CN114924207A - Method for detecting intermittent earth fault based on machine vision - Google Patents
Method for detecting intermittent earth fault based on machine vision Download PDFInfo
- Publication number
- CN114924207A CN114924207A CN202210416361.9A CN202210416361A CN114924207A CN 114924207 A CN114924207 A CN 114924207A CN 202210416361 A CN202210416361 A CN 202210416361A CN 114924207 A CN114924207 A CN 114924207A
- Authority
- CN
- China
- Prior art keywords
- temperature
- intermittent
- fault
- data
- oil
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000001514 detection method Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 238000004364 calculation method Methods 0.000 claims abstract description 18
- 238000012544 monitoring process Methods 0.000 claims abstract description 18
- 238000004458 analytical method Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 16
- 230000000007 visual effect Effects 0.000 claims abstract description 12
- 230000036541 health Effects 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 26
- 230000002159 abnormal effect Effects 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000013178 mathematical model Methods 0.000 claims description 4
- 210000002364 input neuron Anatomy 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 2
- 230000007935 neutral effect Effects 0.000 description 13
- 238000013473 artificial intelligence Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 230000009471 action Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000010438 heat treatment Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000003745 diagnosis Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000033228 biological regulation Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 239000004020 conductor Substances 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
- G01R31/52—Testing for short-circuits, leakage current or ground faults
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Emergency Protection Circuit Devices (AREA)
Abstract
A method for detecting intermittent earth fault based on machine vision belongs to the field of fault detection. The method comprises the steps of collecting surface temperature data of a grounding resistor through a thermal infrared imager; detecting the oil level of a transformer oil conservator and the oil level of a respirator oil cup; detecting the indoor and outdoor temperature of the transformer; transmitting detection data of the thermal infrared imager, the oil conservator oil level camera, the oil cup oil level camera, the transformer indoor temperature sensor and the transformer outdoor temperature sensor to an edge server; constructing an intermittent ground fault detection data set based on temperature data; performing model training based on BP neural network data; based on the data driving model, carrying out early warning and alarming on the intermittent earth fault; and through image processing and calculation of the edge server, a decision basis is provided for the health state of the main transformer in a data driving mode. The system adopts non-contact visual monitoring and analysis as backup protection of the existing ground fault outgoing line protection, and improves the running reliability of the system.
Description
Technical Field
The invention belongs to the field of fault detection, and particularly relates to a method for detecting an intermittent earth fault by using machine vision.
Background
Common neutral point grounding methods in the power system include direct grounding, non-grounding, grounding through an arc suppression coil, grounding through a resistor at a neutral point, and the like.
The neutral point is grounded through a small resistor under the conditions of improvement of the power supply capacity of the modern urban network, enhancement of the continuous working capacity of the breaker and the like. When a system line is grounded, the grounding line in a power distribution system is quickly cut off, so that the repair time of a fault line is favorably shortened, and a neutral point is widely applied in a grounding mode of a small resistor in an urban power distribution network mainly based on cable outgoing lines.
By intermittent arc is meant the multiple repetition of the extinguishing of the earth arc with consequent reignition. Resistive faults and line faults are the causes of intermittent ground faults. Such as partial resistor damage, loses thermal stability after the resistor heats up during long-term operation, resulting in an increase in the resistance value. Or when the overhead insulated conductor is broken and falls to the ground, a small resistance grounding system may generate small grounding current, and when the system has single-phase grounding fault, the intermittent grounding phenomenon is caused by the lap joint of surrounding objects (such as branches and the like) and complex external actions such as wind power, air ionization and the like.
The intermittent single-phase earth fault is represented in the form of intermittent occurrence of zero-sequence abnormal current flowing through a neutral point resistor, and because the single arcing duration is always smaller than the protection action time limit and the zero-sequence abnormal current is smaller than the current quick-break protection condition, the intermittent fault current existing for a long time repeatedly occurs for multiple times until the neutral point resistor is burnt out. The distribution network can even become an ungrounded system after the resistor is burnt out, and the arc overvoltage generated by the resistance can cause great harm to the whole power grid.
In order to solve the problem of intermittent earth faults, various detection schemes are provided based on the characteristics of the intermittent earth faults; for example, in the literature "intermittent arc overvoltage analysis of a small-resistance grounding system" (fairy guest, uke, li shu. "power system and its automated journal", 2012, 24(3): 116-; in documents of "intermittent earth fault analysis of a small-resistance earth system of a 10kV distribution network in shanghai" (yu zhi, shanghai: shanghai university of traffic, 2011) and "intermittent earth fault protection of a neutral point small-resistance earth system based on a thermal stability principle" (schroemering, anwei, tangyi et al. "automation of electrical systems" 2021, 45(18)), protection schemes by arc light overvoltage analysis, resistance value change of small resistance, zero-sequence current analysis, and the like are proposed. The method mostly takes the failure times in a fixed time limit as a judgment basis, and the single arcing time, the failure interval time and the like are less considered; in the literature, "application and prospect of artificial intelligence in high-resistance ground fault detection of a power distribution network" (whitely, lipeng, Yuanzhiyong, etc.. "southern power grid technology", 2019, 13(2)), the characteristics that the neutral point grounding resistor mainly generates heat during fault arcing and radiates heat after fault arcing are provided, and a protection method is provided based on a thermal stability principle, so that the anti-interference performance and the stability are further improved. However, in the above solutions, intermittent faults are identified through electric quantity indexes, and since the judgment based on characteristics of the electric quantity indexes (generally, the size of the ground current or the leakage current is monitored), there is a certain limitation on the judgment and monitoring manner in practical application, and the fault identification effect and the response speed are not ideal.
