CN115047313B - ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN - Google Patents
ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN Download PDFInfo
- Publication number
- CN115047313B CN115047313B CN202210621985.4A CN202210621985A CN115047313B CN 115047313 B CN115047313 B CN 115047313B CN 202210621985 A CN202210621985 A CN 202210621985A CN 115047313 B CN115047313 B CN 115047313B
- Authority
- CN
- China
- Prior art keywords
- data
- circuit signal
- section
- track circuit
- signal data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000003745 diagnosis Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 79
- 238000007781 pre-processing Methods 0.000 claims abstract description 22
- 230000006870 function Effects 0.000 claims description 48
- 238000004364 calculation method Methods 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 238000000926 separation method Methods 0.000 claims description 7
- 238000001514 detection method Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 102100029469 WD repeat and HMG-box DNA-binding protein 1 Human genes 0.000 description 1
- 101710097421 WD repeat and HMG-box DNA-binding protein 1 Proteins 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G01R31/28—Testing of electronic circuits, e.g. by signal tracer
- G01R31/2832—Specific tests of electronic circuits not provided for elsewhere
- G01R31/2836—Fault-finding or characterising
- G01R31/2843—In-circuit-testing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Train Traffic Observation, Control, And Security (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention provides a ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN. The method comprises the following steps: acquiring track circuit signal data of a monitored section at the current moment; identifying the section type and preprocessing track circuit signal data; predicting the preprocessed orbit circuit signal data by adopting an OC-SVM prediction model, and judging whether the orbit circuit signal data belongs to normal data or novel data; if the track circuit signal data belong to the normal data, the track circuit signal data are directly sent to a DNN prediction model for fault classification prediction; if the data belong to novel data, continuously detecting the track circuit signal data at the next moment, if each track circuit signal data acquired in a certain continuous time period from the current moment is novel data, adding the track circuit signal data corresponding to the current moment into a training data set, and setting a new fault label for the track circuit signal data; and training by using the new training data set to obtain a new OC-SVM predictive model and a new DNN predictive model.
Description
Technical Field
The invention relates to the technical field of track circuits, in particular to a ZPW-2000R track circuit fault diagnosis method and device based on OC-SVM and DNN.
Background
The track circuit is an important component of the safe operation of a railroad train. The existing track circuit has various track circuit faults due to a plurality of components and complicated external environment factors, and the current diagnosis of the track circuit faults also depends on manual experience, so that the problems of long maintenance time, complicated flow, large manual workload and the like after the faults occur greatly influence the normal operation of the train.
With the development of artificial intelligence technology, numerous machine learning algorithms exhibit good effects in various fields, and some students introduce intelligent algorithms into the study of track circuit fault diagnosis. However, in the current research of fault diagnosis of track circuits, a neural network method or other machine learning methods are mostly used for performing supervised learning on some fault type data samples, and the method needs a large amount of labeled fault data to perform algorithm training and can be operated, and once unknown or brand new faults are encountered, the model cannot be identified, so that the model cannot be applied to actual working scenes.
Disclosure of Invention
In order to solve the problem of automatic identification of unknown or brand new faults, the invention provides a ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN.
In one aspect, the invention provides a ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN, which comprises the following steps:
step 1: acquiring track circuit signal data of a monitored section at the current moment;
step 2: identifying the section type of the monitored section and carrying out data preprocessing on the track circuit signal data according to a data preprocessing mode corresponding to the section type;
step 3: predicting the preprocessed track circuit signal data by adopting a trained OC-SVM prediction model, and judging whether the track circuit signal data belongs to normal data or novel data according to a prediction result;
Step 4: if the track circuit signal data belong to normal data, the track circuit signal data are directly sent to a DNN prediction model for fault classification prediction;
Step 5: if the track circuit signal data belong to novel data, continuously detecting the track circuit signal data of the monitored section at the next moment, and if each track circuit signal data acquired in a certain continuous time period from the current moment is novel data, adding the track circuit signal data corresponding to the current moment into a training data set, and setting a new fault label for the track circuit signal data;
step 6: and training by using the new training data set to obtain a new OC-SVM prediction model and a new DNN prediction model for subsequent detection of the orbit circuit signal data.
Further, the segment types include no-rear segment, no-front segment, middle segment, and no-front no-rear segment.
