CN116610998A - Switch cabinet fault diagnosis method and system based on multi-mode data fusion - Google Patents
Switch cabinet fault diagnosis method and system based on multi-mode data fusion Download PDFInfo
- Publication number
- CN116610998A CN116610998A CN202310591863.XA CN202310591863A CN116610998A CN 116610998 A CN116610998 A CN 116610998A CN 202310591863 A CN202310591863 A CN 202310591863A CN 116610998 A CN116610998 A CN 116610998A
- Authority
- CN
- China
- Prior art keywords
- switch cabinet
- fault diagnosis
- layer
- data
- data information
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 47
- 230000004927 fusion Effects 0.000 title claims abstract description 21
- 238000012544 monitoring process Methods 0.000 claims abstract description 40
- 239000013598 vector Substances 0.000 claims abstract description 25
- 238000007781 pre-processing Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims description 29
- 238000011176 pooling Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 9
- 238000002372 labelling Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 6
- 238000007499 fusion processing Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000005236 sound signal Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 6
- 230000002159 abnormal effect Effects 0.000 description 4
- 238000012806 monitoring device Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
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
- G06F18/10—Pre-processing; Data cleansing
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
- G06F18/256—Fusion techniques of classification results, e.g. of results related to same input data of results relating to different input data, e.g. multimodal recognition
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a switch cabinet fault diagnosis method based on multi-mode data fusion, which comprises the following steps: monitoring the running state of the switch cabinet to obtain the data information of the switch cabinet, preprocessing the data information to obtain the preprocessed data information, fusing the preprocessed data information to obtain a multi-mode feature vector, and inputting the multi-mode feature vector into a pre-trained fault diagnosis model of the switch cabinet to obtain a fault diagnosis result of the switch cabinet. The invention can solve the technical problems that the existing monitoring method adopting a single type sensor can only reflect the specific state of the equipment, is difficult to comprehensively and accurately reflect the whole state of the equipment, causes incomplete and inaccurate monitoring of the state of the equipment, can not timely discover and process equipment faults, and affects the safety, stability and reliability of the operation of the equipment.
Description
Technical Field
The invention belongs to the technical field of power system monitoring and early warning, and particularly relates to a switch cabinet fault diagnosis method and system based on multi-mode data fusion.
Background
The switch cabinet is very important equipment in the power system, is used for protecting the safe and stable operation of the power system, and as the automation and intelligent degree of the power equipment are improved, the monitoring and early warning of the switch cabinet become more and more important. The existing switch cabinet monitoring and early warning mechanism comprises a monitoring method adopting a single type of sensor and a monitoring method based on an on-line monitoring device.
However, the existing switch cabinet monitoring and early warning mechanisms have some non-negligible defects: (1) The state information of the switch cabinet equipment is more, the monitoring method adopting a single type sensor can only reflect the specific state of the equipment, and the whole state of the equipment is difficult to reflect comprehensively and accurately, which can lead to incomplete and inaccurate equipment state monitoring, and can lead to failure of the equipment to be discovered and processed in time, and influence the safety, stability and reliability of the operation of the equipment; (2) The monitoring method of the single type sensor generally requires special personnel to analyze and judge the data, and consumes time and manpower resources; (3) The existing monitoring method based on the switch cabinet on-line monitoring device is huge in information data to be monitored, easy to be interfered by external factors, and low in fault early warning efficiency and accuracy due to the fact that the existing monitoring method based on the switch cabinet on-line monitoring device lacks of functions of accurate analysis and processing.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the invention provides a switch cabinet intelligent monitoring and fault early warning method and system based on multi-mode data fusion, which aim to solve the technical problems that the existing monitoring method adopting a single type sensor is difficult to fully and accurately reflect the whole state of equipment, the equipment state is incomplete and inaccurate to monitor, the equipment faults cannot be found and processed in time, the operation safety, stability and reliability of the equipment are affected, and time and manpower resources are consumed because special personnel are required to analyze and judge the data; the existing monitoring method based on the switch cabinet on-line monitoring device has the technical problems that information data to be monitored are huge, the information data are easily interfered by external factors, and the functions of accurate analysis and processing are lacking, so that the fault early warning efficiency and the fault early warning accuracy are low.
In order to achieve the above object, according to one aspect of the present invention, there is provided a switchgear fault diagnosis method based on multi-mode data fusion, comprising the steps of:
(1) The operating state of the switch cabinet is monitored to obtain the data information of the switch cabinet.
(2) Preprocessing the data information obtained in the step (1) to obtain preprocessed data information.
