CN113391239A - Transformer abnormality monitoring method and system based on edge calculation - Google Patents
Transformer abnormality monitoring method and system based on edge calculation Download PDFInfo
- Publication number
- CN113391239A CN113391239A CN202110648925.7A CN202110648925A CN113391239A CN 113391239 A CN113391239 A CN 113391239A CN 202110648925 A CN202110648925 A CN 202110648925A CN 113391239 A CN113391239 A CN 113391239A
- Authority
- CN
- China
- Prior art keywords
- data
- edge
- abnormal
- primary side
- center
- 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.)
- Granted
Links
- 238000004364 calculation method Methods 0.000 title claims abstract description 68
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012544 monitoring process Methods 0.000 title claims abstract description 26
- 230000005856 abnormality Effects 0.000 title claims description 16
- 230000002159 abnormal effect Effects 0.000 claims abstract description 118
- 238000013145 classification model Methods 0.000 claims abstract description 87
- 238000000605 extraction Methods 0.000 claims abstract description 63
- 238000013527 convolutional neural network Methods 0.000 claims description 50
- 238000012549 training Methods 0.000 claims description 46
- 238000012545 processing Methods 0.000 claims description 39
- 238000001514 detection method Methods 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012423 maintenance Methods 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
- G01R31/56—Testing of electric apparatus
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R35/00—Testing or calibrating of apparatus covered by the other groups of this subclass
- G01R35/02—Testing or calibrating of apparatus covered by the other groups of this subclass of auxiliary devices, e.g. of instrument transformers according to prescribed transformation ratio, phase angle, or wattage rating
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a transformer abnormity monitoring method and system based on edge calculation, wherein operation data of primary side equipment in normal operation for d days is collected as normal sample data and classified, a CNN characteristic extraction model, an abnormity classification model and a KNN classification model are respectively trained according to classes, abnormity scores of the normal sample data are calculated, and abnormal score distribution parameter estimation of the primary side equipment in normal operation is obtained; collecting real-time operation data of primary side equipment in real time, and performing classification, feature extraction and abnormal score calculation by using each trained model; and comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has faults or not, and sending the prediction result to a centralized data processing center. And the possible faults of the mutual inductor can be predicted in time, so that related maintenance personnel can carry out maintenance work in time.
Description
Technical Field
The invention relates to a mutual inductor abnormity detection technology, in particular to a mutual inductor abnormity monitoring method and system based on edge calculation.
Background
With the continuous development of modern industrial technology and information technology, the traditional power grid gradually develops towards the direction of intellectualization and digitization. The smart grid has advanced measurement and sensing technologies, and the safe and reliable operation of the modern power grid is guaranteed. The mutual inductor is an indispensable ring in the smart power grid, can realize standardization and miniaturization of measuring instruments, protective equipment and automatic control equipment, and can also isolate a high-voltage system to ensure personal safety. Therefore, the running condition of the mutual inductor is monitored, the mutual inductor can run stably, faults are prevented, and the method has very important significance.
The existing transformer fault diagnosis methods are few, and monitoring and evaluation of transformer faults generally adopt manual measurement of transformer states in a power failure state in a certain verification period, but due to the fact that a high-voltage line is powered off, certain economic loss can be caused, so that most transformers in a power grid are in an operating state exceeding the verification period, and state detection is required under the non-power-off condition. In the prior art, most of the mutual inductors are subjected to fault diagnosis by using experience judgment or a simpler machine learning algorithm, and the mutual inductor fault monitoring is carried out by combining primary and secondary side current and voltage information and temperature information acquired by an acquisition module with the machine learning algorithm, so that the deep fault effect with a more complex analysis structure is not ideal. In addition, the diagnostic methods need to perform analysis and judgment on the operation parameters of the mutual inductor in a centralized manner, which causes that a large amount of information acquired by the acquisition module needs to be uploaded to the cloud, occupies a large amount of bandwidth and computing resources, and wastes a large amount of bandwidth resources during data transmission.
Disclosure of Invention
The invention aims to solve the technical problem of how to monitor mutual inductor equipment in real time under the condition of no power failure, and aims to provide a mutual inductor abnormity monitoring method based on edge calculation.
The invention is realized by the following technical scheme:
an edge calculation-based transformer anomaly monitoring method comprises the following steps:
step S1, collecting operation data of the primary side transformer equipment in normal operation for d days as normal sample data, classifying the normal sample data, respectively training a convolutional neural network feature extraction model, an abnormal classification model and a KNN classification model according to classes, calculating abnormal scores of the normal sample data according to the abnormal classification model, and obtaining abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation;
step S2, collecting real-time operation data of the primary side transformer equipment in real time, classifying the operation data by using a trained KNN classification model to obtain data to be detected, sequentially performing data feature extraction and anomaly detection on the data to be detected by using a convolutional neural network feature extraction model and an anomaly classification model, and performing anomaly score calculation according to an anomaly detection result to obtain an anomaly score calculation result;
and step S3, comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has a fault or not, and sending the prediction result to a centralized data processing center.
