CN111191693B - Method for identifying thermal fault state of high-voltage switch cabinet based on convolutional neural network - Google Patents
Method for identifying thermal fault state of high-voltage switch cabinet based on convolutional neural network Download PDFInfo
- Publication number
- CN111191693B CN111191693B CN201911310519.9A CN201911310519A CN111191693B CN 111191693 B CN111191693 B CN 111191693B CN 201911310519 A CN201911310519 A CN 201911310519A CN 111191693 B CN111191693 B CN 111191693B
- Authority
- CN
- China
- Prior art keywords
- data
- convolutional neural
- neural network
- switch cabinet
- data set
- 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
Images
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
Abstract
The invention discloses a method for identifying a thermal fault state of a high-voltage switch cabinet based on a convolutional neural network, which belongs to the technical field of safe operation of the high-voltage switch cabinet. Therefore, the defect that thermal fault states such as heating components, heating types and the like in the high-voltage switch cabinet body cannot be directly indicated by infrared temperature measurement of the high-voltage switch cabinet body through a non-contact temperature measurement method in the prior art is overcome, and the thermal fault states are identified by the temperature of the high-voltage switch cabinet body.
Description
Technical Field
The invention relates to the technical field of safe operation of a high-voltage switch cabinet, in particular to an artificial intelligence method for identifying a thermal fault state of the high-voltage switch cabinet through the temperature of the high-voltage switch cabinet based on a convolutional neural network.
Background
The high-voltage switch cabinet is a complete set of power distribution device formed by combining a circuit breaker, a load switch, a contactor, a fuse, various switches, various mutual inductors, control, measurement, protection, regulation devices and the like. The main function is to distribute the electric energy reasonably and accurately according to the power supply requirement.
The high-voltage switch cabinet is of a metal closed structure, and once internal components are overheated, chain reaction is easily caused, so that heat in the cabinet is quickly accumulated to cause fire accidents. The miniaturization trend of the high-voltage switch cabinet is obvious, the load of electric equipment is continuously increased, and the problem of heating of the high-voltage switch cabinet is increased rapidly. Because the temperature in the cabinet is too high and overheating faults are not found and processed in time, the caused insulation failure accidents also occur frequently, the adverse effect is generated on power consumption units, and the loss is caused to national economy.
Compared with a contact type temperature measurement method for arranging a large number of probes and measurement lines in a cabinet, the non-contact temperature measurement method for performing infrared temperature measurement on the cabinet body of the high-voltage switch cabinet has the advantages of higher safety, convenience and rapidness, but cannot directly indicate thermal fault states of heating components, heating types and the like in the cabinet. A set of mapping relation needs to be established between the thermal fault state of the switch cabinet and the temperature change of the cabinet body, and the thermal fault state of the switch cabinet is deduced according to the temperature of the cabinet body, so that operation and maintenance personnel can find and process heating faults in the cabinet in time. The environmental influence factors of the high-voltage switch cabinet heating are very complex, the heat conduction process in the cabinet is not clear, and the function transformation relation which is accurate and widely applicable between the heating of the internal parts of the high-voltage switch cabinet and the temperature of the cabinet body does not exist. The artificial intelligent neural network has nonlinear large-scale self-adaptive capacity, good nonlinear mapping capacity and self-learning adaptive capacity, and has a particular advantage for the problem of the high-voltage switch cabinet heating fault identification.
Therefore, the invention designs a set of artificial intelligence method for establishing the relationship between the thermal fault state of the high-voltage switch cabinet and the temperature of the cabinet body based on the convolutional neural network, and has important significance for the judgment and the maintenance of the heating fault of the high-voltage switch cabinet.
Disclosure of Invention
The invention aims to provide a method for identifying the thermal fault state of a high-voltage switch cabinet based on a convolutional neural network, thereby overcoming the defect that the existing method for carrying out infrared temperature measurement on a cabinet body of the high-voltage switch cabinet by a non-contact temperature measurement method cannot directly indicate the thermal fault states of heating components, heating types and the like in the cabinet.
