CN113705763B - Real-time detection method and system for distribution network transformer based on nerve computation stick - Google Patents
Real-time detection method and system for distribution network transformer based on nerve computation stick Download PDFInfo
- Publication number
- CN113705763B CN113705763B CN202010429817.6A CN202010429817A CN113705763B CN 113705763 B CN113705763 B CN 113705763B CN 202010429817 A CN202010429817 A CN 202010429817A CN 113705763 B CN113705763 B CN 113705763B
- Authority
- CN
- China
- Prior art keywords
- transformer
- distribution network
- real
- positioning
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000009826 distribution Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000011897 real-time detection Methods 0.000 title claims abstract description 37
- 210000005036 nerve Anatomy 0.000 title claims abstract description 20
- 238000001514 detection method Methods 0.000 claims abstract description 77
- 230000008569 process Effects 0.000 claims abstract description 17
- 238000012795 verification Methods 0.000 claims abstract description 14
- 240000007651 Rubus glaucus Species 0.000 claims abstract description 13
- 235000011034 Rubus glaucus Nutrition 0.000 claims abstract description 13
- 235000009122 Rubus idaeus Nutrition 0.000 claims abstract description 13
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 10
- 230000009466 transformation Effects 0.000 claims abstract description 10
- 238000010207 Bayesian analysis Methods 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000011156 evaluation Methods 0.000 claims abstract description 6
- 238000004817 gas chromatography Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims description 56
- 238000012549 training Methods 0.000 claims description 26
- 230000001537 neural effect Effects 0.000 claims description 12
- 238000003062 neural network model Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 11
- 238000005265 energy consumption Methods 0.000 claims description 11
- 238000004891 communication Methods 0.000 claims description 8
- 238000009792 diffusion process Methods 0.000 claims description 7
- 230000004927 fusion Effects 0.000 claims description 7
- 238000002372 labelling Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 4
- 230000002238 attenuated effect Effects 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims 1
- 238000002203 pretreatment Methods 0.000 claims 1
- 239000004065 semiconductor Substances 0.000 claims 1
- 239000007789 gas Substances 0.000 description 54
- 230000006870 function Effects 0.000 description 10
- 238000012360 testing method Methods 0.000 description 7
- 238000004088 simulation Methods 0.000 description 6
- 238000001237 Raman spectrum Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 5
- 238000003860 storage Methods 0.000 description 5
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 4
- 239000001257 hydrogen Substances 0.000 description 3
- 229910052739 hydrogen Inorganic materials 0.000 description 3
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 description 2
- 238000004057 DFT-B3LYP calculation Methods 0.000 description 2
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 description 2
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 description 2
- 239000005977 Ethylene Substances 0.000 description 2
- HSFWRNGVRCDJHI-UHFFFAOYSA-N alpha-acetylene Natural products C#C HSFWRNGVRCDJHI-UHFFFAOYSA-N 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 229910002092 carbon dioxide Inorganic materials 0.000 description 2
- 239000001569 carbon dioxide Substances 0.000 description 2
- 229910002091 carbon monoxide Inorganic materials 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 125000002534 ethynyl group Chemical group [H]C#C* 0.000 description 2
- 150000002431 hydrogen Chemical class 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- 238000001069 Raman spectroscopy Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 150000001335 aliphatic alkanes Chemical class 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000004868 gas analysis Methods 0.000 description 1
- 238000009396 hybridization Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 229910001220 stainless steel Inorganic materials 0.000 description 1
- 239000010935 stainless steel Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/28—Measuring arrangements characterised by the use of optical techniques for measuring areas
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a real-time detection method and a real-time detection system for a distribution network transformer based on a nerve computation stick, wherein the method comprises the steps of collecting a distribution network transformer picture to be detected by utilizing a yolo3 target detection strategy, converting the distribution network transformer picture into image data, transmitting the image data into an image sensor, and carrying out affine transformation to output a corresponding actual position; collecting transformer oil characteristic gas in a transformer by utilizing a gas chromatography, and preprocessing by fusing a WSMOTE strategy; constructing a positioning model based on a distributed sequential evaluation gas leakage source position strategy to detect the actual position of the distribution network transformer and the preprocessed characteristic gas, and accelerating the positioning judgment process of the positioning model in real time by combining with an AI accelerator to output a positioning result; and importing the positioning result into a Bayesian analysis model for secondary verification and judgment to obtain the position information of the distribution network transformer, and displaying the result in real time by using a mobile terminal or a raspberry group display. The invention achieves the high-speed AI reasoning calculation capability and the low power consumption capability by adopting the nerve calculation rod of the AI accelerator.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a real-time detection method and system for a distribution network transformer based on a nerve computation stick.
Background
In order to realize intelligent management of main power equipment of a transformer substation, safe and reliable operation of a power system is guaranteed, and an electric company strengthens an operation monitoring method and means for the power equipment. At present, massive visible light and infrared images are generated based on the powerful popularization and use of thermal infrared imager live detection, unmanned aerial vehicle inspection, video on-line monitoring and the like, but all problems of main equipment are mainly treated by adopting a manual analysis mode. This approach results in a waste of a lot of human resources and also in errors due to lack of objectivity. Therefore, the intelligent identification of the power equipment target by utilizing the image identification technology is very necessary for the subsequent fault detection and diagnosis of the power equipment, and the mode can not only reduce the manual workload, but also avoid the misjudgment caused by the manual detection of the real object, thereby achieving the purpose of improving the detection accuracy.
At present, artificial intelligence is gradually applied in the power industry, for the transformer identification process of a distribution network, the traditional solution is to perform off-line identification in an image processing mode but has the problem of poor identification effect, and also has the problem of adopting a deep learning mode but arranging a model on a GPU server, so that the cost is high, the provided on-line service AI interface can be used under the condition of good network, but the on-line service AI interface cannot be used under the condition of no network or high network delay and the like in a specific use scene, so that the portable embedded device with high performance, low power consumption and small volume is urgently required to finish the off-line real-time identification task of the distribution network transformer.