Further, the specific process of step S1 is:
step S11, collecting operation data of the primary side transformer equipment during normal operation by using a sensor, collecting data of d days of the primary side transformer equipment during normal operation as normal sample data, and transmitting the data to a collector, wherein d is larger than or equal to 90;
step S12, normal sample data are uploaded to the edge classification data center through the collector, and the edge classification data center forwards the normal sample data to the centralized data processing center;
s13, preprocessing normal sample data by the centralized data processing center, and rapidly clustering the normal sample data by using a CFSFDP algorithm to obtain training data with data labels;
step S14, the centralized data processing center classifies the training data according to the data labels, respectively trains the convolutional neural network feature extraction model and the abnormal classification model under the corresponding category according to the classified training data, obtains convolutional neural network feature extraction model parameters and abnormal classification model parameters corresponding to the edge processing data centers under different categories, and sends the obtained model parameters to different edge processing data centers according to categories; for the edge classification data processing center, training the KNN classification model according to the clustering data with the data labels to obtain model parameters of the KNN classification model, and issuing the model parameters of the KNN classification model to the edge classification data center;
step S15, the centralized data processing center calculates the abnormal score of normal sample data according to the abnormal classification model, obtains the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation and sends the abnormal score distribution parameter estimation to the corresponding edge processing data center;
and taking d days as an updating period, and taking the collected running data of the primary side transformer equipment in normal operation as normal sample data every d days, executing the steps, and finishing the updating of the convolutional neural network feature extraction model, the abnormal classification model and the KNN classification model.
Further, the specific process of step S2 is:
step S21, collecting real-time operation data of the primary side transformer equipment by using a sensor, and uploading the real-time operation data to a collector;
step S22, uploading the real-time operation data to an edge classification data center through a collector, adjusting a KNN classification model in the edge classification data center according to KNN classification model parameters issued by a centralized data processing center by the edge classification data center, classifying the real-time operation data according to the KNN model to obtain to-be-detected data, and distributing the to-be-detected data to an edge processing data center for processing the to-be-detected data according to classification;
step S23, the edge processing data center sequentially performs data feature extraction and anomaly detection on the data to be detected according to the adjusted convolutional neural network feature extraction model and the anomaly classification model, and performs anomaly score calculation according to the anomaly detection result to obtain an anomaly score calculation result;
and step S24, comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has a fault or not, and sending the prediction result to a centralized data processing center.
Further, the training process of the convolutional neural network feature extraction model in step S1 specifically includes:
decomposing the training data into one-dimensional time operation data and two-dimensional daily operation data;
inputting one-dimensional time operation data and two-dimensional day operation data into a convolutional neural network feature extraction model, sequentially performing local sensing, parameter sharing and convolutional calculation on the input data by the convolutional neural network feature extraction model, and extracting data features of normal sample data to obtain training data;
the activation function and the optimization function in the convolutional neural network feature extraction model are respectively the ReLU and sgd algorithms, and dropout is used in the multilayer convolution structure of the convolutional neural network feature extraction model.
Further, the abnormal classification model adopts an isolated forest algorithm, and the specific process of obtaining abnormal score distribution parameter estimation by using the abnormal classification model is as follows:
sampling from training data to obtain a plurality of sample points, respectively forming the plurality of sample points into a plurality of sub-samples, respectively constructing an isolated tree for each sub-sample, testing each isolated tree in a forest, and recording the path length of each isolated tree;
calculating an abnormal score by using an abnormal score calculation formula according to the path length of each tree to obtain an abnormal score of each sample point;
and summarizing the abnormal score of each sample point to obtain the corresponding abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation.
Further, the specific process of testing each isolated tree is as follows:
randomly selecting psi data from the training data as subsamples, and putting the subsamples into a root node of an isolated tree;
randomly appointing a dimension, and randomly generating a cutting point p in the range of the current node data, wherein the cutting point p is generated between the maximum value and the minimum value of the appointed dimension in the current node data;
dividing the data space of the current node by using the cutting point p, and repeating the steps until only one piece of data in the subsample is included in the root node corresponding to the current node or the height of the isolated tree reaches log2(Ψ), and recording the path length of the isolated tree at the moment.
Further, calculating an abnormal score, and calculating the abnormal score of each sample point in the training data according to an abnormal score calculation formula, wherein the calculation formula is as follows:
where x represents a sample point in the subsample, h (x) represents the path length that sample point x passes through to reach the leaf node on the ith isolated tree, and E (h (x)) represents the average path length of x over Nt trees in the forest, expressed as:
c (psi) represents the average path length required when a search failure occurs in the binary search tree, and the calculation formula is as follows:
when the abnormal score s (x, ψ) of x is close to 1, the sample point is a discrete point, and the corresponding primary side transformer device is abnormal in operation state;
when the anomaly score s (x, ψ) of x is close to 0, the sample point is a normal point, and the operation state of the corresponding primary side transformer device is normal.