In order to achieve the aim, the invention provides a method for identifying the thermal fault state of a high-voltage switch cabinet based on a convolutional neural network, which takes the convolutional neural network as a core learning algorithm to form a mapping from cabinet body surface temperature data, switch cabinet operation data and environment data to thermal fault type classification so as to identify the thermal fault state of the high-voltage switch cabinet; the method comprises the following steps:
step 1, continuously acquiring temperature original data of a high-voltage switch cabinet body, operating current original data of the high-voltage switch cabinet and environmental temperature and humidity data of the high-voltage switch cabinet;
step 2, carrying out segmentation processing on each type of original continuous data according to time intervals, and calculating to obtain a plurality of indirect data as extracted characteristic parameters to form a characteristic vector data set; dividing the characteristic vector data set into a training mode data set and an identification mode data set according to the difference of the source and the use of the original continuous data;
step 3, establishing an input and output classification structure based on a convolutional neural network, wherein the input and output classification structure based on the convolutional neural network can carry out forward and backward propagation; the input and output classification structure based on the convolutional neural network adopts the training mode data set to perform network training, performs high-voltage switch cabinet thermal fault state classification on the recognition mode data set, and takes the high-voltage switch cabinet thermal fault state classification result as the output result of the input and output classification structure based on the convolutional neural network;
and 4, outputting the newly acquired original temperature data and original operating current data of the high-voltage switch cabinet and the environment temperature and humidity data of the high-voltage switch cabinet through the input and output classification structure of the convolutional neural network to obtain the thermal fault category of the high-voltage switch cabinet.
Further, in step 2, the method for extracting feature parameters to form a feature vector data set includes the following steps:
segmenting each type of original continuous data according to a segmentation time interval dt to obtain segmented original data, wherein the segmented original data are time segmented data, and numerical values of a starting time and an ending time of each type of segmented original data are reserved;
time indirect data calculated according to each segmented original data is used as a feature vector,
if the number of segments of the original data is i, the obtained feature parameter vector data set IN is a 1 × i × 6 array.
Further, in step 2, the recognition pattern data set is: the identification mode data set is continuously updated according to real-time original data and cannot be infinitely expanded; when the division time interval dt and the number of segments i of the original data are fixed, once new data is generated, the feature parameter vector of the latest segment is supplemented to the last column of the recognition pattern data set, and the first column of the recognition pattern data set is deleted, so that the recognition pattern data set is updated and the data length is unchanged.
Further, in step 3, the input and output classification structure based on the convolutional neural network includes: the network structure of the multilayer convolution layer, the pooling layer and the full-connection layer in the forward propagation process network can perform loss function evaluation and network parameter optimization in the backward propagation process.
Further, in step 3, the network training of the input/output classification structure based on the convolutional neural network by using the training pattern data set includes the following steps:
step 31, initializing a training network, initializing training parameters in the convolutional neural network by using a small random number, and setting a result output vector;
step 32, inputting the training mode data set into the initialized convolutional neural network, performing forward propagation, and outputting a network classification result; calculating loss function values of the network classification result vector and the sample real condition result vector;
step 33, calculating the gradient value of each layer of loss function in the convolutional neural network by adopting a gradient descent algorithm, and updating network parameters by multiplying the gradient value with the original parameters for back propagation;
step 34, repeating the step 33 and the step 34 for multiple times, and adjusting and optimizing the network parameters to minimize the loss function value;
and step 35, taking a plurality of training mode data sets as input data, repeating the steps 31 to 34, and determining network parameters to finish network training when the thermal fault state identification accuracy rate is stable and high.
Further, the segmenting the original data includes: temperature T of switch cabinet body, operation load current I of switch cabinet and ambient temperature TeHumidity He(ii) a The time indirection data includes: the cabinet body average temperature, the cabinet body temperature primary change rate, the cabinet body temperature secondary change rate, the current primary change rate, the environment average temperature and the average humidity.
Further, the data of the training pattern data set is data when a thermal fault state occurs.
Further, the loss function adopts a softmax loss function.
Further, the high-voltage switch cabinet thermal fault state comprises: poor contact, excessive load and fault arcing.