In recent years, artificial intelligence is rapidly developed and applied to various industries, and a new solution is brought to the problem which is not easily solved by the traditional scheme.
In the aspect of ground fault detection, the literature "single-phase ground fault diagnosis of a power distribution network based on an optimized deep neural network" (schottelin, gunn. "electrical automation", 2021, 43(1)) and "temperature field simulation and experiment of a neutral point ground resistor" (rolong fu, zhangjun., hei. "power system and its automation bulletin, 2010, 22(4)) proposes to use an artificial intelligence algorithm to identify the characteristics of intermittent ground faults, so as to achieve the effect of rapidly processing the faults. However, the above technical solutions also have the disadvantages of high requirements for single arcing current amplitude and duration, etc., the actual fault condition is complex, and the reliability of the fault diagnosis algorithm is to be further improved.
Disclosure of Invention
The invention aims to solve the technical problem of providing a method for detecting intermittent earth faults based on machine vision. In order to enable the system to have identification accuracy and rapid processing capability of intermittent ground faults, the technical scheme of the invention extracts relevant characteristics such as ground resistance temperature, indoor temperature, oil conservator oil level and the like in a power distribution network system by adopting non-contact visual monitoring and analysis on main transformer key electrical equipment (including a respirator and an oil conservator) and main transformer ground resistance and the like, arranges and obtains accumulated fault temperature in fixed time based on the thermal stability principle of the ground resistance, and combines an artificial intelligence algorithm to further provide an intermittent ground fault protection method which is used as backup protection of the existing ground fault outgoing line protection, thereby improving the operation reliability of the system.
The technical scheme of the invention is as follows: the method for detecting the intermittent ground fault based on the machine vision is characterized by comprising the following steps of:
1) arranging an infrared thermal imager in a visible range of the grounding resistor, acquiring surface temperature data of the grounding resistor through the infrared thermal imager, and carrying out non-contact visual detection and monitoring instantaneous temperature rise on a main transformer grounding resistor device;
2) arranging an oil conservator oil level camera and an oil cup oil level camera for detecting the oil level of the transformer oil conservator and the oil level of the respirator oil cup;
3) a temperature transmitter or a temperature sensor is arranged for detecting the indoor and outdoor temperatures of the transformer;
4) comparing the ground resistance temperature acquired each time with the indoor environment temperature to acquire a temperature difference for comparing the given temperature difference;
5) the infrared thermal imager, the oil conservator oil level camera, the oil cup oil level camera, the transformer indoor temperature sensor and the transformer outdoor temperature sensor are respectively and correspondingly connected with an edge server arranged in the transformer room;
6) constructing an intermittent earth fault detection data set based on temperature data, and designing a training test data set meeting machine learning requirements;
7) performing model training based on BP neural network data;
8) the method comprises the steps that the surface temperature of a grounding resistor, the oil level of an oil conservator, the oil level of a respirator oil cup, the indoor temperature and the outdoor temperature are detected in real time, and an intermittent grounding fault is early warned based on a data driving model;
9) and through image processing and calculation of the edge server, a decision basis is provided for the health state of the main transformer in a data-driven mode.
Specifically, the thermal infrared imager collects the surface temperature data of the ground resistor at a sampling frequency of 200ms so as to monitor a blind area where the fault current continuously reciprocates and does not reach a trigger setting value threshold, and acquire information of each area including the highest temperature, the lowest temperature and the average temperature.
Further, the threshold of the trigger setting value is 5.5 s.
Specifically, the surface temperature T and the indoor temperature T of the ground resistor room Is Δ T ═ T-T room Setting the temperature difference T set Judging the real-time system state according to the temperature difference value when the temperature is 3 ℃, and when the temperature is delta T<T set The real-time state of the system is 1 and normal; otherwise, it is 0 and is not normal.
Further, the intermittent ground fault detection method based on the machine vision is characterized in that in the obtained intermittent ground fault detection data set, the data set within 5.5S is taken as a sample, a sample label is obtained, and the fault temperature S is calculated T And a low setting Threshold min And a high setting Threshold max The label type of the sample can be judged.
Furthermore, the low setting Threshold is Threshold min 300, said high setting thresholdThreshold max Is 800.
Specifically, the model training includes:
1) data normalization:
the calculation method is as follows:
in the formula, x i As input variables, t i Normalized value, x max And x min The maximum value and the minimum value of the input variable x are respectively;
2) number of layers and number of neural units:
reducing the x sample two-dimensional vector to one dimension according to the row direction, wherein the number of input neurons is 140, and the output neurons is 1; the number of the hidden layers is set to be 3, the number of neurons in each layer is 128, 10% of neurons in each layer are randomly removed for preventing overfitting, and L2 regularization is adopted;
3) learning rate:
an exponential decay learning rate is adopted, the initial learning rate is set to be 0.2, the decay rate of the learning rate is set to be 0.09, the iteration step number of the decay of the learning rate once is set to be 128 steps, and the calculation method of the decay learning rate is as follows:
wherein lr is learning rate after attenuation, lr base For initial learning rate, decay rate For learning the attenuation rate, step is the number of steps of one iteration per attenuation;
4) activation function and cost function:
and selecting a linear rectification function relu as an activation function of the hidden layer, taking a normalized multi-classification function softmax as an activation function of the output layer, and taking a cross entropy loss function as a cost function.
