure the accuracy of the time-providing; it is also used to preprocess the received electric field and magnetic field data, and convert the preprocessed electric field and magnetic field data into frequency domain signals, identify the frequency characteristics of the fault by comparing the changes of high frequency and low frequency signals, and obtain the fault monitoring result.

[0009] Preferably, it further comprises a server end for sending corresponding control instructions to the processing unit and accepting the processing result and fault monitoring result fed back by the processing unit;

[0010] The communication module is configured to communicate between the server and the processing unit.

[0011] The data storage module is configured to store the detected data and data records generated during the processing.

[0012] The infrared temperature measurement module is configured to detect the temperature of the target power line in real time.

[0013] Preferably, the processing unit is specifically configured to,

[0014] According to the trigger time and the recording time length, the electric field and magnetic field data within the first preset proportion threshold before the trigger time and the recording time length and the electric field and magnetic field data within the second preset proportion threshold after the trigger time are integrated into an original waveform.

[0015] The original waveform is sequentially filtered, de-noised, restored, and de-repeated, and the processed waveform is stored in the data storage module.

[0016] Preferably, filtering the original waveform includes,

[0017] The original waveform is moved and translated according to the following formula,

[0018]

[0019] wherein y n represents the output at the nth moment; x(n-k) represents the input signal at the n-k moment; and M represents the window length.

[0020] The original waveform after moving and translating is adaptively filtered according to the following formula,

[0021] y(n) = w T (n)x(n)

[0022] wherein w T (n)x(n) represents the inner product of the coefficient and the input signal; and y(n) represents the waveform after adaptive filtering.

[0023] Preferably, de-noising the original waveform includes,

[0024]

[0025] wherein x represents the position of the trigger time in the waveform; n represents the total number of check points; Y represents the actual sampling value; and h represents the trigger threshold.

[0026] Preferably, restoring the original waveform includes,

[0027]

[0028] |y N |-p aver >T

[0029] wherein, Xi represents the i-th data point in the sequence or signal; x i+1 represents the i+1-th data point in the sequence or signal; N represents the front-end data amount of data points; |x i+1 -x i | represents the absolute difference value between the two consecutive points; y N represents the waveform data value; p aver represents the differential absolute value mean; T represents the mutation coefficient.

[0030] Preferably, the de-duplication of the original waveform comprises,

[0031]

[0032] wherein, x i and y i are the values of the waveform data x(t) and y(t) at the i-th moment, respectively; n is the length of the waveform data.

[0033] Preferably, the processing unit is further configured to determine corresponding time-domain data according to the electric field and magnetic field data, and convert the time-domain signal into a frequency-domain signal.

[0034] The frequency-domain signal is input into a preset convolutional neural network, frequency features in the signal are extracted through a convolutional layer and a pooling layer, and a final fault monitoring result is obtained.

[0035] Preferably, the time-domain signal is converted into a frequency-domain signal through Fourier transform by the following formula,

[0036]

[0037] wherein, x(t) represents the time-domain signal; X(f) represents the frequency spectrum at different frequencies f; e -j2πft is a complex exponential function, representing the frequency component of the signal; dt represents the time interval when sampling the discrete time signal.

[0038] Preferably, the convolutional neural network extracts the frequency features in the signal through the following formula,

[0039]

[0040] wherein, C ij represents the value at the (i, j) position of the output feature map; X represents the output waveform data; W represents the convolution kernel weight matrix; mn represents the position parameter of the convolution kernel weight matrix; b represents the bias term; and σ represents the activation function.

[0041] In summary, the embodiment of the present application has the following beneficial effects:

[0042] The fault monitoring device for the power line provided by the present application realizes fault monitoring of multiple scene power equipment and provides edge measurement artificial intelligence algorithm through the innovation of hardware and algorithm application, and timely and efficiently obtains detection results. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings obtained according to these drawings without creative labor are still within the scope of the present application.

[0044] Figure 1 FIG. 1 is a schematic diagram of a fault monitoring device for a power line in an embodiment of the present application.

[0045] Figure 2 FIG. 1 is a schematic diagram of a fault monitoring device for a power line in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings.