Compared with the prior art, the invention has the following beneficial effects:
the method for identifying the thermal fault state of the high-voltage switch cabinet based on the convolutional neural network comprises the steps of continuously acquiring the original temperature data of the high-voltage switch cabinet, the original operating current data of the high-voltage switch cabinet and the original ambient temperature and humidity data of the high-voltage switch cabinet, forming mapping from the temperature data of the surface of the cabinet body, the operating data of the switch cabinet and the ambient data to thermal fault type classification by taking the convolutional neural network as a core learning algorithm, forming a characteristic parameter data set by carrying out time segmentation processing on various kinds of original data to be used as an input layer of the convolutional neural network, and identifying the thermal fault state of the high-voltage switch cabinet by using the convolutional neural network as a classifier through back propagation optimization and sample data set training to realize the identification of the fault state and the normal state of the switch cabinet. Therefore, the defect that the existing high-voltage switch cabinet body is subjected to infrared temperature measurement by a non-contact temperature measurement method and cannot directly indicate thermal fault states such as heating components, heating types and the like in the cabinet is overcome, the thermal fault state is identified by the temperature of the high-voltage switch cabinet body, and the heating fault of the switch cabinet is favorably found and processed in time.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only one embodiment of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
FIG. 1 is a flow chart of a method of identifying a thermal fault condition of a high voltage switchgear based on a convolutional neural network of the present invention;
fig. 2 is a schematic diagram of an input-output classification structure based on a convolutional neural network according to an embodiment of the present invention.
Detailed Description
The technical solutions in the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The influence process of the heating of the internal parts of the high-voltage switch cabinet on the temperature of the cabinet body is very complex, and the accurate and widely applicable function transformation relation between the heating of the internal parts of the high-voltage switch cabinet and the temperature of the cabinet body does not exist. A set of mapping relation needs to be established between the thermal fault state of the switch cabinet and the temperature change of the cabinet body, the thermal fault state of the switch cabinet is deduced according to the temperature of the cabinet body, and the operation and maintenance personnel can find and process heating faults in the cabinet in time according to the temperature change of the cabinet body. And the artificial intelligence neural network has wide applicability and special advantages for the identification problem.
The method for identifying the thermal fault state of the high-voltage switch cabinet based on the convolutional neural network comprises the steps of forming mapping from cabinet body surface temperature data, switch cabinet operation data and environment data to thermal fault type classification by taking the convolutional neural network as a core learning algorithm, forming a characteristic parameter data set by carrying out time segmentation processing on various kinds of original data to be used as an input layer of the convolutional neural network, and identifying the thermal fault state of the high-voltage switch cabinet by using the convolutional neural network as a classifier through back propagation optimization and sample data set training to realize identification of the fault state and the normal state of the switch cabinet. The switchgear fault state includes: poor contact, excessive load and fault arc.
As shown in fig. 1, the method for identifying a thermal fault state of a high-voltage switch cabinet based on a convolutional neural network provided by the invention comprises the following steps:
step 1: the method comprises the steps of continuously collecting temperature original data of a high-voltage switch cabinet body, operation load current original data of the high-voltage switch cabinet, and ambient temperature and humidity data of the high-voltage switch cabinet. The continuously acquired raw data can reflect the time-varying relationship.
The original data of the temperature of the switch cabinet body is derived from the maximum value of the temperature of a preset area on the surface of the cabinet body acquired by infrared temperature measuring equipment, the original data of the operation load current of the high-voltage switch cabinet is derived from a current transformer of the switch cabinet, and the data of the environment temperature and the environment humidity are derived from automatic acquisition equipment. The sampling interval of the original data is set according to the requirement of convolutional neural network training, but the minimum value is not more than the time interval dt of the convolutional neural network characteristic parameter processing.
Step 2: segmenting each type of original continuous data according to a fixed time interval dt to obtain segmented original data, wherein the segmented original data are time segmented data; calculating to obtain a plurality of time indirect data according to each segmented original data, wherein each time period indirect data is used as an extracted characteristic parameter to form a characteristic vector data set; according to the difference of sources and purposes, the characteristic vector data sets are divided into training pattern data sets for providing training samples for the convolutional neural network and recognition pattern data sets for recognizing the thermal fault types.