According to the method for detecting the intermittent ground fault based on the machine vision, the surface temperature, the indoor and outdoor temperature, the oil conservator oil level and other related information of the ground resistor are obtained through real-time visual analysis and monitoring aiming at the thermal stability of the ground resistor when the ground resistor breaks down, and the edge server is combined with a BP neural network algorithm to predict the equipment fault.
According to the method for detecting the intermittent ground fault based on the machine vision, the thermal infrared imager is used for carrying out non-contact vision detection and monitoring instantaneous temperature rise on a ground resistor device connected with a main transformer at a sampling frequency of 200ms, collecting the surface temperature data of the ground resistor, adopting a visible light edge to calculate a camera to collect the oil level data of an oil conservator, and combining indoor and outdoor temperature numbers based on a data-driven mathematical model to early warn and alarm the intermittent ground fault, so that a decision basis is provided for guaranteeing the normal operation of the main transformer.
The intermittent ground fault detection method based on machine vision is used as backup protection, faults are quickly judged and processed, and the safety and reliability of the system are jointly improved by matching with the existing monitoring protection system.
Compared with the prior art, the invention has the advantages that:
1. according to the technical scheme, non-contact visual detection and instantaneous temperature rise monitoring are carried out on a ground resistance device connected with a main transformer through a thermal infrared imager at a sampling frequency of 200ms, the surface temperature data of the ground resistance are collected, a visible light edge calculation camera is adopted to collect the oil level data of an oil conservator, and an intermittent ground fault is early warned and alarmed on the basis of a data-driven mathematical model by combining indoor and outdoor temperature numbers, so that a decision basis is provided for guaranteeing the normal operation of the main transformer;
2. according to the technical scheme, the thermal stability of the ground resistor in the fault is analyzed and monitored in real time, the related information such as the surface temperature, the indoor and outdoor temperature, the oil level of an oil conservator and the like of the ground resistor is obtained, the edge server is utilized to predict the equipment fault by combining a BP neural network algorithm, the anti-interference performance is high, and the fault is judged and processed quickly by adopting an intelligent algorithm through a large amount of data;
3. according to the technical scheme, through image processing and calculation of the edge server and in a data driving mode, a decision basis is provided for the health state of a main transformer, faults are quickly judged and processed to serve as backup protection, and the safety and the reliability of the system are jointly improved by matching with an existing monitoring protection system.
Drawings
FIG. 1 is a schematic diagram of intermittent ground fault resistance temperature variation;
FIG. 2a, FIG. 2b, FIG. 2c and FIG. 2d are schematic views of the fault area in different states of the system;
FIG. 3 is a schematic diagram of the intermittent ground protection flow of the present invention;
FIG. 4 is a schematic diagram of the overall analysis process for system failure according to the present invention;
FIG. 5 is a schematic diagram of the hardware configuration of the system of the present invention;
FIG. 6 is a schematic diagram of a BP neural network topology;
FIG. 7a is a graph showing the ground resistance temperature after adding simulation data;
FIG. 7b is a sample fraction of different system states after adding simulation data;
FIG. 8a is a graph illustrating loss values of the training set and the prediction set according to the present invention;
FIG. 8b is a graph illustrating the accuracy of the training set and the prediction set according to the present invention.
In the figure, 1 is a PoE switch, 2 is a 4G router, 3 is an edge server, 4 is an indoor temperature and humidity transmitter, 4a is an outdoor temperature and humidity transmitter, 5 is a thermal infrared imager, 6 is an oil conservator oil level detection camera, 7 is an oil cup oil level detection camera, 8 is a grounding resistor, 9 is a transformer chamber, 10 is a transformer oil conservator, and 11 is a respirator oil cup.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
A main transformer (called main transformer for short) is one of the most important devices in a transformer substation. The fault of the auxiliary safety protection devices of the key electrical equipment, such as the respirator, the oil conservator, the main transformer grounding resistor and the like, is discovered in time, and plays an important role in guaranteeing the normal operation of the main transformer.
The volume of the transformer oil changes due to the temperature change of the transformer oil in the conservator under the influence of factors such as air temperature, load and the like, so that the oil level in the conservator changes. The oil level is displayed through an oil level gauge pipe or an oil level gauge beside the oil conservator.
In the accident that the neutral point grounding resistor burns out due to the intermittent grounding fault, the grounding current is small when the line has the intermittent grounding fault, and the intermittent characteristic is achieved. In contrast, the conventional relay protection device does not provide a targeted measure, so that outgoing line zero current protection does not act, a fault line cannot be cut off quickly, and the power distribution network is in a fault state for a long time. Meanwhile, the small neutral point resistor is in a working state for a long time, the phenomenon of serious heating can occur, even the small neutral point resistor is burnt out, analysis shows that part of the resistor in field working is damaged due to various reasons, but the resistor fault cannot be detected under the condition of not working at ordinary times, once the heating phenomenon of the resistor in long-term working occurs, the resistance value of the resistor losing thermal stability can rise, and the advantage that the small resistor inhibits intermittent grounding voltage at the moment can be promoted. The distribution network can even become an ungrounded system after the resistor is burnt out, and a plurality of transformed small resistor systems install the main transformer grounding resistor in the transformer chamber, so that the resistor is burnt out, the transformer is probably burnt out together, and the whole power grid is greatly damaged.