The existing monitoring and evaluation of the fault of the mutual inductor generally adopts the manual measurement of the state of the mutual inductor in a power failure state in a certain verification period, but the power failure of a high-voltage line can cause certain economic loss, so that most of the mutual inductors in a power grid are in an operating state exceeding the verification period, and the state detection is required under the non-power-off condition. In the prior art, fault classification of the mutual inductor is rarely realized by adopting a machine learning algorithm based on an edge computing frame, in a mutual inductor fault monitoring method carried out by using a non-manual method, the fault monitoring of the mutual inductor is mostly carried out by combining primary and secondary side current and voltage information and temperature information acquired by an acquisition module with the machine learning algorithm, so that a large amount of information acquired by the acquisition module needs to be uploaded to a cloud end, and a large amount of bandwidth and computing resources are occupied. Simple fault judgment is through artificial mode, needs the staff to carry equipment to rush to the scene, and the dismouting mutual-inductor is once led wire, and is inefficient, and unable accurate judgement mutual-inductor running state, influences electric power system's safe operation and electric energy measurement's fair and just. According to the invention, the composite convolutional neural network capable of extracting the periodic regularity is selected as the feature extraction model, the isolated forest algorithm which is suitable for carrying out abnormity discrimination and can grade the abnormal state is selected as the fault classification model, the bandwidth pressure is reduced, the information of the primary physical power grid of the mutual inductor is collected, the running state and the running trend of the current mutual inductor are judged, the feature is extracted according to the running parameters of the front mutual inductor and the rear mutual inductor, the mutual inductor with the running trend reduced and the running state poorer is predicted in time, the possible faults of the mutual inductor can be effectively predicted, and the related maintenance personnel can carry out maintenance work in time. Most of calculation tasks are concentrated in the edge calculation center, calculation results are uploaded to the cloud for early warning, transmitted data are reduced, and bandwidth resources can be saved; in addition, the system model is updated regularly, so that errors in the prediction process can be reduced.
In addition, the invention provides a transformer abnormity monitoring system based on edge calculation, which comprises primary side transformer equipment, a sensor, a collector, an edge data center and a centralized data processing center, wherein,
the primary side transformer equipment is used for providing normal sample data and real-time operation data for normal operation;
the sensor is used for collecting real-time operation data and normal sample data of the primary side mutual inductor equipment, and collecting and transmitting the real-time operation data and the normal sample data to the collector;
the collector is used for uploading real-time operation data and normal sample data of the primary side mutual inductor equipment to the edge data center;
the edge data center is used for uploading normal sample data to the centralized data processing center, downloading KNN classification model parameters and convolution neural network characteristics under corresponding categories from the centralized data processing center, extracting model parameters and abnormal classification model parameters, classifying real-time operation data according to the downloaded model, calculating abnormal scores, and sending an alarm to the centralized data processing center according to a calculation result;
the centralized data processing center is used for regularly receiving normal sample data in normal operation from the edge data center, clustering all the normal sample data, respectively training a convolutional neural network feature extraction model and an abnormal classification model according to classes by utilizing the clustered data to obtain convolutional neural network feature extraction model parameters and abnormal classification model parameters under different classes, calculating abnormal scores of the normal sample data according to the abnormal classification model to obtain abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, sending the obtained model parameters and the abnormal score distribution parameter estimation of the primary side transformer equipment to the edge data center according to the classes, training a KNN classification model according to the normal sample data to obtain parameters of the KNN classification model, and sending the parameters to the edge data center.
Further, the edge data center comprises an edge classification data center and an edge processing data center,
the edge classification data center is used for uploading normal sample data to the centralized data processing center, classifying real-time operation data according to the trained KNN classification model to obtain data to be detected, and distributing the data to be detected to the edge processing data center under corresponding classification;
the edge processing data center comprises a model training module, a data processing module and an early warning module,
the model training module is used for downloading convolutional neural network feature extraction model parameters and abnormal classification model parameters of the edge processing data center under corresponding classification from the centralized data processing center, and adjusting the corresponding convolutional neural network feature extraction model and the corresponding abnormal classification model according to the model parameters;
the data processing module is used for carrying out data feature extraction and anomaly detection on the data to be detected distributed from the edge classification data center according to the adjusted convolutional neural network feature extraction model and the anomaly classification model, and carrying out anomaly score calculation according to an anomaly detection result to obtain an anomaly score calculation result;
the early warning module is used for comparing the obtained abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side mutual inductor equipment, predicting whether the corresponding primary side mutual inductor equipment fails or not, and sending the prediction result to the centralized data processing center.
Compared with the prior art, the invention has the following advantages and beneficial effects:
according to the transformer abnormity monitoring method and system based on edge calculation, calculation resources are provided at the network edge, a composite convolutional neural network capable of extracting periodic regularity is selected as a feature extraction model, an isolated forest algorithm which is suitable for abnormity discrimination and can grade an abnormal state is selected as a fault classification model to monitor the operation state of a transformer in real time, so that the bandwidth pressure is reduced, the characteristics are extracted according to the operation parameters of front and rear transformer equipment, the transformer with a reduced operation trend and a poorer operation state is predicted in time, and possible faults are predicted in time, so that related maintenance personnel can perform maintenance work in time.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a schematic diagram of the overall system framework of the present invention;
fig. 3 is a schematic diagram of fault monitoring acquisition.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that: it is not necessary to employ these specific details to practice the present invention. In other instances, well-known structures, circuits, materials, or methods have not been described in detail so as not to obscure the present invention.
