CN-122020504-A - Defect online monitoring method and device based on multi-feature fusion distribution network power equipment

CN122020504ACN 122020504 ACN122020504 ACN 122020504ACN-122020504-A

Abstract

The invention discloses a defect online monitoring method and device based on multi-feature fusion distribution network power equipment, comprising the following steps of S1, multi-source data acquisition, S2, signal filtering and enhancement processing, S3, multi-feature fusion research and judgment, S4, online early warning and data storage, S5, on-site diagnosis verification and defect elimination, wherein core features of different types of defects are covered through multi-mode data complementation, meanwhile, the limitation of a single signal is effectively avoided by adopting a differential processing technology aiming at different signal characteristics, three-level feature fusion research and judgment logic is constructed, compared with a single feature research and judgment or simple fusion mode, the accuracy of defect type identification, accurate position positioning and severity judgment is effectively improved, an architecture of cooperative work of an online monitoring system and a portable diagnosis device is constructed, the efficient cooperation of online large-scale primary screening and on-site accurate verification is realized, and the accuracy and timeliness of defect treatment are ensured.

Inventors

CHEN HANXIANG
WAN HUAN
Zhi Kai
OUYANG XIAOPING
FAN YUTONG
XIONG HAIQIANG
LIU ZHOU
CHEN JIEFENG
FU LIXIANG
YIN JIAN
LUO YANAN
WANG DAN
YING QIANG

Assignees

国网江西省电力有限公司南昌供电分公司

Dates

Publication Date: 20260512
Application Date: 20251212

Claims (10)

