CN112668715A - Turnout switch machine abnormity diagnosis method and system based on machine learning - Google Patents
Turnout switch machine abnormity diagnosis method and system based on machine learning Download PDFInfo
- Publication number
- CN112668715A CN112668715A CN202011578819.8A CN202011578819A CN112668715A CN 112668715 A CN112668715 A CN 112668715A CN 202011578819 A CN202011578819 A CN 202011578819A CN 112668715 A CN112668715 A CN 112668715A
- Authority
- CN
- China
- Prior art keywords
- curve
- data
- working characteristic
- fault
- characteristic curve
- 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
Images
Landscapes
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The invention provides a turnout switch machine abnormity diagnosis method based on machine learning, which comprises the following steps: collecting multiple types of working characteristic curves of the turnout switch machine during action for multiple times to establish multiple working characteristic curve groups, and generating corresponding curve data for each working characteristic curve group based on the selected reference working characteristic curve group; putting curve data into a normal data pool or a fault data pool based on a label set for the curve data; curve data in a normal and fault data pool are sampled to establish a training data set and a testing data set; pre-training a convolutional neural network model through a training data set and a testing data set, wherein the model is used for identifying whether a fault and a fault type occur during the action of a turnout switch machine; and processing the working characteristic curve group to be diagnosed to obtain curve data of the working characteristic curve group, and inputting the curve data into a trained convolutional neural network model to obtain a corresponding turnout switch machine fault judgment result. The invention also provides a turnout switch machine abnormity diagnosis system based on machine learning.
Description
Technical Field
The invention relates to the technical field of rail transit switch machine equipment detection, in particular to a turnout switch machine abnormity diagnosis method and system based on machine learning.
Background
With the rapid development of railways in China, higher requirements are put forward on the safety and reliability of railway signal equipment. Switches are key components of railway signaling equipment, and the state of the switches directly influences the safety and efficiency of railway transportation. The turnout for switching the train from one track to another is a key device for arranging the train route and realizing the route switching. The point switch is the switching equipment of the switch, used for realizing switching the switch, locking the switch and reflecting the position of the switch tongue. The failure is one of the main reasons causing major accidents of railways.
At present, switches are overhauled by combining two modes of monitoring the action state of the switches by a microcomputer monitoring system and manually and periodically checking. The microcomputer monitoring system collects the data of the switch such as the action voltage, the action current and the like and displays the action current curve and the power curve of the switch in the system. The action current curve and the power curve of a turnout switch machine in a microcomputer monitoring system are actively called by individual experienced workers, and the called curves are analyzed to judge whether the switch machine has faults or potential safety hazards.
The microcomputer monitoring system in the prior art can not predict the fault of the turnout switch machine in advance, but the fault diagnosis and prediction method based on the regular troubleshooting and the analysis of curve information stored in the microcomputer monitoring system by workers is very dependent on the experience knowledge and professional level of the workers. The operator generally only has the possibility of identifying the turnout fault under the condition that the difference between the collected voltage, current and power curves of the switch machine and the reference curve is obvious (such as the slope of the current curve of the switch machine is steeply increased). The same collected curve may represent different operation conditions for different devices, and particularly, misjudgment often occurs depending on personal experience for identifying switch machine faults under special working conditions and environments. The precious processing experience of workers can only exist in a certain operation and maintenance unit in an isolated mode or only be known by some people, the turnout fault can occur and be solved on other previous lines, when the problem occurs again, due to personnel flowing or other reasons, previous knowledge cannot be effectively utilized, repeated faults and fault time delay are formed, and therefore the traveling efficiency of passengers is delayed.
In summary, the abnormal diagnosis method for the turnout switch machine in the prior art has the disadvantages of large workload, low efficiency, and low reliability of fault judgment of the turnout switch machine, and cannot meet the requirements of the existing high-speed railway on operation safety and efficiency.
Disclosure of Invention
The invention aims to provide a turnout switch machine abnormity diagnosis method and system based on machine learning, which can intelligently analyze a three-phase current characteristic curve and a three-phase power characteristic curve acquired during the action of the turnout switch machine based on a convolutional neural network model, automatically identify the fault and hidden danger of the turnout switch machine, and greatly improve the efficiency of turnout switch machine abnormity diagnosis and the accuracy of diagnosis results.