Throughout the specification, reference to "one embodiment," "an embodiment," "one example," or "an example" means: the particular features, structures, or characteristics described in connection with the embodiment or example are included in at least one embodiment of the invention. Thus, the appearances of the phrases "one embodiment," "an embodiment," "one example" or "an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combination and/or sub-combination in one or more embodiments or examples. Further, those of ordinary skill in the art will appreciate that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
In the description of the present invention, it is to be understood that the terms "front", "rear", "left", "right", "upper", "lower", "vertical", "horizontal", "high", "low", "inner", "outer", etc. indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplicity of description, and do not indicate or imply that the referenced devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and therefore, are not to be construed as limiting the scope of the present invention.
Example 1
As shown in fig. 1, the method for monitoring transformer abnormality based on edge calculation of the present invention includes the following steps:
step S1, collecting operation data of primary side transformer equipment (such as a current transformer) which normally operates for d days as normal sample data, classifying the normal sample data, respectively training a convolutional neural network feature extraction model, an abnormal classification model and a KNN classification model according to classes, calculating abnormal scores of the normal sample data according to the abnormal classification model, and obtaining abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation; and with d days as an updating period, taking the collected running data of the primary side transformer equipment in normal operation as normal sample data every d days, executing the following steps, and updating the convolutional neural network feature extraction model, the abnormal classification model and the KNN classification model, wherein the specific process is as follows:
step S11, collecting operation data (such as voltage and current actual values of the transformer) of the primary side transformer equipment during normal operation by using a sensor, summarizing data of the primary side transformer equipment during normal operation for d days as normal sample data, and transmitting the data to a collector, wherein d is larger than or equal to 90;
step S12, normal sample data are uploaded to the edge classification data center through the collector, and the edge classification data center forwards the normal sample data to the centralized data processing center;
s13, preprocessing normal sample data by the centralized data processing center, and rapidly clustering the normal sample data by using a CFSFDP algorithm to obtain training data with data labels;
step S14, the centralized data processing center classifies the training data according to the data labels, respectively trains the convolutional neural network feature extraction model and the abnormal classification model under the corresponding category according to the classified training data, obtains convolutional neural network feature extraction model parameters and abnormal classification model parameters corresponding to the edge processing data centers under different categories, and sends the obtained model parameters to different edge processing data centers according to categories; for the edge classification data processing center, training the KNN classification model according to the clustering data with the data labels to obtain model parameters of the KNN classification model, and issuing the model parameters of the KNN classification model to the edge classification data center;
and step S15, the centralized data processing center calculates the abnormal score of normal sample data according to the abnormal classification model, obtains the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation and sends the abnormal score distribution parameter estimation to the corresponding edge processing data center.
Step S2, collecting real-time operation data of the primary side transformer equipment in real time, classifying the operation data by using a trained KNN classification model to obtain data to be detected, sequentially performing data feature extraction and anomaly detection on the data to be detected by using a convolutional neural network feature extraction model and an anomaly classification model, and performing anomaly score calculation according to an anomaly detection result to obtain an anomaly score calculation result;
step S21, collecting real-time operation data of the primary side transformer equipment by using a sensor, and uploading the real-time operation data to a collector;
step S22, uploading the real-time operation data to an edge classification data center through a collector, adjusting a KNN classification model in the edge classification data center according to KNN classification model parameters issued by a centralized data processing center by the edge classification data center, classifying the real-time operation data according to the KNN model to obtain to-be-detected data, and distributing the to-be-detected data to an edge processing data center for processing the to-be-detected data according to classification;
step S23, the edge processing data center sequentially performs data feature extraction and anomaly detection on the data to be detected according to the adjusted convolutional neural network feature extraction model and the anomaly classification model, and performs anomaly score calculation according to the anomaly detection result to obtain an anomaly score calculation result;
and step S24, comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has a fault or not, and sending the prediction result to a centralized data processing center.
And step S3, comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has a fault or not, and sending the prediction result to a centralized data processing center.
Specifically, clustering using the fast search density peak clustering technique (CFSFDP algorithm) at the centralized processing center is divided into 2 steps, including PCA feature extraction, finding density peak points and classifying according to distance.
And obtaining an equipment operation characteristic matrix V by PCA characteristic extraction, wherein the form of V is as follows:
wherein: k represents the characteristic number of the equipment, and the equipment characteristics are the operation parameters of the transformer, such as current, voltage, maximum value, minimum value, standard deviation of a current-voltage sequence and the like; v in matrixuifkRepresenting the value of user i on feature k. Searching density peak points and classifying according to the distance, and defining the density peak points according to the following formula:
δ (i) is the peak distance, the CFSFDP algorithm calculates the local density ρ and the higher density distance δ, maps the data set into a two-dimensional graph and constructs a decision graph for selection; in the decision diagram, the point where both ρ and δ are large (the point on the upper right) is the cluster center. And after the clustering center is selected, distributing the remaining points to the nearest clustering center to complete clustering.