(3) Carrying out fusion processing on the data information preprocessed in the step (2) to obtain a multi-mode feature vector;
(4) Inputting the multi-mode feature vector obtained in the step (3) into a pre-trained fault diagnosis model of the switch cabinet to obtain a fault diagnosis result of the switch cabinet.
Preferably, the data information of the switch cabinet includes an image, a sound signal, a temperature signal, etc. of the switch cabinet;
firstly, converting sound in data information from a time domain signal to a frequency domain signal by adopting short-time Fourier transform (STFT), then extracting local features in a temperature signal by adopting a sliding window method, and finally denoising an image by adopting a Gaussian filtering method.
Preferably, step (3) specifically comprises the following sub-steps:
and (3-1) sorting the data information preprocessed in the step (2) according to a time sequence, and sampling and interpolating the sorted data information according to a fixed time interval to obtain a set of regular time sequence information.
And (3-2) carrying out normalization processing on the time series information obtained in the step (3-1) to obtain time series information after normalization processing.
And (3-3) classifying the data information subjected to the normalization processing in the step (3-2) to obtain a plurality of channels.
(3-4) performing element-by-element product processing on the plurality of channels obtained in the step (3-3) to obtain the multi-modal feature vector.
Preferably, the fault diagnosis model of the switch cabinet is a CNN-LSTM network, and the specific structure is as follows:
the first layer is a convolution layer which is input as a three-dimensional tensor X E R containing continuous data (e.g., image, sound, temperature, etc.) B×T×I Outputting a group of three-dimensional tensor feature graphs X epsilon R generated after extracting the spatial features B×C×F Wherein X represents input continuous data, R represents a real number field of vectors, B represents the number of continuous data, T represents the length of each continuous data, I represents the feature vector size of each time step, C represents the new number of time steps obtained after convolution, and F represents the number of convolution kernels.
The second layer is a pooling layer, the input of which is a three-dimensional tensor feature map X epsilon 788R output by a convolution layer B×C×F Outputting a three-dimensional tensor feature graph X epsilon R with reduced size B1×C1×F1 Wherein B1 represents the data amount, C1 represents the new time step number obtained after pooling, F1 represents the number of convolution kernels
The third layer is an LSTM layer, and the input of the LSTM layer is a three-dimensional characteristic diagram X epsilon R output by the pooling layer B1×C1×F1 Outputting a two-dimensional tensor feature map X epsilon R modeled for time sequence information B×S Wherein S representsThe length of the sequence output by the LSTM layer.
The fourth layer is a fully connected layer, and the input is a two-dimensional tensor feature map X epsilon R output by the LSTM layer B×S The output is a two-dimensional tensor feature graph X epsilon R B×D Where D represents the output dimension. 5
Preferably, the switch cabinet fault diagnosis model is trained by the following steps:
(4-1) acquiring a switch cabinet monitoring data set, preprocessing the switch cabinet monitoring data set, dividing the preprocessed switch cabinet monitoring data set into a training set, a verification set and a test set according to the proportion of 8:1:1, and labeling all samples in the switch cabinet monitoring data set.
(4-2) classifying all samples in the training set obtained in the step (4-1) according to the labeling result of the step (4-1), sampling each type of samples according to a fixed time interval to obtain a plurality of sampled samples, sorting and integrating each sampled sample, the sampling time of the sample and the class of the sample into triples according to a time sequence, wherein the triples corresponding to all sampled samples form a triplet set;
(4-3) dividing the triplet set obtained in the step (4-2) through a time window method to obtain a plurality of triples comprising a plurality of time steps and corresponding input features.
And (4-4) inputting the multiple triplets obtained in the step (4-3) into a convolution layer and a pooling layer of the fault diagnosis model of the switch cabinet to perform feature extraction processing so as to obtain multiple feature graphs.
And (4-5) inputting the plurality of feature images obtained in the step (4-4) into an LSTM layer of a fault diagnosis model of the switch cabinet to perform sequence modeling processing so as to obtain output results, wherein the output results corresponding to all the feature images form an output sequence.
(4-6) comparing the output sequence obtained in the step (4-5) with the triplet set of the step (4-2) to obtain a loss function L MSE 。
(4-7) applying a back propagation algorithm to the loss function value L obtained in the step (4-6) MSE Performing update processing to obtain a convolution layer andupdated parameters in the pooling layer, and updated parameters in the LSTM layer.
(4-8) performing classifier direct connection processing on the output sequence of the step (4-5) to obtain a classification result.