In order to achieve the above object, the present invention provides a method for diagnosing abnormality of a point switch machine based on machine learning, comprising the steps of:
collecting multiple types of working characteristic curves of the turnout switch machine during action for multiple times, wherein the collected multiple types of working characteristic curves at one time are used as a working characteristic curve group; selecting a group of working characteristic curves as a reference working characteristic curve group corresponding to the action type based on the action type of the point switch; normalizing the operating characteristic curve; preprocessing curve data of each group of working characteristic curves in a matrix form based on the action type of the point switch and the corresponding reference working characteristic curve group;
manually setting a label for the curve data, wherein the label is used for indicating whether the turnout switch machine is in fault and the fault type; based on the label, putting curve data into a normal data pool or a fault data pool;
sampling curve data in the normal and fault data pools to establish a training data set and a testing data set;
pre-training a convolutional neural network model through the training data set and the testing data set, wherein the convolutional neural network model is used for identifying whether a fault and a fault type occur during the action of a turnout switch machine;
and processing the working characteristic curve group to be diagnosed, acquiring corresponding curve data, and inputting the curve data into the trained convolutional neural network model to obtain a corresponding turnout switch machine fault judgment result.
Preferably, each set of operating characteristics includes a three-phase current characteristic and a power characteristic.
Preferably, the normalizing the operating characteristic curve specifically includes the steps of:
sampling the working characteristic curve according to a set frequency f to obtain a plurality of point data x1,…,xr(ii) a X is to beiIs updated to
Wherein x'iIs to xiNormalized dot data, xmax、xminAre respectively x1,…,xrMaximum and minimum values of (a).
Preferably, the preprocessing of the curve data in which each set of operating characteristic curves is in the form of a matrix based on the reference operating characteristic curve set includes:
selecting a group of point data after working characteristic curve normalization: x'1，1,…,x′1,r；x′2,1,…,x′2,r；x′3,1,…,x′3,r；x′4,1,…,x′4,r；
Wherein x'1，1,…,x′1,r；x′2,1,…,x′2,r；x′3,1,…,x′3,r(ii) a Respectively obtaining point data after three-phase current characteristic curves of the group of working characteristic curves are normalized; x'4,1,…,x′4,rThe point data is obtained after the power characteristic curve of the group of working characteristic curves is normalized;
obtaining the point data after the normalization of the corresponding reference working characteristic curve group: y'1,1,…,y′1,r；y′2,1,…,y′2,r；y′3,1,…,y′3,r；y′4,1,…,y′4,r；；
Wherein y'1,1,…,y′1,r；y′2,1,…,y′2,r；y′3,1,…,y′3,r(ii) a Respectively obtaining point data after normalizing the three-phase current characteristic curves of the reference working characteristic curve group; y'4,1,…,y′4,r(ii) a The point data is obtained by normalizing the power characteristic curve of the reference working characteristic curve group; the reference working characteristic curve group and the selected working characteristic curve group correspond to the same action type;
point data x'i1,…,x′irCurve and point data y 'formed'i1,…,y′irThe constructed curves are spliced to contain point data x'i1,…,x′ir,y′i1,…,y′irWherein i e [1,4 ]](ii) a Curve data in a matrix form into which the first to fourth curves are converted;
and repeating the steps until curve data corresponding to all the working characteristic curve groups are obtained.
Preferably, the sampling curve data in the normal and fault data pools to establish a training data set and a testing data set includes:
making curve data in the normal/fault data pool be normal/fault curve data; and (3) up-sampling curve data in the fault data pool, so that the same quantity of normal curve data and fault curve data in the training data set is realized, and the same quantity of normal curve data and fault curve data in the testing data set is realized.
Preferably, the convolutional neural network model is a ResNet network model; the ResNet network model comprises: the network model comprises a plurality of convolutional layers which are connected in sequence and a full-connection layer which is arranged at the tail end of a ResNet network model; the convolutional layer comprises a plurality of convolutional units which are connected in sequence and a pooling layer arranged at the tail end of the convolutional layer; the activation function of the convolutional layer is a Relu function; normalizing the output result of each convolution layer by batch normalization processing after each convolution layer; cross-layer connection is used between the convolutional layers to realize transmission of residual errors; and outputting the fault probability and the fault type of the characteristic curve group to be diagnosed by the full connection layer.
Preferably, the convolutional layer includes three convolution units, and the three convolution units are stacked by using convolution kernels of 1 × 1, 3 × 3, and 1 × 1, respectively.