The training process using the CNN convolutional neural network feature extraction model in step S1 specifically includes:
decomposing the training data into one-dimensional time operation data, and converting the one-dimensional time operation data into two-dimensional daily operation data; the features are convenient to extract, wherein the one-dimensional time operation data input shape is as follows:
M1×d＝[P1 P2 … Pd] (5)
wherein h is the total operating hours of the primary side equipment;
the two-dimensional daily operation data input shape is as follows:
wherein d is the total operation days of the primary side equipment, and d is more than or equal to 90;
inputting one-dimensional time operation data and two-dimensional day operation data into a convolutional neural network feature extraction model, sequentially performing local sensing, parameter sharing and convolutional calculation on the input data by the convolutional neural network feature extraction model, greatly reducing the number of parameters by utilizing the parameter sharing to obtain an effective feature extraction result, and extracting data features of normal sample data to obtain training data;
the activation function and the optimization function in the convolutional neural network feature extraction model are respectively ReLU and sgd algorithms, and dropout is used in a multilayer convolution structure of the convolutional neural network feature extraction model; the probability of the occurrence of the over-fitting problem is reduced; wherein the formula for ReLU is as follows:
wherein, yjIs the output of the complete connection layer in the jth neuron, n is the one-dimensional input data length, wi，jRepresenting the neuron weight between the 1 st input value and the jth neuron, b1Is a deviation.
In this embodiment, the abnormal classification model adopts an isolated forest algorithm, and the specific process of obtaining the abnormal score distribution parameter estimation by using the abnormal classification model is as follows:
sampling from training data to obtain a plurality of sample points, respectively constructing an isolated tree for each sample point, testing each isolated tree in a forest, and recording the path length of each isolated tree;
calculating an abnormal score by using an abnormal score calculation formula according to the path length of each tree to obtain an abnormal score of each sample point;
and summarizing the abnormal score of each sample point to obtain the corresponding abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation.
Specifically, the specific process of testing each isolated tree is as follows:
randomly selecting psi data from the training data as subsamples, and putting the subsamples into a root node of an isolated tree;
randomly appointing a dimension, and randomly generating a cutting point p in the range of the current node data, wherein the cutting point p is generated between the maximum value and the minimum value of the appointed dimension in the current node data;
dividing the data space of the current node by using the cutting point p, and repeating the steps until only one piece of data in the subsample is included in the root node corresponding to the current node or the height of the isolated tree reaches log2(Ψ), and recording the path length of the isolated tree at the moment.
Calculating an abnormal score, and calculating the abnormal score of each sample point in the training data according to an abnormal score calculation formula, wherein the calculation formula is as follows:
where h (x) represents the path length through to the leaf node on the ith isolated tree, and E (h (x)) represents the average of the path lengths of x on Nt trees, and is expressed as:
c (psi) represents the average path length required when a search failure occurs in the binary search tree, and the calculation formula is as follows:
when the abnormal score s (x, ψ) of x is close to 1, the sample point is a discrete point, and the corresponding primary side transformer device is abnormal in operation state;
when the anomaly score s (x, ψ) of x is close to 0, the sample point is a normal point, and the operation state of the corresponding primary side transformer device is normal.
When the transformer parameters are collected, for example, for a current transformer and a voltage transformer, the system needs to collect current signals and voltage signals as the basis for judging whether the transformer normally operates, and the fault monitoring and collecting principle is as shown in fig. 3:
in fig. 3, CT is a current transformer, PT is a voltage transformer, and the current of the a phase of the transmission line is IAThe current of the C phase is ICThe metering unit can count the voltage UaAnd UcSecondary monitor signal u for CT1 and CT2aAnd ucThen uploading the counted current and voltage signals to an edge data center;
before model training, data needs to be preprocessed, generally, input and output of a neural network are within [0, 1], but parameters such as system current and voltage are not between [0, 1], so that normalization processing needs to be performed on the acquired data, and a normalization processing method is shown as the following formula:
wherein, Z represents specific input data, such as current, voltage and the like, so that the current and voltage values after data processing can improve the efficiency and accuracy of fault judgment.
Example 2
As shown in fig. 2, the present embodiment is different from embodiment 1 in that the present invention provides an edge calculation-based transformer abnormality monitoring system, which includes a primary-side transformer device, a sensor, a collector, an edge data center, and a centralized data processing center, wherein,
the primary side transformer equipment is used for providing normal sample data and real-time operation data for normal operation;
the sensor is used for collecting real-time operation data and normal sample data of the primary side mutual inductor equipment, and collecting and transmitting the real-time operation data and the normal sample data to the collector;
the collector is used for uploading real-time operation data and normal sample data of the primary side mutual inductor equipment to the edge data center;
the edge data center comprises an edge classification data center and an edge processing data center, the edge classification data center is used for uploading normal sample data to the centralized data processing center, classifying real-time operation data according to the trained KNN classification model to obtain data to be detected, and distributing the data to be detected to the edge processing data center under the corresponding classification;
the edge processing data center comprises a model training module, a data processing module and an early warning module,
the model training module is used for downloading convolutional neural network feature extraction model parameters and abnormal classification model parameters of the edge processing data center under corresponding classification from the centralized data processing center, and adjusting the corresponding convolutional neural network feature extraction model and the corresponding abnormal classification model according to the model parameters;
the data processing module is used for carrying out data feature extraction and anomaly detection on the data to be detected distributed from the edge classification data center according to the adjusted convolutional neural network feature extraction model and the anomaly classification model, and carrying out anomaly score calculation according to an anomaly detection result to obtain an anomaly score calculation result;
the early warning module is used for comparing the obtained abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment, predicting whether the corresponding primary side transformer equipment fails or not, and sending the prediction result to the centralized data processing center;