Disclosure of Invention
This section is intended to outline some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section as well as in the description summary and in the title of the application, to avoid obscuring the purpose of this section, the description summary and the title of the invention, which should not be used to limit the scope of the invention.
The present invention has been made in view of the above-described problems occurring in the prior art.
Therefore, the invention provides a real-time detection method and a real-time detection system for a distribution network transformer based on a nerve computation stick, which can solve the problem that the real-time detection task of the distribution network transformer can not be completed in an off-line state.
In order to solve the technical problems, the invention provides the following technical scheme: collecting a distribution network transformer picture to be detected by utilizing a yolo3 target detection strategy, converting the distribution network transformer picture into image data, transmitting the image data into an image sensor, and carrying out affine transformation to output a corresponding actual position; collecting transformer oil characteristic gas in a transformer by utilizing a gas chromatography, and preprocessing by fusing a WSMOTE strategy; constructing a positioning model based on a distributed sequential evaluation gas leakage source position strategy to detect the actual position of the distribution network transformer and the preprocessed characteristic gas, and accelerating the positioning judgment process of the positioning model in real time by combining an AI accelerator to output a positioning result; and importing the positioning result into a Bayesian analysis model for secondary verification and judgment to obtain the position information of the distribution network transformer, and displaying the result in real time by using a mobile terminal or a raspberry group display.
As a preferable scheme of the real-time detection method of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: collecting the distribution network transformer picture and carrying out affine transformation to output the actual position comprises the steps of continuously detecting the cross-sectional area characteristics of the transformer pile head by using a camera, and carrying out real-time detection by combining with the yolo3 target detection strategy; when the cross-sectional area characteristic of the transformer pile head is detected, if the detection result is negative, the characteristic detection is continued until the cross-sectional area characteristic of the transformer pile head is detected, and then the detection of the transformer nameplate is carried out; when the transformer nameplate is detected, if the detection result is negative, continuing to detect the transformer nameplate until the transformer nameplate is detected, and then detecting the environment of the distribution network equipment; when detecting whether the distribution network transformer exists in the distribution network equipment environment, if the detection result is no, directly returning to the detection of the cross-sectional area characteristic of the transformer pile head and re-detecting, and if the detection result is yes, returning to the transformer pile head coordinate; calculating the coordinates of the center point of the bbox of the pile head of the transformer, and storing the coordinates according to the acquired picture sequence in a point2f format; and carrying out affine transformation on the coordinate values to convert the coordinate values into actual coordinates of the relative distribution network equipment, and calculating a track by utilizing the transformed actual coordinates to obtain the position of the transformer in the distribution network equipment.
As a preferable scheme of the real-time detection method of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: the calculation of the track requires training a neural network model in advance, and comprises the steps of acquiring the environment video data of the distribution network equipment by using the camera and converting the environment video data into picture data of a frame to be stored; generating png pictures with the same size for each frame of the picture data, and marking the section position of the transformer pile head, the position of a nameplate and the position of the transformer in each electrical device; 1000 pictures with 1920 x 1080 resolution and the labeling data of the same number are respectively obtained, and are input into the constructed neural network model; setting training parameters, and training the neural network model until the track data is accurately output.
As a preferable scheme of the real-time detection method of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: preprocessing the characteristic gas comprises marking and dividing the collected transformer oil characteristic gas to construct a sample data set; generating a classifier by utilizing an oversampling strategy based on a weight, and performing N rounds of iterative training on the classifier; the sample data sets with different weight distribution are used in each round of iterative training, and are updated, learned and modified until a new classifier and a new sample are generated, and the new sample is added into the original data set; and when the N rounds of iterative training are finished, integrating all the generated classifiers together and outputting a processing result.
As a preferable scheme of the real-time detection method of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: the positioning model is built by utilizing a turbulence diffusion theory to build a time-averaged gas diffusion model, and meanwhile, an observation model of the sensor node at the time t is built by combining the characteristic that the gas concentration is attenuated along with the increase of the propagation distance; constructing an information fusion objective function by using information gain parameters among the sensor nodes and network communication link energy consumption parameters, and calculating a distributed minimum mean square error prediction value to determine a gas leakage source position parameter; and constructing the positioning model based on the information fusion objective function and the yolo3 multi-objective detection strategy, and adjusting the size of the adjacent node set in real time according to the predicted mean square error.
As a preferable scheme of the real-time detection method of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: solving extremum of objective function between current node and adjacent node by the positioning model to select route node; outputting and positioning the position coordinates of the gas leakage source in real time according to the position parameters of the gas leakage source and the track data output by the neural network model; the AI accelerator accelerates the positioning judgment process of the positioning model in real time and outputs the positioning result.
As a preferable scheme of the real-time detection method of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: the secondary verification judgment comprises the steps of inputting the positioning result into the Bayesian analysis model; the frequency quantity, pulse quantity and gas volatilization quantity of the transformer are fused and the error probability is calculated, as follows,
P(B i |A i )＝B i |A i ，i＝1,2,……n
wherein B is i : the number of correctly identified characteristic parameters in the ith cooperative analysis factor, A i : the number of the characteristic parameters identified in the ith cooperative analysis factor; if the probability of the verification result is more than or equal to 0.5, the positioning result is correct; and if the probability of the verification result is smaller than 0.5, the positioning result is wrong.
As a preferable scheme of the real-time detection system of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: the detection module is used for collecting the distribution network transformer pictures and the characteristic gas, and comprises a camera and a sensor, wherein the camera is used for detecting and shooting the transformer pictures in real time, and the sensor is used for capturing the transformer oil characteristic gas in real time; the image sensing module is connected with the camera, and is used for acquiring the shot transformer picture, carrying out characteristic processing on the shot transformer picture, identifying, converting the shot transformer picture into image data and transmitting the image data into the core processing module; the core processing module is used for uniformly calculating and processing the image data and the characteristic gas acquired by the detection module, and comprises a data operation unit, a database and an input/output management unit, wherein the data operation unit is used for calculating acquired data information and giving an operation result, the database is used for providing a sample data set for the data operation unit and storing the operation result, and the input/output management unit is used for connecting each unit for information transmission interaction; the AI accelerator is connected with the core processing module and is used for accelerating the operation processing speed of the core processing module and simultaneously controlling and reducing the energy consumption of the core processing module.