At present, the monitoring of power equipment is mainly completed by depending on manual inspection, fixedly mounting a high-definition camera and an inspection robot. And the site partially deploying the artificial intelligence algorithm deploys the computing power algorithm to the private cloud server. Foreign manufacturers such as the intex dada and the saix provide software and hardware tools for deep learning. Domestic enterprises such as the sea XX view to provide cloud-edge-end full-chain service, the HuaX cloud algorithm mall provides downloadable algorithm service to a high-computing camera and the hundred X provides flying X system tool service.
With the improvement of the computing power of the artificial intelligence chip, edge computing aiming at the end side becomes a core technology of an artificial intelligence landing scheme.
1. Protection principle of intermittent earth fault:
1.1, intermittent earth fault resistance temperature characteristic:
when the small-resistance grounding system has intermittent grounding fault, the effective value of the fault current is small and the duration is short, so that the microcomputer type relay protection is continuously started, reset, started and reset due to insufficient setting time, the tripping protection effect cannot reach the setting value requirement, and the fault current and voltage always exist in the system.
The grounding resistor heats up during fault arcing due to the sudden increase in neutral current. The ground resistance surface temperature rises.
After the arc is extinguished due to the fault, the ground resistor continuously radiates heat outwards due to the temperature difference with the ambient environment, and the surface temperature of the ground resistor continuously decreases until the temperature is close to the ambient environment temperature.
As shown in fig. 1, a certain time is required for the ground resistance temperature to reach the ambient temperature after the fault is extinguished, and a continuous and cyclic heating and cooling process or a continuous heating process exists for the ground resistance surface temperature before the fault is permanently disappeared or the fault outgoing line is reliably cut off.
1.2, basic principle of protection action:
referring to 'operating setting regulation of 3 kV-110 kV kilovolt power grid relay protection device' intermittent earth protection setting value, when the fault current exceeds the setting value and lasts for more than one cycle, and the interval time between the fault current and the second trigger setting value threshold is within 1000ms, the fault current is determined to be intermittent current, and when the fault current continuously triggers the setting value threshold to reciprocate for more than 5.5s, the relay is required to send a signal or trip out.
According to the characteristics of intermittent earth fault current and resistance temperature, a protection idea is provided, each cycle (0.02s) collects relevant data such as earth resistance temperature, room temperature and the like, whether the resistance temperature collected each time is normal or not is judged, and the indoor temperature plus a set temperature difference value is used as a normal temperature threshold value, as shown in formula 1.
And when the surface temperature of the grounding resistor is higher than the normal temperature threshold value, the abnormal temperature is considered. If the abnormal temperature is generated, the abnormal zero sequence current of the cycle is represented. The surface temperature of the grounding resistor rises sharply, and the surface temperature of the resistor drops slowly after the abnormal zero-sequence current disappears, so that the surface temperature change of the grounding resistor is a continuous process changing along with time.
In a certain time period (5.5s is adopted in the present technical solution), the fault temperature is obtained (as shown by the shaded portion in fig. 2b to 2 d). The calculation formula of the fault temperature is shown in formula 2.
T th ＝T room +T set (1)
In the formula: t is a unit of th Normal temperature threshold, T room Is the room temperature, T set To set the temperature difference, S T To fault temperature, T unusual Is an abnormal temperature, t 0 The time when the abnormal temperature occurs for the first time in the time period.
Each acquisition time is 0.02s, in order to simplify calculation, only continuous abnormal temperatures are screened out in the technical scheme, and all abnormal temperatures and values in a time period are accumulated to be used as fault temperatures, as shown in a formula 3.
S T ＝∑(T unusual -T th ) (3)
According to the comparison between the fault temperature value and the set setting Threshold value, two setting Threshold values are set, namely a low setting Threshold value min And a high setting Threshold max And when the fault temperature is lower than the low setting threshold, the system is normal, and when the fault temperature is between the two setting thresholds, the system gives an abnormal alarm without actual operation.
And when the fault temperature is higher than the high setting threshold value, the system performs breaker tripping protection, as shown in formula 4.
As long as the abnormal temperature system appears in the sample of every collection and all can report to the police earlier and indicate, carry out the circuit breaker trip when the system state judgement of present sample is fault type, if because of sample time node problem (around 5.5s at sample time t), system fault type appears, and only judge in the present sample as alarm state, initial temperature is higher in next sample, and the temperature drops slowly, still can judge fault type, and the fault area is shown in fig. 2b to fig. 2d for the different states of system.
The system state is judged once in a certain time period, certain hysteresis is achieved, and the real-time performance of system fault identification is poorer due to the fact that the whole processing flow is relatively complex and needs a large amount of calculation.
Based on this, adopt neural network to train a large amount of historical data in the past in this technical scheme, obtain a comparatively stable model, consider relevant characteristic factor in this model simultaneously, increase the accuracy of model. And sending the data acquired in the time period into the trained model, and returning to the state of the system. The intermittent grounding protection is completed, and the protection action can be not earlier than the wire outlet protection action and not later than the overheating protection of the grounding resistor by using the method. Can be used as a backup protection mode for processing the ground fault.