the centralized data processing center is used for receiving normal sample data in normal operation from the edge data center at regular intervals (for example, 90 days to 120 days), clustering all normal sample data, respectively training a convolutional neural network feature extraction model and an abnormal classification model according to classes by utilizing the clustered data to obtain convolutional neural network feature extraction model parameters and abnormal classification model parameters under different classes, calculating the abnormal score of normal sample data according to the abnormal classification model to obtain the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, and the obtained model parameters and the abnormal score distribution parameter estimation of the primary side transformer equipment are sent to an edge processing data center according to the classes, and training the KNN classification model according to the normal sample data to obtain parameters of the KNN classification model and issuing the parameters to the edge classification data center.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (9)
1. An edge calculation-based transformer anomaly monitoring method is characterized by comprising the following steps:
step S1, collecting operation data of the primary side transformer equipment in normal operation for d days as normal sample data, wherein d is larger than or equal to 90, classifying the normal sample data, respectively training a convolutional neural network feature extraction model, an abnormal classification model and a KNN classification model according to classes, calculating abnormal scores of the normal sample data according to the abnormal classification model, and obtaining abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation;
step S2, collecting real-time operation data of the primary side transformer equipment in real time, classifying the operation data by using a trained KNN classification model to obtain data to be detected, sequentially performing data feature extraction and anomaly detection on the data to be detected by using a convolutional neural network feature extraction model and an anomaly classification model, and performing anomaly score calculation according to an anomaly detection result to obtain an anomaly score calculation result;
and step S3, comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has a fault or not, and sending the prediction result to a centralized data processing center.
2. The transformer abnormality monitoring method based on the edge calculation according to claim 1, wherein the specific process of step S1 is as follows:
step S11, collecting operation data of the primary side transformer equipment during normal operation by using a sensor, collecting data of d days of the primary side transformer equipment during normal operation as normal sample data, and transmitting the data to a collector, wherein d is larger than or equal to 90;
step S12, normal sample data are uploaded to the edge classification data center through the collector, and the edge classification data center forwards the normal sample data to the centralized data processing center;
s13, preprocessing normal sample data by the centralized data processing center, and rapidly clustering the normal sample data by using a CFSFDP algorithm to obtain training data with data labels;
step S14, the centralized data processing center classifies the training data according to the data labels, respectively trains the convolutional neural network feature extraction model and the abnormal classification model under the corresponding category according to the classified training data, obtains convolutional neural network feature extraction model parameters and abnormal classification model parameters corresponding to the edge processing data centers under different categories, and sends the obtained model parameters to different edge processing data centers according to categories; for the edge classification data processing center, training the KNN classification model according to the clustering data with the data labels to obtain model parameters of the KNN classification model, and issuing the model parameters of the KNN classification model to the edge classification data center;
and step S15, the centralized data processing center calculates the abnormal score of normal sample data according to the abnormal classification model, obtains the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation and sends the abnormal score distribution parameter estimation to the corresponding edge processing data center.
3. The transformer abnormality monitoring method based on the edge calculation according to claim 1, wherein the specific process of step S2 is as follows:
step S21, collecting real-time operation data of the primary side transformer equipment by using a sensor, and uploading the real-time operation data to a collector;
step S22, uploading the real-time operation data to an edge classification data center through a collector, adjusting a KNN classification model in the edge classification data center according to KNN classification model parameters issued by a centralized data processing center by the edge classification data center, classifying the real-time operation data according to the KNN model to obtain to-be-detected data, and distributing the to-be-detected data to an edge processing data center for processing the to-be-detected data according to classification;
step S23, the edge processing data center sequentially performs data feature extraction and anomaly detection on the data to be detected according to the adjusted convolutional neural network feature extraction model and the anomaly classification model, and performs anomaly score calculation according to the anomaly detection result to obtain an anomaly score calculation result;
and step S24, comparing the abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, predicting whether the primary side transformer equipment has a fault or not, and sending the prediction result to a centralized data processing center.
4. The method for monitoring transformer abnormality based on edge calculation according to claim 1, wherein the training process of the convolutional neural network feature extraction model in step S1 is specifically as follows:
decomposing the training data into one-dimensional time operation data and two-dimensional daily operation data;
inputting one-dimensional time operation data and two-dimensional day operation data into a convolutional neural network feature extraction model, sequentially performing local sensing, parameter sharing and convolutional calculation on the input data by the convolutional neural network feature extraction model, and extracting data features of normal sample data to obtain training data;
the activation function and the optimization function in the convolutional neural network feature extraction model are respectively the ReLU and sgd algorithms, and dropout is used in the multilayer convolution structure of the convolutional neural network feature extraction model.
5. The transformer abnormality monitoring method based on edge calculation as claimed in claim 4, wherein the abnormality classification model adopts an isolated forest algorithm, and the specific process of obtaining the abnormal score distribution parameter estimation by using the abnormality classification model is as follows:
sampling from training data to obtain a plurality of sample points, respectively forming the plurality of sample points into a plurality of sub-samples, respectively constructing an isolated tree for each sub-sample, testing each isolated tree in a forest, and recording the path length of each isolated tree;
calculating an abnormal score by using an abnormal score calculation formula according to the path length of each tree to obtain an abnormal score of each sample point;
and summarizing the abnormal score of each sample point to obtain the corresponding abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation.