As a preferable scheme of the real-time detection system of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: the network module is connected with the core processing module and is used for distributing a network for the core processing module; the power supply module is connected with the core processing module and the network module and is used for 5v power supply and charging of the core processing module and the network module; the expansion interface module is connected with the core processing module and is used for expanding an interface for the core processing module.
As a preferable scheme of the real-time detection system of the distribution network transformer based on the nerve computation stick, the invention comprises the following steps: the AI accelerator is an intel second-generation nerve computation stick and comprises an openvino platform for optimally compressing the positioning model; the network module is a wireless router, and is connected with the core processing module through a network cable to distribute a network and provide WiFi, so that equipment connected with the WiFi accesses the picture real-time detection service provided by the core processing module.
The invention has the beneficial effects that: according to the invention, the neural computing rod of the AI accelerator is adopted to achieve the capabilities of high-speed AI reasoning calculation capability and low power consumption, meanwhile, the coordinate judgment is respectively carried out on the acquisition detection of the transformer image and the gas, the two coordinate judgment results are uniformly integrated and positioned by utilizing the positioning model, finally, the accuracy of the positioning result is ensured by secondary verification analysis, and the network-free accurate real-time identification detection transformer is realized under the basic condition of long endurance time.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Wherein:
fig. 1 is a flow chart of a real-time detection method for a distribution network transformer based on a neural computing rod according to a first embodiment of the present invention;
fig. 2 is a schematic diagram of simulation test results of a conventional transformer fault infrared image recognition and detection method of a distribution network transformer real-time detection method based on a neural computing rod according to a first embodiment of the present invention;
fig. 3 is a schematic diagram of simulation test results of a conventional transformer characteristic gas raman spectrum detection method based on a neural computing rod distribution network transformer real-time detection method according to a first embodiment of the present invention;
fig. 4 is a schematic diagram of simulation test results of a method according to the present invention of a real-time detection method of a distribution network transformer based on a neural computing rod according to a first embodiment of the present invention;
Fig. 5 is a schematic diagram illustrating a module structure distribution of a real-time detection system for a distribution network transformer based on a neural computing rod according to a second embodiment of the present invention.
Detailed Description
So that the manner in which the above recited objects, features and advantages of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to the embodiments, some of which are illustrated in the appended drawings. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
While the embodiments of the present invention have been illustrated and described in detail in the drawings, the cross-sectional view of the device structure is not to scale in the general sense for ease of illustration, and the drawings are merely exemplary and should not be construed as limiting the scope of the invention. In addition, the three-dimensional dimensions of length, width and depth should be included in actual fabrication.
Also in the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper, lower, inner and outer", etc. are based on the orientation or positional relationship shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the apparatus or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
The terms "mounted, connected, and coupled" should be construed broadly in this disclosure unless otherwise specifically indicated and defined, such as: can be fixed connection, detachable connection or integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, or may be indirectly connected through an intermediate medium, or may be a communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
Example 1
Referring to fig. 1, a method for detecting a distribution network transformer in real time based on a neural computing rod is provided in a first embodiment of the present invention, including:
s1: and collecting the picture of the distribution network transformer to be detected by utilizing a yolo3 target detection strategy, converting the picture into image data, transmitting the image data into an image sensor, and carrying out affine transformation to output a corresponding actual position. The method for acquiring the distribution network transformer picture and carrying out affine transformation to output the actual position comprises the following steps:
continuously detecting the cross-sectional area characteristics of the transformer pile head by using a camera, and carrying out real-time detection by combining a yolo3 target detection strategy;
when the cross-sectional area characteristic of the transformer pile head is detected, if the detection result is negative, the characteristic detection is continued until the cross-sectional area characteristic of the transformer pile head is detected, and then the detection of the transformer nameplate is carried out;
when the transformer nameplate is detected, if the detection result is negative, continuing to detect the transformer nameplate until the transformer nameplate is detected, and detecting the environment of the distribution network equipment;
when detecting whether a distribution network transformer exists in the distribution network equipment environment, if the detection result is negative, directly returning to the action of detecting the cross-sectional area characteristic of the pile head of the transformer and re-detecting, and if the detection result is positive, returning to the pile head coordinate of the transformer;
Calculating coordinates of a central point of a bbox of the pile head of the transformer, and storing the coordinates according to the acquired picture sequence in a point2f format;
and carrying out affine transformation on the coordinate values to convert the coordinate values into actual coordinates of the relative distribution network equipment, and calculating a track by utilizing the transformed actual coordinates to obtain the position of the transformer in the distribution network equipment.
Further, the calculation of the trajectory requires training the neural network model in advance, including:
collecting the video data of the distribution network equipment environment by using a camera, converting the video data into picture data of a frame, and storing the picture data;
generating png pictures with the same size for each frame of picture data, and marking the section position of a pile head of the transformer, the position of a nameplate and the position of the transformer in each electrical device;
1000 pictures with 1920 x 1080 resolution and the labeling data of the same number are respectively obtained, and are input into a constructed neural network model;
setting training parameters, and training a neural network model until track data is accurately output.
Still further, feature extraction of the image is required, including:
the input picture is an original picture acquired by a camera, the resolution is 1920 x 1080, the number of channels is 3 (r, g and b), and the input picture is input into a neural network preprocessing unit;
Carrying out the resolution processing on the input picture, wherein the resolution of the adjusted picture is 512 x 512, and the number of channels is unchanged;
performing convolution operation on the picture, wherein the convolution kernel size 3*3, the convolution kernel number 64 and the convolution mode are same, the original image size after convolution is unchanged, and the channel number is changed to 64;
and carrying out Relu activation operation on the characteristic images of 512 x 64, and then carrying out maxpooling with the step length of 2, wherein the size of the characteristic image obtained after the pooling is halved, and the channel number is unchanged.
