CN110262463A - A kind of rail traffic platform door fault diagnosis system based on deep learning - Google Patents
A kind of rail traffic platform door fault diagnosis system based on deep learning Download PDFInfo
- Publication number
- CN110262463A CN110262463A CN201910613949.1A CN201910613949A CN110262463A CN 110262463 A CN110262463 A CN 110262463A CN 201910613949 A CN201910613949 A CN 201910613949A CN 110262463 A CN110262463 A CN 110262463A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- fault diagnosis
- training
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0262—Confirmation of fault detection, e.g. extra checks to confirm that a failure has indeed occurred
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/24—Pc safety
- G05B2219/24065—Real time diagnostics
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The rail traffic platform door fault diagnosis system based on deep learning that the invention discloses a kind of, including rail traffic platform door body system, door machine transmission system, unit accelerator control system and fault diagnosis system, door machine transmission system by unit accelerator control system timing control, for controlling closing the switch for rail traffic platform door body system;Fault diagnosis system acquires the feedback signal of monitoring sensor in real time, while generating fault diagnosis model using server training and carrying out real-time fault diagnosis to gate operating status by fault diagnosis model.The present invention takes full advantage of neural network can learn implicit feature from large capacity multi-modal data automatically, rather than the characteristic model of engineer.
Description
Technical field
The rail traffic platform door fault diagnosis system based on deep learning that the present invention relates to a kind of, belongs to rail traffic station
Platform door fault diagnosis field.
Background technique
Rail traffic platform door is mounted on platform edge, and orbital region and platform are waited zone isolation, prevent passenger from falling
Track is fallen, reduction Si Ji looks at number, reduces the number of dropouts of the cold and hot gas of platform, reduces platform air-conditioning system energy loss, reduces
Noise, dust that train operation generates etc. influences, and improves safety and comfort index that passenger waits, has safety, energy conservation, ring
The functions such as guarantor, beauty.In addition, the agility of the safety of rail traffic platform door system, reliability and O&M will also be direct
Influence the transportation safety and efficiency of urban track traffic.Therefore ensure the normal operation of rail traffic platform door system to Guan Chong
It wants, research rail traffic platform door fault diagnosis system is very urgent, really realizes the operating status to rail traffic platform door
Real-time monitoring is carried out, failure is timely responded to and carries out assault repairing.
China Patent No.: CN108268023 discloses a kind of rail traffic platform door remote fault diagnosis method and is
System.The fault diagnosis system generates diagnosis report in the event of a failure by collecting the parameter of each operative state of gate
Be sent to remote terminal, professional research and development technology personnel can long-range real time inspection Trouble Report, then maintenance personal is assigned to carry out
Troubleshooting.
China Patent No.: CN202008608 discloses the failure of urban railway system platform shielding door redundancy PEDC a kind of
Detection device.The fault detection means is used to issue high level, triggering redundancy PEDC switch when exploitation door failure occurs for PEDC
The switching of door control logic.
Above-mentioned patent the degree of automation is not high, and some need artificially participates in, and is unable to real-time response track complicated and changeable and hands over
Logical platform door system.In addition, the failure of track gate has randomness and uncertainty, only by increase shield door redundancy
The fault detection means of PEDC can not solve many failure problems.
When the operation of rail traffic platform door is broken down, the information and track gate of all kinds of monitoring sensor acquisitions are just
The data often acquired when operation are objectively that there are some differences.Due to the more sum numbers of data class of various monitoring sensor acquisitions
Amount is big, and traditional fault diagnosis system can not integrate multi-modal data collected and carry out feature association and feature extraction, usually
Analysis on Fault Diagnosis is carried out according only to certain several significant index of artificial objective setting, it is potentially a large amount of to have ignored multi-modal data
Efficient association information.In addition, data analysis algorithm, during application, the quality of selection and the definition of multi-modal data determines
The superiority and inferiority and order of accuarcy of testing result.
Summary of the invention
Analysis on Fault Diagnosis is carried out using certain several significant index of artificial objective setting for the prior art, there is event
Hinder the waste problem of time and efforts caused by the inaccuracy and artificial data analysis of diagnosis, the present invention provides one kind
Rail traffic platform door fault diagnosis system based on deep learning, it is automatic using the convolutional neural networks (CNN) of deep learning
The multidimensional characteristic for extracting monitoring sensor data collected is realized and is excavated to the profound level of a large amount of multi-modal data information,
Fast and effective reaction can be made to abnormal data well, so as to respond rail traffic platform door system complicated and changeable well
System.
The technical solution mainly used in the present invention are as follows:
A kind of rail traffic platform door fault diagnosis system based on deep learning, including rail traffic platform door body system
System, door machine transmission system, unit accelerator control system and fault diagnosis system, wherein
The rail traffic platform door body system includes upper support structure, sliding door, fixed door, escape door, end door, consolidates
Determine side box, top case, threshold;
The door machine transmission system includes door machine beam, driving motor, guide rail, gear changing group, driven wheel;
The unit accelerator control system includes controller, monitors sensor and data transmission set;
The door machine transmission system by unit accelerator control system timing control, for controlling rail traffic platform door body system
System closes the switch, and the monitoring sensor of the unit accelerator control system acquires items when door machine transmission system is run respectively
Data, and fed back to fault diagnosis system;
The fault diagnosis system acquires the feedback signal of monitoring sensor in real time, while generating event using server training
Hinder diagnostic model, and real-time fault diagnosis is carried out by fault diagnosis model gate operating status.
The specific knot in rail traffic platform door body system, door machine transmission system, unit accelerator control system in the present invention
Structure, connection relationship and operation logic relationship belong to the prior art, therefore specific detailed description.