6. The transformer abnormality monitoring method based on edge calculation according to claim 5, characterized in that the specific process of testing each isolated tree is as follows:
randomly selecting psi data from the training data as subsamples, and putting the subsamples into a root node of an isolated tree;
randomly appointing a dimension, and randomly generating a cutting point p in the range of the current node data, wherein the cutting point p is generated between the maximum value and the minimum value of the appointed dimension in the current node data;
dividing the data space of the current node by using the cutting point p, and repeating the steps until only one piece of data in the subsample is included in the root node corresponding to the current node or the height of the isolated tree reaches log2(Ψ), and recording the path length of the isolated tree at the moment.
7. The transformer abnormality monitoring method based on the edge calculation is characterized in that an abnormality score is calculated, and an abnormality score of each sample point in training data is calculated according to an abnormality score calculation formula, wherein the calculation formula is as follows:
where x represents a sample point in the subsample, h (x) represents the path length that sample point x passes through to reach the leaf node on the ith isolated tree, and E (h (x)) represents the average path length of x over Nt trees in the forest, expressed as:
c (psi) represents the average path length required when a search failure occurs in the binary search tree, and the calculation formula is as follows:
when the abnormal score s (x, ψ) of x is close to 1, the sample point is a discrete point, and the corresponding primary side transformer device is abnormal in operation state;
when the anomaly score s (x, ψ) of x is close to 0, the sample point is a normal point, and the operation state of the corresponding primary side transformer device is normal.
8. An edge calculation-based transformer anomaly monitoring system is characterized by comprising primary side transformer equipment, a sensor, a collector, an edge data center and a centralized data processing center, wherein,
the primary side transformer equipment is used for providing normal sample data and real-time operation data for normal operation;
the sensor is used for collecting real-time operation data and normal sample data of the primary side mutual inductor equipment, and collecting and transmitting the real-time operation data and the normal sample data to the collector;
the collector is used for uploading real-time operation data and normal sample data of the primary side mutual inductor equipment to the edge data center;
the edge data center is used for uploading normal sample data to the centralized data processing center, downloading KNN classification model parameters and convolution neural network characteristics under corresponding categories from the centralized data processing center, extracting model parameters and abnormal classification model parameters, classifying real-time operation data according to the downloaded model, calculating abnormal scores, and sending an alarm to the centralized data processing center according to a calculation result;
the centralized data processing center is used for regularly receiving normal sample data in normal operation from the edge data center, clustering all the normal sample data, respectively training a convolutional neural network feature extraction model and an abnormal classification model according to classes by utilizing the clustered data to obtain convolutional neural network feature extraction model parameters and abnormal classification model parameters under different classes, calculating abnormal scores of the normal sample data according to the abnormal classification model to obtain abnormal score distribution parameter estimation of the primary side transformer equipment in normal operation, sending the obtained model parameters and the abnormal score distribution parameter estimation of the primary side transformer equipment to the edge data center according to the classes, training a KNN classification model according to the normal sample data to obtain parameters of the KNN classification model, and sending the parameters to the edge data center.
9. The system for monitoring transformer abnormality based on edge calculation according to claim 8,
the edge data center comprises an edge classification data center and an edge processing data center,
the edge classification data center is used for uploading normal sample data to the centralized data processing center, classifying real-time operation data according to the trained KNN classification model to obtain data to be detected, and distributing the data to be detected to the edge processing data center under corresponding classification;
the edge processing data center comprises a model training module, a data processing module and an early warning module,
the model training module is used for downloading convolutional neural network feature extraction model parameters and abnormal classification model parameters of the edge processing data center under corresponding classification from the centralized data processing center, and adjusting the corresponding convolutional neural network feature extraction model and the corresponding abnormal classification model according to the model parameters;
the data processing module is used for carrying out data feature extraction and anomaly detection on the data to be detected distributed from the edge classification data center according to the adjusted convolutional neural network feature extraction model and the anomaly classification model, and carrying out anomaly score calculation according to an anomaly detection result to obtain an anomaly score calculation result;