The invention also provides a turnout switch machine abnormity diagnosis system based on machine learning, which is used for realizing the turnout switch machine abnormity diagnosis method based on machine learning, and the system comprises:
the working characteristic curve storage unit is used for storing a plurality of groups of working characteristic curves to be trained, a plurality of groups of working characteristic curves to be diagnosed and curve information corresponding to each group of working characteristic curves;
the preprocessing module is used for selecting a reference working characteristic curve group corresponding to the action type of the switch machine from the plurality of groups of working characteristic curves to be trained, generating corresponding curve data for the plurality of groups of working characteristic curves to be trained and diagnosed, and generating a training data set and a testing data set based on the generated curve data;
the convolutional neural network model training module is used for training a convolutional neural network model based on curve data in a training data set, and the convolutional neural network model is used for identifying whether a fault and a fault type occur during the action of a turnout switch machine;
and the point switch fault identification module outputs a corresponding fault judgment result for the working characteristic curve to be diagnosed based on the trained convolutional neural network model.
Preferably, the machine learning-based switch machine abnormality diagnosis system further includes:
the working characteristic curve inquiring unit is used for inputting curve information and acquiring a plurality of corresponding groups of working characteristic curves from the working characteristic curve storage unit based on the curve information;
the working characteristic curve display unit is used for displaying a plurality of groups of working characteristic curves obtained by query and corresponding reference working characteristic curve groups;
and the curve information display unit is used for displaying the curve information corresponding to each group of working characteristic curves and the fault judgment result of the group of working characteristic curves.
Preferably, the curve information includes: any one or more of the name of the switch station, the name of the switch, the switching direction of the switch, the insertion time of the action rod of the switch and the pulling time of the action rod of the switch.
Compared with the prior art, the invention has the beneficial effects that:
1) the turnout switch machine abnormity diagnosis method and system based on machine learning can effectively find potential safety hazards of switch machine equipment and reduce equipment faults; the collected three-phase current characteristic curve and power characteristic curve of the turnout switch machine during action are intelligently analyzed based on the convolutional neural network model, so that the fault and hidden danger of the turnout switch machine are automatically identified, the fault diagnosis process of the turnout switch machine is compressed, and the abnormal diagnosis efficiency and the correct rate of the diagnosis result of the turnout switch machine are greatly improved;
2) the switch machine fault diagnosis result based on the invention can effectively guide the switch machine fault maintenance field operation, and improves the emergency handling capability of the switch machine maintenance.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings used in the description will be briefly introduced, and it is obvious that the drawings in the following description are an embodiment of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts according to the drawings:
FIG. 1 is a flow chart of the method for diagnosing the abnormality of the point switch machine based on machine learning according to the present invention;
FIG. 2 is a schematic diagram of curve data obtained in a matrix form by curve splicing in an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a machine learning-based abnormal diagnosis system for a turnout switch machine according to the present invention;
FIG. 4 is a schematic interface diagram of a work characteristic curve query unit and a work characteristic curve display unit according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a set of operating characteristics of a switch machine, a set of reference operating characteristics, according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of an interface through a curved information display unit according to an embodiment of the present invention;
FIG. 7 is a schematic interface diagram of an operating characteristic query unit in an embodiment of the present invention;
in the figure: 1. an operating characteristic curve storage unit; 2. a preprocessing module; 3. a convolutional neural network model training module; 4. a switch machine fault identification module; 5. a working characteristic curve query unit; 6. a working characteristic curve display unit; 7. and a curve information display unit.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention relates to a turnout switch machine abnormity identification and diagnosis method based on machine learning. When the switch is needed to be switched, an action circuit in the switch machine is switched on, and the switch is driven by the power output by the motor to be switched. For turnout switch machine equipment, a subway maintenance support system acquires characteristic curves such as current, voltage and power of a point switch machine on site by additionally arranging a voltage sensor and a current sensor, and the working state of the point switch machine can be known to a certain extent by analyzing the characteristic curves. Under the condition that the equipment state is healthy, the driving turnout conversion of the point switch is divided into three stages of unlocking, conversion between positioning and reverse position and locking, and corresponding reaction can be carried out on a characteristic curve of the point switch. For the same point switch of the same turnout, the operating characteristic curve reflected under the normal operating condition has similar characteristics, but when the point switch is in failure or abnormal, some differences can be shown on the curve, so whether the point switch characteristic curve is normal or not is an important basis for judging whether the point switch works normally or not.