Preferably, specific step is as follows for the data set construction method of the fault diagnosis system:
Step 2-1: the feedback signal of all kinds of monitoring sensors in acquisition unit accelerator control system;
Step 2-2: uploading to Cloud Server for the collected feedback signal of step 2-1 and handle, and is sensed according to monitoring
The signal producing method difference of device is classified, and is divided into the monitoring sensor for generating continuous data and is generated discrete data
Monitor sensor, wherein generate when the monitoring sensor of discrete data breaks down according to track gate signal wave emotionally
Condition is divided into the monitoring sensor easily to break down in the case where numerical value is low and is easy to happen the monitoring sensing of failure in numerical value height
Device;
Step 2-3: all kinds of monitoring sensors are acquired in a period of time TSecondary data, to constituteTwo dimension
Matrix data, as a training sample, wherein M indicates the quantity of monitoring sensor, and the column of matrix transverse direction represent each unit
The data of sensor are monitored on timing node, longitudinal row represents all kinds of monitoring sensors;
Step 2-4: dimensional matrix data obtained in step 2-3 is normalized;
Step 2-5: to the data and fortune of operation troubles in the dimensional matrix data after step 2-4 is normalized
The normal data of row are labeled, thus training dataset and test data set needed for obtaining.
Preferably, the model training generating process of the fault diagnosis system includes model training, model verifying and model
Test, wherein
Specific step is as follows for the model training method:
Step 3-1: convolutional Neural net is input to using each of training dataset training sample as input layer data
In network;
Step 3-2: training sample is by each convolutional layer, pond layer, open and flat layer and each full connection in convolutional neural networks
The extraction feature and classification processing of layer calculate the prediction classification of gate operating status, export the predicted value of training sample;
Step 3-3: judge whether the error function between the real output value and predicted value of all training samples converges to
Whether the required precision of setting and training the number of iterations reach specified the number of iterations, when the not converged extremely setting of error function
Required precision, and training the number of iterations then carries out backpropagation, adjustment to convolutional neural networks when reaching specified the number of iterations
Network parameter;When error function converges to the required precision of setting, and the iteration that training the number of iterations reaches or not up to specifies
When number, then training terminates;
Specific step is as follows for the model verifying:
Step 3-4: during model training, concentrate a certain proportion of training sample as verifying number training data
According to;
Step 3-5: the model accuracy rate being calculated according to verify data judges the quality of the model, if reach target
It is required that then obtaining gate fault diagnosis convolutional neural networks model;
Specific step is as follows for the model measurement mode:
Step 3-6: selection gate operation data is concentrated to be input to trained gate fault diagnosis test data
The input layer of model exports the classification results of prediction;
Step 3-7: according to the classification results of the prediction of gate fault diagnosis convolutional neural networks model output, with test
Data label in data set is compared, wherein data label includes normal operation and the two categories that break down, to count
The accuracy rate of fault diagnosis model is calculated, if the accuracy rate of gate fault diagnosis convolutional neural networks model does not reach essence
Degree requires, then optimizes to gate fault diagnosis convolutional neural networks model, until reaching expected required precision.
Preferably, the track gate operating status assorting process are as follows: by real-time gate operation data to be sorted
It is input in model training generating process stage obtained gate fault diagnosis convolutional neural networks model, obtains diagnosis mould
Classification results of the type to real-time gate operating status.
Preferably, the optimization structure of convolutional neural networks described in the step 3-2 includes that convolution kernel size is different
Convolutional layer, pond layer, developer layer, full articulamentum, quick connection structure and dropout regularization layer, the training sample it is defeated
It is divided into three paths by the different convolutional layer of convolution kernel size after entering to convolutional neural networks respectively to be learnt;Institute
Convolutional neural networks are stated by introducing and expand the quick connection structure of ResNet, after each path and each pond layer,
The data of network shallow-layer are transferred directly to deep layer before and after path by the connection type before and after build path between path;?
Between path, the layer in each paths is connected to other path and assigns certain weight, makes to connect each other between each path
And share Partial Feature, after three paths, output is cascaded by back-propagation algorithm self study weight parameter, is formed
Data then are unfolded to export by multi-dimensional matrix data by multi-dimensional matrix data input developer layer, and the output data of developer layer is into one
Step is input to dropout regularization layer, over-fitting occurs for reducing model, finally by close between full articulamentum
Connection carries out the classification of data, and the error between computation model real output value and desired value carries out backpropagation to network,
Adjust network weight parameter.
Preferably, shown in the calculation formula such as formula (1) of the accuracy rate A of the fault diagnosis model in the step 3-7:
In formula (1), m is correct amount of test data of classifying, and n is test data sum.
The utility model has the advantages that the present invention provides a kind of rail traffic platform door fault diagnosis system based on deep learning, with original
Beginning method for diagnosing faults is compared, and implicit spy can be learnt from large capacity multi-modal data automatically by taking full advantage of neural network
Sign, rather than the characteristic model of engineer.In addition, the characteristic model of engineer, due to its processing capacity and analyzes data
Ability is limited, and the parameter utilized is also less, and deep learning may include thousands of parameter, and with data set
Increase, the robustness of deep learning algorithm, generalization ability are stronger.It is more convenient, for data sample, it is only necessary to carry out
Normalized carries out complicated feature selecting and extraction to data sample using algorithm without other.Therefore, the present invention exists
Guarantee high degree of automation while the high-accuracy of fault diagnosis, there is important market value.
Detailed description of the invention
Fig. 1 is present system disposed of in its entirety flow chart;
Fig. 2 is present system overall architecture block diagram;
Fig. 3 is data set building process flow chart of the present invention;
Fig. 4 is New model pond of the present invention process flow diagram;
Fig. 5 is model training process flow diagram flow chart of the present invention;
Fig. 6 is model measurement process flow diagram flow chart of the present invention;
Fig. 7 is the structural design drawing of the novel convolutional neural networks model of the present invention;
Fig. 8 is the structural design drawing of the novel convolutional neural networks Optimized model of the present invention;
Convolution-pond function structure chart when Fig. 9 in the novel convolutional neural networks model of the present invention and Optimized model.
Specific embodiment
In order to make those skilled in the art better understand the technical solutions in the application, below to the embodiment of the present application
In technical solution be clearly and completely described, it is clear that described embodiments are only a part of embodiments of the present application,
Instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not making creative labor
Every other embodiment obtained under the premise of dynamic, shall fall within the protection scope of the present application.