the early warning module is used for comparing the obtained abnormal score calculation result with the abnormal score distribution parameter estimation of the primary side mutual inductor equipment, predicting whether the corresponding primary side mutual inductor equipment fails or not, and sending the prediction result to the centralized data processing center.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110648925.7A CN113391239B (en) | 2021-06-10 | 2021-06-10 | Mutual inductor anomaly monitoring method and system based on edge calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110648925.7A CN113391239B (en) | 2021-06-10 | 2021-06-10 | Mutual inductor anomaly monitoring method and system based on edge calculation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113391239A true CN113391239A (en) | 2021-09-14 |
| CN113391239B CN113391239B (en) | 2024-01-23 |
Family
ID=77620272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110648925.7A Active CN113391239B (en) | 2021-06-10 | 2021-06-10 | Mutual inductor anomaly monitoring method and system based on edge calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113391239B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116256690A (en) * | 2023-02-15 | 2023-06-13 | 广州市德珑电子器件有限公司 | Current transformer safety detection method and system |
| CN116699318A (en) * | 2023-07-27 | 2023-09-05 | 成都汉度科技有限公司 | Electric leakage detection method and system based on edge calculation |
| CN117148775A (en) * | 2023-10-31 | 2023-12-01 | 中国电建集团山东电力管道工程有限公司 | Remote monitoring method, system, equipment and medium for pipeline production process |
| CN117216673A (en) * | 2023-11-08 | 2023-12-12 | 国网江西省电力有限公司电力科学研究院 | Current transformer monitoring evaluation overhauls platform |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020133639A1 (en) * | 2018-12-29 | 2020-07-02 | 东北大学 | Intelligent analysis system for magnetic flux leakage detection data in pipeline |
| CN111522804A (en) * | 2020-04-15 | 2020-08-11 | 国网江苏省电力有限公司 | Cleaning method for abnormal data of transformer equipment state monitoring |
-
2021
- 2021-06-10 CN CN202110648925.7A patent/CN113391239B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020133639A1 (en) * | 2018-12-29 | 2020-07-02 | 东北大学 | Intelligent analysis system for magnetic flux leakage detection data in pipeline |
| CN111522804A (en) * | 2020-04-15 | 2020-08-11 | 国网江苏省电力有限公司 | Cleaning method for abnormal data of transformer equipment state monitoring |
Non-Patent Citations (3)
| Title |
|---|
| 余翔等: "基于孤立森林算法的用电数据异常检测研究", 《信息技术》, pages 88 * |
| 刘凤魁等: "基于改进快速密度峰值聚类算法的 电力大数据异常值检测", 《电力信息与通信技术》, vol. 15, no. 6, pages 36 * |
| 白昱阳等: "云边智能: 电力系统运行控制的边缘计算方法 及其应用现状与展望", 《自动化学报》, vol. 46, no. 3, pages 397 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116256690A (en) * | 2023-02-15 | 2023-06-13 | 广州市德珑电子器件有限公司 | Current transformer safety detection method and system |
| CN116699318A (en) * | 2023-07-27 | 2023-09-05 | 成都汉度科技有限公司 | Electric leakage detection method and system based on edge calculation |
| CN116699318B (en) * | 2023-07-27 | 2023-10-20 | 成都汉度科技有限公司 | Electric leakage detection method and system based on edge calculation |
| CN117148775A (en) * | 2023-10-31 | 2023-12-01 | 中国电建集团山东电力管道工程有限公司 | Remote monitoring method, system, equipment and medium for pipeline production process |
| CN117148775B (en) * | 2023-10-31 | 2024-01-23 | 中国电建集团山东电力管道工程有限公司 | Remote monitoring method, system, equipment and medium for pipeline production process |
| CN117216673A (en) * | 2023-11-08 | 2023-12-12 | 国网江西省电力有限公司电力科学研究院 | Current transformer monitoring evaluation overhauls platform |
| CN117216673B (en) * | 2023-11-08 | 2024-03-12 | 国网江西省电力有限公司电力科学研究院 | Current transformer monitoring evaluation overhauls platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113391239B (en) | 2024-01-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113391239B (en) | Mutual inductor anomaly monitoring method and system based on edge calculation | |
| CN113255795B (en) | Equipment state monitoring method based on multi-index cluster analysis | |
| CN113156917B (en) | Power grid equipment fault diagnosis method and system based on artificial intelligence | |
| CN110837866A (en) | XGboost-based electric power secondary equipment defect degree evaluation method | |
| CN103699698A (en) | Method and system for track traffic failure recognition based on improved Bayesian algorithm | |
| CN111639467A (en) | Aero-engine service life prediction method based on long-term and short-term memory network | |
| CN111209934B (en) | Fan fault pre-alarm method and system | |
| CN111650451A (en) | Converter transformer fault reason identification method and system | |
| CN115409131B (en) | Production line abnormity detection method based on SPC process control system | |
| CN114723285A (en) | Power grid equipment safety evaluation prediction method | |
| CN108334894B (en) | Unsupervised machine learning-based transformer oil temperature abnormity identification method | |
| CN114856941A (en) | Offshore wind power plant and unit fault diagnosis operation and maintenance system and diagnosis operation and maintenance method thereof | |
| CN110737976A (en) | mechanical equipment health assessment method based on multi-dimensional information fusion | |
| CN110580492A (en) | Track circuit fault precursor discovery method based on small fluctuation detection | |
| CN112990656A (en) | Health evaluation system and health evaluation method for IT equipment monitoring data | |
| CN111767657B (en) | Nuclear power system fault diagnosis method and system | |
| CN113191585A (en) | Typhoon disaster risk assessment method for power transmission line | |
| CN114487705A (en) | Power grid equipment fault positioning detection method | |
| CN108446202B (en) | Method for judging safety condition of machine room equipment | |
| CN117277566B (en) | Power grid data analysis power dispatching system and method based on big data | |
| CN117151684A (en) | Wind power fan data analysis early warning method, system, device and readable storage medium | |
| CN115941807B (en) | Efficient data compression method for park security system | |
| CN117113135A (en) | Carbon emission anomaly monitoring and analyzing system capable of sorting and classifying anomaly data | |
| CN111459697A (en) | Excitation system fault monitoring method based on deep learning network | |
| CN115713027A (en) | Transformer state evaluation method, device and system |
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 |