The flow block of the whole protection method is shown in fig. 3.
Specifically, the technical scheme of the invention adopts the following mode to realize the detection of the intermittent earth fault:
1) installing an infrared thermal imager, acquiring surface temperature data of the grounding resistor through the infrared thermal imager, and carrying out non-contact visual detection and monitoring instantaneous temperature rise on a main transformer grounding resistor device; acquiring grounding resistance surface temperature data by the thermal infrared imager at a sampling frequency of 200ms so as to monitor a blind area where a fault current continuously reciprocates and does not reach a trigger setting value threshold (5.5 s);
2) arranging an oil conservator oil level camera and an oil cup oil level camera for detecting the oil level of the transformer oil conservator and the oil level of the respirator oil cup; the oil level of the oil conservator correspondingly indicates the internal temperature of the main transformer;
3) a temperature transmitter or a temperature sensor is arranged for detecting the indoor and outdoor temperatures of the transformer;
4) the infrared thermal imager, the oil conservator oil level camera, the oil cup oil level camera, the transformer indoor temperature sensor and the transformer outdoor temperature sensor are correspondingly connected with an edge server arranged in the transformer room respectively;
5) constructing an intermittent earth fault detection data set based on temperature data, and designing a training test data set meeting machine learning requirements;
6) performing model training based on BP neural network data;
7) through image processing and calculation of an edge server, a decision basis is provided for the health state of a main transformer in a data driving mode;
8) by detecting the oil level change of the oil cup, the oil level change of the oil conservator and the temperature change of the grounding resistor in real time, the intermittent grounding fault is early warned and alarmed based on the data driving model.
Referring to 'operating setting regulation of 3 kV-110 kV kilovolt power grid relay protection device' intermittent earth protection setting value, when the fault current exceeds the setting value and lasts for more than one cycle, and the interval time between the fault current and the second trigger setting value threshold is within 1000ms, the fault current is determined to be intermittent current, and when the fault current continuously triggers the setting value threshold to reciprocate for more than 5.5s, the relay is required to send a signal or trip out.
1.3, system fault processing flow:
the overall system fault analysis flow is shown in fig. 4.
2. Acquiring and processing field data:
2.1, analyzing system state characteristics:
in the small-resistor grounding distribution network system, the following 5 characteristic parameters are selected to represent the state of the system, wherein the characteristic parameters are the surface temperature of a grounding resistor, the oil level of an oil conservator, the oil level of a breather oil cup, the indoor temperature and the outdoor temperature, and the characteristics have certain relevance in the distribution network system. Under the normal condition of the system state, the surface temperature of the grounding resistor is close to the indoor temperature (the ambient temperature) and the oil level temperature of the oil conservator, and the temperature difference between the outdoor temperature and the indoor temperature is in a certain range.
2.2, hardware composition and deployment:
in the technical scheme, a small-resistance grounding distribution network system of a 35Kv transformer substation of a certain power supply company is taken as a test object, and as shown in fig. 5, the main hardware components of the system include a PoE switch 1, a 4G router 2, an edge server (also called a field on-site server) 3, an indoor temperature and humidity transmitter 4, an outdoor temperature and humidity transmitter 4a, an infrared thermal imager 5, an oil conservator oil level detection camera 6 and an oil cup oil level detection camera 7.
The system detects surface temperature data of the grounding resistor 8 through a non-contact thermal infrared imager, the thermal infrared imager is deployed in a visible distance range of the grounding resistor on site, and the thermal infrared imager collects the surface temperature data of the grounding resistor once every 200ms so as to acquire information such as the highest temperature, the lowest temperature and the average temperature of each area.
Indoor temperature and humidity data of the transformer chamber 9 are collected through the indoor temperature transmitter. And comparing the ground resistance temperature acquired each time with the indoor environment temperature to acquire a temperature difference, comparing the temperature difference with a given temperature difference, and judging whether the acquisition has abnormal temperature or not, wherein if the acquisition does not have abnormal temperature, the state is set as normal and is represented by '1'. Otherwise, it is set to be abnormal and is represented by "0".
In addition, other related data are fully considered and used as auxiliary features, so that the accuracy of the model is improved:
the system uses a visible light camera (represented by an oil conservator oil level detection camera 6 and an oil cup oil level detection camera 7 in the figure) to acquire an oil level image of a transformer oil conservator 10 and an oil level image of a respirator oil cup 11 in real time, utilizes the advantages of real-time performance, intelligence and the like of an edge server, is matched with an artificial intelligence visual algorithm to acquire oil conservator oil level data and oil cup oil level data from the images, and meanwhile acquires outdoor temperature data through an outdoor temperature and humidity transmitter 4 a.
2.3, data processing:
the surface temperature T and the indoor temperature T of the grounding resistor in the technical scheme room Is Δ T ═ T-T room Setting a temperature difference T set Judging the real-time system state according to the temperature difference when the temperature is 3 ℃, and when the temperature is delta T<T set The real-time state of the system is 1 (normal), otherwise, the real-time state of the system is 0 (abnormal). The temperature portion data obtained are shown in table 1:
table 1 part temperature data
According to the analysis of the intermittent earth protection principle, a data set (28 pieces of data are selected in the technical scheme) within 5.5S is taken as a sample, a sample label is obtained, and the fault temperature S is calculated T And a low setting Threshold min And a high setting Threshold max The label type of the sample can be judged.