The step 2 comprises the following steps:
and step 21, setting the total time length t of the original continuous data to be 600s, dividing the original continuous data into 60 time segments according to a fixed time interval dt to be 10s, reserving the values of the starting time and the ending time of each segment of the original data, and discarding other intermediate data.
Segmenting the original data includes: temperature T of switch cabinet body, operation load current I of switch cabinet and ambient temperature TeHumidity He(ii) a Wherein, the 2 nd segment of the original temperature data in the segmented original data is at the starting time t010s and end time t1A value of 20s may be denoted T20、T21(ii) a Other segmented raw data comprise the operation load current value I and the environment temperature T of the switch cabineteHumidity He。
Step 22, the time indirect data calculated from the ith segmented original data includes: average temperature TA of cabineti＝(Ti1-Ti0) /2, one-time change rate dT of cabinet body temperaturei/dt＝(Ti1-Ti0) dT, cabinet body temperature secondary change rate dTi/dt2＝(Ti1-Ti0)/dt2First rate of change of current dIi/dt＝(Ii1-Ii0) Dt, ambient mean temperature TAeiAnd average humidity HAei；
Then the cabinet average temperature characteristic vector TA ═ TA1 TA2…TA60]The characteristic vector dT/dT of the primary rate of change of the temperature of the cabinet is [ dT ═ dT1/dt dT2/dt…dT60/dt]Characteristic vector dT/dT of cabinet body temperature secondary change rate2＝[dT1/dt2 dT2/dt2…dT60/dt2]And setting the current primary change rate characteristic vector dI/dt as [ dI ═ dI1/dtdI2/dt…dI60/dt]Ambient average temperature feature vector TAe＝[TAe1 TAe2…TAe60]And average humidity feature vector HAe＝[HAe1 HAe2…HAe60]。
And 23, obtaining a characteristic parameter data set IN [ TA dT/dT ] by 60 sections2 dI/dt TAe HAe]TAnd IN is a 1 × 60 × 6 array.
The training pattern data set is different from the recognition pattern data set. The training pattern data set is derived from the original data of the training sample, and the data of a fixed time period when the thermal fault state occurs is selected. When the time division interval dt is 10s and the number of segments i is 60, IN order to ensure that the feature parameter data set does not expand indefinitely, the latest 600s to 610s segmented data feature vector is needed to update the data set, that is, the original 0s to 10s segmented data feature vector (the first column of the feature parameter data set IN) is deleted, the latest 600s to 610s segmented data feature vector is supplemented to the last column, and the data length of IN is still maintained at 1 × 60 × 6. By adopting the processing method, the identification mode data set can reflect the latest operation state of the switch cabinet, and the thermal fault of the switch cabinet can be identified at any time.
And step 3: the method comprises the steps of constructing an input and output classification structure based on a convolutional neural network, setting network structures such as the sequence and the number of convolutional layers, pooling layers, full-link layers and output layers in a convolutional neural training network in a forward propagation process, setting a loss function in a backward propagation process, and optimizing network parameters. Evaluating the quality of the network classification result by using a loss function, if the function value is overlarge, the classification result is poor, the recognition rate is low, and a back propagation process is required to optimize the network parameters; if the function value is smaller, the classification result is better, the recognition rate is high, and the network classification result is output. The forward and backward propagation training is repeatedly carried out by a plurality of training samples, the network classification result is better in accordance with the real result, the recognition rate is kept stable and higher, and then the network training is completed. The trained convolutional neural network is used as a classification recognizer (namely an input and output classification structure based on the convolutional neural network), and a thermal fault classification result vector is output at a high accuracy rate through inputting a recognition mode data set.