1. The defect online monitoring method based on the multi-feature fusion distribution network power equipment is characterized by comprising the following steps of: the method comprises the steps of S1, multi-source data acquisition, namely, deploying a sensing unit containing multi-type sensing components in a key monitoring area of power equipment of a distribution network, and synchronously acquiring multi-source data related to equipment defects, namely, ultra-high frequency electric signals, infrared thermal imaging, partial discharge ultrasonic signals and visible light images; S2, signal filtering and enhancement processing, namely removing interference noise from the ultrahigh frequency electric signal and the partial discharge ultrasonic signal acquired in the S1 by adopting a noise suppression algorithm, and reserving a partial discharge characteristic frequency band; S3, multi-feature fusion research and judgment, namely, based on the multi-source data processed in the S2, fusing multi-dimensional features through a three-level feature fusion algorithm to generate an integrated diagnosis map, and determining defect information through the integrated diagnosis map; S4, on-line early warning and data storage, namely transmitting the defect information and the integrated diagnostic map output by the step S3 to an on-line monitoring host, triggering grading early warning according to the defect severity degree, and simultaneously storing the related data of the steps S1 to S3 together into a database; S5, performing on-site diagnosis verification and defect elimination, namely enabling operation and maintenance personnel to reach the site with a portable diagnosis device, re-picking multi-source data of the defect position, processing the multi-source data through a built-in fusion algorithm of the device to generate on-site diagnosis results, comparing the on-site diagnosis results with the early warning results of S4, verifying defect information accuracy and generating a defect elimination scheme.
2. The online defect monitoring method based on the multi-feature fusion distribution network power equipment, which is disclosed in claim 1, is characterized in that in S1, the multi-type sensing assembly comprises an ultrahigh frequency sensor, an infrared thermal imaging sensor, an ultrasonic sensor array and a visible light camera, the sensing assembly is arranged in one-to-one correspondence with a key monitoring area of the distribution network power equipment, wherein the online monitoring ultrasonic sensor array is a 128-channel sensor array and is used for collecting partial discharge ultrasonic signals.
3. The online defect monitoring method based on the multi-feature fusion distribution network power equipment is characterized in that an ultrahigh frequency electric signal is collected by an ultrahigh frequency sensor, the collecting frequency is covered by 300MHz-3GHz, the power frequency repeatability features of signal peaks and phases are reserved, the infrared thermal imaging is collected by an infrared thermal imaging sensor, the temperature measuring range is-20 ℃ to 150 ℃, the resolution is more than or equal to 384×288, the partial discharge ultrasonic signal is collected by the 128-channel ultrasonic sensor array, the frequency is covered by 2kHz-35kHz, the visible light image is collected by a visible light camera, the resolution is more than or equal to 1080P, the collecting process is carried out synchronously, and data are transmitted to a data preprocessing module through an Ethernet, and the transmission delay is not more than 1s.
4. The online defect monitoring method based on the multi-feature fusion distribution network power equipment according to claim 1, wherein in the step S2, the noise suppression algorithm is an FIR frequency domain filtering algorithm, and the frequency response function of the noise suppression algorithm meets the following conditions: Wherein, the In order to be a coefficient of the FIR filter, In order to be a filter order number, The frequency is angular frequency and is used for pertinently restraining low-frequency noise in the environment below 50Hz power frequency interference and 1 kHz; the airspace enhancing algorithm is a delay-sum beam forming algorithm, and the directional diagram function of the airspace enhancing algorithm meets the following conditions: Wherein, the Is the number of array elements of the ultrasonic sensor array, Is the first The received signals of the individual array elements are received, For the frequency of the ultrasonic signal, To signal arrive at the first The delay time of each array element is set, Is the first The position coordinates of the individual array elements, Is a unit vector of the direction of the sound source, For the azimuth and elevation of the sound source, The signal to noise ratio is improved by focusing the direction of the local sound source; the Gaussian spatial domain denoising algorithm is adopted for denoising the infrared thermal imaging and the visible light image, and the filtering formula of the Gaussian spatial domain denoising algorithm meets the following conditions: Wherein, the Is the center coordinate of the Gaussian kernel, Is the standard deviation; the contrast enhancement process adopts a histogram equalization algorithm, and the transformation function of the histogram equalization algorithm meets the following conditions: Wherein, the For the gray level of the original image, In order to enhance the gray level of the post-image, As the total number of gray levels of the image, 、 For the number of rows and columns of image pixels, Is of gray scale Is a number of pixels of a display panel.
5. The online defect monitoring method based on the multi-feature fusion distribution network power equipment according to claim 1, wherein in the step S3, the specific process of the three-level feature fusion algorithm is as follows: firstly, extracting temperature characteristics of the infrared thermal imaging after S2 processing, and identifying an abnormal high-temperature region through a threshold judgment formula: Wherein, the Is a pixel point in infrared thermal imaging Is used for the temperature value of (a), For the normal operating temperature of the device corresponding to the pixel, Generating a temperature map for the abnormal temperature threshold, and extracting parameters such as abnormal area, temperature gradient, highest temperature value and the like of the map as temperature map features Preliminary locking of potential defect areas; step two, converting the partial discharge ultrasonic signal processed in the step 2 into an acoustic spectrum through Fourier transformation, wherein the transformation formula is as follows: Wherein, the For the time-space domain ultrasonic signal, For frequency domain acoustic signals, overlapping the acoustic spectrum with the visible light image processed by the S2 one by one according to pixel coordinates, positioning the partial-discharge acoustic source position, extracting parameters such as energy peak value, acoustic source positioning coordinates, signal duration and the like of the acoustic spectrum as ultrasonic-visible light overlapping spectrum characteristics ; Thirdly, extracting the power frequency repeatability characteristics such as peak value, phase, spectrum distribution and the like of the ultrahigh frequency electric signal after S2 processing as the characteristics of the ultrahigh frequency electric signal Will be 、 And The intelligent diagnosis model is input in advance, the multidimensional features are fused through the attention mechanism feature fusion algorithm, and the fusion formula meets the following conditions: Wherein, the Is attention weight coefficient, and The weight coefficient is obtained through self-adaptive learning of an intelligent diagnosis model, and an integrated diagnosis map is generated according to the dynamic distribution of the contribution degree of each characteristic to the defect diagnosis.