Further detailed description has been done to technical solution of the present invention with reference to the accompanying drawing:
A kind of rail traffic platform door fault diagnosis system based on deep learning, including rail traffic platform door body system
System, door machine transmission system, unit accelerator control system and fault diagnosis system, wherein
The rail traffic platform door body system includes upper support structure, sliding door, fixed door, escape door, end door, consolidates
Determine side box, top case, threshold;
The door machine transmission system includes door machine beam, driving motor, guide rail, gear changing group, driven wheel;
The unit accelerator control system includes controller, monitors sensor and data transmission set, further include industrial personal computer,
PLC etc. controls component, belongs to routine techniques, is described in detail so not adding;
The door machine transmission system by unit accelerator control system timing control, for controlling rail traffic platform door body system
System closes the switch, and the monitoring sensor of the unit accelerator control system acquires items when door machine transmission system is run respectively
Data, and fed back to fault diagnosis system;
The fault diagnosis system acquires the feedback signal of monitoring sensor in real time, while generating event using server training
Hinder diagnostic model, and real-time fault diagnosis is carried out by fault diagnosis model gate operating status.Gate is transported in the present invention
Row state is fault diagnosis model to be monitored the feedback signal of sensor and passes through deep learning mould by collecting in gate operational process
Type analysis data potential information judges.
Preferably, specific step is as follows for the data set construction method of the fault diagnosis system:
Step 2-1: the feedback signal of all kinds of monitoring sensors in acquisition unit accelerator control system;
Step 2-2: uploading to Cloud Server for the collected feedback signal of step 2-1 and handle, and is sensed according to monitoring
The signal producing method difference of device is classified, and is divided into the monitoring sensor for generating continuous data and is generated discrete data
Monitor sensor, wherein generate when the monitoring sensor of discrete data breaks down according to track gate signal wave emotionally
Condition is divided into the monitoring sensor easily to break down in the case where numerical value is low and is easy to happen the monitoring sensing of failure in numerical value height
Device;
Step 2-3: all kinds of monitoring sensors are acquired in a period of time TSecondary data, to constituteTwo
Matrix data is tieed up, as a training sample, wherein M indicates the quantity of monitoring sensor, and the column of matrix transverse direction represent each list
The data of sensor are monitored on the timing node of position, longitudinal row represents all kinds of monitoring sensors;
Step 2-4: dimensional matrix data obtained in step 2-3 is normalized;
Step 2-5: to the data and fortune of operation troubles in the dimensional matrix data after step 2-4 is normalized
The normal data of row are labeled, thus training dataset and test data set needed for obtaining.
Preferably, the model training generating process of the fault diagnosis system includes model training, model verifying and model
Test, wherein
Specific step is as follows for the model training method:
Step 3-1: convolutional Neural net is input to using each of training dataset training sample as input layer data
In network;
Step 3-2: training sample is by each convolutional layer, pond layer, open and flat layer and each full connection in convolutional neural networks
The extraction feature and classification processing of layer calculate the prediction classification of gate operating status, export the predicted value of training sample;
Step 3-3: judge whether the error function between the real output value and predicted value of all training samples converges to
Whether the required precision of setting and training the number of iterations reach specified the number of iterations, when the not converged extremely setting of error function
Required precision, and training the number of iterations then carries out backpropagation, adjustment to convolutional neural networks when reaching specified the number of iterations
Network parameter;When error function converges to the required precision of setting, and the iteration that training the number of iterations reaches or not up to specifies
When number, then training terminates;
Specific step is as follows for the model verifying:
Step 3-4: during model training, concentrate a certain proportion of training sample as verifying number training data
According to;
Step 3-5: the model accuracy rate being calculated according to verify data judges the quality of the model, if reach target
It is required that then obtaining gate fault diagnosis convolutional neural networks model;
Specific step is as follows for the model measurement mode:
Step 3-6: selection gate operation data is concentrated to be input to trained gate fault diagnosis test data
The input layer of model exports the classification results of prediction;
Step 3-7: according to the classification results of the prediction of gate fault diagnosis convolutional neural networks model output, with test
Data label in data set is compared, wherein data label includes normal operation and the two categories that break down, to count
The accuracy rate of fault diagnosis model is calculated, if the accuracy rate of gate fault diagnosis convolutional neural networks model does not reach essence
Degree requires, then optimizes to gate fault diagnosis convolutional neural networks model, until reaching expected required precision.
Preferably, the track gate operating status assorting process are as follows: by real-time gate operation data to be sorted
It is input in model training generating process stage obtained gate fault diagnosis convolutional neural networks model, obtains diagnosis mould
Classification results of the type to real-time gate operating status.In the present invention, all kinds of monitorings during gate real time execution are passed
The monitoring signals of sensor feedback, in input fault diagnostic model, model according to data itself and mutual inherent can join
System, predicts whether these data instantly show that failure has occurred.
Preferably, the optimization structure of convolutional neural networks described in the step 3-2 includes that convolution kernel size is different
Convolutional layer, pond layer, developer layer, full articulamentum, quick connection structure and dropout regularization layer, the training sample it is defeated
It is divided into three paths by the different convolutional layer of convolution kernel size after entering to convolutional neural networks respectively to be learnt;Institute
Convolutional neural networks are stated by introducing and expand the quick connection structure of ResNet, after each path and each pond layer,
The data of network shallow-layer are transferred directly to deep layer before and after path by the connection type before and after build path between path;?
Between path, the layer in each paths is connected to other path and assigns certain weight, makes to connect each other between each path
And share Partial Feature, after three paths, output is cascaded by back-propagation algorithm self study weight parameter, is formed
Data then are unfolded to export by multi-dimensional matrix data by multi-dimensional matrix data input developer layer, and the output data of developer layer is into one
Step is input to dropout regularization layer, over-fitting occurs for reducing model, finally by close between full articulamentum
Connection carries out the classification of data, and the error between computation model real output value and desired value carries out backpropagation to network,
Adjust network weight parameter.
Preferably, shown in the calculation formula such as formula (1) of the accuracy rate A of the fault diagnosis model in the step 3-7:
In formula (1), m is correct amount of test data of classifying, and n is test data sum.
Data set constructs in the present invention:
The training and generation of convolutional neural networks model, which are built upon, carries out feature extraction to a large amount of data with label
It is realized on the basis of study.Proposed by the present invention is a kind of rail traffic platform door fault diagnosis system based on deep learning
System, the basis which establishes are a large amount of gates by the monitoring sensor acquisition on unit door system
Operation troubles data and normal operation data, then all a large amount of multi-modal datas are uploaded to built in real time in distributed cloud
On high-performance server on, pass through the normalized to different achievement datas, construct data set.
As shown in figure 3, the present invention is directed to the specific practical application of rail traffic platform door, platform door system is fully taken into account
In monitoring sensor type it is varied, difference monitoring sensors acquisition, generate signal data modes be not quite similar.Cause
This, is in order to allow fault diagnosis model in the training process, monitoring sensor data collected that can preferably to different attribute
The feature for being adapted to monitoring attribute sensor is extracted, to be divided according to the difference of monitors sensor signal producing method
Class is divided into the monitoring sensor for generating continuous data and discrete data, for generating the monitoring sensor of discrete data
Signal fluctuation situation is finely divided when further being broken down according to track gate, is divided into and being easy in the lower situation of numerical value
The monitoring sensor that breaks down and the monitoring sensor that failure is easy to happen when numerical value is higher, in conclusion monitoring sensing
Device has been divided into three classes.But it can monitoring sensor itself the attribute progress according to used in specific different rail traffic platform systems
The adjustment of classification number.Numerical value height does not need explicitly to describe, and can be referred to by specifically monitoring the items of normal operation of sensor
Mark numerical value height is judged.The height of the numerical value, those skilled in the art can saying by monitoring normal operation of sensor
Bright book judged, or can be by breaking down when monitoring sensor it is higher or inclined relative to belonging to when operating normally
It is low to be judged.The present invention is embodied in case, in order to better illustrate problem using above-mentioned classification, thus by 100
It monitors sensor and (30,30,40) three classes is divided into according to above-mentioned classification method.
The acquired data of sensor are respectively monitored whithin a period of time to preferably analyze when track gate failure occurs
Difference because usually to the analysis of failure being made a decision by the variation tendency to data, 100 monitoring sensors
30 data in 60 minutes are acquired respectively, to constitute the dimensional matrix data of (30+30+40) * 30.Since each monitoring senses
The dimension of the acquired data of device is different, to also need that the data of acquisition are normalized, by all data all normalizings
Change to [0,1], then the data of the data of operation troubles and normal operation are labeled, thus the data set needed for obtaining.
Model training generating process of the invention includes model training, model verifying and model measurement.
(1) model training
The present invention fully takes into account the particularity of rail traffic platform door fault diagnosis system, by traditional convolutional neural networks
Model is combined with the fault diagnosis system self attributes of gate, a kind of novel convolutional neural networks is designed, such as Fig. 7 institute
Show, can achieve a large amount of real-time multimode state data collected in input track traffic platform door operational process, accurately and efficiently
The current operating conditions of prediction output gate, are to break down also to be up.
Model training process flow is as shown in figure 5, during model training, for three kinds different types of (30,30,40)
Sensor is monitored, as shown in figure 4, the pond layer in the novel convolutional neural networks of the present invention, respectively to 1-30 monitoring sensor
Data use average value pond method, preferably to extract the individual features of continuous type signal;To 31-60 monitoring sensor number
According to maximum value pond method is used, preferably to extract the individual features for being easy to happen failure in the case where numerical value is larger;To 61-
100 monitoring sensing datas use minimum value pond method, are easy to happen failure in the case where numerical value is smaller preferably to extract
Individual features.
Further, since track platform door system is a huge and complicated system, monitoring sensor is deposited between each other
In various connections.Then, when carrying out fault diagnosis, cannot only consider the monitoring sensor number of each classification
It according to separated carry out feature extraction, also should consider to be combined to be judged, in conjunction with above-mentioned consideration, be designed in the present invention
A kind of novel convolutional neural networks model, as shown in figure 5, the connection between passage path is increasing data on each path
Correlation.Firstly, training dataset is input in network, by the different convolutional layer of convolution kernel size (in such as Fig. 7
Convolution 1, convolution 2, convolution 3) after be divided into three paths and respectively learnt.Then, the present invention is introduced and is opened up in a model
The Novel rapid connection structure of ResNet has been opened up, as shown in figure 9, after each path and each pond layer, through this structure
Connection type before and after build path between path.By the application to fast connecting, before and after path, by network shallow-layer
Data are transferred directly to deep layer;And between path, the layer in each paths is connected to other path and assigns certain weight,
Make to connect and share Partial Feature between each path each other, to reach the mesh that various types of monitoring sensing data is contacted
's.Output for each path, the output in this path will occupy biggish weight, and the output bypassed is then relatively secondary.
After three paths, output will carry out grade by back-propagation algorithm self study weight parameter (w1, w2, w3 in such as Fig. 7)
Connection, so that fault diagnosis model more accurately can carry out mining analysis to multi-modal data, also make with deeper time, more effectively
The novel convolutional neural networks that must be designed more rationally and meet reality.
In each paths, as shown in figure 9, multiple convolutional layers and pond layer module-cascade carry out convolution to the data of input
Operation automatically extracts the extraction of data implicit features.Multilayer convolution can acquire the feature being more globalized, but convolutional layer is not yet
It is The more the better.Finally, the data that each path exports are cascaded to form multi-dimensional matrix data, developer layer is subsequently input by data
Expansion.In order to reduce the over-fitting of training pattern, the output data of developer layer is further input by Hinton proposition
Dropout regularization layer, to reduce the possibility that over-fitting occurs for model.Finally, by close between full articulamentum
Connection carries out the classification of data, and the error between computation model real output value and desired value carries out backpropagation to network, adjusts
Whole network weight parameter.Judge to be by whether converging to error function required precision and the training the number of iterations of setting
The no requirement for having reached setting terminates the training of model if reaching requirement, if specified precision is not achieved in the output of model
It is required that the optimization processing of model is then carried out, re -training model.
(2) model is verified
Model verifying is during model training, using the certain proportion of training data as verify data and model
The unity of thinking of test, the data for only first passing through verifying collection carry out Performance Evaluation to trained model, survey for subsequent model
Examination provides reference.
(3) model measurement
After model training generates, needing further to utilize, the training sample that test data is concentrated carries out model performance test,
To verify the generalization ability and robustness of model.
As shown in fig. 6, model measurement is substantially exactly to carry out forward calculation to model using test data, input number is obtained
According to prediction result output.Firstly, trained fault diagnosis model is imported according to the model name saved during model training,
It is concentrated from test data and chooses gate operation data, entered data into the input layer of model, utilize novel volume of the invention
Convolutional layer and pond layer in product neural network model carry out the Automatic signature extraction of data, finally pass through the open and flat place of developer layer
Reason, the reduction data over-fitting processing of Dropout regularization layer, the classification of full articulamentum, the classification to output test data, according to
As a result classification judges whether rail traffic platform door has occurred failure at this time.To all test sample data, by by model
The label information of the classification and data itself of predicting output is compared, and calculates the accuracy rate A of model measurement, calculation formula is such as
Shown in formula (1):
In formula (1), accuracy rate A=m (correct amount of test data of classifying)/n (test data sum).
If accuracy rate A is lower than preset accuracy rate value, illustrate that the obtained network model generalization ability of training is weaker, performance compared with
Difference, it is therefore desirable to continue to optimize network model to improve its test accuracy rate.
For the present invention by the optimization of following three method implementation models, the network architecture after optimization is as shown in Figure 8.
Method 1: the extension of training dataset
Training dataset is the basis that convolutional neural networks model carries out feature learning, and size has model training precision
Great influence.Data set is too small, and the feature extracted is not enough, and model just will receive limit for the abstracting power of feature learning
System, and then influence test accuracy rate, the deficiency of generalization and robustness of model.Therefore, it in order to improve the accuracy rate of model, needs
Data set is extended.For the training dataset of rail traffic platform door fault diagnosis model, broken down by gate
When monitoring sensor acquisition signal data and the signal data of monitoring sensor acquisition is formed when normal operation, especially
Data collected when for breaking down, since in most cases gate is all under normal operating condition, so adopting
The fault data collected may be less, and the present invention is by establishing huge distributed cloud service platform, by the platform of each section
The data that door breaks down upload to Cloud Server, are stored, so as to effective extended model training set.
The variation of 2 prototype network structure of method
The network hierarchical structure of model will affect the test accuracy rate of model, especially to convolutional layer, pond layer, full connection
The setting of layer and Dropout layers of number, according to "ockham's razor" (Occam ' s razor) principle, for the weight of different models
Parameter, under the premise of can achieve same precision, the more simple weight parameter of optimum selecting.The tool handled in training pattern
Body method is exactly to increase a weight regularization term in loss function, including L1 regularization, L2 regularization are used herein
Be the Dropout regularization proposed by Hinton, a possibility that reduce generation model over-fitting, to improve test
The accuracy rate of collection.The present invention optimizes calculation formula after network model increases Dropout layers are as follows:
Wherein Bernoulli function in formula (2) is in order to Probability p, the random vector for generating one 0,1.By right
Measuring accuracy under heterogeneous networks structure compares, the highest model structure of choice accuracy, to obtain relatively good model structure.
The adjustment of 3 prototype network parameter of method
Due to convolution kernel size and step-length, training the number of iterations, the calculation method of loss function, optimizer selection and
Learning rate etc. all affects the test accurate rate that training generates model.Initialization for the parameter of network model, master of the present invention
What is relied on is that empirical value is adjusted with continuous by result.
The classification of track gate operating status
In track gate operating status assorting process, each monitoring sensor of unit door system is acquired in real time
Signal data uploads to cloud service platform in real time and carries out data prediction, then calls trained high-precision and high extensive energy
The fault diagnosis model of power judges current track gate operating status, breaks down or normal operation.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (6)
1. a kind of rail traffic platform door fault diagnosis system based on deep learning, which is characterized in that including rail traffic station
Platform door body system, door machine transmission system, unit accelerator control system and fault diagnosis system, wherein
The rail traffic platform door body system includes upper support structure, sliding door, fixed door, escape door, end door, affixed side
Box, top case, threshold;
The door machine transmission system includes door machine beam, driving motor, guide rail, gear changing group, driven wheel;
The unit accelerator control system includes controller, monitors sensor and data transmission set;
The door machine transmission system by unit accelerator control system timing control, for controlling rail traffic platform door body system
It closes the switch, and the monitoring sensor of the unit accelerator control system acquires each item number when door machine transmission system is run respectively
According to, and fed back to fault diagnosis system;
The fault diagnosis system acquires the feedback signal of monitoring sensor in real time, while generating failure using server training and examining
Disconnected model, and real-time fault diagnosis is carried out by fault diagnosis model gate operating status.
2. a kind of rail traffic platform door fault diagnosis system based on deep learning according to claim 1, feature
It is, specific step is as follows for the data set construction method of the fault diagnosis system:
Step 2-1: the feedback signal of all kinds of monitoring sensors in acquisition unit accelerator control system;
Step 2-2: uploading to Cloud Server for the collected feedback signal of step 2-1 and handle, according to monitoring sensor
Signal producing method difference is classified, and the monitoring sensor for generating continuous data and the monitoring for generating discrete data are divided into
Sensor, wherein generate signal fluctuation situation point when the monitoring sensor of discrete data breaks down according to track gate
The monitoring sensor of failure is easy to happen for the monitoring sensor easily to break down in the case where numerical value is low and in numerical value height;
Step 2-3: all kinds of monitoring sensors are acquired in a period of time TSecondary data, to constituteTwo-dimensional matrix
Data, as a training sample, wherein M indicates the quantity of monitoring sensor, and the column of matrix transverse direction represent each unit time
The data of sensor are monitored on node, longitudinal row represents all kinds of monitoring sensors;
Step 2-4: dimensional matrix data obtained in step 2-3 is normalized;
Step 2-5: in the dimensional matrix data after step 2-4 is normalized just to the data of operation troubles and operation
Normal data are labeled, thus training dataset and test data set needed for obtaining.
3. a kind of rail traffic platform door fault diagnosis system based on deep learning according to claim 2, feature
It is, the model training generating process of the fault diagnosis system includes model training, model verifying and model measurement, wherein
Specific step is as follows for the model training method:
Step 3-1: it is input to each of training dataset training sample as input layer data in convolutional neural networks;
Step 3-2: training sample passes through each convolutional layer in convolutional neural networks, pond layer, open and flat layer and each full articulamentum
Feature and classification processing are extracted, the prediction classification of gate operating status is calculated, exports the predicted value of training sample;
Step 3-3: judge whether the error function between the real output value and predicted value of all training samples converges to setting
Required precision and training the number of iterations whether reach specified the number of iterations, when error function it is not converged to setting precision
It is required that and training the number of iterations backpropagation then is carried out to convolutional neural networks when reaching specified the number of iterations, adjust network
Parameter;When error function converges to the required precision of setting, and the number of iterations that training the number of iterations reaches or not up to specifies
When, then training terminates;
Specific step is as follows for the model verifying:
Step 3-4: during model training, concentrate a certain proportion of training sample as verify data training data;
Step 3-5: the model accuracy rate that is calculated according to verify data judges the quality of the model, wants if reaching target
It asks, then obtains gate fault diagnosis convolutional neural networks model;
Specific step is as follows for the model measurement mode:
Step 3-6: selection gate operation data is concentrated to be input to trained gate fault diagnosis model test data
Input layer, export the classification results of prediction;
Step 3-7: according to the classification results of the prediction of gate fault diagnosis convolutional neural networks model output, with test data
The data label of concentration is compared, wherein data label includes normal operation and the two categories that break down, to calculate
The accuracy rate of fault diagnosis model, if the accuracy rate of gate fault diagnosis convolutional neural networks model does not reach precision and wants
It asks, then gate fault diagnosis convolutional neural networks model is optimized, until reaching expected required precision.
4. a kind of rail traffic platform door fault diagnosis system based on deep learning according to claim 3, feature
It is, the track gate operating status assorting process are as follows: real-time gate operation data to be sorted is input to model
In training generating process stage obtained gate fault diagnosis convolutional neural networks model, diagnostic model is obtained to real-time station
The classification results of platform door operating status.
5. a kind of rail traffic platform door fault diagnosis system based on deep learning according to claim 3, feature
It is, the optimization structure of convolutional neural networks described in the step 3-2 includes the different convolutional layer of convolution kernel size, pond
Change layer, developer layer, full articulamentum, quick connection structure and dropout regularization layer, the training sample and is input to convolutional Neural
It is divided into three paths by the different convolutional layer of convolution kernel size after network respectively to be learnt;The convolutional Neural net
Network passes through the quick connection structure for introducing and expanding ResNet, after each path and each pond layer, before and after build path
The data of network shallow-layer are transferred directly to deep layer before and after path by the connection type between path;It, will between path
Layer in each paths is connected to other path and assigns certain weight, makes to connect and share part spy between each path each other
Sign, after three paths, output is cascaded by back-propagation algorithm self study weight parameter, forms multi-dimensional matrix number
According to then by multi-dimensional matrix data input developer layer by data expansion output, the output data of developer layer is further input into
For reducing model over-fitting occurs for dropout regularization layer, finally by the close connection between full articulamentum, into
The classification of row data, the error between computation model real output value and desired value carry out backpropagation to network, adjust network
Weight parameter.
6. a kind of rail traffic platform door fault diagnosis system based on deep learning according to claim 3, feature
It is, shown in the calculation formula such as formula (1) of the accuracy rate A of the fault diagnosis model in the step 3-7:
In formula (1), m is correct amount of test data of classifying, and n is test data sum.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910613949.1A CN110262463B (en) | 2019-07-09 | 2019-07-09 | Rail transit platform door fault diagnosis system based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910613949.1A CN110262463B (en) | 2019-07-09 | 2019-07-09 | Rail transit platform door fault diagnosis system based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110262463A true CN110262463A (en) | 2019-09-20 |
CN110262463B CN110262463B (en) | 2021-12-10 |
Family
ID=67925126
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910613949.1A Active CN110262463B (en) | 2019-07-09 | 2019-07-09 | Rail transit platform door fault diagnosis system based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110262463B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110822768A (en) * | 2019-11-18 | 2020-02-21 | 河南大学 | Intelligent recovery system for waste hot water heat of residential area |
CN110908359A (en) * | 2019-11-01 | 2020-03-24 | 日立楼宇技术(广州)有限公司 | Function test equipment, method and device |
CN111487950A (en) * | 2020-04-24 | 2020-08-04 | 西安交通大学 | 'prediction-verification-feedback-optimization' closed-loop system for online early warning and offline diagnosis |