[0047] As shown in FIGS. 1 and 2, FIG. 1 is a schematic diagram of an embodiment of a fault monitoring device for a power line provided by the present application. In this embodiment, the device comprises: Figure 1 Figure 2 As shown in FIGS. 1 and 2, FIG. 1 is a schematic diagram of an embodiment of a fault monitoring device for a power line provided by the present application. In this embodiment, the device comprises:

[0048] ​A plurality of probes, time module, processing unit; each probe is isolated from each other and is connected with the processing unit; the probe at least includes electric field probe, magnetic field probe; the probe is used for detecting the electric field and magnetic field data corresponding to the power line; the time module is used for providing the corresponding clock signal for the processing unit; the processing unit is used for maintaining the local time according to the built-in timer, and calibrating the long period according to the received clock signal, to ensure the accuracy of the time service; also used for pre-processing the received electric field and magnetic field data, and converting the pre-processed electric field and magnetic field data into frequency domain signal, identifying the frequency characteristics of the fault by comparing the changes of high frequency and low frequency signals, obtaining the fault monitoring result. Server, used for sending corresponding control instructions to the processing unit, and accepting the processing result and fault monitoring result fed back by the processing unit; the communication module is used for the communication connection between the server and the processing unit; the data storage module is used for storing the detected data and the data record generated in the processing process; the infrared temperature measurement module is used for real-time detection of the temperature of the target power line.

[0049] The probe is composed of four separate sensors and processing circuits, magnetic field traveling wave, magnetic field power frequency, electric field traveling wave, and electric field power frequency. The advantages are that the signals are isolated from each other to avoid interference, and even if an abnormality or damage occurs in a certain sensor, the other sensors can still work normally. By combining the signals of the electric field and magnetic field probes, the fault type and location can be more accurately identified. The operation state of the power line can be comprehensively captured from multiple dimensions, thereby reducing the misjudgment caused by a single signal source. The time service module adopts a Beidou time service module to provide a high-precision clock for the central processing unit. The central processing unit adopts a combined architecture of CPU and GPU, which can ensure that the waveform processing capability is still efficient when multiple tasks are processed in real time. The core functions include: ① clock self-maintenance function, which is responsible for maintaining the local time by an internal high-precision timer, and is calibrated by the Beidou time service module for a long period to ensure the accuracy of time service. This function effectively avoids the time authorization error caused by the fluctuation of the Beidou signal, and reduces the occupation of CPU resources by external devices. ② signal capture function, the sampling rate is adjustable in the range of 2k-5mhz. By setting the ADC (analog-to-digital converter) threshold, when the signal exceeds the set threshold, the system can automatically capture the original waveform data and record the capture time. This mechanism can ensure effective recording when a fault signal occurs. ③ data preprocessing, including deburring (filtering noise interference), removing clutter (removing unnecessary stray signals), removing repeated waveforms (deleting repeated signal waveform data), and restoring waveform start time (correcting the timestamp of the signal to restore its start time). After preprocessing, the obtained data is valid data that can be used for subsequent algorithm analysis. ④ data management, which stores valid data in an external flash and maintains a certain amount of historical data. Through the circular storage method, the storage space is avoided from being used up, and the data is uploaded and used according to the connection state of the wireless communication. ⑤ algorithm processing, which is through fast Fourier transform (FFT) and convolutional neural network (CNN). The FFT is used to convert the time domain signal to the frequency domain signal, and by comparing the changes of high-frequency and low-frequency signals, the frequency characteristics of the fault are identified. For high-frequency faults (such as electrical short circuit), FFT will show higher frequency components; while for low-frequency faults (such as device failure or electrical fault), FFT will show lower frequency components. Through the deep learning algorithm, the frequency spectrum data processed by FFT is classified and pattern recognized, and combined with temperature data to determine the fault type (such as short circuit, overload, ground fault, etc.). CNN can automatically extract key features in the signal and establish a model based on training data for real-time identification and classification of faults. The data storage module stores valid data waveforms. The infrared temperature measurement module detects the target temperature in real time. The wireless communication module provides a wireless communication network for the whole machine, which is used for communication with the server and can be used for real-time alarm, remote control, remote maintenance, etc.