According to the statistics of fault data, the low setting Threshold value in the technical scheme min And a high setting Threshold max Set to 300, 800, respectively.
For example: the system fault sample acquisition and tag determination in table 2 is as follows, where the sum of fault temperatures in the accumulated time period in the system state of 0 in the table is 1036, which is greater than Threshold max Therefore, it is determined as a fault type.
TABLE 2 System Fault samples and labels
Where sample X can be represented as:
sample X corresponds to label Y: -1
2.4, system fault type:
there are three types of states according to the above analysis system, as shown in Table 3
TABLE 3 System State types
3. The failure prediction algorithm based on the BP neural network comprises the following steps:
3.1, structural design of the BP neural network:
the BP neural network is a feedforward network based on error inverse propagation and having no feedback between layers and no interconnection between layers. The BP neural network has a structure with three or more layers, namely an input layer, one or more hidden layers and an output layer.
Each hidden layer is generally acted by an activation function to obtain an output layer. The weight values are adjusted by continuously learning through a back propagation method.
And finally determining the weight value and the threshold value through repeated iteration. Experiments show that: the algorithm has the advantages of high convergence speed, high error analysis precision and the like in the process of predicting the faults of the processing equipment.
(1) Data normalization:
the BP neural network adjusts the connection weight according to a gradient descent method to minimize the training error, in order to enable the input data to adapt to the transfer function and improve the convergence rate of the calculation process, the training data needs to be normalized, the value range of the processed variable is between [ -1, 1], and the calculation method is as shown in formula (6):
in formula (6), x i As input variables, t i Normalized value, x max And x min The maximum and minimum values of the input variable x, respectively.
(2) Number of layers and number of neural units:
here, the x-sample two-dimensional vector is reduced to one dimension in the row direction, the number of input neurons is 140, and the output is 1. The number of the hidden layers is set to be 3, the number of neurons in each layer is 128, 10% of neurons in each layer are randomly removed for preventing overfitting, and L2 regularization is adopted.
(3) Learning rate:
in the gradient descent method, the learning rate is a super parameter for adjusting the network weight, the initial learning rate is set to be relatively large, the convergence rate of the calculation process can be increased, the learning rate is gradually reduced along with the increase of the iteration times, and the accuracy of fitting is ensured.
Here, an exponential decay learning rate is adopted, the initial learning rate is set to 0.2, the decay rate of the learning rate is set to 0.09, the iteration step number of the learning rate decay once is set to 128 steps (here, the sample training of each batch is one step), and the calculation method of the decay learning rate is shown in formula (7):
in the formula (7), lr is the learning rate after attenuation, and lr is base As initial learning rate, decay rate To learn the rate of decay, step is the number of steps per decay iteration.
(4) Activation function and cost function:
and selecting a linear rectification function relu as an activation function of the hidden layer, taking a normalized multi-classification function softmax as an activation function of the output layer, and taking a cross entropy loss function as a cost function.
The constructed BP neural network topology is shown in fig. 6.
3.2, training the BP neural network:
because of few fault samples, if the fault detection network is directly used for BP neural network training, the fault detection network cannot cover all state types, and the unbalanced learning set influences the accuracy of the result.
Wherein the sum of the simulated system fault and the system early warning sample accounts for about 5 percent of all samples. The ground resistance temperature and the different system state samples after adding the simulation data are shown in fig. 7a and 7 b.
According to the BP neural network model established above, the technical scheme uses python3.6 software, constructs a model based on a tensierflow 2.0 deep learning framework, and obtains 51365 samples and labels through data preprocessing; wherein 41365 sample data are used as training set, 10000 sample data are used as testing set.
Each time 128 sample data are sent into the model as a group, weight updating is carried out, and 25 rounds of training are carried out. According to the training result, the accuracy of the test set and the training set is continuously improved through the training of the model, the loss value is continuously reduced, and the accuracy is continuously improved, so that the model is properly established.
3.3, BP neural network prediction precision and error analysis:
after 25 rounds of training, as shown in fig. 8a and 8b, it can be seen that the loss values of the test set of the training set gradually decrease and gradually approach, and finally are both less than 0.1. The accuracy of the training set and the testing set gradually rises and finally tends to be stable, and the accuracy ranges are more than 98%.
After the network training is completed, new samples in various states are selected to test the network result, and the test result is shown in table 4, and the network identification rate of the new samples is up to 98%.
TABLE 4 System State identification results
In conclusion, when the line has an intermittent ground fault, the grounding current is small and the line has the characteristic of instantaneity. The traditional relay protection device does not process the scene that fault current continuously reciprocates but does not trigger a setting value threshold, so that protection does not act, a fault line cannot be quickly cut off, the power distribution network can be in a fault state for a long time, and finally the condition that the ground resistance is burnt is occasionally caused.
According to the characteristic that the temperature of the grounding resistor is increased by instantaneous grounding current, the technical scheme of the invention obtains the surface temperature of the grounding resistor, the indoor and outdoor temperature, the oil level of an oil conservator and other related information by real-time visual analysis and monitoring aiming at the thermal stability of the grounding resistor when a fault occurs, and predicts the equipment fault by combining an edge server and a BP neural network algorithm.