The invention provides a turnout switch machine abnormity diagnosis method based on machine learning, which comprises the following steps of:
and S1, collecting a plurality of working characteristic curves of the turnout switch machine during action for a plurality of times. The plurality of classes of operating characteristic curves include: the three-phase current characteristic curve and the power characteristic curve of the point switch, and the three-phase current characteristic curve and the power characteristic curve acquired at one time are used as a working characteristic curve group (also called a group of working characteristic curves); one operating characteristic curve group corresponds to one action of the switch machine.
Selecting a group of working characteristic curves as a reference working characteristic curve group corresponding to the action type based on the action type of the point switch; for example, a set of operation characteristic curves at the time of normal operation of the switch machine when the switch machine is switched from the positioning state to the reverse state is selected from the plurality of sets of operation characteristic curves as a reference operation characteristic curve set corresponding to the operation type of "positioning state to the reverse state". Because the operating characteristics of different motion types may differ, corresponding reference operating characteristic sets are selected for different motion types.
Normalizing the working characteristic curve (which may be a three-phase current characteristic curve or a power characteristic curve), specifically comprising the steps of:
sampling the working characteristic curve according to a set frequency f to obtain a plurality of point data x1,…,xr(ii) a X is to beiIs updated to
Wherein x'iIs to xiNormalized dot data, xmax、xminAre respectively x1,…,xrMaximum and minimum values of (a).
Preprocessing curve data of each group of working characteristic curves in a matrix form based on the action type of the point switch and a corresponding reference working characteristic curve group, and specifically comprising the following steps:
firstly, for each working characteristic curve, selecting the first 200 point data of the curve through sampling;
arbitrarily selecting a group of normalized point data of the working characteristic curve (corresponding to the action type of the positioning reversal position): x'1，1,…,x′1,200；x′2,1,…,x′2,200；x′3,1,…,x′3,200；x′4,1,…,x′4,200；
Wherein x'1，1,…,x′1,200；x′2,1,…,x′2,200；x′3,1,…,x′3,200(ii) a Respectively obtaining point data after three-phase current characteristic curves of the group of working characteristic curves are normalized; x'4,1,…,x′4,200The point data is obtained after the power characteristic curve of the group of working characteristic curves is normalized;
obtaining the point data after normalization of the corresponding reference working characteristic curve group (corresponding to the action type of the positioning reversal position): y'1,1,…,y′1,200；y′2,1,…,y′2,200；y′3,1,…,y′3,200；y′4,1,…,y′4,200；
Wherein y'1,1,…,y′1,r；y′2,1,…,y′2,r；y′3,1,…,y′3,r(ii) a Respectively obtaining point data after normalizing the three-phase current characteristic curves of the reference working characteristic curve group; y'4,1,…,y′4,r(ii) a The point data is obtained by normalizing the power characteristic curve of the reference working characteristic curve group;
point data x'i1,…,x′irCurve and point data y 'formed'i1,…,y′irThe constructed curves are spliced to contain point data x'i1,…,x′ir,y′i1,…,y′irWherein i e [1,4 ]](ii) a Curve data in a matrix form into which the first to fourth curves are converted; as shown in FIG. 2, the first curve is sequentially divided to include point data x'i1,…,x′ir,y′i1,…,y′irThe grids of (2), containing point data x'1，1,…,x′1,40As the first row, x ', of the curve data in the form of a matrix after splicing'1，41,…,x′1,80As the second row of curve data, and so on.
And repeating the steps until curve data corresponding to all the working characteristic curve groups are obtained.
After splicing is completed, pseudo picture data of 40 x 40 is generated for each group of working characteristic curves, the pseudo picture data is used as input of a convolutional neural network model, and the trained convolutional neural network can be suitable for fault scenes of various action types through internal connection between a convolutional neural network model self-adaptive learning working characteristic curve group and a corresponding reference working characteristic curve group.
S2, manually setting a label for the curve data, wherein the label is used for indicating whether the turnout switch machine is in fault and fault type; based on the label, putting curve data into a normal data pool or a fault data pool;
s3, sampling curve data in the normal and fault data pools to establish a training data set and a testing data set, and specifically comprising:
making curve data in the normal/fault data pool be normal/fault curve data; the curve data in the normal data pool is extracted, and the curve data in the fault data pool is up-sampled (which is the prior art and is not the key point of the description of the invention), so that the same quantity of the normal curve data and the fault curve data in the training data set is realized, and the same quantity of the normal curve data and the fault curve data in the testing data set is realized.