[0050] One specific embodiment of the present invention, electric field probe: for sensing the electric field signal around the power line, especially when the line fault (such as short circuit, grounding, etc.), the electric field distribution will change significantly. Electric field probe can capture these changes in real time and convert them into signals for analysis. Common effects include:

[0051] Sensing voltage anomalies: for example, when a short circuit fault occurs, the electric field distribution of the power line will change due to voltage instability, and the electric field probe can detect this change.

[0052] Locating the source of the fault: by measuring the strength and distribution of the electric field, the electric field probe can help determine the possible location of the fault point.

[0053] Real-time monitoring of power line status: the electric field probe can continuously monitor the electric field changes of the power line, providing real-time data for fault diagnosis.

[0054] Magnetic field probe: for detecting the magnetic field changes around the power line. The current of the power line will produce a magnetic field, and the fault (such as short circuit, grounding) will cause the current to change or fluctuate, causing abnormal changes in the magnetic field. The role of the magnetic field probe includes:

[0055] Sensing current anomalies: transient changes in current, such as a surge in fault current, will cause changes in magnetic field strength. The magnetic field probe can detect these changes and reflect the fault of the power line in time.

[0056] Fault type identification: different types of faults (such as short circuit, poor contact, etc.) will usually produce different characteristics in the magnetic field signal. By analyzing the frequency, amplitude and phase changes of the magnetic field signal, the magnetic field probe can help identify the fault type.

[0057] Improve fault detection sensitivity: compared with traditional contact detection methods, the magnetic field probe can efficiently detect faults without contacting the power line, especially in high-voltage power systems, avoiding the possibility of personnel contacting dangerous.

[0058] One specific embodiment of the present invention, the processing unit is specifically used for integrating the electric field and magnetic field data within the first preset proportion threshold before the trigger time and the recording time and the electric field and magnetic field data within the second preset proportion threshold after the trigger time for a recording time into an original waveform according to the trigger time and the recording time; sequentially filtering, de-noising, restoring the wave head, and removing the repeated waveform of the original waveform, and storing the processed waveform into the data storage module.

[0059] Specifically, the adaptive filtering combination moving average algorithm is used for filtering. First, the original data is filtered by moving average, and then adaptive filtering is performed. Filtering the original waveform includes moving the original waveform according to the following formula,

[0060]

[0061] wherein, y n represents the output at the nth moment; x(n-k) represents the input signal at the n-k moment; M represents the window length;

[0062] The original waveform after moving translation is adaptively filtered according to the following formula,

[0063] y(n)=w T (n)x(n)

[0064] wherein, w T (n)x(n) represents the inner product of the coefficient and the input signal; y(n) represents the waveform after adaptive filtering.

[0065] The waveform after deburring is judged whether it is a fault data waveform. The original waveform is de-noised, including,

[0066]

[0067] wherein, x represents the position in the waveform at the triggering moment; n represents the total number of check points; Y represents the actual sampling value; h represents the triggering threshold.

[0068] The starting point of the waveform data mutation is the moment corresponding to the occurrence of the fault. The mutation point is identified by a combination of a sliding window of a differential absolute value mean. The original waveform is restored to the wave head, including,

[0069]

[0070] |y N |-p aver >T

[0071] wherein, Xi represents the i-th data point in the sequence or signal; x i+1 represents the i+1-th data point in the sequence or signal; N represents the front data amount of the data points; |x i+1 -x i | represents the absolute difference between the two consecutive points; y N represents the waveform data value; p aver represents the differential absolute value mean; T represents the mutation coefficient.

[0072] The fault waveform has a large duration and a long duration (more than 20 / 1 of the recording time), and a single fault will appear multiple times. The Euclidean distance algorithm is used to judge whether the waveform is consistent. The original waveform is de-duplicated, including,

[0073]

[0074] where x i and y i are the values of the waveform data x(t) and y(t) at the i-th time instant, respectively; n is the length of the waveform data.

[0075] The effective waveform data is stored in an external flash, and the flash is dynamically divided into corresponding storage spaces for different platform servers, an annular storage file system is internally implemented, and the reliability of data storage is ensured through read-write flags and network connection states. When the communication connection with the server is abnormal, the unsent data will be stored in the space all the time, which does not affect the storage of new data, and when the communication is restored, the previously captured data will be sent one by one.