6. The online defect monitoring method based on the multi-feature fusion distribution network power equipment of claim 5, wherein the intelligent diagnosis model is a CNN-LSTM deep learning model, the model structure comprises a CNN feature extraction layer and an LSTM time sequence analysis layer, and the training process adopts a cross entropy loss function: Wherein, the As the number of defect classes to be used, Is the first The true label of the class defect, Predicting the first for the model Probability of class defects; the training samples comprise typical defect types of distribution network power equipment, the number of the samples is not less than 500 groups, and the defect information comprises the defect types, the accurate positions and the severity.
7. The method for online monitoring defects of multi-feature fusion distribution network based power equipment according to claim 1, wherein in S4, the data stored in the database comprises original multi-source data of S1, signal processing results of S2, integrated diagnostic patterns of S3 and defect information, and time stamps are synchronously associated when the data are stored Device identification The data storage format adopts a JSON structured format, and supports SQL query, historical data backtracking and batch export.
8. The online defect monitoring method based on the multi-feature fusion distribution network power equipment is characterized in that in the S5, a miniature infrared sensing module, a 32-channel portable ultrasonic sensing array and a visible light acquisition module are integrated in the portable diagnosis device, wherein the temperature measuring range of the miniature infrared sensing module is-20 ℃ to 120 ℃, the resolution is more than or equal to 256 multiplied by 192, the frequency of the portable ultrasonic sensing array is 2kHz to 35kHz, and the resolution of the visible light acquisition module is more than or equal to 720P; The built-in fusion algorithm of the device is homologous with the attention mechanism characteristic fusion algorithm of S3, and weight coefficients are obtained The calculation logic of the system is consistent, the standard uniformity of field diagnosis and on-line monitoring is ensured, and the processing delay of the repeated mining data is not more than 30s.
9. The online defect monitoring method based on the multi-feature fusion distribution network power equipment is characterized by further comprising S6, performing data closed loop and model optimization, namely transmitting the on-site diagnosis result, the defect elimination scheme and state data of equipment continuously operated for 72 hours after defect elimination of S5 back to a database, updating a defect sample data set, and performing iterative training on a CNN-LSTM deep learning model of S3 based on a new added sample to continuously optimize monitoring precision.
10. The defect online monitoring device based on the multi-feature fusion distribution network power equipment is characterized by comprising an online monitoring system and a portable diagnosis device, wherein the online monitoring system and the portable diagnosis device realize bidirectional data intercommunication through a 4G or 5G wireless communication module to cooperatively complete online monitoring and field diagnosis of the distribution network power equipment defects; The on-line monitoring system includes: The sensing unit is deployed in a key monitoring area of the distribution network power equipment, comprises an ultrahigh frequency sensor, an infrared thermal imaging sensor, a 128-channel ultrasonic sensor array and a visible light camera, and is used for synchronously acquiring multi-source data and transmitting the multi-source data to the data preprocessing module in real time; The data preprocessing module is connected with the sensing unit through an Ethernet signal, is internally provided with an FIR frequency domain filtering module, a delay-sum beam forming module and an image enhancement module, and respectively executes the filtering, airspace enhancement and image processing algorithm described in S2 to output pure ultrahigh frequency electric signals, enhanced ultrasonic signals, denoised and enhanced infrared thermal imaging and visible light images; The fusion judging module is in signal connection with the data preprocessing module through a PCIE3.0 interface, is internally provided with a three-level feature fusion algorithm and a CNN-LSTM deep learning model which are used for carrying out multi-dimensional feature fusion on pure signals and clear imaging data, generating an integrated diagnosis map and outputting defect information; The on-line monitoring host is connected with the fusion judging module through an industrial bus, is internally provided with a grading early warning module and is used for receiving defect information and an integrated diagnosis map, triggering red and yellow secondary early warning according to the severity, wherein the red light corresponds to the severe defect, the yellow light corresponds to the light and medium defect, and meanwhile, data forwarding and instruction issuing are realized; the database is connected with the online monitoring host computer through optical fiber data, the storage capacity is more than or equal to 1TB, and a relational database management system is adopted for structurally storing original multi-source data, signal processing results, integrated diagnostic patterns, defect information, associated time stamps and equipment identifiers and supporting concurrent inquiry and historical data backtracking; The portable diagnostic device includes: the miniature sensing module is integrated with the miniature infrared sensing module, the 32-channel portable ultrasonic sensing array and the visible light acquisition module and is used for multi-source data of the site repeated mining defect position; The embedded processing module is connected with the micro sensing module through an SPI interface signal, is provided with an ARMCortex-A9 processor and the attention mechanism feature fusion algorithm described in S3, and is used for processing the repeated mining data in real time and generating a field diagnosis result; The interaction and communication module is electrically connected with the embedded processing module and comprises a 7-inch touch display screen and a 4G or 5G communication unit, and is used for displaying field diagnosis results and defect maps, supporting data interaction with an online monitoring host and realizing comparison verification of the field diagnosis results and early warning results; And the power supply module is used for supplying power to each module of the portable diagnosis device and adopts 10000mAh rechargeable lithium batteries.