Further, the convolution operation is performed on the extracted characteristic image, including:
performing convolution operation on the feature map, wherein the convolution kernel size 3*3 and the convolution kernel number 512 are used for obtaining feature map 64×64×512, and the convolution operation is repeated for 3 times;
upsampling the 64×64×512 feature map, wherein the scaling factor is 2, and the sampling mode is bilinear, so as to obtain the feature map with the size of 128×128×512;
and stacking the feature graphs together, wherein the first two dimensions are unchanged during stacking, and the last channel number is added to obtain the feature graphs of 128 x 768.
S2: and (3) collecting transformer oil characteristic gas in the transformer by utilizing a gas chromatography, and preprocessing by fusing a WSMOTE strategy. The step is to say that the pretreatment characteristic gas comprises:
Marking and dividing the collected transformer oil characteristic gas (hydrogen, oxygen, methane, ethylene, ethane, acetylene, carbon monoxide and carbon dioxide content full analysis) to construct a sample data set;
generating a classifier by utilizing an oversampling strategy based on the weight, and performing N rounds of iterative training on the classifier;
sample data sets with different weight distribution are used in each round of iterative training, and are updated, learned and modified until a new classifier and a new sample are generated, and the new sample is added into the original data set;
when the N rounds of iterative training are finished, all the generated classifiers are integrated together and processing results are output.
Specifically, performing iterative training by using a WSMOTE strategy includes:
(1) Initializing parameters, wherein t=0;
(2) Synthesizing new samples by using the similarity measure and the oversampling rate R, constructing a new training set,
X d ＝X i +δ*(X a -X i )
where x= (X1, X2, …, xm) and y= (Y1, Y2, …, ym) represent two samples, m: number of sample features, X i : few samples, W i : weights of ith feature, X d : novel synthetic samples, X a : one neighbor sample of a few samples, δ: random parameters between values (0, 1);
(3) Learning the new training set to obtain a base classifier and calculating the weight of the base classifier by using the following formula,
Wherein f t : base classifier, a t : weight, e t : a feature weight;
(4) Calculating characteristic weight according to the classification error rate, updating the existing characteristic weight, and modifying the sample weight as follows
(5) If T is more than or equal to T, exiting the iterative training, calculating a final classifier by using a weighted combination, and outputting a result;
(6) If T is less than T, re-synthesizing a new sample, constructing a new training set, and continuing to perform iterative training until the requirement is met, and ending the iterative training.
S3: and constructing a positioning model based on a distributed sequential evaluation gas leakage source position strategy to detect the actual position of the distribution network transformer and the preprocessed characteristic gas, accelerating the positioning judgment process of the positioning model in real time by combining with an AI accelerator, and outputting a positioning result. It should also be noted that constructing the positioning model includes,
a time-averaged gas diffusion model is built by utilizing a turbulent diffusion theory, an observation model of the sensor node at the time t is built by combining the characteristic that the gas concentration is attenuated along with the increase of the propagation distance,
Z k (t)＝θ k (X s )+V k (t),k＝1,2，......n
wherein C is k (X s ) Coordinate X k Sensor node S of (a) k Concentration value at n: total number of sensor nodes in the network, q: gas release rate, k: turbulence diffusion coefficient, U: wind speed value, theta k (X s ): random quantity, V, containing gas leakage source position information k (t): the sensor measures noise;
an information fusion objective function is constructed by using information gain parameters (measured values of a current node and an adjacent node and position information of a gas leakage source) among sensor nodes and network communication link energy consumption parameters (bandwidth, delay, node information acquisition and processing energy consumption and information transmission among nodes), and the position parameters of the gas leakage source are determined by calculating the distributed minimum mean square error prediction, wherein the information fusion objective function is as follows:
R(S j ,S k )＝βR i (θ k ,Z K ,Y k )+(1-β)R c (S j ,S k )
wherein R is i (θ k ,Z K ,Y k ): node S being operated on j Selecting a next router node S k Gain of information generated at the time, R c (S j ,S k ): node S j And S is k The energy consumption of the communication link between them, beta E [0,1 ]]Is a coefficient for balancing the influence of two parameters on an objective function, when beta=1, the selection of the next routing node is based on the information gain and ignores the influence of the energy consumption of the communication link, when beta=0, the selection is based on the reduction of the energy consumption of the communication link and ignores the influence of the information gain, Y k : the current node transmits the prediction information containing noise to the next route node;
constructing a positioning model based on an information fusion objective function and a yolo3 multi-objective detection strategy, and solving extremum of the objective function between a current node and adjacent nodes thereof to select a routing node, wherein the positioning model comprises the following steps:
When k=1, define θ 1 (X s ) And observation noise V 1 Independent of each other, based on the observed value Z 1 Is of node s of (2) 1 The obtained predicted amount is
Mean square error of
When k > 1, node s k Predicted quantity θ of (2) k And corresponding mean square error M k Respectively is
The size of the adjacent node set is adjusted in real time by utilizing the predicted mean square error, and the error precision is eliminated;
outputting and positioning the position coordinates of the gas leakage source in real time according to the position parameters of the gas leakage source and the track data output by the neural network model;
furthermore, the AI accelerator accelerates the positioning judgment process of the positioning model in real time and outputs a positioning result.
In particular, routing node selection based on distributed sequential evaluation of gas leakage source location policies, including,
starting initial evaluation prediction operation, activating a node and receiving the prediction information of a front node;
updating the minimum mean square error of the predicted quantity and the predicted quantity, inputting the node observation value of the observation model, and judging whether the node observation value reaches the performance index threshold;
if yes, outputting final operation selection result data;
if not, selecting the next route node from the adjacent node set according to the operation result to transmit the prediction information, adjusting the selection radius of the adjacent node set according to the prediction variance value of the previous node and updating the candidate node set, transferring the current node into the sleep mode, and reactivating the node to receive the prediction information of the previous node until the operation result meets the performance index threshold requirement.