CN111751655A (en) * | 2020-05-29 | 2020-10-09 | 深圳供电局有限公司 | Fault self-healing method and device for distribution line, computer equipment and storage medium |
CN111982256A (en) * | 2020-08-18 | 2020-11-24 | 上海陆杰电子科技有限公司 | Automobile balance pressure sensor fault early warning method based on multichannel CNN model |
CN112198799A (en) * | 2020-10-28 | 2021-01-08 | 北京交通大学 | High-speed train parking control method and system based on deep learning |
CN112381308A (en) * | 2020-11-20 | 2021-02-19 | 广州新科佳都科技有限公司 | Training of current prediction model, current prediction method and related device |
CN112432754A (en) * | 2020-10-14 | 2021-03-02 | 北京市地铁运营有限公司地铁运营技术研发中心 | Subway platform door impact monitoring method and device and readable storage medium |
CN112464958A (en) * | 2020-12-11 | 2021-03-09 | 沈阳芯魂科技有限公司 | Multi-modal neural network information processing method and device, electronic equipment and medium |
CN112993345A (en) * | 2021-04-21 | 2021-06-18 | 北京氢澜科技有限公司 | Artificial intelligence-based fuel cell control system and control method |
CN113255764A (en) * | 2021-05-21 | 2021-08-13 | 池测(上海)数据科技有限公司 | Method, system and device for detecting electrochemical energy storage system fault by using machine learning |
CN113267344A (en) * | 2021-02-05 | 2021-08-17 | 深圳大雷汽车检测股份有限公司 | Automobile abs detection device and method |
CN113568393A (en) * | 2021-07-01 | 2021-10-29 | 日立楼宇技术(广州)有限公司 | Platform door fault detection method and device, computer equipment and storage medium |
CN113625094A (en) * | 2021-09-11 | 2021-11-09 | 沈阳工业大学 | Subway platform screen door fault diagnosis system |
CN114021621A (en) * | 2021-10-13 | 2022-02-08 | 北京和利时系统集成有限公司 | Fault diagnosis method, system, storage medium and edge computing device |
CN115140102A (en) * | 2022-05-18 | 2022-10-04 | 卡斯柯信号有限公司 | Urban rail transit platform door linkage control fault detection method and device |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN202008608U (en) * | 2011-03-25 | 2011-10-12 | 谭铁仁 | Fault detection device for redundant PEDC (platform edge door controller) of urban rail transit platform screen door |
CN102407867A (en) * | 2011-08-09 | 2012-04-11 | 河南辉煌科技股份有限公司 | Urban rail transit signal maintenance supporting system |
CN106650919A (en) * | 2016-12-23 | 2017-05-10 | 国家电网公司信息通信分公司 | Information system fault diagnosis method and device based on convolutional neural network |
CN107219837A (en) * | 2016-03-21 | 2017-09-29 | 刘春梅 | One kind is used for subway shield door failure monitoring and early warning intelligence system |
CN108268023A (en) * | 2016-12-30 | 2018-07-10 | 上海嘉成轨道交通安全保障系统股份公司 | A kind of rail traffic platform door remote fault diagnosis method and system |
CN108319962A (en) * | 2018-01-29 | 2018-07-24 | 安徽大学 | A kind of Tool Wear Monitoring method based on convolutional neural networks |
CN108614548A (en) * | 2018-04-03 | 2018-10-02 | 北京理工大学 | A kind of intelligent failure diagnosis method based on multi-modal fusion deep learning |
EP3407151A1 (en) * | 2017-05-24 | 2018-11-28 | Tata Consultancy Services Limited | Systems and methods for cognitive control of data acquisition for efficient fault diagnosis |
CN108960337A (en) * | 2018-07-18 | 2018-12-07 | 浙江大学 | A kind of multi-modal complicated activity recognition method based on deep learning model |
CN109116203A (en) * | 2018-10-31 | 2019-01-01 | 红相股份有限公司 | Power equipment partial discharges fault diagnostic method based on convolutional neural networks |
CN208502530U (en) * | 2017-11-03 | 2019-02-15 | 上海嘉成轨道交通安全保障系统股份公司 | A kind of rail traffic platform accelerator control system |
KR20190018798A (en) * | 2017-08-16 | 2019-02-26 | 강병수 | car noise sound with Convolution Nueral Network classification method |
CN109814523A (en) * | 2018-12-04 | 2019-05-28 | 合肥工业大学 | Method for diagnosing faults based on CNN-LSTM deep learning method and more attribute time series datas |
CN109902399A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN |
-
2019
- 2019-07-09 CN CN201910613949.1A patent/CN110262463B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN202008608U (en) * | 2011-03-25 | 2011-10-12 | 谭铁仁 | Fault detection device for redundant PEDC (platform edge door controller) of urban rail transit platform screen door |
CN102407867A (en) * | 2011-08-09 | 2012-04-11 | 河南辉煌科技股份有限公司 | Urban rail transit signal maintenance supporting system |
CN107219837A (en) * | 2016-03-21 | 2017-09-29 | 刘春梅 | One kind is used for subway shield door failure monitoring and early warning intelligence system |
CN106650919A (en) * | 2016-12-23 | 2017-05-10 | 国家电网公司信息通信分公司 | Information system fault diagnosis method and device based on convolutional neural network |
CN108268023A (en) * | 2016-12-30 | 2018-07-10 | 上海嘉成轨道交通安全保障系统股份公司 | A kind of rail traffic platform door remote fault diagnosis method and system |
EP3407151A1 (en) * | 2017-05-24 | 2018-11-28 | Tata Consultancy Services Limited | Systems and methods for cognitive control of data acquisition for efficient fault diagnosis |
KR20190018798A (en) * | 2017-08-16 | 2019-02-26 | 강병수 | car noise sound with Convolution Nueral Network classification method |
CN208502530U (en) * | 2017-11-03 | 2019-02-15 | 上海嘉成轨道交通安全保障系统股份公司 | A kind of rail traffic platform accelerator control system |
CN108319962A (en) * | 2018-01-29 | 2018-07-24 | 安徽大学 | A kind of Tool Wear Monitoring method based on convolutional neural networks |
CN108614548A (en) * | 2018-04-03 | 2018-10-02 | 北京理工大学 | A kind of intelligent failure diagnosis method based on multi-modal fusion deep learning |
CN108960337A (en) * | 2018-07-18 | 2018-12-07 | 浙江大学 | A kind of multi-modal complicated activity recognition method based on deep learning model |
CN109116203A (en) * | 2018-10-31 | 2019-01-01 | 红相股份有限公司 | Power equipment partial discharges fault diagnostic method based on convolutional neural networks |
CN109814523A (en) * | 2018-12-04 | 2019-05-28 | 合肥工业大学 | Method for diagnosing faults based on CNN-LSTM deep learning method and more attribute time series datas |
CN109902399A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN |
Non-Patent Citations (1)
Title |
---|
佘博 等: "基于深度卷积变分自编码网络的故障诊断方法", 《仪器仪表学报》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110908359A (en) * | 2019-11-01 | 2020-03-24 | 日立楼宇技术(广州)有限公司 | Function test equipment, method and device |
CN110908359B (en) * | 2019-11-01 | 2021-03-09 | 日立楼宇技术(广州)有限公司 | Function test equipment, method and device |
CN110822768A (en) * | 2019-11-18 | 2020-02-21 | 河南大学 | Intelligent recovery system for waste hot water heat of residential area |
CN111487950A (en) * | 2020-04-24 | 2020-08-04 | 西安交通大学 | 'prediction-verification-feedback-optimization' closed-loop system for online early warning and offline diagnosis |
CN111487950B (en) * | 2020-04-24 | 2021-11-16 | 西安交通大学 | 'prediction-verification-feedback-optimization' closed-loop system for online early warning and offline diagnosis |
CN111751655A (en) * | 2020-05-29 | 2020-10-09 | 深圳供电局有限公司 | Fault self-healing method and device for distribution line, computer equipment and storage medium |
CN111751655B (en) * | 2020-05-29 | 2022-07-15 | 深圳供电局有限公司 | Fault self-healing method and device for distribution line, computer equipment and storage medium |
CN111982256B (en) * | 2020-08-18 | 2021-11-16 | 上海陆杰电子科技有限公司 | Automobile balance pressure sensor fault early warning method based on multichannel CNN model |
CN111982256A (en) * | 2020-08-18 | 2020-11-24 | 上海陆杰电子科技有限公司 | Automobile balance pressure sensor fault early warning method based on multichannel CNN model |
CN112432754A (en) * | 2020-10-14 | 2021-03-02 | 北京市地铁运营有限公司地铁运营技术研发中心 | Subway platform door impact monitoring method and device and readable storage medium |
CN112198799A (en) * | 2020-10-28 | 2021-01-08 | 北京交通大学 | High-speed train parking control method and system based on deep learning |
CN112381308A (en) * | 2020-11-20 | 2021-02-19 | 广州新科佳都科技有限公司 | Training of current prediction model, current prediction method and related device |
CN112464958A (en) * | 2020-12-11 | 2021-03-09 | 沈阳芯魂科技有限公司 | Multi-modal neural network information processing method and device, electronic equipment and medium |
CN113267344A (en) * | 2021-02-05 | 2021-08-17 | 深圳大雷汽车检测股份有限公司 | Automobile abs detection device and method |
CN112993345A (en) * | 2021-04-21 | 2021-06-18 | 北京氢澜科技有限公司 | Artificial intelligence-based fuel cell control system and control method |
CN112993345B (en) * | 2021-04-21 | 2022-05-03 | 北京氢澜科技有限公司 | Artificial intelligence-based fuel cell control system and control method |
CN113255764A (en) * | 2021-05-21 | 2021-08-13 | 池测(上海)数据科技有限公司 | Method, system and device for detecting electrochemical energy storage system fault by using machine learning |
CN113568393A (en) * | 2021-07-01 | 2021-10-29 | 日立楼宇技术(广州)有限公司 | Platform door fault detection method and device, computer equipment and storage medium |
CN113568393B (en) * | 2021-07-01 | 2023-10-27 | 日立楼宇技术(广州)有限公司 | Platform door fault detection method, device, computer equipment and storage medium |
CN113625094A (en) * | 2021-09-11 | 2021-11-09 | 沈阳工业大学 | Subway platform screen door fault diagnosis system |
CN114021621A (en) * | 2021-10-13 | 2022-02-08 | 北京和利时系统集成有限公司 | Fault diagnosis method, system, storage medium and edge computing device |
CN115140102A (en) * | 2022-05-18 | 2022-10-04 | 卡斯柯信号有限公司 | Urban rail transit platform door linkage control fault detection method and device |
CN115140102B (en) * | 2022-05-18 | 2024-03-29 | 卡斯柯信号有限公司 | Urban rail transit platform door linkage control fault detection method and device |
Also Published As
Publication number | Publication date |
---|---|
CN110262463B (en) | 2021-12-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110262463A (en) | A kind of rail traffic platform door fault diagnosis system based on deep learning | |
Capozzoli et al. | Fault detection analysis using data mining techniques for a cluster of smart office buildings | |
CN102520697B (en) | Onsite information preprocessing method of remote cooperative diagnosis | |
CN110287552B (en) | Motor bearing fault diagnosis method and system based on improved random forest algorithm | |
CN110455537A (en) | A kind of Method for Bearing Fault Diagnosis and system | |
CN107146004A (en) | A kind of slag milling system health status identifying system and method based on data mining | |
CN105846780A (en) | Decision tree model-based photovoltaic assembly fault diagnosis method | |
CN108304661A (en) | Diagnosis prediction method based on TDP models | |
CN108053052B (en) | A kind of oil truck oil and gas leakage speed intelligent monitor system | |
Wang et al. | A remaining useful life prediction model based on hybrid long-short sequences for engines | |
CN105842607A (en) | Test point quantitative selection method and device in testing design | |
CN108334902A (en) | A kind of track train equipment room smog fireproof monitoring method based on deep learning | |
Li et al. | Framework and case study of cognitive maintenance in Industry 4.0 | |
CN113962308A (en) | Aviation equipment fault prediction method | |
Zhou et al. | Fault classification for on-board equipment of high-speed railway based on attention capsule network | |
Jung | Engine fault diagnosis combining model-based residuals and data-driven classifiers | |
Liu et al. | Towards safety-risk prediction of CBTC systems with deep learning and formal methods | |
Zhang et al. | Grouped time series networks for grade monitoring of zinc tailings with multisource features | |
Lyu et al. | Examination on avionics system fault prediction technology based on ashy neural network and fuzzy recognition | |
CN114118288A (en) | Transformer state evaluation method based on deep learning | |
Chen et al. | Text mining of power secondary equipment based on BiLSTM-attention | |
Wang et al. | Time series fragmental variation trend anomaly detection method based on a temporal sequential modeling approach | |
Zhang et al. | Research on intelligent health management technology of opto-electronic equipment | |
Chen et al. | Exploration on-orbit management work of satellite payload | |
CN111832808A (en) | Flood early warning monitoring method based on big data |
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 | ||
TR01 | Transfer of patent right |
Effective date of registration: 20221207 Address after: 213000 Buyi industrial concentration area, Zou District, Zhonglou District, Changzhou City, Jiangsu Province Patentee after: Jiangsu haokai Machinery Co.,Ltd. Address before: 213022, No. 200, Jinling North Road, Xinbei District, Jiangsu, Changzhou Patentee before: CHANGZHOU CAMPUS OF HOHAI University |
|
TR01 | Transfer of patent right |