According to the technical scheme, in order to enable the system to have identification accuracy and rapid processing capacity of intermittent ground faults, non-contact visual detection and instantaneous temperature rise monitoring are carried out on devices such as a main transformer ground resistor and the like at a sampling frequency of 200ms, surface temperature data of the ground resistor are collected through a thermal infrared imager, oil level data of an oil conservator are collected through a visible light edge computing camera, and early warning and alarming are carried out on the intermittent ground faults by combining an indoor temperature number and an outdoor temperature number based on a data-driven mathematical model, so that a decision basis is provided for guaranteeing normal operation of the main transformer.
According to the technical scheme, single arcing time, single abnormal conditions or fault characteristic times in simple statistical limit time are not used as judgment bases, and the anti-interference performance is high; the fault diagnosis is carried out by adopting an intelligent algorithm through a large amount of data, and the advantages of quickly judging and processing the fault are achieved.
The intermittent grounding protection is used as backup protection, and can be matched with the existing monitoring protection system to jointly improve the safety and reliability of the system.
The invention can be widely applied to the field of intermittent earth fault detection and protection.
Claims (10)
1. A method for detecting intermittent earth faults based on machine vision is characterized in that:
1) arranging an infrared thermal imager in a visible range of the grounding resistor, acquiring surface temperature data of the grounding resistor through the infrared thermal imager, and carrying out non-contact visual detection and monitoring instantaneous temperature rise on a main transformer grounding resistor device;
2) arranging an oil conservator oil level camera and an oil cup oil level camera for detecting the oil level of the transformer oil conservator and the oil level of the respirator oil cup;
3) arranging a temperature transmitter or a temperature sensor for detecting the indoor and outdoor temperatures of the transformer;
4) comparing the ground resistance temperature acquired each time with the indoor environment temperature to acquire a temperature difference for comparing the given temperature difference;
5) the infrared thermal imager, the oil conservator oil level camera, the oil cup oil level camera, the transformer indoor temperature sensor and the transformer outdoor temperature sensor are respectively and correspondingly connected with an edge server arranged in the transformer room;
6) constructing an intermittent earth fault detection data set based on temperature data, and designing a training test data set meeting machine learning requirements;
7) performing model training based on BP neural network data;
8) the method comprises the steps that the surface temperature of a grounding resistor, the oil level of an oil conservator, the oil level of a respirator oil cup, the indoor temperature and the outdoor temperature are detected in real time, and an intermittent grounding fault is early warned based on a data driving model;
9) and through image processing and calculation of the edge server, a decision basis is provided for the health state of the main transformer in a data driving mode.
2. The machine-vision-based intermittent ground fault detection method according to claim 1, wherein the thermal infrared imager collects ground resistance surface temperature data at a sampling frequency of 200ms to monitor a blind area where a fault current continuously reciprocates without reaching a trigger setting threshold, and obtains information including a maximum temperature, a minimum temperature and an average temperature of each area.
3. The machine-vision based intermittent ground fault detection method as claimed in claim 2, wherein said trigger setting threshold is 5.5 s.
4. The method for machine vision based detection of intermittent earth faults according to claim 1, characterised in that the surface temperature T and the room temperature T of the earth resistance room Is Δ T ═ T-T room Setting a temperature difference T set Judging the real-time system state according to the temperature difference value when the temperature is 3 ℃, and when the temperature is delta T<T set The real-time state of the system is 1 and normal; otherwise 0, it is abnormal.
5. The method for machine-vision-based intermittent ground fault detection as claimed in claim 1, wherein the method for machine-vision-based intermittent ground fault detection takes a data set within 5.5S as a sample in the obtained intermittent ground fault detection data set, obtains a sample label, and calculates a fault temperature S T And a low setting Threshold min And a high setting Threshold max The label type of the sample can be judged.
6. The machine-vision-based intermittent earth fault detection method as claimed in claim 5, wherein said low setting Threshold is set min 300, the high setting Threshold is Threshold max Is 800.
7. The machine-vision based intermittent ground fault detection method as claimed in claim 1, wherein said model training comprises:
1) data normalization:
the calculation method is as follows:
in the formula, x i As input variables, t i Normalized value, x max And x min The maximum value and the minimum value of the input variable x are respectively;
2) number of layers and number of neural units:
reducing the x sample two-dimensional vector to one dimension according to the row direction, wherein the number of input neurons is 140, and the output neurons is 1; the number of the hidden layers is set to be 3, the number of neurons in each layer is 128, 10% of neurons in each layer are randomly removed for preventing overfitting, and L2 regularization is adopted;
3) learning rate:
an exponential decay learning rate is adopted, the initial learning rate is set to be 0.2, the decay rate of the learning rate is set to be 0.09, the iteration step number of the decay of the learning rate once is set to be 128 steps, and the calculation method of the decay learning rate is as follows:
where lr is the learning rate after attenuation, lr base For initial learning rate, decay rate For learning the attenuation rate, step is the number of steps of one iteration per attenuation;
4) activation function and cost function:
and selecting a linear rectification function relu as an activation function of the hidden layer, taking a normalized multi-classification function softmax as an activation function of the output layer, and taking a cross entropy loss function as a cost function.
8. The method for detecting the intermittent ground fault based on the machine vision as claimed in claim 1, wherein the method for detecting the intermittent ground fault based on the machine vision obtains the surface temperature of the ground resistor, the indoor and outdoor temperature, the oil conservator oil level and other relevant information through real-time visual analysis and monitoring aiming at the thermal stability of the ground resistor when the fault occurs, and predicts the equipment fault by using an edge server in combination with a BP neural network algorithm.