S4: and importing the positioning result into a Bayesian analysis model for secondary verification and judgment to obtain the position information of the distribution network transformer, and displaying the result in real time by using a mobile terminal or a raspberry group display. The step further includes the steps of:
inputting the positioning result into a Bayesian analysis model;
the frequency quantity, pulse quantity and gas volatilization quantity of the transformer are fused and the error probability is calculated, as follows,
P(B i |A i )＝B i |A i ，i＝1,2,……n
wherein B is i : the number of correctly identified characteristic parameters in the ith cooperative analysis factor, A i : the number of the characteristic parameters identified in the ith cooperative analysis factor;
if the probability of the verification result is more than or equal to 0.5, the positioning result is correct;
if the probability of the verification result is less than 0.5, the positioning result is wrong.
Preferably, the embodiment also needs to explain that, in the detection method for the distribution network transformer, image recognition is generally adopted or characteristic gas is utilized to perform equipment fault detection independently, for example, the traditional raman spectrum detection method for the characteristic gas of the transformer mainly uses the raman scattering effect to directly measure alkane gas and hydrogen in transformer insulating oil, and deduces the property and content of substances, and the problem solved by the spectrum analysis method mainly aims at trace detection of complex fault characteristic gas and gas cross interference, but in real life application, the phenomena that a sensor is easy to age and the minimum detection concentration analysis is limited exist, so that the accuracy and operability of gas detection are affected; the traditional transformer fault infrared image recognition method mainly utilizes a thermal infrared imaging technology to scan images through windows with different sizes, and finishes target classification recognition detection according to a manual shallow image feature training classification model, and solves the problems mainly aiming at the poor adaptability of manually extracting description features and subjective judgment on the running state of equipment, wherein in real life application, the shallow image features do not have the characteristics on a semantic level, the scanning of the whole image consumes a large amount of calculation time, and meanwhile, a large amount of repeated phenomena occur in the processing mode, namely a plurality of overlapped positioning windows exist in the same target part, and the situations of image deletion by mistake, lower detection precision and lower efficiency are easily caused; the method combines image recognition detection and characteristic gas detection, performs primary positioning and secondary verification positioning respectively by using a positioning model and a Bayesian analysis model, greatly improves the accuracy of recognition results, promotes operation efficiency and balance power consumption in real time by using an AI accelerator, improves working efficiency and performs real-time detection under the condition of realizing lower energy consumption operation and random network switching, and avoids the defects of the two methods when being independently applied.
In order to better verify and explain the technical effects adopted in the method, the traditional transformer fault infrared image identification detection method and the traditional transformer characteristic gas Raman spectrum detection method are adopted to respectively carry out comparison test with the method, and the test results are compared by a scientific demonstration means to verify the true effects of the method.
The traditional transformer fault infrared image recognition detection method has the advantages of lower detection precision, lower efficiency, lower accuracy and operability, and in order to verify that the method has higher detection precision, lower power consumption performance, higher operation speed and better instantaneity compared with the two traditional methods, the two traditional methods and the method are adopted to respectively measure and compare the transformer of a certain transformer substation in real time.
Test conditions: (1) The traditional transformer fault infrared image identification and detection method needs to carry out weighted gray scale treatment, then carries out edge detection by utilizing a Canny operator, obtains filtering to enable the subsequent identification analysis to achieve the detection purpose, respectively adopts infrared thermal images of VS2008, openCV and 650 x 590, and carries out simulation by utilizing a MATLB platform;
(2) The traditional transformer characteristic gas Raman spectrum detection method optimizes a characteristic gas molecular model (hydrogen, carbon monoxide, carbon dioxide, methane, ethane, ethylene and acetylene) based on a density functional B3LYP strategy, utilizes Gaussian software to analyze the gas molecular motion state and the energy level in transformer oil by adopting a B3LYP hybridization function formed by generalized gradient approximation of three parameters Becke exchange potential and LYP related potential, respectively adopts a 270mm x 195mm stainless steel oil gas permeation device, a tubular hollow permeation membrane temperature regulator, a motor and a plunger pump to be combined, and utilizes a MATLB platform to simulate;
(3) The method starts automatic test equipment (cameras, sensors, raspberry pie and AI accelerator) and simulates by utilizing a MATLB platform, and simulation data are obtained according to experimental results.
The simulation output results refer to fig. 2, 3 and 4 respectively, fig. 2 is the detection accuracy obtained by the traditional transformer fault infrared image recognition detection method under different iteration optimization times, as shown in fig. 2, the accuracy fluctuates with the increase of the detection iteration times, the instability trend is shown, the accuracy is high and low, and therefore, the detection accuracy is not good in practicability; FIG. 3 shows the detection accuracy obtained by the conventional method for detecting the characteristic gas Raman spectrum of the transformer under the conditions of different detection distances and gas concentrations, as shown in FIG. 3, the detection accuracy is lower and lower as the detection distance increases and the gas concentration decreases, so that the method is not a detection study of substantial significance; and FIG. 4 shows that the method of the present invention is a detection method for performing integrated positioning and secondary verification determination according to image recognition at different distance angles and gas analysis at different distance concentrations, as shown in FIG. 4, along with the change (i.e. increase or decrease) of the gas concentration, the detection precision value does not change in a fluctuation manner, and along with the increase of the detection workload, the detection time is not increased, so that the method of the present invention overcomes the problem of influence of the gas concentration on the accuracy, and has higher efficiency compared with two conventional methods.
Example 2
Referring to fig. 5, in a second embodiment of the present invention, unlike the first embodiment, there is provided a real-time detection system for a distribution network transformer based on a neural computing stick, including:
the detection module 100 is used for collecting pictures and characteristic gases of the distribution network transformer and comprises a camera 101 and a sensor 102, wherein the camera 101 is used for detecting and shooting the pictures of the transformer in real time, and the sensor 102 is used for capturing the characteristic gases of the transformer oil in real time;
the image sensing module 200 is connected with the camera 101, and is used for acquiring and performing characteristic processing on the shot transformer picture, identifying, converting the shot transformer picture into image data, and transmitting the image data into the core processing module 300;
the core processing module 300 is configured to uniformly calculate and process the image data and the feature gas collected by the detection module 100, and includes a data operation unit 301, a database 302, and an input/output management unit 303, where the data operation unit 301 is configured to calculate the obtained data information, give an operation result, the database 302 is configured to provide a sample data set for the data operation unit 301 and store the operation result, and the input/output management unit 303 is configured to connect each unit for information transmission interaction;
the AI accelerator 400 is connected to the core processing module 300, and is configured to accelerate the operation processing speed of the core processing module 300, and simultaneously control and reduce the energy consumption of the core processing module 300;
The network module 500 is connected to the core processing module 300, and is configured to allocate a network to the core processing module 300;
the power supply module 600 is connected to the core processing module 300 and the network module 500, and is used for 5v power supply and charging of the core processing module 300 and the network module 500;
the expansion interface module 700 is connected to the core processing module 300, and is used for expanding an interface for the core processing module 300.