(4-9) repeating the steps (4-4) to (4-8) until the preset training round number or the loss function converges, thereby obtaining a preliminarily trained fault diagnosis model of the switch cabinet.
Specifically, the preset training wheel number in the invention is 100000 times.
And (4-10) verifying the switch cabinet fault diagnosis model which is preliminarily trained in the step (4-9) by using the test set obtained in the step (4-1) until the accuracy of the obtained switch cabinet fault diagnosis model reaches the optimum, thereby obtaining the trained switch cabinet fault diagnosis model.
Preferably, in step (4-4), the output of the ith (where i e [1, the total number of triples obtained in step (4-3 ]) triplet through the convolution and pooling layers of the switchgear fault diagnosis model may be expressed as:
wherein Wj represents the weight matrix of the jth convolution kernel, X i,j Representing the jth input in the ith neuron, b representing the bias vector, σ representing the activation function, pool representing the pooling operation, j e [1 ], the total number of convolution kernels in the convolution layer of the input switch cabinet fault diagnosis model]。
Preferably, step (4-6) is performed using the following formula:
wherein y is reg,j Representing the true value of the ith triplet of all triples obtained in step (4-3),representing model pair stepsAnd (3) predicting target variable values of the ith triplet in all the triples obtained in the step (4-3).
Preferably, step (4-7) is performed using the following formula:
wherein, alpha represents learning rate, W is updated parameters in the convolution layer and the pooling layer, and U is updated parameters in the LSTM layer.
According to another aspect of the present invention, there is provided a switchgear fault diagnosis system based on multi-mode data fusion, including:
the first module is used for monitoring the operation state of the switch cabinet so as to acquire the data information of the switch cabinet.
And the second module is used for preprocessing the data information acquired by the first module to acquire the preprocessed data information.
The third module is used for carrying out fusion processing on the data information preprocessed by the second module so as to obtain a multi-mode feature vector;
and the fourth module is used for inputting the multi-mode feature vector obtained by the third module into a pre-trained fault diagnosis model of the switch cabinet so as to obtain a fault diagnosis result of the switch cabinet.
In general, the above technical solutions conceived by the present invention, compared with the prior art, enable the following beneficial effects to be obtained:
because the invention adopts the step (2) and the step (3), the invention can comprehensively and accurately reflect the equipment state by combining different types of sensor signals such as sound, temperature, image and the like in a multi-mode data fusion mode, thereby improving the monitoring accuracy;
(2) Because the invention adopts the step (4), the comprehensive analysis is carried out through the machine learning algorithm, and the monitoring and the early warning of the equipment state can be automatically completed, thereby improving the monitoring efficiency and saving the time and the manpower resources.
(3) Because the invention adopts the step (4), the characteristics and the rules in the data can be effectively extracted by adopting the CNN-LSTM neural network, and the prediction precision is improved.
(4) Because the invention adopts the step (4) and adopts the LSTM layer modeling time sequence information, the invention has better interpretability and can help the user to interpret and locate the abnormal state of the equipment.
(5) Because the invention adopts the steps (1) to (4), the abnormal condition can be found in time and early warning can be carried out by carrying out real-time monitoring and analysis on each parameter of the switch cabinet.
Drawings
FIG. 1 is a flow chart of a switch cabinet fault diagnosis method based on multi-mode data fusion.
Fig. 2 is a block diagram of a fault diagnosis model of the switchgear of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
As shown in fig. 1, the invention provides a switch cabinet fault diagnosis method based on multi-mode data fusion, which comprises the following steps:
(1) The operating state of the switch cabinet is monitored to obtain the data information of the switch cabinet.
Specifically, the operation state of the 35KV switch cabinet is monitored on line through various sensors, so that data information such as images, sound signals and temperature signals of the 35KV switch cabinet can be obtained.
(2) Preprocessing the data information obtained in the step (1) to obtain preprocessed data information.
Specifically, the method comprises the steps of firstly converting sound in data information from a time domain signal to a frequency domain signal by Short-time Fourier transform (STFT), then extracting local features in a temperature signal by a sliding window method, and finally denoising an image by a Gaussian filtering method.
The purpose of this step is to perform different preprocessing for each type of data to ensure data stability and handleability.
(3) Carrying out fusion processing on the data information preprocessed in the step (2) to obtain a multi-mode feature vector;
the method has the advantages that through fusion of different types of data information, more comprehensive, rich and complete information is provided, and the analysis result is more comprehensive and accurate.
The method specifically comprises the following substeps:
and (3-1) sorting the data information preprocessed in the step (2) according to a time sequence, and sampling and interpolating the sorted data information according to a fixed time interval to obtain a set of regular time sequence information.