9. The method for detecting intermittent ground faults based on machine vision as claimed in claim 1, wherein the method for detecting intermittent ground faults based on machine vision is characterized in that a thermal infrared imager is used for carrying out non-contact visual detection on a ground resistance device of a main transformer at a sampling frequency of 200ms and monitoring instantaneous temperature rise, surface temperature data of the ground resistance is collected, a visible light edge calculation camera is used for collecting oil level data of an oil conservator, and an indoor and outdoor temperature number is combined to be based on a data-driven mathematical model, so that early warning and alarming are carried out on the intermittent ground faults, and a decision basis is provided for guaranteeing normal operation of the main transformer.
10. The method for detecting intermittent ground fault based on machine vision as claimed in claim 1, wherein said method for detecting intermittent ground fault based on machine vision is used as backup protection, fast judging and processing fault, and is matched with existing monitoring protection system to improve system safety and reliability.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210416361.9A CN114924207A (en) | 2022-04-20 | 2022-04-20 | Method for detecting intermittent earth fault based on machine vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210416361.9A CN114924207A (en) | 2022-04-20 | 2022-04-20 | Method for detecting intermittent earth fault based on machine vision |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114924207A true CN114924207A (en) | 2022-08-19 |
Family
ID=82806026
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210416361.9A Pending CN114924207A (en) | 2022-04-20 | 2022-04-20 | Method for detecting intermittent earth fault based on machine vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114924207A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
| CN110598736A (en) * | 2019-08-06 | 2019-12-20 | 西安理工大学 | Power equipment infrared image fault positioning, identifying and predicting method |
| CN111122195A (en) * | 2019-12-25 | 2020-05-08 | 应急管理部天津消防研究所 | Experimental platform and experimental method for simulating fire of extra-high voltage converter transformer |
| CN113532753A (en) * | 2021-07-15 | 2021-10-22 | 西安华控智能系统工程有限公司 | Wind power plant gear box oil leakage detection method based on machine vision |
-
2022
- 2022-04-20 CN CN202210416361.9A patent/CN114924207A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
| CN110598736A (en) * | 2019-08-06 | 2019-12-20 | 西安理工大学 | Power equipment infrared image fault positioning, identifying and predicting method |
| CN111122195A (en) * | 2019-12-25 | 2020-05-08 | 应急管理部天津消防研究所 | Experimental platform and experimental method for simulating fire of extra-high voltage converter transformer |
| CN113532753A (en) * | 2021-07-15 | 2021-10-22 | 西安华控智能系统工程有限公司 | Wind power plant gear box oil leakage detection method based on machine vision |
Non-Patent Citations (1)
| Title |
|---|
| 王坚;蒋绍华;陈雄;程素霞;姜杏辉;: "基于油气试验车和神经网络的变压器故障诊断方法", 科技创新导报, no. 20, 11 July 2018 (2018-07-11) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109839559B (en) | Lightning protection detection robot system | |
| CN105866638B (en) | A kind of city net cable connector state of insulation on-line monitoring prior-warning device and method | |
| CN104977087B (en) | A kind of method of the electrical equipment fault automatic early-warning based on infrared imaging temperature measuring | |
| CN107367673B (en) | A kind of arrester valve piece resistance operating status diagnostic method | |
| CN109061463A (en) | A kind of monitoring of mechanical state of high-voltage circuit breaker and method for diagnosing faults | |
| CN103292989B (en) | Breaker mechanical characteristic performance testing method and system | |
| CN106856322B (en) | A kind of flexible direct current power distribution network intelligent protection system based on neural network | |
| CN104990629B (en) | A kind of electrical equipment fault automatic early-warning system of infrared imaging temperature measuring | |
| CN107390049A (en) | A kind of power cable fault monitoring method and device | |
| CN110376514A (en) | A kind of DC high-speed switch synthetic performance evaluation method | |
| CN106291201A (en) | The lightning monitoring of a kind of lightning protection box and deterioration state monitoring system and method | |
| CN105092058A (en) | On-line temperature monitoring system for high tension switchgear, and monitoring method thereof | |
| CN106771716A (en) | Lightning protected intelligent management system | |
| CN110515333B (en) | Intelligent Internet of things terminal box, ubiquitous power Internet of things platform and application method thereof | |
| CN109374995A (en) | Low-voltage distribution system, aging circuit and oppose electricity-stealing monitoring method and correspondence system | |
| CN106646130A (en) | Active power distribution network fault positioning method and system based on current polarity comparison | |
| CN111579934A (en) | Method and device for realizing electric safety monitoring of external power network of base station and base station | |
| CN109470983B (en) | Power protection equipment fault early warning method and system | |
| CN117612345A (en) | Power equipment state monitoring and alarming system and method | |
| CN204881871U (en) | High tension switchgear's temperature on -line monitoring system | |
| Hamatwi et al. | Comparative analysis of high impedance fault detection techniques on distribution networks | |
| CN109713788A (en) | A kind of remote failure monitoring method for power equipment | |
| CN105098993A (en) | Network message analysis and fault recording system of intelligent substation | |
| CN114924207A (en) | Method for detecting intermittent earth fault based on machine vision | |
| CN117609841A (en) | Cable fault pre-judging analysis method based on neural network algorithm |
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 |