Preferably, it should be noted that, the detection module 100 is internally provided with a tiny-yolo3 model trained for labeling data pictures through a distribution network transformer, so that the calculation amount is small, and the detection module can be operated at a mobile terminal or a device terminal and deployed on a raspberry group 3b+ to provide a picture real-time detection service; the core processing module 300 adopts an ARM processing chip raspberry group 3B+, wherein the raspberry group 3B+ is supplied with power for 5v voltage, and is loaded with linux to run a tensorflow framework; the AI accelerator 400 is an intel second generation neural computing rod, which comprises an openvino platform for optimally compressing a positioning model; the network module 500 is a wireless router, and is connected with the core processing module 300 through a network cable to distribute a network and provide WiFi, so that equipment connected with the WiFi accesses the picture real-time detection service provided by the core processing module 300; the power supply module 600 adopts a charging device with larger capacity to provide two 5V charging interfaces, so that the raspberry group is continuously powered for a long time in a low power consumption state.
Preferably, the embodiment also needs to explain that the process of detecting the distribution network transformer in real time by the system of the invention is as follows:
the camera 101 is used for collecting photos in real time and sending each frame of image to the raspberry group for preprocessing through the image sensor;
the sensor 102 collects the characteristic gas of the transformer oil in real time and transmits the characteristic gas to the raspberry group for pretreatment;
the data operation unit 301 reads and analyzes the preprocessed characteristic gas data and the preprocessed image data, and meanwhile, the AI accelerator 400 is connected with the raspberry pie through the USB port of the image sensing module 200, so that the data operation unit 301 is supported to accelerate the operation process;
the input/output management unit 303 outputs the location information of the distribution network transformer and returns to the terminal or the raspberry group display connected to the network module 500 to display the result.
It should be appreciated that embodiments of the invention may be implemented or realized by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer readable storage medium configured with a computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner, in accordance with the methods and drawings described in the specific embodiments. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.
Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes (or variations and/or combinations thereof) described herein may be performed under control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications), by hardware, or combinations thereof, collectively executing on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.
Further, the method may be implemented in any type of computing platform operatively connected to a suitable computing platform, including, but not limited to, a personal computer, mini-computer, mainframe, workstation, network or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and so forth. Aspects of the invention may be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and/or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, which when read by a computer, is operable to configure and operate the computer to perform the processes described herein. Further, the machine readable code, or portions thereof, may be transmitted over a wired or wireless network. When such media includes instructions or programs that, in conjunction with a microprocessor or other data processor, implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media. The invention also includes the computer itself when programmed according to the methods and techniques of the present invention. The computer program can be applied to the input data to perform the functions described herein, thereby converting the input data to generate output data that is stored to the non-volatile memory. The output information may also be applied to one or more output devices such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects produced on a display.
As used in this application, the terms "component," "module," "system," and the like are intended to refer to a computer-related entity, either hardware, firmware, a combination of hardware and software, or software in execution. For example, the components may be, but are not limited to: a process running on a processor, an object, an executable, a thread of execution, a program, and/or a computer. By way of example, both an application running on a computing device and the computing device can be a component. One or more components may reside within a process and/or thread of execution and a component may be localized on one computer and/or distributed between two or more computers. Furthermore, these components can execute from various computer readable media having various data structures thereon. The components may communicate by way of local and/or remote processes such as in accordance with a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and/or across a network such as the internet with other systems by way of the signal).
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention may be modified or substituted without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered in the scope of the claims of the present invention.
Claims (5)
1. A real-time detection method for a distribution network transformer based on a nerve computation stick is characterized by comprising the following steps of: comprising the steps of (a) a step of,
collecting a distribution network transformer picture to be detected by utilizing a yolo3 target detection strategy, converting the picture into image data, transmitting the image data into an image sensor, carrying out affine transformation, and outputting a corresponding actual position;
collecting transformer oil characteristic gas in a transformer by utilizing a gas chromatography, and preprocessing by fusing a WSMOTE strategy;
constructing a positioning model based on a distributed sequential evaluation gas leakage source position strategy to detect the actual position of the distribution network transformer and the preprocessed characteristic gas, and accelerating the positioning judgment process of the positioning model in real time by combining an AI accelerator to output a positioning result;
the positioning result is imported into a Bayesian analysis model for secondary verification and judgment, so that the position information of the distribution network transformer is obtained, and the result is displayed in real time by utilizing a mobile terminal or a raspberry group display;
Collecting the distribution network transformer picture and carrying out affine transformation to output the actual position comprises,
continuously detecting the cross-sectional area characteristics of the transformer pile head by using a camera, and carrying out real-time detection by combining the yolo3 target detection strategy;
when the cross-sectional area characteristic of the transformer pile head is detected, if the detection result is negative, the characteristic detection is continued until the cross-sectional area characteristic of the transformer pile head is detected, and then the detection of the transformer nameplate is carried out;