Further, step 2 specifically includes:
If the monitored section is a section without a rear section, combining the track circuit signal data of the monitored section with the track circuit signal data of a section in front of the monitored section, and deleting the data item with the characteristic value not being a number;
If the monitored section is a section without a front section, combining the track circuit signal data of the monitored section with the track circuit signal data of a rear section thereof, and deleting the data item with the characteristic value not being a number;
If the monitored section is a middle section, combining the track circuit signal data of the monitored section with the track circuit signal data of the front section and the track circuit signal data of the rear section, and deleting the data item with the characteristic value not being digital;
If the monitored section is a section without front and rear, directly deleting the data item with the characteristic value not being the number in the track circuit signal data.
Further, the training process of the OC-SVM predictive model comprises the following steps:
Step A1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
step A2: mapping each sample to a high-dimensional feature space through a kernel function, and solving an optimal hypersphere in the feature space by using a formula (1);
s.t.ωΤφ(χi)≥ρ-ξi
ξi≥0,i＝1,2,……,l (1)
Wherein ω and ρ represent normal vector and offset of the separation hyperplane in the feature space, respectively; ζ i denotes the relaxation variable of the ith sample, v is a parameter controlling the upper limit of the number of outliers and the lower limit of the number of total support vectors, l is the number of samples, and φ (χ) is a function mapping sample x to feature space;
Step A3: obtaining a classification decision function shown in a formula (2) by using a Lagrange multiplier method, wherein the classification decision function is an OC-SVM prediction model, if f (x) =1, the sample x is normal data, and if f (x) = -1, the sample x is novel data;
Wherein κ (χ i, χ) represents the kernel function; α i is the Lagrangian coefficient corresponding to sample x i.
Further, the training process of the DNN prediction model comprises the following steps:
step B1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
Step B2: constructing a deep neural network framework, setting network parameters, and randomly initializing a weight value w and a bias term b; wherein the network parameters comprise the number of hidden layers, the number of nodes in each hidden layer and the learning rate;
step B3: setting a gradient optimization method, an implicit layer activation function and an output layer activation function;
Step B4: and carrying out forward calculation, feedback calculation and parameter updating on all samples, and determining the optimal numerical values of the weight value w and the bias term b, thereby obtaining the DNN prediction model.
In another aspect, the present invention provides a ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN, comprising:
The data acquisition module is used for acquiring track circuit signal data of the monitored section;
The data preprocessing module is used for identifying the section type of the monitored section and preprocessing the track circuit signal data according to a data preprocessing mode corresponding to the section type;
The novel data prediction module is used for predicting the preprocessed track circuit signal data by adopting a trained OC-SVM prediction model, and judging whether the track circuit signal data belongs to normal data or novel data according to a prediction result;
the fault classification module is used for directly transmitting the track circuit signal data belonging to the normal data to the DNN prediction model for fault classification prediction;
The training data set updating module is used for adding the track circuit signal data at the current moment meeting the preset condition to the training data set and setting a new fault label for the training data set; the preset conditions include: each track circuit signal data collected in a certain period of time from the current moment is novel data;
The model training module is used for training to obtain an OC-SVM prediction model and a DNN prediction model by using the training data set.
Further, the segment types include no-rear segment, no-front segment, middle segment, and no-front no-rear segment.
Further, the data preprocessing module is specifically configured to:
If the monitored section is a section without a rear section, combining the track circuit signal data of the monitored section with the track circuit signal data of a section in front of the monitored section, and deleting the data item with the characteristic value not being a number;
If the monitored section is a section without a front section, combining the track circuit signal data of the monitored section with the track circuit signal data of a rear section thereof, and deleting the data item with the characteristic value not being a number;
If the monitored section is a middle section, combining the track circuit signal data of the monitored section with the track circuit signal data of the front section and the track circuit signal data of the rear section, and deleting the data item with the characteristic value not being digital;
If the monitored section is a section without front and rear, directly deleting the data item with the characteristic value not being the number in the track circuit signal data.
Further, the training process of the model training module to obtain the OC-SVM predictive model comprises the following steps:
Step A1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
step A2: mapping each sample to a high-dimensional feature space through a kernel function, and solving an optimal hypersphere in the feature space by using a formula (1);
s.t.ωΤφ(χi)≥ρ-ξi
ξi≥0,i＝1,2,……,l (1)
Wherein ω and ρ represent normal vector and offset of the separation hyperplane in the feature space, respectively; ζ i denotes the relaxation variable of the ith sample, v is a parameter controlling the upper limit of the number of outliers and the lower limit of the number of total support vectors, l is the number of samples, and φ (χ) is a function mapping sample x to feature space;
Step A3: obtaining a classification decision function shown in a formula (2) by using a Lagrange multiplier method, wherein the classification decision function is an OC-SVM prediction model, if f (x) =1, the sample x is normal data, and if f (x) = -1, the sample x is novel data;
Wherein κ (χ i, χ) represents the kernel function; α i is the Lagrangian coefficient corresponding to sample x i.