In particular, the fixed time interval in this step is 1ms to 1s, preferably 500ms,
the interpolation processing is carried out by adopting a spline interpolation method, firstly, an interpolation interval is divided into a plurality of small sections, data points in each small section form a local spline interpolation function, then, continuity and smoothness among the small sections are ensured through constraint conditions, and finally, the function value of the target point is calculated by utilizing the constructed local spline interpolation function.
And (3-2) carrying out normalization processing on the time series information obtained in the step (3-1) to obtain time series information after normalization processing.
Specifically, the step is to perform normalization processing by a z-score normalization method to obtain a group of dimensionless data information.
The purpose of this step is to make the data information comparable and analyzable for better data analysis and model construction.
And (3-3) classifying the data information subjected to the normalization processing in the step (3-2) to obtain a plurality of channels.
(3-4) performing element-by-element product processing on the plurality of channels obtained in the step (3-3) to obtain the multi-modal feature vector.
The sub-steps (3-2) to (3-4) have the advantages that the interaction information among different feature vectors can be well captured by adopting the element-by-element product fusion method, and the problem of over-fitting can be effectively avoided, so that the training effect is more stable and reliable.
(4) Inputting the multi-mode feature vector obtained in the step (3) into a pre-trained fault diagnosis model of the switch cabinet to obtain a fault diagnosis result of the switch cabinet.
As shown in fig. 1, the fault diagnosis model of the switch cabinet of the invention is a CNN-LSTM network, which comprises a convolutional neural network (Convolutional neutral network, abbreviated as CNN) network and a long-short-time memory network (Long short term memory, abbreviated as LSTM) network which are directly connected.
Specifically, the CNN-LSTM network comprises four layers, and the specific structure is as follows:
the first layer is a convolution layer which is input as a three-dimensional tensor X E R containing continuous data (e.g., image, sound, temperature, etc.) B×T×I Outputting a group of three-dimensional tensor feature graphs X epsilon R generated after extracting the spatial features B×C×F Wherein X represents input continuous data, R represents a real number field of vectors, B represents the number of continuous data, T represents the length of each continuous data, I represents the feature vector size of each time step, C represents the new number of time steps obtained after convolution, and F represents the number of convolution kernels.
The second layer is a pooling layer, the input of which is a three-dimensional tensor feature map X epsilon 788R output by a convolution layer B×C×F Outputting a three-dimensional tensor feature graph X epsilon R with reduced size B1×C1×F1 Wherein B1 represents the data amount, C1 represents the new time step number obtained after pooling, F1 represents the number of convolution kernels
The third layer is an LSTM layer, and the input of the LSTM layer is a three-dimensional characteristic diagram X epsilon R output by the pooling layer B1×C1×F1 Outputting a two-dimensional tensor feature map X epsilon R modeled for time sequence information B×S Where S represents the sequence length of the LSTM layer output.
The fourth layer is a fully connected layer, and the input is a two-dimensional tensor feature map X epsilon R output by the LSTM layer B×S The output is a two-dimensional tensor feature graph X epsilon R B×D Where D represents the output dimension.
Specifically, the fault diagnosis model of the switch cabinet is obtained through training the following steps:
(4-1) acquiring a switch cabinet monitoring data set, preprocessing the switch cabinet monitoring data set, dividing the preprocessed switch cabinet monitoring data set into a training set, a verification set and a test set according to the ratio of 8:1:1, and labeling all samples in the switch cabinet monitoring data set (namely labeling all samples as normal, abnormal, fault and the like).
In this step, the switchgear monitoring dataset is obtained from the open database kagle.
The pretreatment process of the switch cabinet monitoring data set in the step is identical to the pretreatment process in the step (2), and is not repeated here;
(4-2) classifying all samples in the training set obtained in the step (4-1) according to the labeling result of the step (4-1), sampling each type of samples according to a fixed time interval to obtain a plurality of sampled samples, sorting and integrating each sampled sample, the sampling time of the sample and the class of the sample into triples according to a time sequence, wherein the triples corresponding to all sampled samples form a triplet set;
specifically, the fixed time interval in this step is 1min to 10min, preferably 5min.
The purpose of this step is to provide a better understanding of the values measured by the various sensors at a certain point in time and to provide a more accurate determination of the correlation between them by arranging these data in time series. In addition, the time series data may also be used to predict future trends or detect abnormal situations.