when the transformer nameplate is detected, if the detection result is negative, continuing to detect the transformer nameplate until the transformer nameplate is detected, and then detecting the environment of the distribution network equipment;
when detecting whether the distribution network transformer exists in the distribution network equipment environment, if the detection result is no, directly returning to the detection of the cross-sectional area characteristic of the transformer pile head and re-detecting, and if the detection result is yes, returning to the transformer pile head coordinate;
calculating the coordinates of the center point of the bbox of the pile head of the transformer, and storing the coordinates according to the acquired picture sequence in a point2f format;
carrying out affine transformation on the coordinate values to convert the coordinate values into actual coordinates of relative distribution network equipment, and calculating a track by utilizing the transformed actual coordinates to obtain the position of the transformer in the distribution network equipment;
The calculation of the trajectory requires training a neural network model in advance, including,
acquiring the environment video data of the distribution network equipment by using the camera, and converting the environment video data into picture data of a frame to be stored;
generating png pictures with the same size for each frame of the picture data, and marking the section position of the transformer pile head, the position of a nameplate and the position of the transformer in each electrical device;
1000 pictures with 1920 x 1080 resolution and the labeling data of the same number are respectively obtained, and are input into the constructed neural network model;
setting training parameters, and training the neural network model until the track data is accurately output;
the pre-treatment of the feature gas includes,
labeling and dividing the collected transformer oil characteristic gas to construct a sample data set;
generating a classifier by utilizing an oversampling strategy based on a weight, and performing N rounds of iterative training on the classifier;
the sample data sets with different weight distribution are used in each round of iterative training, and are updated, learned and modified until a new classifier and a new sample are generated, and the new sample is added into the original data set;
when the N rounds of iterative training are finished, integrating all the generated classifiers together and outputting a processing result;
The construction of the positioning model includes the steps of,
constructing a time-averaged gas diffusion model by using a turbulence diffusion theory, and simultaneously constructing an observation model of the sensor node at the time t by combining the characteristic that the gas concentration is attenuated along with the increase of the propagation distance;
constructing an information fusion objective function by using information gain parameters among the sensor nodes and network communication link energy consumption parameters, and calculating a distributed minimum mean square error prediction value to determine a gas leakage source position parameter;
constructing the positioning model based on the information fusion objective function and the yolo3 multi-objective detection strategy, and adjusting the size of the adjacent node set in real time according to the predicted mean square error;
the positioning model solves extremum of objective function between current node and adjacent node to select route node;
outputting and positioning the position coordinates of the gas leakage source in real time according to the position parameters of the gas leakage source and the track data output by the neural network model;
the AI accelerator accelerates the positioning judgment process of the positioning model in real time and outputs the positioning result.
2. The neural computing rod-based real-time detection method for distribution network transformers, as set forth in claim 1, wherein: the secondary authentication decision includes,
Inputting the positioning result into the Bayesian analysis model;
the frequency quantity, pulse quantity and gas volatilization quantity of the transformer are fused and the error probability is calculated, as follows,
P(B i |A i )＝B i |A i ，i＝1,2,……n
wherein B is i : the number of correctly identified characteristic parameters in the ith cooperative analysis factor, A i : the number of the characteristic parameters identified in the ith cooperative analysis factor;
if the probability of the verification result is more than or equal to 0.5, the positioning result is correct;
and if the probability of the verification result is smaller than 0.5, the positioning result is wrong.
3. A real-time detection system of distribution network transformers based on a nerve computation stick, which is based on the real-time detection method of the distribution network transformers based on the nerve computation stick according to any one of claims 1-2, and is characterized in that: comprising the steps of (a) a step of,
the detection module (100) is used for acquiring the distribution network transformer picture and the characteristic gas and comprises a camera (101) and a sensor (102), wherein the camera (101) is used for detecting and shooting the transformer picture in real time, and the sensor (102) is used for capturing the transformer oil characteristic gas in real time;
the image sensing module (200) is connected with the camera (101) and is used for acquiring the shot transformer picture, carrying out characteristic processing on the shot transformer picture, identifying and converting the shot transformer picture into image data, and transmitting the image data into the core processing module (300);
The core processing module (300) is used for uniformly calculating and processing the image data and the characteristic gas acquired by the detection module (100), and comprises a data operation unit (301), a database (302) and an input/output management unit (303), wherein the data operation unit (301) is used for calculating acquired data information and giving an operation result, the database (302) is used for providing a sample data set for the data operation unit (301) and storing the operation result, and the input/output management unit (303) is used for connecting each unit for information transmission interaction;
the AI accelerator (400) is connected to the core processing module (300) and is used for accelerating the operation processing speed of the core processing module (300) and simultaneously controlling and reducing the energy consumption of the core processing module (300).
4. A neural computing wand based real-time detection system for distribution network transformers as recited in claim 3, wherein: also included is a method of manufacturing a semiconductor device,
the network module (500) is connected with the core processing module (300) and is used for distributing a network for the core processing module (300);
a power supply module (600) connected to the core processing module (300) and the network module (500) for 5v powering and charging the core processing module (300) and the network module (500);
An expansion interface module (700) is connected to the core processing module (300) for expanding an interface for the core processing module (300).