Description

Defect online monitoring method and device based on multi-feature fusion distribution network power equipment Technical Field The invention relates to the technical field of distribution network power equipment state monitoring, in particular to a defect online monitoring method and device based on multi-feature fusion distribution network power equipment. Background The defect monitoring of the current distribution network equipment mainly depends on technologies such as sensor acquisition, signal processing, feature analysis and the like, but complex conditions such as power frequency electromagnetic interference, environmental noise, equipment distribution dispersion and the like commonly exist on the distribution network site, and the traditional monitoring technology needs to adapt to scene requirements of multiple types of defects, complex interference and dispersion equipment so as to improve the accuracy of defect identification and operation and maintenance efficiency, so that certain defects exist in practical application: Firstly, the defect monitoring of the existing distribution network equipment usually adopts a single-mode sensor to collect data, for example, partial discharge is monitored only by an ultrahigh frequency sensor, or heating is detected only by infrared thermal imaging detection equipment; in part of the technology, a plurality of sensors are adopted, but only simple filtering processing is carried out on signals, the field environment of a distribution network is complex, 50Hz power frequency interference, environmental noise and the like exist around equipment such as a switch cabinet, a cable head and the like, the single ultrahigh frequency signal is easy to generate characteristic distortion due to interference, and single infrared thermal imaging is easy to misjudge temperature fluctuation caused by normal heat dissipation of the equipment as a defect; Secondly, the defect diagnosis of the existing distribution network equipment mostly adopts single characteristic research and judgment, or carries out simple decision-level splicing and fusion on the multi-source characteristics; for example, defects are judged only according to the temperature characteristics of infrared thermal imaging, or ultrasonic signals and the results of infrared images are displayed in parallel, but the distribution network equipment has various defect types, such as partial discharge is accompanied by ultra-high frequency and ultrasonic signal characteristics, insulation aging takes temperature abnormality as a core characteristic, and single characteristic judgment cannot cover the multidimensional characteristics of different defects; The method comprises the steps of carrying out on-line monitoring on defects of a current distribution network device by adopting a single on-line monitoring system or a single portable on-site detection device, wherein the on-line monitoring system can realize large-scale device coverage, but is influenced by deployment distance and on-site interference, misjudgment is easy to occur when partial defect signals are weak; Therefore, it is necessary to design a defect online monitoring method and device based on multi-feature fusion distribution network power equipment. Disclosure of Invention The invention aims to provide a defect online monitoring method and device based on multi-feature fusion distribution network power equipment, which are used for solving the problems that in the distribution network equipment defect monitoring process provided in the background art, the quality of basic data is insufficient, defect missed judgment misjudgment is easy to occur, defect type identification is inaccurate and position positioning deviation is large caused by single feature research judgment or simple feature splicing fusion, and the operation and maintenance treatment is low-efficiency caused by lack of data linkage between a single online monitoring system and an independent portable detection device, and a diagnosis model cannot continuously and iteratively adapt new defect features based on-site actual data and the long-term monitoring precision is reduced. In order to achieve the above purpose, the present invention provides the following technical solutions: In a first aspect, a defect online monitoring method based on multi-feature fusion distribution network power equipment is provided, including the following steps: The method comprises the steps of S1, multi-source data acquisition, namely, arranging a sensing unit containing multi-type sensing components in a key monitoring area of distribution network power equipment, synchronously acquiring multi-source data related to equipment defects, namely, ultra-high frequency electric signals, infrared thermal imaging, partial discharge ultrasonic signals and visible light images, selecting the ultra-high frequency, infrared, ultrasonic and visible light multi-type sensing components, wherein the multi-type sensin