S4, pre-training a convolutional neural network model through the training data set and the testing data set, wherein the convolutional neural network model is used for identifying whether faults occur or not and fault types when the turnout switch machine acts;
the convolutional neural network model is a ResNet network model; the ResNet network model comprises: the network model comprises a plurality of convolutional layers which are connected in sequence and a full-connection layer which is arranged at the tail end of a ResNet network model; the convolutional layer comprises a plurality of convolutional units which are connected in sequence and a pooling layer arranged at the tail end of the convolutional layer; the activation function of the convolutional layer is a Relu function; normalizing the output result of each convolution layer by batch normalization processing after each convolution layer; cross-layer connection is used between the convolutional layers to realize transmission of residual errors; and outputting the fault probability and the fault type of the characteristic curve group to be diagnosed by the full connection layer.
In an embodiment of the present invention, the convolutional layer comprises three convolutional units, which are stacked by three convolutional units, respectively using convolutional kernels of 1 × 1, 3 × 3, and 1 × 1.
And S5, processing the working characteristic curve group to be diagnosed, acquiring corresponding curve data, and inputting the curve data into the trained convolutional neural network model to obtain a corresponding turnout switch machine fault judgment result.
The present invention also provides a machine learning-based abnormal diagnosis system for a turnout switch machine, which is used for implementing the machine learning-based abnormal diagnosis method for a turnout switch machine of the present invention, as shown in fig. 3, the system comprises: the device comprises a working characteristic curve storage unit 1, a preprocessing module 2, a convolutional neural network model training module 3, a point switch fault identification module 4, a working characteristic curve query unit 5, a working characteristic curve display unit 6 and a curve information display unit 7.
The working characteristic curve storage unit 1 is used for storing a plurality of groups of working characteristic curves to be trained, a plurality of groups of working characteristic curves to be diagnosed and curve information corresponding to each group of working characteristic curves;
the preprocessing module 2 is configured to select a reference operating characteristic curve group (a group of operating characteristic curves acquired by the point machine during normal execution of the action) corresponding to the action type of the point machine from the plurality of groups of operating characteristic curves to be trained, generate corresponding curve data for the plurality of groups of operating characteristic curves to be trained and diagnosed, and generate a training data set and a test data set based on the generated curve data;
the convolutional neural network model training module 3 trains a convolutional neural network model based on curve data in a training data set, and the convolutional neural network model is used for identifying whether a fault and a fault type occur during the action of a turnout switch machine;
and the point switch fault identification module 4 outputs a corresponding fault judgment result for the working characteristic curve to be diagnosed based on the trained convolutional neural network model.
As shown in fig. 7, the operating characteristic curve querying unit 5 is configured to input curve information, and obtain a plurality of corresponding sets of operating characteristic curves from the operating characteristic curve storage unit 1 based on the curve information; in an embodiment of the present invention, the curve information includes: any one or more of the name of the switch station, the name of the switch, the switching direction of the switch, the insertion time of the action rod of the switch and the pulling time of the action rod of the switch. The working characteristic curve query unit 5 screens data according to conditions selected by a user, which is substantially database condition query using SQL statements, so that the user can conveniently query a desired characteristic curve from a large number of characteristic curves.
As shown in fig. 4 and 5, the operating characteristic curve display unit 6 is configured to display a plurality of sets of operating characteristic curves obtained by query and corresponding reference operating characteristic curve sets;
the graphs in fig. 4 and 5 have time on the abscissa and power or current on the ordinate. In addition, the reference working characteristic curve group corresponding to the displayed working characteristic curve group can be displayed in the control by clicking the check button, so that the comparison by a user is facilitated.
And the curve information display unit 7 is used for displaying the curve information corresponding to each group of working characteristic curves and the fault judgment result of the group of working characteristic curves.
As shown in fig. 6, the curve information is in the form of a list, and includes all fields of the curve information existing in the operation characteristic curve storage unit 1. On the one hand, the information of the current curve is provided for the user, and on the other hand, the user can check the intelligent detection result or judge the training effect of the current training model.