[0076] In one specific embodiment of the application, the processing unit is further configured to determine corresponding time-domain data based on the electric field and magnetic field data, and convert the time-domain signal to a frequency-domain signal; take the frequency-domain signal as the input of a preset convolutional neural network, extract the frequency features in the signal through a convolutional layer and a pooling layer, and obtain the final fault monitoring result. The fault category is determined through fast Fourier transform (FFT) and convolutional neural network (CNN). First, the time-domain data of the equipment is collected through the sensor, and the FFT is applied to convert the time-domain signal to a frequency-domain signal. The frequency-domain signal reflects the vibration characteristics of the equipment and can reveal potential fault patterns. The frequency-domain signal after the FFT is taken as the input of the CNN, and the frequency features in the signal are extracted through the convolutional layer and the pooling layer. These frequency features may be associated with different fault types (short circuit, overload, partial discharge, etc.), and the final result is obtained.

[0077] Specifically, the time-domain signal is converted to a frequency-domain signal through Fourier transform by the following formula,

[0078]

[0079] where x(t) represents the time-domain signal; X(f) represents the frequency spectrum at different frequencies f; e -j2πft is a complex exponential function, representing the frequency component of the signal; dt represents the time interval when sampling the discrete time signal.

[0080] The convolutional neural network extracts the frequency features in the signal through the following formula,

[0081]

[0082] where C ij represents the value of the (i, j) position of the output feature map; X represents the output waveform data; W represents the convolution kernel weight matrix; mn represents the position parameter of the convolution kernel weight matrix; b represents the bias term; and sigma represents the activation function.

[0083] The embodiment of the present application integrates the electric field and magnetic field sensors into the probe and is independent of each other, ensures that a single point failure does not affect the overall function, and the reliability is significantly improved. Multi-dimensional sensing fusion: through edge algorithm and artificial intelligence cross verification of four independent sensors and temperature, the false positive rate is reduced, and the redundant design of the sensor ensures that a single point failure does not affect the overall function, and the reliability is significantly improved. The Beidou timing module is combined with the local clock calibration, the time error is controlled in the nanosecond level, which is better than the traditional GPS / Beidou direct timing scheme, and the occurrence position of the transient event can be accurately positioned through multiple detection devices. Effective data accuracy: through multiple preprocessing data algorithms, the data is ensured to be effective, and the interference and subsequent calculation amount are reduced. Edge intelligent diagnosis algorithm: FFT frequency domain analysis combined with CNN pattern recognition, the fault classification is reliable and accurate. The device itself has intelligent computing power and does not depend on the cloud server computing power, and local alarm can still be performed in the case of no network. It has multiple adjustable sampling rates (2kHz-5Mhz), which can adapt to various working scenes (high-voltage power line fault detection, industrial equipment monitoring, partial discharge detection, power system short circuit or overload detection, high-frequency signal interference monitoring). It has communication detection and data storage functions, and ensures that the detection data is not lost.

[0084] Compared with the prior art, the following advantages and effects are achieved: Redundant multi-physical quantity sensing: the probe integrates four independent detection channels of electric field (traveling wave / power frequency) and magnetic field (traveling wave / power frequency), adopts electromagnetic isolation design, single-channel failure does not affect system operation. Through multi-dimensional signal cross verification, the false positive rate is reduced. Nanosecond time synchronization system: the Beidou satellite timing module and the local high-precision timer constitute a double clock source, and nanosecond time synchronization accuracy is achieved through dynamic calibration, supporting multi-device cooperative positioning of transient faults. Adaptive data management: multi-level preprocessing process: noise filtering, clutter suppression, wave head restoration, and waveform deduplication. The cycle storage mechanism retains the data of the last 30 days, and supports breakpoint resume. Edge intelligent diagnosis engine: FFT-CNN hybrid analysis architecture is adopted, FFT extracts signal frequency domain features (sampling rate 2k-5MHz adjustable) in real time, and CNN model automatically classifies multiple faults such as short circuit and overload. Local deployment of lightweight algorithm can maintain autonomous operation and local alarm in the case of network interruption. Support for infrared temperature measurement module extension, through temperature-electromagnetic signal joint analysis, early warning of overload fault is realized.