Further, the process of training the model training module to obtain the DNN prediction model comprises the following steps:
step B1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
Step B2: constructing a deep neural network framework, setting network parameters, and randomly initializing a weight value w and a bias term b; wherein the network parameters comprise the number of hidden layers, the number of nodes in each hidden layer and the learning rate;
step B3: setting a gradient optimization method, an implicit layer activation function and an output layer activation function;
Step B4: and carrying out forward calculation, feedback calculation and parameter updating on all samples, and determining the optimal numerical values of the weight value w and the bias term b, thereby obtaining the DNN prediction model.
The invention has the beneficial effects that:
Aiming at the problems that the current track circuit has various faults and high-quality tag data are difficult to acquire, the ZPW-2000R track circuit fault diagnosis method and device based on the OC-SVM and DNN provided by the invention use the OC-SVM to carry out single classification and identification on 48 monitoring data acquired by the ZPW-2000R track circuit so as to collect signal data with different fault types; the collected fault data is then trained using the DNN method to automatically classify the known fault data. A large number of experiments are carried out on the method by utilizing ZPW-2000R track circuit data, and the experimental results show that the fault diagnosis method provided by the invention can effectively and automatically collect novel fault data and accurately detect the fault type.
Drawings
FIG. 1 is a schematic flow chart of a ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a segment type according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a deep neural network training process according to an embodiment of the present invention;
Fig. 4 is a schematic structural diagram of a ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
As shown in FIG. 1, the embodiment of the invention provides a ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN, which comprises the following steps:
s101: acquiring track circuit signal data of a monitored section at the current moment;
S102: identifying the section type of the monitored section and carrying out data preprocessing on the track circuit signal data according to a data preprocessing mode corresponding to the section type;
Specifically, different sections at different positions of different types of stations form different circuit structures by adopting different equipment at the outdoor rail side, so that the characteristic properties of different section types are different, and differentiation processing is needed. The zone types include four broad categories of no-rear zone, no-front zone, intermediate zone, and no-front no-rear zone. Wherein, (1) no rear section: forward no backward sector data (neighbor jurisdiction cannot acquire), as in section a, e, c, g in fig. 2; (2) no front section: forward no forward zone data (neighbor jurisdiction cannot acquire), as in section d, h, b, f in fig. 2; (3) intermediate section: the data of the rear and front sections are complete; (4) no front and no rear section: forward no forward and no backward section data, such as the intra-station section in fig. 2.
As an implementation manner, the data preprocessing method for the track circuit signal data according to the data preprocessing mode corresponding to the section type specifically includes:
If the monitored section is a section without a rear section, combining the track circuit signal data of the monitored section with the track circuit signal data of a section in front of the monitored section, and deleting the data item with the characteristic value not being a number;
For example, if the monitored segment is the 1485BG segment in fig. 1, the track circuit signal data of the two segments 1485BG and 1485AG need to be combined, and then the data item in which the feature value is not digital is deleted, so as to be input into the OC-SVM prediction model for processing.
If the monitored section is a section without a front section, combining the track circuit signal data of the monitored section with the track circuit signal data of a rear section thereof, and deleting the data item with the characteristic value not being a number;
For example, if the monitored segment is 1510AG in fig. 1, the track circuit signal data of the two segments 1510AG and 1510BG need to be combined, and then the data item in which the characteristic value is not digital is deleted, so that the data item is subsequently input into the OC-SVM prediction model for processing.
If the monitored section is a middle section, combining the track circuit signal data of the monitored section with the track circuit signal data of the front section and the track circuit signal data of the rear section, and deleting the data item with the characteristic value not being digital;
For example, if the monitored segment is the 1505AG segment in fig. 1, the track circuit signal data of the three segments 1505AG, 1505BG, 1525BG need to be combined, and then the data item whose characteristic value is not digital is deleted, so as to be input into the OC-SVM prediction model for processing.
If the monitored section is a section without front and rear, directly deleting the data item with the characteristic value not being the number in the track circuit signal data.
S103: predicting the preprocessed track circuit signal data by adopting a trained OC-SVM prediction model, and judging whether the track circuit signal data belongs to normal data or novel data according to a prediction result;
Specifically, the predicted result of the OC-SVM prediction model includes two decision values (e.g., 1 and-1), and it is assumed that the predicted result corresponding to the decision value 1 indicates that the orbit circuit signal data belongs to normal data, and the predicted result corresponding to the decision value-1 indicates that the orbit circuit signal data belongs to novel data, so, if the decision value after the orbit circuit signal data preprocessed at the current time is input to the OC-SVM prediction model is 1, the orbit circuit signal data is indicated as normal data, and otherwise is novel data.
S104: if the track circuit signal data belong to normal data, the track circuit signal data are directly sent to a DNN prediction model for fault classification prediction;
Specifically, the track circuit signal data are input into a DNN prediction model for forward calculation to obtain the corresponding fault category.
S105: if the track circuit signal data belong to novel data, continuously detecting the track circuit signal data of the monitored section at the next moment, and if each track circuit signal data acquired in a certain continuous time period from the current moment is novel data, adding the track circuit signal data corresponding to the current moment into a training data set, and setting a new fault label for the track circuit signal data;
As an implementation manner, if each track circuit signal data acquired at ten consecutive sampling moments from the current moment is novel data, the track circuit signal data corresponding to the current moment is added to a training data set, and a new fault label is set for the track circuit signal data.
For example, from the time t, 10 decision values continuously output by the OC-SVM predictive model are all-1, the orbit circuit signal data corresponding to the time t is considered to be added into the training data set, and a new fault label is set for the orbit circuit signal data.
S106: and training by using the new training data set to obtain a new OC-SVM prediction model and a new DNN prediction model for subsequent detection of the orbit circuit signal data.
On the basis of the above embodiment, the training process of the OC-SVM prediction model includes:
Step A1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
step A2: mapping each sample to a high-dimensional feature space through a kernel function, and solving an optimal hypersphere in the feature space by using a formula (1);
s.t.ωΤφ(χi)≥ρ-ξi
ξi≥0,i＝1,2,……,l (1)
Wherein ω and ρ represent normal vector and offset of the separation hyperplane in the feature space, respectively; ζ i denotes the relaxation variable of the ith sample, v is a parameter controlling the upper limit of the number of outliers and the lower limit of the number of total support vectors, l is the number of samples, and φ (χ) is a function mapping sample x to feature space;
Step A3: obtaining a classification decision function shown in a formula (2) by using a Lagrange multiplier method, wherein the classification decision function is an OC-SVM prediction model, if f (x) =1, the sample x is normal data, and if f (x) = -1, the sample x is novel data;
Wherein, κ (χ i, χ) represents the kernel function, replacing the inner product calculation in the feature space; α i is the Lagrangian coefficient corresponding to sample x i.
As an embodiment, as shown in fig. 3, the training process of the DNN prediction model includes:
step B1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
Step B2: constructing a deep neural network framework, setting network parameters, and randomly initializing a weight value w and a bias term b; wherein the network parameters comprise the number of hidden layers, the number of nodes in each hidden layer and the learning rate;
For example, the hidden layer number is 2, the hidden layer node number is 20, and the learning rate is 0.01.
Step B3: setting a gradient optimization method, an implicit layer activation function and an output layer activation function;
For example, the gradient optimization method is an Admin algorithm, the implicit layer activation function is a tanh function, and the output layer activation function is a sigmoid function.
Step B4: and carrying out forward calculation, feedback calculation and parameter updating on all samples, and determining the optimal numerical values of the weight value w and the bias term b, thereby obtaining the DNN prediction model.
Example 2
As shown in fig. 4, an embodiment of the present invention provides a ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN, including: the system comprises a data acquisition module, a data preprocessing module, a novel data prediction module, a fault classification module, a training data set updating module and a model training module. Wherein:
the data acquisition module is used for acquiring track circuit signal data of the monitored section; the data preprocessing module is used for identifying the section type of the monitored section and preprocessing the track circuit signal data according to a data preprocessing mode corresponding to the section type; the novel data prediction module is used for predicting the preprocessed track circuit signal data by adopting a trained OC-SVM prediction model, and judging whether the track circuit signal data belongs to normal data or novel data according to a prediction result; the fault classification module is used for directly transmitting the track circuit signal data belonging to the normal data to the DNN prediction model for fault classification prediction; the training data set updating module is used for adding the track circuit signal data at the current moment meeting the preset condition to the training data set and setting a new fault label for the training data set; the preset conditions include: each track circuit signal data collected in a certain period of time from the current moment is novel data; the model training module is used for training to obtain an OC-SVM prediction model and a DNN prediction model by using the training data set.