The method can intelligently analyze the collected three-phase current characteristic curve and power characteristic curve when the turnout switch machine acts based on the convolutional neural network model, automatically identify the fault and hidden danger of the turnout switch machine, and greatly improve the efficiency of abnormal diagnosis of the turnout switch machine and the accuracy of the diagnosis result.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A turnout switch machine abnormity diagnosis method based on machine learning is characterized by comprising the following steps:
collecting multiple types of working characteristic curves of the turnout switch machine during action for multiple times, wherein the collected multiple types of working characteristic curves at one time are used as a working characteristic curve group; selecting a group of working characteristic curves as a reference working characteristic curve group corresponding to the action type based on the action type of the point switch; normalizing the operating characteristic curve; preprocessing curve data of each group of working characteristic curves in a matrix form based on the action type of the point switch and the corresponding reference working characteristic curve group;
manually setting a label for the curve data, wherein the label is used for indicating whether the turnout switch machine is in fault and the fault type; based on the label, putting curve data into a normal data pool or a fault data pool;
sampling curve data in the normal and fault data pools to establish a training data set and a testing data set;
pre-training a convolutional neural network model through the training data set and the testing data set, wherein the convolutional neural network model is used for identifying whether a fault and a fault type occur during the action of a turnout switch machine;
and processing the working characteristic curve group to be diagnosed, acquiring corresponding curve data, and inputting the curve data into the trained convolutional neural network model to obtain a corresponding turnout switch machine fault judgment result.
2. The method of diagnosing abnormality of a point switch machine based on machine learning as claimed in claim 1, wherein each set of operation characteristic curves includes a three-phase current characteristic curve and a power characteristic curve.
3. The method of machine learning-based switch machine anomaly diagnosis for a switch machine as claimed in claim 2, wherein said normalizing said operating characteristic curve specifically comprises:
sampling the working characteristic curve according to a set frequency f to obtain a plurality of point data x1,…,xr(ii) a X is to beiIs updated to
Wherein x'iIs to xiNormalized dot data, xmax、xminAre respectively x1,…,xrMaximum and minimum values of (a).
4. The method of machine learning-based switch machine anomaly diagnosis for a switch machine as claimed in claim 3, wherein said preprocessing curve data for each set of operating characteristic curves in matrix form based on said set of reference operating characteristic curves comprises:
selecting a group of point data after working characteristic curve normalization: x'1，1,…,x′1,r；x′2,1,…,x′2,r；x′3,1,…,x′3,r；x′4,1,…,x′4,r；
Wherein x'1，1,…,x′1,r；x′2,1,…,x′2,r；x′3,1,…,x′3,r(ii) a Respectively obtaining point data after three-phase current characteristic curves of the group of working characteristic curves are normalized; x'4,1,…,x′4,rThe point data is obtained after the power characteristic curve of the group of working characteristic curves is normalized;
obtaining the point data after the normalization of the corresponding reference working characteristic curve group: y'1,1,…,y′1,r；y′2,1,…,y′2,r；y′3,1,…,y′3,r；y′4,1,…,y′4,r；
Wherein y'1,1,…,y′1,r；y′2,1,…,y′2,r；y′3,1,…,y′3,r(ii) a Respectively obtaining point data after normalizing the three-phase current characteristic curves of the reference working characteristic curve group; y'4,1,…,y′4,r(ii) a The point data is obtained by normalizing the power characteristic curve of the reference working characteristic curve group; the reference working characteristic curve group and the selected working characteristic curve group correspond to the same action type;
point data x'i,1,…,x′i,rCurve and point data y 'formed'i,1,…,y′i,rThe constructed curves are spliced to contain point data x'i,1,…,x′i,r,y′i,1,…,y′i,rWherein i e [1,4 ]](ii) a Curve data in a matrix form into which the first to fourth curves are converted;
and repeating the steps until curve data corresponding to all the working characteristic curve groups are obtained.
5. The method of machine learning-based switch machine anomaly diagnosis for a switch machine as claimed in claim 1, wherein said sampling curve data in said normal and fault data pools to create training data sets and test data sets comprises:
making curve data in the normal/fault data pool be normal/fault curve data; and (3) up-sampling curve data in the fault data pool, so that the same quantity of normal curve data and fault curve data in the training data set is realized, and the same quantity of normal curve data and fault curve data in the testing data set is realized.
6. The method of machine learning-based diagnosis of abnormalities in switch machines of points, as set forth in claim 1, characterized in that said convolutional neural network model is the ResNet network model; the ResNet network model comprises: the network model comprises a plurality of convolutional layers which are connected in sequence and a full-connection layer which is arranged at the tail end of a ResNet network model; the convolutional layer comprises a plurality of convolutional units which are connected in sequence and a pooling layer arranged at the tail end of the convolutional layer; the activation function of the convolutional layer is a Relu function; normalizing the output result of each convolution layer by batch normalization processing after each convolution layer; cross-layer connection is used between the convolutional layers to realize transmission of residual errors; and outputting the fault probability and the fault type of the characteristic curve group to be diagnosed by the full connection layer.
7. The method of diagnosing abnormality of a point switch based on machine learning as claimed in claim 6, wherein said convolutional layer comprises three convolution units, said three convolution units respectively using convolution kernels of 1 x 1, 3 x 3 and 1 x 1, and said convolutional layer is stacked by said three convolution units.
8. A machine learning-based turnout switch machine abnormality diagnosis system for implementing the machine learning-based turnout switch machine abnormality diagnosis method according to any one of claims 1 to 7, comprising:
the working characteristic curve storage unit is used for storing a plurality of groups of working characteristic curves to be trained, a plurality of groups of working characteristic curves to be diagnosed and curve information corresponding to each group of working characteristic curves;
the preprocessing module is used for selecting a reference working characteristic curve group corresponding to the action type of the switch machine from the plurality of groups of working characteristic curves to be trained, generating corresponding curve data for the plurality of groups of working characteristic curves to be trained and diagnosed, and generating a training data set and a testing data set based on the generated curve data;
the convolutional neural network model training module is used for training a convolutional neural network model based on curve data in a training data set, and the convolutional neural network model is used for identifying whether a fault and a fault type occur during the action of a turnout switch machine;
and the point switch fault identification module outputs a corresponding fault judgment result for the working characteristic curve to be diagnosed based on the trained convolutional neural network model.
9. The machine learning-based switch machine anomaly diagnostic system for a turnout switch machine of claim 8, further comprising:
the working characteristic curve inquiring unit is used for inputting curve information and acquiring a plurality of corresponding groups of working characteristic curves from the working characteristic curve storage unit based on the curve information;
the working characteristic curve display unit is used for displaying a plurality of groups of working characteristic curves obtained by query and corresponding reference working characteristic curve groups;
and the curve information display unit is used for displaying the curve information corresponding to each group of working characteristic curves and the fault judgment result of the group of working characteristic curves.
10. The machine learning-based switch machine anomaly diagnostic system for a turnout switch machine of claim 9, wherein the curve information comprises: any one or more of the name of the switch station, the name of the switch, the switching direction of the switch, the insertion time of the action rod of the switch and the pulling time of the action rod of the switch.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011578819.8A CN112668715A (en) | 2020-12-28 | 2020-12-28 | Turnout switch machine abnormity diagnosis method and system based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011578819.8A CN112668715A (en) | 2020-12-28 | 2020-12-28 | Turnout switch machine abnormity diagnosis method and system based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112668715A true CN112668715A (en) | 2021-04-16 |
Family
ID=75410710
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011578819.8A Pending CN112668715A (en) | 2020-12-28 | 2020-12-28 | Turnout switch machine abnormity diagnosis method and system based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668715A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113421017A (en) * | 2021-07-12 | 2021-09-21 | 北京昊鹏智能技术有限公司 | System state detection method, device, equipment, system and storage medium |
| CN113799855A (en) * | 2021-11-16 | 2021-12-17 | 通号工程局集团电气工程有限公司 | Railway signal data acquisition simulation analysis processing system based on visual interface |
| CN114742297A (en) * | 2022-04-11 | 2022-07-12 | 中国第一汽车股份有限公司 | Power battery treatment method |
| CN115859090A (en) * | 2023-02-23 | 2023-03-28 | 华东交通大学 | Turnout fault diagnosis method and system based on 1DCNN-LSTM |
| CN114742297B (en) * | 2022-04-11 | 2024-05-24 | 中国第一汽车股份有限公司 | Method for processing power battery |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108445861A (en) * | 2018-02-05 | 2018-08-24 | 华南理工大学 | A kind of goat fault detection method and system based on convolutional neural networks algorithm |
| CN110728300A (en) * | 2019-09-09 | 2020-01-24 | 交控科技股份有限公司 | Method and system for identifying fault type based on turnout action current curve |
| CN111046583A (en) * | 2019-12-27 | 2020-04-21 | 中国铁道科学研究院集团有限公司通信信号研究所 | Switch machine fault diagnosis method based on DTW algorithm and ResNet network |
| CN111914320A (en) * | 2020-06-06 | 2020-11-10 | 同济大学 | No-sample turnout fault diagnosis method based on deep learning |
-
2020
- 2020-12-28 CN CN202011578819.8A patent/CN112668715A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108445861A (en) * | 2018-02-05 | 2018-08-24 | 华南理工大学 | A kind of goat fault detection method and system based on convolutional neural networks algorithm |
| CN110728300A (en) * | 2019-09-09 | 2020-01-24 | 交控科技股份有限公司 | Method and system for identifying fault type based on turnout action current curve |
| CN111046583A (en) * | 2019-12-27 | 2020-04-21 | 中国铁道科学研究院集团有限公司通信信号研究所 | Switch machine fault diagnosis method based on DTW algorithm and ResNet network |
| CN111914320A (en) * | 2020-06-06 | 2020-11-10 | 同济大学 | No-sample turnout fault diagnosis method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 程翔: "《基于本体数据库的多标签预测模型及生物医药数据挖掘研究》", 31 May 2019, 天津：天津大学出版社 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113421017A (en) * | 2021-07-12 | 2021-09-21 | 北京昊鹏智能技术有限公司 | System state detection method, device, equipment, system and storage medium |
| CN113799855A (en) * | 2021-11-16 | 2021-12-17 | 通号工程局集团电气工程有限公司 | Railway signal data acquisition simulation analysis processing system based on visual interface |
| CN113799855B (en) * | 2021-11-16 | 2022-03-22 | 通号工程局集团电气工程有限公司 | Railway signal data acquisition simulation analysis processing system based on visual interface |
| CN114742297A (en) * | 2022-04-11 | 2022-07-12 | 中国第一汽车股份有限公司 | Power battery treatment method |
| CN114742297B (en) * | 2022-04-11 | 2024-05-24 | 中国第一汽车股份有限公司 | Method for processing power battery |
| CN115859090A (en) * | 2023-02-23 | 2023-03-28 | 华东交通大学 | Turnout fault diagnosis method and system based on 1DCNN-LSTM |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112668715A (en) | Turnout switch machine abnormity diagnosis method and system based on machine learning | |
| CN108569607B (en) | Elevator fault early warning method based on bidirectional gating cyclic neural network | |
| CN110929898B (en) | Hydropower station start-stop equipment operation and maintenance and fault monitoring online evaluation system and method | |
| CN106354118A (en) | Fault diagnosis system and method for train based on fault tree | |
| CN103699698A (en) | Method and system for track traffic failure recognition based on improved Bayesian algorithm | |
| CN110941918B (en) | Intelligent substation fault analysis system | |
| CN108709744B (en) | Motor bearings method for diagnosing faults under a kind of varying load operating condition | |
| CN111046583A (en) | Switch machine fault diagnosis method based on DTW algorithm and ResNet network | |
| CN110375983A (en) | Failsafe valve real-time diagnosis system and diagnostic method based on time series analysis | |
| CN112115618A (en) | Power equipment fault diagnosis method and system based on matrix chart and confidence | |
| CN110824280B (en) | Diagnosis method and visualization method of turnout health state based on feature similarity | |
| CN114066220A (en) | Ship fault prediction and health management system | |
| CN110132617B (en) | Rail vehicle on-line monitoring system | |
| JPH06186140A (en) | Plant facility diagnosis device | |
| CN115859090A (en) | Turnout fault diagnosis method and system based on 1DCNN-LSTM | |
| CN109030976B (en) | Power electronic equipment monitoring and diagnosing system and method | |
| CN112542892B (en) | Substation regulation and control integrated control method and control device | |
| CN109592525A (en) | Elevator frequency converter fault diagnosis system and method | |
| CN112434979B (en) | Switch system health assessment method | |
| JP2502766B2 (en) | Elevator failure diagnostic device | |
| CN115186694A (en) | Turnout fault expert knowledge base, establishment method, fault detection method, diagnosis method and device | |
| CN114280383A (en) | Health detection device and health detection method for large ground station antenna | |
| CN117131354A (en) | Real-time on-line monitoring system for turnout faults | |
| CN117235514A (en) | Elevator fault diagnosis system using cyclic neural network model | |
| CN211904656U (en) | Monitoring system based on rail vehicle and rail vehicle |
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 | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40044498 Country of ref document: HK |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210416 |