[0085] In summary, the embodiment of the present application has the following beneficial effects:

[0086] The fault monitoring device for the power line provided by the present application realizes fault monitoring of multiple scene power equipment through the innovation of hardware and algorithm application, and provides edge measurement artificial intelligence algorithm to obtain detection results in time and efficiently.

[0087] The above merely provides the preferred embodiment of the application, and cannot allude the protection scope of the application, therefore any equivalent changes made according to the claims of the application shall be within the scope of the application.

Claims

1. A fault monitoring device for an electrical power line, characterized in that The utility model relates to a kind of power line fault monitoring system, including: Multiple probes, time module, processing unit;Each probe is isolated from each other and is connected with the processing unit respectively;The probe at least includes electric field probe, magnetic field probe; The probe is used to detect the electric field and magnetic field data corresponding to power line; The time module is used to provide corresponding clock signal for processing unit; The processing unit is used to maintain local time according to internal timer, and long-period calibration is carried out according to the received clock signal, to ensure the accuracy of time service;It is also used to pre-process the received electric field and magnetic field data, and convert the pre-processed electric field and magnetic field data into frequency domain signal, identify the frequency characteristics of fault by comparing the changes of high-frequency and low-frequency signals, and obtain fault monitoring result.

2. The apparatus of claim 1, wherein, It also includes a server end for sending corresponding control instructions to the processing unit and accepting the processing results and fault monitoring results fed back by the processing unit; The communication module is used for communication connection between the server and the processing unit; Data storage module is used to store the detected data and data records generated in the processing process; Infrared temperature measurement module is used to detect the temperature of target power line in real time.

3. The apparatus of claim 2, wherein, The processing unit is specifically used for, According to trigger time and recording broadcast length, the electric field and magnetic field data within the first preset proportion threshold before trigger time and the electric field and magnetic field data within the second preset proportion threshold after trigger time are integrated into original waveform according to recording broadcast length; Original waveform is filtered, de-noised, restored and repeated, and the processed waveform is stored in the data storage module.

4. The apparatus of claim 3, wherein, Filtering original waveform includes, According to the following formula, the original waveform is moved and translated, where y n represents the output at the nth moment; x(n-k) represents the input signal at the n-k moment; M represents the window length; According to the following formula, the original waveform is moved and translated, y(n) = w T (n)x(n) where w T (n)x(n) represents the inner product of the coefficients and the input signal; y(n) represents the waveform after adaptive filtering.

5. The apparatus of claim 3, wherein, De-noising original waveform includes, Wherein, x represents the position of trigger time in waveform;N represents the total number of check points;Y represents actual sampling value;H represents trigger threshold.

6. The apparatus of claim 3, wherein, Restoring original waveform includes, | y N | -p aver > T wherein Xi represents the i-th data point in a sequence or signal; x i+1 Xi+1 represents the i+1-th data point in a sequence or signal; N represents the front data amount of data points; |x i+1 -x i | represents the absolute difference value between two consecutive points; y N represents the waveform data value; p aver represents the differential absolute value mean; T represents the mutation coefficient.

7. The apparatus of claim 3, wherein, De-repeating original waveform includes, where x i and y i are the values of the waveform data x(t) and y(t) at the i-th time instant, respectively; n is the length of the waveform data.

8. The apparatus of claim 4, wherein, The processing unit is also used to determine the corresponding time domain data according to the electric field and magnetic field data, and convert the time domain signal into frequency domain signal; Frequency domain signal is used as the input of preset convolutional neural network, and frequency characteristics in signal are extracted through convolution layer and pooling layer to obtain final fault monitoring result.

9. The apparatus of claim 8, wherein, Time domain signal is converted into frequency domain signal by Fourier transform through the following formula, where x(t) represents the time-domain signal; X(f) represents the frequency spectrum at different frequencies f; e -j2πft is a complex exponential function, representing the frequency component of the signal; dt represents the time interval when sampling the discrete-time signal.

10. The apparatus of claim 8, wherein, The convolutional neural network extracts frequency characteristics in signal through the following formula, wherein C ij represents the value of the (i, j) position of the output feature map; X represents the output waveform data; W represents the convolution kernel weight matrix; mn represents the position parameter of the convolution kernel weight matrix; b represents the bias term; and σ represents the activation function.

Images