The convolutional neural training network in the step 3 comprises the following main layers: several convolution layers, a pooling layer, several full-connection layers and an output layer. Aiming at different specific problems, the convolutional neural network structure is changed, a plurality of groups of convolutional neural networks with different levels are configured, and the output effects of different network structures are compared to optimize the efficient convolutional neural network. As shown in fig. 2, taking VGG-net-like network as an example, the convolutional neural network for identifying the thermal fault state of the high-voltage switch cabinet is set as a four-layer convolutional structure: the first convolution layer comprises two convolution sublayers, each sublayer uses 64 convolution kernels, the size of each convolution kernel is 1 multiplied by 3 multiplied by 6, the moving step length is 1, and each sublayer uses a modified linear unit (ReLU) activation function to perform nonlinear transformation; the first convolution layer and the second convolution layer are connected by using a pooling layer, and the pooling method is a max-pooling method; the second convolution layer contains two convolution sublayers, each sublayer uses 128 convolution kernels, the size of each convolution kernel is 1 multiplied by 3 multiplied by 6, the moving step length is 1, and each sublayer uses a ReLU activation function to perform nonlinear transformation; the second convolution layer and the third convolution layer are connected by using a pooling layer, wherein the pooling method is a max-pooling method; the third layer of convolution layer contains two convolution sublayers, wherein each sublayer uses 256 convolution kernels, the size of each convolution kernel is 1 multiplied by 3 multiplied by 6, the moving step length is 1, and each sublayer uses a ReLU activation function to perform nonlinear transformation; the third convolution layer and the fourth convolution layer are connected by using a pooling layer, wherein the pooling method is a max-pooling method; the fourth convolution layer contains two convolution sublayers, each sublayer uses 512 convolution kernels, the size of each convolution kernel is 1 multiplied by 3 multiplied by 6, the moving step length is 1, and each sublayer uses a ReLU activation function to perform nonlinear transformation; the fourth layer of convolution layer is connected with the full connecting layer; and connecting the output layer behind the full connection layer, and obtaining a network classification result vector by the output layer, wherein the size of the result vector is 1 multiplied by 4, and the number of elements is consistent with the number of thermal fault state classifications.
In step 3, training the convolutional neural network model (network training is performed by using a training pattern data set based on the input-output classification structure of the convolutional neural network) includes the following steps:
step 31, initializing a training network, and setting a random number between the weight values of the convolutional layer and the full connection layer as [0,1] to initialize training parameters in the convolutional neural network; setting a result output vector OUT to be [ R1R 2R 3R 4], wherein R1, R2, R3 and R4 respectively represent three fault states of poor contact, overload and fault arc and a normal state, and the element value corresponding to one state is 1 and the other element values are 0 in a real situation. For example, a training sample generating thermal fault due to poor busbar contact, the real condition result vector is [ 1000 ]; for a training sample in normal state, the real case result vector is [ 0001 ].
Step 32, inputting a training mode data set IN with the size of 1 × 60 × 6 into the four-layer structure convolutional neural network, performing forward propagation, and outputting a network classification result; and calculating the values of a loss function L of the network classification result vector and the sample real condition result vector, wherein the loss function L uses softmax. The loss function is used for evaluating the quality of the network classification result, the larger the difference between the network classification result and the real condition result is, the larger the loss function value is, and otherwise, the loss function value is minimum.
Step 33, calculating the gradient value Δ L of each layer loss function in the convolutional neural network by using a gradient descent algorithm, and using the gradient value Δ L and the original network parameter NoldMultiplying to obtain an updated network parameter NnewThe counter-propagation is performed. Of a loss function LThe gradient Δ L, expressed as the partial derivative of the loss function L on the network parameter N:
updated network parameter NnewExpressed as the product of the loss function gradient value Δ L and the original network parameter N: n is a radical ofnew＝Noldα · Δ L, where α is the learning rate, set to 0.1.
And 34, repeating the step 33 and the step 34 for multiple times, and adjusting and optimizing the network parameter N to minimize the loss function value L.
Step 35, using n training pattern data sets IN1、IN2…INnAnd (5) as input data, repeating the steps 31 to 34, and fixing the network parameters to finish network training when the accuracy of thermal fault state identification is stable and high. The trained convolutional neural network is used as a classifier for identifying the thermal fault state of the pattern data set, namely, data are collected IN real time according to the temperature of the high-voltage switch cabinet and are processed to form an identification pattern data set INi，INiContinuously updated over time, each INiThe thermal fault state is identified through the convolutional neural network, and the state classification result is continuously output.