As one embodiment, the section types include no-rear section, no-front section, middle section, and no-front no-rear section.
As an implementation manner, the data preprocessing module is specifically configured to:
If the monitored section is a section without a rear section, combining the track circuit signal data of the monitored section with the track circuit signal data of a section in front of the monitored section, and deleting the data item with the characteristic value not being a number;
If the monitored section is a section without a front section, combining the track circuit signal data of the monitored section with the track circuit signal data of a rear section thereof, and deleting the data item with the characteristic value not being a number;
If the monitored section is a middle section, combining the track circuit signal data of the monitored section with the track circuit signal data of the front section and the track circuit signal data of the rear section, and deleting the data item with the characteristic value not being digital;
If the monitored section is a section without front and rear, directly deleting the data item with the characteristic value not being the number in the track circuit signal data.
As one possible implementation, the training process of the model training module to obtain the OC-SVM predictive model includes:
Step A1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
step A2: mapping each sample to a high-dimensional feature space through a kernel function, and solving an optimal hypersphere in the feature space by using a formula (1) to realize maximum separation of training data and a coordinate origin (namely novel data);
s.t.ωΤφ(χi)≥ρ-ξi
ξi≥0,i＝1,2,……,l (1)
Wherein ω and ρ represent normal vector and offset of the separation hyperplane in the feature space, respectively; ζ i denotes the relaxation variable of the ith sample, v is a parameter controlling the upper limit of the number of outliers and the lower limit of the number of total support vectors, l is the number of samples, and φ (χ) is a function mapping sample x to feature space;
Step A3: obtaining a classification decision function shown in a formula (2) by using a Lagrange multiplier method, wherein the classification decision function is an OC-SVM prediction model, if f (x) =1, the sample x is normal data, and if f (x) = -1, the sample x is novel data;
Wherein, κ (χ i, χ) represents the kernel function, replacing the inner product calculation in the feature space; α i is the Lagrangian coefficient corresponding to sample x i.
As an implementation manner, the process of training the model training module to obtain the DNN prediction model includes:
step B1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
Step B2: constructing a deep neural network framework, setting network parameters, and randomly initializing a weight value w and a bias term b; wherein the network parameters comprise the number of hidden layers, the number of nodes in each hidden layer and the learning rate;
step B3: setting a gradient optimization method, an implicit layer activation function and an output layer activation function;
Step B4: and carrying out forward calculation, feedback calculation and parameter updating on all samples, and determining the optimal numerical values of the weight value w and the bias term b, thereby obtaining the DNN prediction model.
It should be noted that, the track circuit fault diagnosis device provided in the embodiment of the present invention is for implementing the above method embodiment, and the function thereof may specifically refer to the above method embodiment, which is not described herein again.
Aiming at the problems that the current track circuit has various faults and high-quality tag data are difficult to acquire, the ZPW-2000R track circuit fault diagnosis method and device based on the OC-SVM and DNN provided by the invention use the OC-SVM to carry out single classification and identification on 48 monitoring data acquired by the ZPW-2000R track circuit so as to collect signal data with different fault types; the collected fault data is then trained using the DNN method to automatically classify the known fault data.
In addition, a great deal of experiments are carried out on the method by utilizing ZPW-2000R track circuit data, and the experimental results show that the fault diagnosis method provided by the invention can effectively and automatically collect novel fault data and accurately detect the fault type.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (8)
1. The ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN is characterized by comprising the following steps:
step 1: acquiring track circuit signal data of a monitored section at the current moment;
step 2: identifying the section type of the monitored section and carrying out data preprocessing on the track circuit signal data according to a data preprocessing mode corresponding to the section type;
Step 3: predicting the preprocessed track circuit signal data by adopting a trained OC-SVM prediction model, and judging whether the track circuit signal data belongs to normal data or novel data according to a prediction result; the training process of the OC-SVM prediction model comprises the following steps:
Step A1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
step A2: mapping each sample to a high-dimensional feature space through a kernel function, and solving an optimal hypersphere in the feature space by using a formula (1);
s.t.ωΤφ(χi)≥ρ-ξi
ξi≥0,i＝1,2,……,l (1)
Wherein ω and ρ represent normal vector and offset of the separation hyperplane in the feature space, respectively; ζ i denotes the relaxation variable of the ith sample, v is the parameter controlling the upper limit of the number of outliers and the lower limit of the number of total support vectors, l is the number of samples, φ (χ) is the function mapping sample x to feature space;
Step A3: obtaining a classification decision function shown in a formula (2) by using a Lagrange multiplier method, wherein the classification decision function is an OC-SVM prediction model, if f (x) =1, the sample x is normal data, and if f (x) = -1, the sample x is novel data;
Wherein κ (χ i, χ) represents the kernel function; α i is the Lagrangian coefficient corresponding to sample x i;
Step 4: if the track circuit signal data belong to normal data, the track circuit signal data are directly sent to a DNN prediction model for fault classification prediction;
Step 5: if the track circuit signal data belong to novel data, continuously detecting the track circuit signal data of the monitored section at the next moment, and if each track circuit signal data acquired in a certain continuous time period from the current moment is novel data, adding the track circuit signal data corresponding to the current moment into a training data set, and setting a new fault label for the track circuit signal data;
step 6: and training by using the new training data set to obtain a new OC-SVM prediction model and a new DNN prediction model for subsequent detection of the orbit circuit signal data.
2. The ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN according to claim 1, wherein the section types include no-rear section, no-front section, middle section and no-front no-rear section.
3. The ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN according to claim 2, wherein step 2 specifically comprises:
If the monitored section is a section without a rear section, combining the track circuit signal data of the monitored section with the track circuit signal data of a section in front of the monitored section, and deleting the data item with the characteristic value not being a number;
If the monitored section is a section without a front section, combining the track circuit signal data of the monitored section with the track circuit signal data of a rear section thereof, and deleting the data item with the characteristic value not being a number;
If the monitored section is a middle section, combining the track circuit signal data of the monitored section with the track circuit signal data of the front section and the track circuit signal data of the rear section, and deleting the data item with the characteristic value not being digital;
If the monitored section is a section without front and rear, directly deleting the data item with the characteristic value not being a number in the track circuit signal data;
Wherein κ (χ i, χ) represents the kernel function; α i is the Lagrangian coefficient corresponding to sample x i.
4. The ZPW-2000R orbit circuit fault diagnosis method based on OC-SVM and DNN according to claim 1, wherein the training process of the DNN prediction model comprises:
step B1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
Step B2: constructing a deep neural network framework, setting network parameters, and randomly initializing a weight value w and a bias term b; wherein the network parameters comprise the number of hidden layers, the number of nodes in each hidden layer and the learning rate;
step B3: setting a gradient optimization method, an implicit layer activation function and an output layer activation function;
Step B4: and carrying out forward calculation, feedback calculation and parameter updating on all samples, and determining the optimal numerical values of the weight value w and the bias term b, thereby obtaining the DNN prediction model.
5. ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN, which is characterized by comprising:
The data acquisition module is used for acquiring track circuit signal data of the monitored section;
The data preprocessing module is used for identifying the section type of the monitored section and preprocessing the track circuit signal data according to a data preprocessing mode corresponding to the section type;
The novel data prediction module is used for predicting the preprocessed track circuit signal data by adopting a trained OC-SVM prediction model, and judging whether the track circuit signal data belongs to normal data or novel data according to a prediction result;
the fault classification module is used for directly transmitting the track circuit signal data belonging to the normal data to the DNN prediction model for fault classification prediction;
The training data set updating module is used for adding the track circuit signal data at the current moment meeting the preset condition to the training data set and setting a new fault label for the training data set; the preset conditions include: each track circuit signal data collected in a certain period of time from the current moment is novel data;
the model training module is used for training by using the training data set to obtain an OC-SVM prediction model and a DNN prediction model; the process of obtaining the OC-SVM predictive model by training the model training module comprises the following steps:
Step A1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
step A2: mapping each sample to a high-dimensional feature space through a kernel function, and solving an optimal hypersphere in the feature space by using a formula (1);
s.t.ωΤφ(χi)≥ρ-ξi
ξi≥0,i＝1,2,……,l (1)
Wherein ω and ρ represent normal vector and offset of the separation hyperplane in the feature space, respectively; ζ i denotes the relaxation variable of the ith sample, v is the parameter controlling the upper limit of the number of outliers and the lower limit of the number of total support vectors, l is the number of samples, φ (χ) is the function mapping sample x to feature space;
Step A3: obtaining a classification decision function shown in a formula (2) by using a Lagrange multiplier method, wherein the classification decision function is an OC-SVM prediction model, if f (x) =1, the sample x is normal data, and if f (x) = -1, the sample x is novel data;
Wherein κ (χ i, χ) represents the kernel function; α i is the Lagrangian coefficient corresponding to sample x i.
6. The ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN according to claim 5, wherein the section types include no rear section, no front section, middle section and no front no rear section.
7. The ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN according to claim 6, wherein the data preprocessing module is specifically configured to:
If the monitored section is a section without a rear section, combining the track circuit signal data of the monitored section with the track circuit signal data of a section in front of the monitored section, and deleting the data item with the characteristic value not being a number;
If the monitored section is a section without a front section, combining the track circuit signal data of the monitored section with the track circuit signal data of a rear section thereof, and deleting the data item with the characteristic value not being a number;
If the monitored section is a middle section, combining the track circuit signal data of the monitored section with the track circuit signal data of the front section and the track circuit signal data of the rear section, and deleting the data item with the characteristic value not being digital;
If the monitored section is a section without front and rear, directly deleting the data item with the characteristic value not being the number in the track circuit signal data.
8. The ZPW-2000R orbit circuit fault diagnosis device based on OC-SVM and DNN according to claim 5, wherein the process of training the model training module to obtain the DNN prediction model comprises:
step B1: loading a training data set, wherein each piece of track circuit signal data in the training data set is taken as one sample, and each sample corresponds to a fault label;
Step B2: constructing a deep neural network framework, setting network parameters, and randomly initializing a weight value w and a bias term b; wherein the network parameters comprise the number of hidden layers, the number of nodes in each hidden layer and the learning rate;
step B3: setting a gradient optimization method, an implicit layer activation function and an output layer activation function;
Step B4: and carrying out forward calculation, feedback calculation and parameter updating on all samples, and determining the optimal numerical values of the weight value w and the bias term b, thereby obtaining the DNN prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210621985.4A CN115047313B (en) | 2022-06-02 | 2022-06-02 | ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210621985.4A CN115047313B (en) | 2022-06-02 | 2022-06-02 | ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115047313A CN115047313A (en) | 2022-09-13 |
| CN115047313B true CN115047313B (en) | 2024-04-26 |
Family
ID=83158856
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210621985.4A Active CN115047313B (en) | 2022-06-02 | 2022-06-02 | ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115047313B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116520817B (en) * | 2023-07-05 | 2023-08-29 | 贵州宏信达高新科技有限责任公司 | ETC system running state real-time monitoring system and method based on expressway |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5483170A (en) * | 1993-08-24 | 1996-01-09 | New Mexico State University Technology Transfer Corp. | Integrated circuit fault testing implementing voltage supply rail pulsing and corresponding instantaneous current response analysis |
| CN103714348A (en) * | 2014-01-09 | 2014-04-09 | 北京泰乐德信息技术有限公司 | Rail transit fault diagnosis method and system based on decision-making tree |
| CN103914735A (en) * | 2014-04-17 | 2014-07-09 | 北京泰乐德信息技术有限公司 | Failure recognition method and system based on neural network self-learning |
| CN110286668A (en) * | 2019-07-15 | 2019-09-27 | 广东毓秀科技有限公司 | A kind of rail friendship signal system VIM board faults prediction technique based on big data |
| CN110852365A (en) * | 2019-10-31 | 2020-02-28 | 北京交通大学 | ZPW-2000A type non-insulated rail circuit fault diagnosis method |
| CN111626416A (en) * | 2020-04-24 | 2020-09-04 | 黑龙江瑞兴科技股份有限公司 | Automatic rail circuit fault diagnosis method based on deep convolutional neural network |
| CN112949715A (en) * | 2013-12-31 | 2021-06-11 | 北京泰乐德信息技术有限公司 | SVM (support vector machine) -based rail transit fault diagnosis method |
| CN113139335A (en) * | 2021-04-09 | 2021-07-20 | 郑州宥新算法智能科技有限公司 | BP neural network-based track circuit fault intelligent diagnosis method |
| CN114462475A (en) * | 2021-12-21 | 2022-05-10 | 南京邮电大学 | Unsupervised machine abnormal sound detection method and unsupervised machine abnormal sound detection device based on single classification algorithm |
-
2022
- 2022-06-02 CN CN202210621985.4A patent/CN115047313B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5483170A (en) * | 1993-08-24 | 1996-01-09 | New Mexico State University Technology Transfer Corp. | Integrated circuit fault testing implementing voltage supply rail pulsing and corresponding instantaneous current response analysis |
| CN112949715A (en) * | 2013-12-31 | 2021-06-11 | 北京泰乐德信息技术有限公司 | SVM (support vector machine) -based rail transit fault diagnosis method |
| CN103714348A (en) * | 2014-01-09 | 2014-04-09 | 北京泰乐德信息技术有限公司 | Rail transit fault diagnosis method and system based on decision-making tree |
| CN103914735A (en) * | 2014-04-17 | 2014-07-09 | 北京泰乐德信息技术有限公司 | Failure recognition method and system based on neural network self-learning |
| CN110286668A (en) * | 2019-07-15 | 2019-09-27 | 广东毓秀科技有限公司 | A kind of rail friendship signal system VIM board faults prediction technique based on big data |
| CN110852365A (en) * | 2019-10-31 | 2020-02-28 | 北京交通大学 | ZPW-2000A type non-insulated rail circuit fault diagnosis method |
| CN111626416A (en) * | 2020-04-24 | 2020-09-04 | 黑龙江瑞兴科技股份有限公司 | Automatic rail circuit fault diagnosis method based on deep convolutional neural network |
| CN113139335A (en) * | 2021-04-09 | 2021-07-20 | 郑州宥新算法智能科技有限公司 | BP neural network-based track circuit fault intelligent diagnosis method |
| CN114462475A (en) * | 2021-12-21 | 2022-05-10 | 南京邮电大学 | Unsupervised machine abnormal sound detection method and unsupervised machine abnormal sound detection device based on single classification algorithm |
Non-Patent Citations (3)
| Title |
|---|
| 基于人工智能的ZPW-2000A轨道电路 故障诊断方法研究;文武臣;《铁路通信信号工程技术》;20211025;第第18卷卷(第第10期期);第34-38页 * |
| 基于卷积神经网络的ZPW-2000R轨道电路智能故障诊断方法;卢皎，等;《工业工程》;20210831;第第24卷卷(第第4期期);第128-134页 * |
| 基于深度学习的无绝缘轨道电路故障诊断研究;谢旭旭，等;《铁道学报》;20200630;第第42卷卷(第第6期期);第79-85页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115047313A (en) | 2022-09-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112784965B (en) | Large-scale multi-element time series data anomaly detection method oriented to cloud environment | |
| CN111178553A (en) | Industrial equipment health trend analysis method and system based on ARIMA and LSTM algorithms | |
| KR102627062B1 (en) | Abnormal equipment trace detection and classification | |
| CN115047313B (en) | ZPW-2000R orbit circuit fault diagnosis method and device based on OC-SVM and DNN | |
| CN112561383A (en) | Real-time anomaly detection method based on generation countermeasure network | |
| CN110210169A (en) | A kind of shield machine failure prediction method based on LSTM | |
| CN108961460B (en) | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization | |
| CN114266301A (en) | Intelligent power equipment fault prediction method based on graph convolution neural network | |
| CN110757510A (en) | Method and system for predicting remaining life of robot | |
| Cao et al. | Multimodal process monitoring based on variational Bayesian PCA and Kullback-Leibler divergence between mixture models | |
| CN111222798B (en) | Complex industrial process key index soft measurement method | |
| CN114048546B (en) | Method for predicting residual service life of aeroengine based on graph convolution network and unsupervised domain self-adaption | |
| CN117077870B (en) | Water resource digital management method based on artificial intelligence | |
| CN117035013A (en) | Method for predicting dynamic network link by adopting impulse neural network | |
| CN114330089B (en) | Rare earth element content change prediction method and system | |
| CN116415485A (en) | Multi-source domain migration learning residual service life prediction method based on dynamic distribution self-adaption | |
| CN114819350A (en) | RUL prediction method under sensor fault condition based on multiple tasks | |
| CN115184734A (en) | Power grid line fault detection method and system | |
| CN114565051A (en) | Test method of product classification model based on neuron influence degree | |
| CN113469013A (en) | Motor fault prediction method and system based on transfer learning and time sequence | |
| Zhou et al. | A Bearing Remaining Useful Life Prediction Method based on Residual Convolutional Network and LSTM | |
| CN110837932A (en) | Thermal power prediction method of solar heat collection system based on DBN-GA model | |
| CN112801815B (en) | Power communication network fault early warning method based on federal learning | |
| CN117874612A (en) | Power system equipment model anomaly detection method based on artificial intelligence | |
| CN117333726B (en) | Quartz crystal cutting abnormality monitoring method, system and device based on deep learning |
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 |