(4-3) dividing the triplet set obtained in the step (4-2) through a Time window (Time Windows) method to obtain a plurality of triples comprising a plurality of Time steps and corresponding to the plurality of input features.
Specifically, in the partitioning process of the time window method in this step, the size and the stride of the window are set first, then the window is created, and finally the created window is used to partition the triplet set obtained in the step (4-2).
Specifically, the window size and the step size determined in this step are generally set to 1min to 20min and 1min to 2min, preferably 10min and 2min, respectively.
The advantage of the sub-step (4-3) described above is that the time series information is divided into a plurality of fixed length time windows, reducing the dimensionality of the data, making the data processing and analysis more efficient.
And (4-4) inputting the multiple triplets obtained in the step (4-3) into a convolution layer and a pooling layer of the fault diagnosis model of the switch cabinet to perform feature extraction processing so as to obtain multiple feature graphs.
In this step, the output of the ith (where i e [1, the total number of triples obtained in step (4-3 ]) triplet through the convolution and pooling layers of the switchgear fault diagnosis model can be expressed as:
wherein Wj represents the weight matrix of the jth convolution kernel, X i,j Representing the jth input in the ith neuron, b representing the bias vector, σ representing the activation function, pool representing the pooling operation, j e [1 ], the total number of convolution kernels in the convolution layer of the input switch cabinet fault diagnosis model]。
And (4-5) inputting the plurality of feature images obtained in the step (4-4) into an LSTM layer of a fault diagnosis model of the switch cabinet to perform sequence modeling processing so as to obtain output results, wherein the output results corresponding to all the feature images form an output sequence.
In this step, the output Hi of the j-th feature map Zi is: hi=lstm (Zi)
Wherein LSTM represents the LSTM layer of the switchgear fault diagnosis model.
(4-6) comparing the output sequence obtained in the step (4-5) with the triplet set of the step (4-2) to obtain a loss function L MSE 。
Specifically, the following formula is adopted in the present step:
wherein k is the total number of triples obtained in the step (4-3), y reg,j Representing the true value of the ith triplet of all triples obtained in step (4-3),representing the predicted target variable value of the model for the ith triplet among all triples obtained in step (4-3).
The substep (4-6) has the advantage that the prediction effect of the model on the normal state of the switch cabinet can be better evaluated by adopting the loss function of the mean square error.
(4-7) applying a back propagation algorithm to the loss function value L obtained in the step (4-6) MSE And performing updating processing to obtain updated parameters in the convolution layer and the pooling layer and updated parameters in the LSTM layer.
Specifically, the following formula is adopted in the present step:
wherein, alpha represents learning rate, W is updated parameters in the convolution layer and the pooling layer, and U is updated parameters in the LSTM layer.
The purpose of this step is to update the parameters in the CNN and LSTM layers by a back propagation algorithm so that the model can better fit the data.
(4-8) performing classifier direct connection processing on the output sequence of the step (4-5) to obtain a classification result.
(4-9) repeating the steps (4-4) to (4-8) until the preset training round number or the loss function converges, thereby obtaining a preliminarily trained fault diagnosis model of the switch cabinet.
Specifically, the preset training wheel number in the invention is 100000 times.
And (4-10) verifying the switch cabinet fault diagnosis model which is preliminarily trained in the step (4-9) by using the test set obtained in the step (4-1) until the accuracy of the obtained switch cabinet fault diagnosis model reaches the optimum, thereby obtaining the trained switch cabinet fault diagnosis model.
According to the invention, a CNN-LSTM network based on multi-mode data fusion is firstly constructed, the network extracts spatial characteristics by introducing different types of sensor signals such as sound, temperature, images and the like and adopts a convolution layer and a pooling layer, modeling time sequence information is further utilized by the LSTM layer, and finally, classification or regression results are output through a full connection layer, so that accurate monitoring and early warning of the internal state of equipment are realized. Compared with the traditional single-type sensor monitoring method, the intelligent switch cabinet monitoring and fault early warning method based on the multi-mode data fusion has the advantages of being strong in comprehensive analysis capability, high in accuracy, high in efficiency and the like, and the accuracy rate reaches 95% by conducting experiments on a real switch cabinet data set. Meanwhile, the method has good interpretability, and can help a user to interpret and locate equipment state anomalies.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (9)
1. A switch cabinet fault diagnosis method based on multi-mode data fusion is characterized by comprising the following steps:
(1) The operating state of the switch cabinet is monitored to obtain the data information of the switch cabinet.
(2) Preprocessing the data information obtained in the step (1) to obtain preprocessed data information.
(3) Carrying out fusion processing on the data information preprocessed in the step (2) to obtain a multi-mode feature vector;
(4) Inputting the multi-mode feature vector obtained in the step (3) into a pre-trained fault diagnosis model of the switch cabinet to obtain a fault diagnosis result of the switch cabinet.
2. The method for diagnosing a fault of a switchgear based on multi-modal data fusion as claimed in claim 1, wherein,
the data information of the switch cabinet comprises images, sound signals, temperature signals and the like of the switch cabinet;
firstly, converting sound in data information from a time domain signal to a frequency domain signal by adopting short-time Fourier transform (STFT), then extracting local features in a temperature signal by adopting a sliding window method, and finally denoising an image by adopting a Gaussian filtering method.
3. The method for diagnosing a fault of a switchgear based on multi-modal data fusion according to claim 1 or 2, wherein the step (3) specifically comprises the following sub-steps:
and (3-1) sorting the data information preprocessed in the step (2) according to a time sequence, and sampling and interpolating the sorted data information according to a fixed time interval to obtain a set of regular time sequence information.
And (3-2) carrying out normalization processing on the time series information obtained in the step (3-1) to obtain time series information after normalization processing.
And (3-3) classifying the data information subjected to the normalization processing in the step (3-2) to obtain a plurality of channels.
(3-4) performing element-by-element product processing on the plurality of channels obtained in the step (3-3) to obtain the multi-modal feature vector.
4. A method for diagnosing a fault in a switchgear based on multi-modal data fusion as claimed in any one of claims 1 to 3,
the fault diagnosis model of the switch cabinet is a CNN-LSTM network, and the specific structure is as follows:
the first layer is a convolution layer which is input as a three-dimensional tensor X E R containing continuous data (e.g., image, sound, temperature, etc.) B ×T×I Outputting a group of three-dimensional tensor feature graphs X epsilon R generated after extracting the spatial features B×C×F Wherein X represents input continuous data, R represents a real number field of vectors, B represents the number of continuous data, T represents the length of each continuous data, I represents the feature vector size of each time step, C represents the new number of time steps obtained after convolution, and F represents the number of convolution kernels.
The second layer is a pooling layer, the input of which is a three-dimensional tensor feature map X epsilon 788R output by a convolution layer B×C×F Outputting a three-dimensional tensor feature graph X epsilon R with reduced size B1×C1×F1 Wherein B1 represents the data amount, C1 represents the new time step number obtained after pooling, F1 represents the number of convolution kernels
The third layer is an LSTM layer, and the input of the LSTM layer is a three-dimensional characteristic diagram X epsilon R output by the pooling layer B1×C1×F1 Outputting a two-dimensional tensor feature map X epsilon R modeled for time sequence information B×S Where S represents the sequence length of the LSTM layer output.
The fourth layer is a fully connected layer, and the input is a two-dimensional tensor feature map X epsilon R output by the LSTM layer B×S The output is a two-dimensional tensor feature graph X epsilon R B×D Where D represents the output dimension.
5. The switch cabinet fault diagnosis method based on multi-mode data fusion according to claim 4, wherein the switch cabinet fault diagnosis model is obtained through training by the following steps:
(4-1) acquiring a switch cabinet monitoring data set, preprocessing the switch cabinet monitoring data set, dividing the preprocessed switch cabinet monitoring data set into a training set, a verification set and a test set according to the proportion of 8:1:1, and labeling all samples in the switch cabinet monitoring data set.
(4-2) classifying all samples in the training set obtained in the step (4-1) according to the labeling result of the step (4-1), sampling each type of samples according to a fixed time interval to obtain a plurality of sampled samples, sorting and integrating each sampled sample, the sampling time of the sample and the class of the sample into triples according to a time sequence, wherein the triples corresponding to all sampled samples form a triplet set;
(4-3) dividing the triplet set obtained in the step (4-2) through a time window method to obtain a plurality of triples comprising a plurality of time steps and corresponding input features.
And (4-4) inputting the multiple triplets obtained in the step (4-3) into a convolution layer and a pooling layer of the fault diagnosis model of the switch cabinet to perform feature extraction processing so as to obtain multiple feature graphs.
And (4-5) inputting the plurality of feature images obtained in the step (4-4) into an LSTM layer of a fault diagnosis model of the switch cabinet to perform sequence modeling processing so as to obtain output results, wherein the output results corresponding to all the feature images form an output sequence.
(4-6) comparing the output sequence obtained in the step (4-5) with the triplet set of the step (4-2) to obtain a loss function L MSE 。
(4-7) applying a back propagation algorithm to the loss function value L obtained in the step (4-6) MSE And performing updating processing to obtain updated parameters in the convolution layer and the pooling layer and updated parameters in the LSTM layer.
(4-8) performing classifier direct connection processing on the output sequence of the step (4-5) to obtain a classification result.
(4-9) repeating the steps (4-4) to (4-8) until the preset training round number or the loss function converges, thereby obtaining a preliminarily trained fault diagnosis model of the switch cabinet.
And (4-10) verifying the switch cabinet fault diagnosis model which is preliminarily trained in the step (4-9) by using the test set obtained in the step (4-1) until the accuracy of the obtained switch cabinet fault diagnosis model reaches the optimum, thereby obtaining the trained switch cabinet fault diagnosis model.
6. The method for diagnosing a fault in a switchgear based on multi-modal data fusion as recited in claim 5, wherein in step (4-4), the output of the ith (where i e [1, total number of triples obtained in step (4-3 ]) triples after passing through the convolution layer and pooling layer of the switchgear fault diagnosis model is expressed as:
wherein Wj represents the weight matrix of the jth convolution kernel, X i,j Representing the jth input in the ith neuron, b representing the bias vector, σ representing the activation function, pool representing the pooling operation, j e [1 ], the total number of convolution kernels in the convolution layer of the input switch cabinet fault diagnosis model]。
7. The method for diagnosing a fault in a switchgear based on multi-modal data fusion as recited in claim 6, wherein the steps (4-6) are performed by using the following formula:
wherein y is reg,j Representing the true value of the ith triplet of all triples obtained in step (4-3),representing the predicted target variable value of the model for the ith triplet among all triples obtained in step (4-3).
8. The method for diagnosing a fault in a switchgear based on multi-modal data fusion as recited in claim 7, wherein the steps (4-7) are performed by using the following formula:
wherein, alpha represents learning rate, W is updated parameters in the convolution layer and the pooling layer, and U is updated parameters in the LSTM layer.
9. The utility model provides a cubical switchboard fault diagnosis system based on multimodal data fuses which characterized in that includes:
the first module is used for monitoring the operation state of the switch cabinet so as to acquire the data information of the switch cabinet.
And the second module is used for preprocessing the data information acquired by the first module to acquire the preprocessed data information.
The third module is used for carrying out fusion processing on the data information preprocessed by the second module so as to obtain a multi-mode feature vector;
and the fourth module is used for inputting the multi-mode feature vector obtained by the third module into a pre-trained fault diagnosis model of the switch cabinet so as to obtain a fault diagnosis result of the switch cabinet.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310591863.XA CN116610998A (en) | 2023-05-24 | 2023-05-24 | Switch cabinet fault diagnosis method and system based on multi-mode data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310591863.XA CN116610998A (en) | 2023-05-24 | 2023-05-24 | Switch cabinet fault diagnosis method and system based on multi-mode data fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116610998A true CN116610998A (en) | 2023-08-18 |
Family
ID=87679605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310591863.XA Pending CN116610998A (en) | 2023-05-24 | 2023-05-24 | Switch cabinet fault diagnosis method and system based on multi-mode data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116610998A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116930666A (en) * | 2023-09-15 | 2023-10-24 | 深圳凯升联合科技有限公司 | Intelligent diagnosis system and diagnosis method for low-voltage complete switch cabinet |
| CN117095188A (en) * | 2023-10-19 | 2023-11-21 | 中国南方电网有限责任公司 | Electric power safety strengthening method and system based on image processing |
| CN117200458A (en) * | 2023-10-25 | 2023-12-08 | 盛隆电气集团有限公司 | Digital management method and device based on intelligent low-voltage switch cabinet |
| CN117407817A (en) * | 2023-10-25 | 2024-01-16 | 国网四川省电力公司绵阳供电公司 | Abnormality monitoring system of power distribution automation machine room |
| CN117590142A (en) * | 2024-01-19 | 2024-02-23 | 杭州万禾电力科技有限公司 | Switch cabinet fault diagnosis method and system based on deep learning |
| CN117805637A (en) * | 2023-12-29 | 2024-04-02 | 浙江启扬智能科技有限公司 | Battery safety monitoring method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113780060A (en) * | 2021-07-26 | 2021-12-10 | 华北电力大学（保定） | High-voltage switch cabinet situation sensing method based on multi-mode deep learning |
| CN115204272A (en) * | 2022-06-24 | 2022-10-18 | 中南大学 | Industrial system fault diagnosis method and equipment based on multi-sampling rate data |
-
2023
- 2023-05-24 CN CN202310591863.XA patent/CN116610998A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113780060A (en) * | 2021-07-26 | 2021-12-10 | 华北电力大学（保定） | High-voltage switch cabinet situation sensing method based on multi-mode deep learning |
| CN115204272A (en) * | 2022-06-24 | 2022-10-18 | 中南大学 | Industrial system fault diagnosis method and equipment based on multi-sampling rate data |
Non-Patent Citations (3)
| Title |
|---|
| 张昱等: "基于Attention机制的CNN-LSTM时序预测方法研究与应用", 《内蒙古大学学报(自然科学版)》, vol. 53, no. 5, pages 1 - 2 * |
| 柳毅等: "《机器学习与Python实践》", 西安电子科技大学出版社, pages: 202 - 203 * |
| 迷途老书虫: "多模态梳理", 《CSDN:HTTPS://BLOG.CSDN.NET/U011490237/ARTICLE/DETAILS/126928793》, pages 10 - 11 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116930666A (en) * | 2023-09-15 | 2023-10-24 | 深圳凯升联合科技有限公司 | Intelligent diagnosis system and diagnosis method for low-voltage complete switch cabinet |
| CN117095188A (en) * | 2023-10-19 | 2023-11-21 | 中国南方电网有限责任公司 | Electric power safety strengthening method and system based on image processing |
| CN117095188B (en) * | 2023-10-19 | 2023-12-29 | 中国南方电网有限责任公司 | Electric power safety strengthening method and system based on image processing |
| CN117200458A (en) * | 2023-10-25 | 2023-12-08 | 盛隆电气集团有限公司 | Digital management method and device based on intelligent low-voltage switch cabinet |
| CN117407817A (en) * | 2023-10-25 | 2024-01-16 | 国网四川省电力公司绵阳供电公司 | Abnormality monitoring system of power distribution automation machine room |
| CN117805637A (en) * | 2023-12-29 | 2024-04-02 | 浙江启扬智能科技有限公司 | Battery safety monitoring method and system |
| CN117590142A (en) * | 2024-01-19 | 2024-02-23 | 杭州万禾电力科技有限公司 | Switch cabinet fault diagnosis method and system based on deep learning |
| CN117590142B (en) * | 2024-01-19 | 2024-03-22 | 杭州万禾电力科技有限公司 | Switch cabinet fault diagnosis method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116610998A (en) | Switch cabinet fault diagnosis method and system based on multi-mode data fusion | |
| Moradzadeh et al. | Hybrid CNN-LSTM approaches for identification of type and locations of transmission line faults | |
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| CN110829417B (en) | Electric power system transient stability prediction method based on LSTM double-structure model | |
| CN114282579A (en) | Aviation bearing fault diagnosis method based on variational modal decomposition and residual error network | |
| CN105606914A (en) | IWO-ELM-based Aviation power converter fault diagnosis method | |
| CN113702895A (en) | Online quantitative evaluation method for error state of voltage transformer | |
| CN108491931B (en) | Method for improving nondestructive testing precision based on machine learning | |
| CN111273125A (en) | RST-CNN-based power cable channel fault diagnosis method | |
| Son et al. | Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge | |
| CN114237046B (en) | Partial discharge pattern recognition method based on SIFT data feature extraction algorithm and BP neural network model | |
| Xu et al. | Global attention mechanism based deep learning for remaining useful life prediction of aero-engine | |
| Li et al. | Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy | |
| Du et al. | A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals | |
| Yu et al. | Data-feature-driven nonlinear process monitoring based on joint deep learning models with dual-scale | |
| CN115049026A (en) | Regression analysis method of space non-stationarity relation based on GSNNR | |
| CN113705695A (en) | Power distribution network fault data identification method based on convolutional neural network | |
| CN116259161B (en) | Power failure early warning system | |
| CN117219124A (en) | Switch cabinet voiceprint fault detection method based on deep neural network | |
| CN116627116A (en) | Process industry fault positioning method and system and electronic equipment | |
| Liu et al. | A sudden fault detection network based on Time-sensitive gated recurrent units for bearings | |
| Yu et al. | Knowledge reasoning of transmission line component detection using CRITIC and TOPSIS approaches | |
| CN115880472A (en) | Intelligent diagnosis and analysis system for electric power infrared image data | |
| Xu et al. | A graph convolutional shrinkage network-based fault diagnosis method for industrial process | |
| CN114137915A (en) | Fault diagnosis method for industrial equipment |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230818 |