And 4, step 4: and (3) outputting result vectors of newly acquired temperature original data of the high-voltage switch cabinet body, operation current original data of the high-voltage switch cabinet and environment temperature and humidity data of the high-voltage switch cabinet through the convolutional neural network input-output classification structure to obtain the thermal fault category of the high-voltage switch cabinet, always prompting a normal state if no thermal fault is generated, and outputting a corresponding fault state type if the thermal fault is generated.
The above disclosure is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of changes or modifications within the technical scope of the present invention, and shall be covered by the scope of the present invention.
Claims (9)
1. The method for identifying the thermal fault state of the high-voltage switch cabinet based on the convolutional neural network is characterized by comprising the following steps of: a convolutional neural network is used as a core learning algorithm to form a mapping from cabinet body surface temperature data, switch cabinet operation data and environment data to thermal fault type classification so as to identify the thermal fault state of the high-voltage switch cabinet; the method comprises the following steps:
step 1, continuously acquiring temperature original data of a high-voltage switch cabinet body, operating current original data of the high-voltage switch cabinet, and environmental temperature and humidity data of the high-voltage switch cabinet;
step 2, carrying out segmentation processing on each type of original continuous data according to time intervals, and calculating to obtain a plurality of indirect data serving as extracted characteristic parameters to form a characteristic vector data set; dividing the characteristic vector data set into a training mode data set and an identification mode data set according to the difference of the source and the use of the original continuous data;
the method comprises the steps that a training mode data set is derived from original data of a training sample, selected data in a fixed time period when a thermal fault state occurs, an identification mode data set can obtain real-time original data, when a time division interval dt =10s and a segment number i =60 are fixed, in order to ensure that a feature parameter data set cannot expand infinitely, the data set needs to be updated by using a latest 600 s-610 s segment data feature vector, namely, the original 0 s-10 s segment data feature vector is deleted, namely, a first column of the feature parameter data set is deleted, the latest 600 s-610 s segment data feature vector is supplemented to a last column of the feature parameter data set, and the data length of the feature parameter data set is still kept to be 1 × 60 × 6;
step 3, establishing an input and output classification structure based on a convolutional neural network, wherein the input and output classification structure based on the convolutional neural network can carry out forward and backward propagation; the input and output classification structure based on the convolutional neural network adopts the training mode data set to perform network training, performs high-voltage switch cabinet thermal fault state classification on the recognition mode data set, and takes the high-voltage switch cabinet thermal fault state classification result as the output result of the input and output classification structure based on the convolutional neural network;
and 4, outputting the newly acquired original temperature data and original operating current data of the high-voltage switch cabinet and the environment temperature and humidity data of the high-voltage switch cabinet through the input and output classification structure of the convolutional neural network to obtain the thermal fault category of the high-voltage switch cabinet.
2. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 1, wherein: in the step 2, the method for extracting the feature parameters to form the feature vector data set includes the following steps:
segmenting each type of original continuous data according to a segmentation time interval dt to obtain segmented original data, wherein the segmented original data are time segmented data, and numerical values of a starting time and an ending time of each type of segmented original data are reserved;
time indirect data calculated according to each segmented original data is used as a feature vector,
if the number of segments of the original data is i, the obtained feature parameter vector data set IN is a 1 × i × 6 array.
3. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 1, wherein: in step 2, the recognition pattern data set is: the identification mode data set is continuously updated according to real-time original data and cannot be infinitely expanded; when the division time interval dt and the number of segments i of the original data are fixed, once new data is generated, the feature parameter vector of the latest segment is supplemented to the last column of the recognition pattern data set, and the first column of the recognition pattern data set is deleted, so that the recognition pattern data set is updated and the data length is unchanged.
4. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 1, wherein: in step 3, the input and output classification structure based on the convolutional neural network includes: the network structure of the multilayer convolution layer, the pooling layer and the full-connection layer in the forward propagation process network can perform loss function evaluation and network parameter optimization in the backward propagation process.
5. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 1, wherein: in step 3, the network training of the input/output classification structure based on the convolutional neural network by using the training pattern data set includes the following steps:
step 31, initializing a training network, initializing training parameters in the convolutional neural network by using a small random number, and setting a result output vector;
step 32, inputting the training mode data set into the initialized convolutional neural network, performing forward propagation, and outputting a network classification result; calculating loss function values of the network classification result vector and the sample real condition result vector;
step 33, calculating the gradient value of each layer of loss function in the convolutional neural network by adopting a gradient descent algorithm, and updating network parameters by multiplying the gradient value with the original parameters for back propagation;
step 34, repeating step 33 and step 34 for multiple times, and adjusting and optimizing the network parameters to minimize the loss function value;
and step 35, taking a plurality of training mode data sets as input data, repeating the steps 31 to 34, and determining network parameters to finish network training when the thermal fault state identification accuracy rate is stable and high.
6. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 2, wherein: the segmented raw data includes: temperature T of switch cabinet body, operation load current I of switch cabinet and ambient temperature TeHumidity He(ii) a The time indirection data includes: the cabinet body average temperature, the cabinet body temperature primary change rate, the cabinet body temperature secondary change rate, the current primary change rate, the environment average temperature and the average humidity.
7. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 1, wherein: the data of the training pattern data set is data when a thermal fault condition occurs.
8. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 4, wherein: the loss function adopts a softmax loss function.
9. The convolutional neural network-based method for identifying a thermal fault condition of a high voltage switchgear as claimed in claim 1, wherein: the high-voltage switch cabinet thermal fault state comprises the following steps: poor contact, excessive load and fault arcing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911310519.9A CN111191693B (en) | 2019-12-18 | 2019-12-18 | Method for identifying thermal fault state of high-voltage switch cabinet based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911310519.9A CN111191693B (en) | 2019-12-18 | 2019-12-18 | Method for identifying thermal fault state of high-voltage switch cabinet based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111191693A CN111191693A (en) | 2020-05-22 |
| CN111191693B true CN111191693B (en) | 2022-06-24 |
Family
ID=70707378
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911310519.9A Active CN111191693B (en) | 2019-12-18 | 2019-12-18 | Method for identifying thermal fault state of high-voltage switch cabinet based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111191693B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112001417B (en) * | 2020-07-17 | 2022-07-05 | 国网宁夏电力有限公司检修公司 | Monitoring method, medium and system for transformer oil conservator |
| CN112630562A (en) * | 2020-12-04 | 2021-04-09 | 国网四川省电力公司广安供电公司 | Switch cabinet fault identification method and device based on deep neural network |
| CN113340438B (en) * | 2021-07-08 | 2022-05-03 | 电子科技大学 | Method for calibrating distance error of hot end component of aero-engine in non-contact temperature field |
| CN113838269A (en) * | 2021-09-18 | 2021-12-24 | 国网河北省电力有限公司电力科学研究院 | Low-voltage switch cabinet temperature and humidity monitoring and early warning method and system and storage medium |
| CN115951002B (en) * | 2023-03-10 | 2023-05-16 | 山东省计量科学研究院 | Gas chromatograph-mass spectrometer fault detection device |
| CN116893336B (en) * | 2023-07-11 | 2024-04-02 | 宁夏隆基电气有限公司 | High-low voltage switching device and method thereof |
| CN116740472B (en) * | 2023-08-09 | 2023-10-20 | 深圳市超业电力科技有限公司 | Power switch cabinet switch state identification method, device, equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108052946A (en) * | 2017-12-11 | 2018-05-18 | 国网上海市电力公司 | A kind of high pressure cabinet switch automatic identifying method based on convolutional neural networks |
| US10140820B1 (en) * | 2015-07-25 | 2018-11-27 | Gary M. Zalewski | Devices for tracking retail interactions with goods and association to user accounts for cashier-less transactions |
| CN109612708A (en) * | 2018-12-28 | 2019-04-12 | 东北大学 | Based on the power transformer on-line detecting system and method for improving convolutional neural networks |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW201915727A (en) * | 2017-09-18 | 2019-04-16 | 元智大學 | Fault detection and classification method of multi-sensors |
| CN109031069A (en) * | 2018-08-16 | 2018-12-18 | 国网山西省电力公司临汾供电公司 | The diagnostic system of high-tension switch cabinet Partial discharge signal |
| CN109918752B (en) * | 2019-02-26 | 2022-12-16 | 华南理工大学 | Mechanical fault diagnosis method, equipment and medium based on migration convolution neural network |
-
2019
- 2019-12-18 CN CN201911310519.9A patent/CN111191693B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10140820B1 (en) * | 2015-07-25 | 2018-11-27 | Gary M. Zalewski | Devices for tracking retail interactions with goods and association to user accounts for cashier-less transactions |
| CN108052946A (en) * | 2017-12-11 | 2018-05-18 | 国网上海市电力公司 | A kind of high pressure cabinet switch automatic identifying method based on convolutional neural networks |
| CN109612708A (en) * | 2018-12-28 | 2019-04-12 | 东北大学 | Based on the power transformer on-line detecting system and method for improving convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| Dhiradj Djairam."Influence of environmental and operational conditions on breakdown voltage of oil in switchgear".《2012 IEEE International Conference on Condition Monitoring and Diagnosis》.2013, * |
| 母卓元.基于深度信念网络的开关柜故障诊断.《电力安全技术》.2019,第第21卷卷(第11期),35-41. * |
| 王飞凤 等."高压开关柜异常局放故障电场数值计算与分析".《电气工程学报》.2019,第14卷(第2期), * |
| 许高峻."基于信息融合技术的高压开关柜温度预警研究".《电工技术》.2016, * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111191693A (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111191693B (en) | Method for identifying thermal fault state of high-voltage switch cabinet based on convolutional neural network | |
| Zollanvari et al. | Transformer fault prognosis using deep recurrent neural network over vibration signals | |
| CN104914327B (en) | Transformer fault maintenance Forecasting Methodology based on real-time monitoring information | |
| Almeida et al. | Intelligent thermographic diagnostic applied to surge arresters: a new approach | |
| CN102663412B (en) | Power equipment current-carrying fault trend prediction method based on least squares support vector machine | |
| CN103886374A (en) | Cable joint wire temperature prediction method based on RBF neural network | |
| Illias et al. | Estimation of transformer parameters from nameplate data by imperialist competitive and gravitational search algorithms | |
| CN108518804A (en) | A kind of machine room humiture environmental forecasting method and system | |
| CN109871660B (en) | Early warning method and fault positioning method for heating fault of main transformer | |
| CN111401657B (en) | Transformer hot spot temperature time sequence prediction method based on support vector regression algorithm | |
| CN106932712A (en) | A kind of circuit breaker failure diagnostic method based on improvement Fuzzy Petri Net | |
| CN116893336B (en) | High-low voltage switching device and method thereof | |
| CN108197396A (en) | A kind of high voltage isolator superheat state Forecasting Methodology based on PSO-SVM | |
| CN108562835A (en) | A kind of fault arc detection method based on BP neural network | |
| CN116529740A (en) | Load forecasting for electrical devices using machine learning | |
| CN110021929B (en) | Electromagnetic transient time domain simulation modeling method for fast switch type fault current limiter | |
| CN106707153A (en) | FOA-RBF based high-voltage circuit breaker fault diagnosis method | |
| CN112630562A (en) | Switch cabinet fault identification method and device based on deep neural network | |
| Cheng et al. | Fault diagnosis method based on Petri nets considering service feature of information source devices | |
| CN116345340A (en) | Intelligent monitoring system for switch cabinet | |
| CN112580232B (en) | Contact resistance online monitoring method for multiple key contact points in medium-voltage switch cabinet | |
| CN112329809A (en) | High-voltage circuit breaker fault diagnosis method based on decision tree algorithm | |
| CN106093130A (en) | A kind of test product to scribbling electric force compounded grease carries out the method and system of combined heat and power test | |
| CN113326663A (en) | Railway relay running state evaluation method based on extreme learning machine | |
| Qiao et al. | Temperature Rise Prediction of GIS Electrical Contact Using an Improved Kalman Filter |
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 |