5. The neural computing rod-based distribution network transformer real-time detection system according to claim 4, wherein: comprising the steps of (a) a step of,
the AI accelerator (400) is an intel second-generation nerve computation stick and comprises an openvino platform for optimally compressing a positioning model;
the network module (500) is a wireless router, and is connected with the core processing module (300) through a network cable to distribute a network and provide WiFi, so that equipment connected with the WiFi accesses the picture real-time detection service provided by the core processing module (300).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010429817.6A CN113705763B (en) | 2020-05-20 | 2020-05-20 | Real-time detection method and system for distribution network transformer based on nerve computation stick |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010429817.6A CN113705763B (en) | 2020-05-20 | 2020-05-20 | Real-time detection method and system for distribution network transformer based on nerve computation stick |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113705763A CN113705763A (en) | 2021-11-26 |
| CN113705763B true CN113705763B (en) | 2024-02-20 |
Family
ID=78645500
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010429817.6A Active CN113705763B (en) | 2020-05-20 | 2020-05-20 | Real-time detection method and system for distribution network transformer based on nerve computation stick |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705763B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114462555B (en) * | 2022-04-13 | 2022-08-16 | 国网江西省电力有限公司电力科学研究院 | Multi-scale feature fusion power distribution network equipment identification method based on raspberry group |
| CN117829822A (en) * | 2024-03-04 | 2024-04-05 | 合肥工业大学 | Power transformer fault early warning method and system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201402257Y (en) * | 2009-02-20 | 2010-02-10 | 国网电力科学研究院 | Gas content on-line monitoring system in transformer oil |
| CN101701940A (en) * | 2009-10-26 | 2010-05-05 | 南京航空航天大学 | On-line transformer fault diagnosis method based on SVM and DGA |
| CN106052962A (en) * | 2016-05-13 | 2016-10-26 | 重庆大学 | Gas source positioning technology in wireless sensor network without wind speed and wind direction measuring |
| CN107063349A (en) * | 2017-04-17 | 2017-08-18 | 云南电网有限责任公司电力科学研究院 | A kind of method and device of Fault Diagnosis Method of Power Transformer |
| CN107332898A (en) * | 2017-06-27 | 2017-11-07 | 中国矿业大学(北京) | A kind of mine gas source of release localization method of Multiple Source Sensor fusion |
| CN109101906A (en) * | 2018-07-27 | 2018-12-28 | 中国南方电网有限责任公司超高压输电公司贵阳局 | A kind of converting station electric power equipment infrared image exception real-time detection method and device |
| CN109711437A (en) * | 2018-12-06 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of transformer part recognition methods based on YOLO network model |
| CN109856299A (en) * | 2018-11-26 | 2019-06-07 | 国家电网有限公司 | A kind of transformer online monitoring differentiation threshold value dynamic setting method, system |
| CN110705395A (en) * | 2019-09-18 | 2020-01-17 | 北京邮电大学 | Detection system based on raspberry group |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6498552B2 (en) * | 2015-07-15 | 2019-04-10 | 株式会社日立製作所 | Oil leak detection system |
| CN108536648B (en) * | 2018-03-30 | 2021-07-06 | 武汉大学 | Partial discharge nonlinear model conversion solving and optimizing method based on multiple ultrasonic sensors |
-
2020
- 2020-05-20 CN CN202010429817.6A patent/CN113705763B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201402257Y (en) * | 2009-02-20 | 2010-02-10 | 国网电力科学研究院 | Gas content on-line monitoring system in transformer oil |
| CN101701940A (en) * | 2009-10-26 | 2010-05-05 | 南京航空航天大学 | On-line transformer fault diagnosis method based on SVM and DGA |
| CN106052962A (en) * | 2016-05-13 | 2016-10-26 | 重庆大学 | Gas source positioning technology in wireless sensor network without wind speed and wind direction measuring |
| CN107063349A (en) * | 2017-04-17 | 2017-08-18 | 云南电网有限责任公司电力科学研究院 | A kind of method and device of Fault Diagnosis Method of Power Transformer |
| CN107332898A (en) * | 2017-06-27 | 2017-11-07 | 中国矿业大学(北京) | A kind of mine gas source of release localization method of Multiple Source Sensor fusion |
| CN109101906A (en) * | 2018-07-27 | 2018-12-28 | 中国南方电网有限责任公司超高压输电公司贵阳局 | A kind of converting station electric power equipment infrared image exception real-time detection method and device |
| CN109856299A (en) * | 2018-11-26 | 2019-06-07 | 国家电网有限公司 | A kind of transformer online monitoring differentiation threshold value dynamic setting method, system |
| CN109711437A (en) * | 2018-12-06 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of transformer part recognition methods based on YOLO network model |
| CN110705395A (en) * | 2019-09-18 | 2020-01-17 | 北京邮电大学 | Detection system based on raspberry group |
Non-Patent Citations (4)
| Title |
|---|
| Weighted-SMOTE: A modification to SMOTE for event classification in sodium cooled fast reactors;Manas Ranjan Prusty等;《Progress in Nuclear Energy》;355-364 * |
| 基于不平衡数据分类的中药肝毒性检测;郭玉萱;《计算应用与软件》;第35卷(第8期);226-230 * |
| 基于分布传感技术的变压器油色谱在线监测技术应用研究;王伦宾;《中国优秀硕士学位论文全文数据库（电子期刊）》;C042-47 * |
| 基于油气量和超高频局放信号的变压器故障诊断研究;洪浪;《中国优秀硕士学位论文全文数据库（电子期刊）》;C042-271 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113705763A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103324937B (en) | The method and apparatus of label target | |
| CN111325347B (en) | Automatic danger early warning description generation method based on interpretable visual reasoning model | |
| CN109447156B (en) | Method and apparatus for generating a model | |
| CN109145828B (en) | Method and apparatus for generating video category detection model | |
| CN104573706A (en) | Object identification method and system thereof | |
| CN109242033A (en) | Wafer defect method for classifying modes and device, storage medium, electronic equipment | |
| CN113705763B (en) | Real-time detection method and system for distribution network transformer based on nerve computation stick | |
| CN110009614A (en) | Method and apparatus for output information | |
| CN109544548A (en) | Defect inspection method, device, server, equipment and the storage medium of cutlery box | |
| CN111598942A (en) | Method and system for automatically positioning electric power facility instrument | |
| CN114332578A (en) | Image anomaly detection model training method, image anomaly detection method and device | |
| CN113435282B (en) | Unmanned aerial vehicle image ear recognition method based on deep learning | |
| US20220366244A1 (en) | Modeling Human Behavior in Work Environment Using Neural Networks | |
| Wang | An improved machine learning and artificial intelligence algorithm for classroom management of English distance education | |
| CN113252701A (en) | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method | |
| CN111242922A (en) | Protein image classification method, device, equipment and medium | |
| CN107992873A (en) | Object detection method and device, storage medium, electronic equipment | |
| CN110909889A (en) | Training set generation and model training method and device based on feature distribution | |
| KR20210142864A (en) | Apparatus and method for recognizing number of measuring intrument | |
| Kopanas et al. | Improving nerf quality by progressive camera placement for unrestricted navigation in complex environments | |
| CN111161238A (en) | Image quality evaluation method and device, electronic device, and storage medium | |
| CN116681962A (en) | Power equipment thermal image detection method and system based on improved YOLOv5 | |
| Pu et al. | AutoRepo: A general framework for multi-modal LLM-based automated construction reporting | |
| CN110738208A (en) | efficient scale-normalized target detection training method | |
| CN115937662A (en) | Intelligent household system control method and device, server and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |