CN113569491A - Analysis and correction method and device for wheel set size detection data - Google Patents
Analysis and correction method and device for wheel set size detection data Download PDFInfo
- Publication number
- CN113569491A CN113569491A CN202110929385.XA CN202110929385A CN113569491A CN 113569491 A CN113569491 A CN 113569491A CN 202110929385 A CN202110929385 A CN 202110929385A CN 113569491 A CN113569491 A CN 113569491A
- Authority
- CN
- China
- Prior art keywords
- data
- prediction model
- wheel set
- model
- training
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012937 correction Methods 0.000 title claims description 33
- 238000004458 analytical method Methods 0.000 title claims description 12
- 238000012549 training Methods 0.000 claims abstract description 76
- 238000013210 evaluation model Methods 0.000 claims abstract description 61
- 238000007405 data analysis Methods 0.000 claims abstract description 22
- 238000011156 evaluation Methods 0.000 claims abstract description 15
- 238000005259 measurement Methods 0.000 claims description 93
- 238000013500 data storage Methods 0.000 claims description 27
- 238000012545 processing Methods 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 8
- 238000007726 management method Methods 0.000 claims description 7
- 230000007175 bidirectional communication Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims 2
- 230000008859 change Effects 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 6
- 238000012423 maintenance Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000006854 communication Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000005299 abrasion Methods 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention discloses a method and a device for analyzing and correcting wheel set size detection data, wherein the method comprises the following steps: and training an evaluation model based on the data set offline, carrying out validity judgment on online data measured in real time on site, dynamically updating an online prediction model, and analyzing and correcting the wheel set size detection data by evaluating a model evaluation result and comparing a prediction result and a measured value of the online prediction model. According to the method and the device for analyzing and correcting the wheel set size detection data, the wheel set size change prediction model and the wheel set size change evaluation model are established based on the big data analysis technology, the validity of online data can be automatically judged, the data are analyzed and screened, and the data are determined to be valid, invalid or required to be corrected; the grading evaluation method combining the evaluation model with the online prediction model is provided, and the prediction model is judged step by step from coarse to fine and dynamically adjusted and optimized, so that the reliability of detection is improved, the false alarm rate is reduced, and the detection rate is improved.
Description
Technical Field
The invention belongs to the technical field of rail transit safety detection monitoring equipment, and particularly relates to a method and a device for analyzing and correcting wheel set size detection data.
Background
The wheel set is used as a key part of the train, the abrasion and the operation state of the wheel set directly influence the train driving safety, and the wheel set dimension detection, namely the measurement and the acquisition of the relevant geometric dimension information of the wheel set, is one of the key links of the rail transit operation and maintenance safety detection. With the rapid development of the rail transit industry in China, the number of vehicles is increased day by day, the operation time is prolonged continuously, the operation and maintenance pressure of the vehicles is increased day by day, the conventional manual parking detection cannot meet the safety requirement, and the automatic detection equipment is gradually applied. At present, most of wheel set size detection equipment adopts a visual measurement technology, a visual sensor is easily influenced by an external environment, particularly under the complex and severe outdoor environment of a railway, the high-speed movement, the vibration impact, the strong and weak complex light and the interference of rain, snow and fog weather of a train can influence the imaging of the sensor, and the measurement precision of a visual system is greatly reduced.
When the measurement data obtained by the safety detection equipment has large errors, the operation and maintenance work of the rail transit is seriously interfered, and if the errors occur, the error report causes unnecessary manual recheck and recheck, so that the labor waste is caused; if the condition is too serious, the condition of overrun cannot be found by missed detection, and potential safety hazards are formed.
Due to the fact that the wheel set measuring environment is complex when the train runs, the measured value is often large in error occasionally and fluctuates to a certain extent when the vision measuring system is interfered. Meanwhile, dirt or foreign matters on the wheel can influence the measurement result, so that the measurement value is unstable. For this reason, it is necessary to analyze and evaluate the measurement data, remove abnormal values, and correct them. And relative submillimeter-level measurement accuracy, the numerical range of the measured value is usually large, multiple levels exist in the transfinite safety limit, and whether the measured data is effective or not is judged by a conventional simple threshold value, so that the reliable requirement is difficult to reach. The railway lines of China are numerous, the operating environments have certain differences, the vehicle types and the purposes are different, and it is difficult to acquire full sample data under the actual operating conditions of all wheel sets and establish a stable prediction model, so that the effective and reliable judgment and correction of measured data are difficult to realize by the conventional big data analysis and data prediction method.
In addition, in the prior art, a prediction model is usually trained offline by using historical data, then the prediction model is deployed on field equipment to calculate and process online data, and then a prediction result is output.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention aims to provide a method and a device for analyzing and correcting wheel set size detection data.
In order to achieve the purpose and achieve the technical effect, the invention adopts the technical scheme that:
a method for analyzing and correcting wheel set size detection data, comprising the steps of:
step one, establishing a wheel set dimension measurement value prediction model, and training the prediction model after manually marking historical data of different fields;
establishing an evaluation model for evaluating the effectiveness of the wheel set size detection data and judging whether the error is overlarge or not and whether the error is effective data or not;
step three, establishing a reference prediction model, wherein the reference prediction model is consistent with the prediction model in the step one in principle, and the evaluation model parameters are used as the super-parameters for reference prediction model training;
calculating the effectiveness of each item of data by an evaluation model for historical data sets of different sites, and updating effective attributes in the manual labeling information before updating to obtain an updated training data set;
step five, training the reference prediction model by adopting the training data set updated in the step four, and taking a predicted value obtained by calculation of the prediction model in the training as a data true value;
step six, repeating the step four to the step five, and performing iterative optimization on the super parameters of the reference prediction model, namely the parameters of the evaluation model to obtain the trained evaluation model;
establishing an online prediction model for the current equipment, determining the effectiveness of the current equipment after the field historical data of the current equipment is processed by an evaluation model, taking the current equipment as training data of the online prediction model, and continuously training and updating the online prediction model;
step eight, comparison and correction: and calculating to obtain a predicted value of the current measurement according to the online prediction model, inputting the current measurement value into the evaluation model, correcting the measurement value into the predicted value if the output of the evaluation model is invalid, comparing the deviation between the predicted value and the measurement value if the output of the evaluation model is valid, and correcting the measurement value into the predicted value if the output of the evaluation model is larger than a set threshold value.
Further, in the step one and the step four, a part of data in historical data sets of different sites is respectively selected as a training data set of the prediction model and a training set of the reference prediction model.
Further, in the first step, the prediction model is established as follows:
for each wheel of each train, respectively establishing a prediction model according to each size parameter:
p=Fcn,wn(x)
wherein cn is the vehicle number; wn is the number of the round; x is the driving mileage of the vehicle wheel set, and when the driving mileage cannot be obtained, the detected times or time of the vehicle wheel set passing through the detection equipment is taken by x; p is a wheel set size predicted value;
the prediction model is trained as follows:
in a training data set of the prediction model, selecting an effective geometric measurement value y detected by wheel pair detection equipment of a vehicle and a corresponding running mileage x of a vehicle wheel pair, and training the prediction model to obtain a trained model; and after x is input, calculating by using the trained model to obtain the wheel set size prediction p corresponding to each geometric parameter.
Further, a training data set of the prediction model is formed by collecting field wheel set size detection data in different time periods and different vehicle types, and the method is labeled by adopting the following steps:
the geometric measured value y detected by wheel pair detection equipment of the vehicle and the manually measured geometric parameter value ymOr comparing the empirical values and marking, wherein each measurement data needs to record the affiliated vehicle number and wheel position or wheel number, and when the geometric measurement value y detected by the wheel pair detection equipment of the vehicle and the manually measured geometric parameter value y are usedmWhen the deviation between the geometric measurement values is smaller than a set threshold value, the geometric measurement value y is marked as an effective value, otherwise, the geometric measurement value y is marked as an invalid value; and each geometric measurement value correspondingly records the driving mileage of the wheel set of the vehicle during measurement, and when the driving mileage cannot be acquired, the accumulated detected times or time of the wheel set passing through the detection equipment can be recorded.
Further, in the second step, based on comparison and analysis between the historical effective data and the actual measurement value, an evaluation model is suggested, and the specific steps are as follows:
calculating the deviation between the current actual measurement value and the weighted arithmetic mean value of the historical effective data according to the following formula, and carrying out threshold judgment on the deviation between the current actual measurement value and the historical effective data to determine the effectiveness of the current actual measurement value:
wherein ,ynIs the current actual measured value; q (y)n) For the evaluation results, 1 represents valid, and 0 represents invalid; y isiIs a valid measurement value before the current actual measurement value, i<n and Q (y)i)=1;wiFor weight, M is the number of historical valid data participating in the calculation, T is the evaluation threshold, wiM, T are used as the super parameter of the reference prediction model in whole or in part, and are determined by the super parameter optimization of the subsequent steps.
And further, in the seventh step, the online prediction model is respectively established according to each size parameter and each wheel of each train, effective data in part or all historical data of the measured size parameters are selected from training data of current equipment according to the number of the train and the number of the wheels and used as training data, and the online prediction model is trained online.
Further, in the first step, a part of data in the wheel set size data set is randomly selected as a training data set of the prediction model, and the rest of data or a part of data is used as a training data set of the evaluation model for training the evaluation model.
The invention discloses a wheel set size detection data analysis and correction device, which analyzes and corrects by adopting a wheel set size detection data analysis and correction method, and comprises a wheel set size detection acquisition measurement unit, a data storage unit, a data analysis and correction unit and a data display and service processing unit, wherein the wheel set size detection acquisition measurement unit, the data storage unit and the data analysis and correction unit are sequentially connected, the data storage unit is in bidirectional communication with the data analysis and correction unit, the data storage unit is in bidirectional communication with the data display and service processing unit, wheel set size, vehicle number and wheel number information are acquired and analyzed and processed by the wheel set size detection acquisition measurement unit to obtain three-dimensional information of wheels, measured values of various geometric parameters are obtained according to wheel set size indexes in railway standards, and then the measured values are uploaded to the data storage unit for storage and management, and the data analysis and correction unit extracts the data in the data storage unit and performs model training and calculation processing, the processing result is returned to the data storage unit, and the data display and service processing unit acquires and displays the analyzed and corrected measurement data in the data storage unit.
Compared with the prior art, the invention has the beneficial effects that:
1. the wheel set size change prediction model is established based on a big data analysis technology, and then an evaluation model for data validity evaluation is established, so that the validity of online data is automatically judged, the data is analyzed and screened, and the data is determined to be valid, invalid or required to be corrected; the method has the advantages that the prediction value of the trained prediction model is used as a true value, the evaluation model parameters are super-parameters, the training of the evaluation model parameters is realized by adopting a mode of super-parameter tuning and training a reference prediction model, the evaluation model is obtained to realize the evaluation of the measured value, the out-of-limit false detection rate can be effectively reduced, and the detection rate is improved; in the actual on-line application, the prediction model does not participate in calculation and is only used for matching with a training evaluation model;
2. the method processes the on-line detection data of the field equipment through an on-line prediction model, and judges the validity of the data by comparing a predicted value with an actual value; the training data input into the online prediction model is only data of single equipment, and through the screening of the evaluation model, the effectiveness of the training data is better, the online prediction model is easy to converge, and the prediction model which can meet practical requirements can be trained without large data accumulation, so that the problem that the prediction model is inaccurate by adopting a conventional method when no large amount of data exists in the early stage of equipment application is solved;
3. according to the method, full-sample labeled data of various conditions is not required to be covered, under the condition that sample data is not very large, the accuracy of data prediction is improved by means of the evaluation model and the online prediction model, and the measurement reliability under the conditions of different rail traffic lines and road conditions is improved, so that the overrun false detection rate is reduced, and the overrun detectable rate is improved;
4. the invention focuses on the analysis and correction of data validity, not on the prediction of data; the effectiveness analysis is primarily judged through an evaluation model, and then is further judged through an online prediction model; replacing the prediction result of the online prediction model when correction is needed; the reliability of wheel pair size measurement is promoted, especially the online dynamic measurement of the high-speed train main line, and the practical requirement of high-precision measurement is met under the complex severe environment.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flow chart of the evaluation model establishment of the present invention;
fig. 3 is a schematic block diagram of the present invention.
Detailed Description
The present invention is described in detail below with reference to the attached drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby clearly defining the protection scope of the present invention.
The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.
As shown in fig. 1-3, a wheel set size detection data analysis and correction method combines an evaluation model and an online prediction model for grading evaluation, trains the evaluation model offline based on an accumulated data set, further performs validity judgment on online data measured on site in real time, dynamically updates the online prediction model, and realizes analysis and correction of wheel set size detection data by comparing a prediction result and a measurement value of the online prediction model; the method for improving the reliability of detection by adopting a mode of gradually judging from coarse to fine and dynamically optimizing a prediction model, reducing the false alarm rate and improving the detection rate mainly comprises the following steps:
step one, establishing a wheel set dimension measurement value prediction model, and training the prediction model after manually marking historical data of different fields;
establishing an evaluation model for evaluating the effectiveness of the wheel set size detection data and judging whether the error is overlarge or not and whether the error is effective data or not;
step three, establishing a reference prediction model, wherein the reference prediction model is consistent with the prediction model in the step one in principle, and the evaluation model parameters are used as the super-parameters for reference prediction model training;
calculating the effectiveness of each item of data by an evaluation model for historical data sets of different sites, and updating effective attributes in the manual labeling information before updating to obtain an updated training data set;
step five, training the reference prediction model by adopting the training data set updated in the step four, and taking a predicted value obtained by calculation of the prediction model in the training as a data true value;
step six, repeating the step four to the step five, and performing iterative optimization on the super parameters of the reference prediction model, namely the parameters of the evaluation model to obtain the trained evaluation model; the method specifically comprises the following steps:
evaluating the parameter weight w of the modeliTaking the number M of the historical effective data participating in calculation and an evaluation threshold value T as super parameters of a reference prediction model, and repeatedly performing the fourth step to the fifth step to perform iterative optimization;
the super-parameter optimization algorithm can select algorithms such as random search, grid search, Bayesian search and the like, and can also be carried out in a traditional manual parameter adjustment mode;
establishing an online prediction model for the current equipment, determining the effectiveness of the current equipment after the field historical data of the current equipment is processed by an evaluation model, taking the current equipment as training data of the online prediction model, and continuously training and updating the online prediction model; inputting a geometric measurement value y detected by wheel set detection equipment of a vehicle into an evaluation model to obtain an evaluation result, specifically, taking the value as a record, and storing the validity evaluation result, the vehicle number, the wheel number and the wheel pair driving mileage in a data set when the value corresponds to the measurement, and when the driving mileage cannot be obtained, recording the accumulated detected times or time of the wheel set passing through the detection equipment, or not distinguishing the vehicle and the wheel position, and only grouping and storing all wheel data according to the geometric measurement value; when new data exists after the equipment detects the vehicle, aiming at each type of measurement data, acquiring effective data in part or all historical data from training data of the current equipment according to the vehicle number and the wheel number to serve as training data, and training a prediction model on line; part of historical effective data can be selected for model training to improve the training speed, and the selection mode can be random selection or selection of the nearest data;
step eight, comparison and correction: and calculating to obtain a predicted value of the current measurement according to the online prediction model, inputting the current measurement value into the evaluation model, correcting the measurement value into the predicted value if the output of the evaluation model is invalid, comparing the deviation between the predicted value and the measurement value if the output of the evaluation model is valid, and correcting the measurement value into the predicted value if the output of the evaluation model is larger than a set threshold value.
In the first step and the fourth step, part of data in historical data sets of different sites are respectively selected as a training data set of a prediction model and a training set of a reference prediction model.
In the first step, a prediction model is established according to the following modes:
for each wheel of each train, respectively establishing a prediction model according to each size parameter:
p=Fcn,wn(x)
wherein cn is the vehicle number; wn is the number of the round; x is the driving mileage of the vehicle wheel set, and when the driving mileage cannot be obtained, the detected times or time of the vehicle wheel set passing through the detection equipment is taken by x; p is a wheel set size predicted value;
the prediction model is trained as follows:
in a training data set of the prediction model, selecting an effective geometric measurement value y detected by wheel pair detection equipment of a vehicle and a corresponding running mileage x of a vehicle wheel pair, and training the prediction model to obtain a trained model; and after x is input, calculating by using the trained model to obtain the wheel set size prediction p corresponding to each geometric parameter.
The prediction model can be a time series prediction model, such as linear models like ARMA and ARIMA, and deep learning networks like LSTM, prophet or LSTNet; common prediction models such as regression models can also be selected;
the prediction models may be respectively established for each size parameter without distinguishing the vehicle, the wheel number, and the like, but the prediction accuracy is slightly poor.
In the first step, a training data set of a prediction model is formed by collecting field wheel set size detection data in different time periods and different vehicle types, and the method is labeled by adopting the following steps:
the collected and accumulated data fields are different as much as possible, the quantity of the collected and accumulated data is as much as possible, and the detected data has different vehicle types and acquisition times as much as possible, wherein the vehicle types and the acquisition times comprise different time periods in the day and at night, different seasons in the year and different weather conditions;
the geometric measured value y detected by wheel pair detection equipment of the vehicle and the manually measured geometric parameter value ymOr comparing the empirical values and marking, wherein each measurement data needs to record the affiliated vehicle number and wheel position or wheel number, and when the geometric measurement value y detected by the wheel pair detection equipment of the vehicle and the manually measured geometric parameter value y are usedmWhen the deviation between the measured value of the device and the measured value of the manual is less than the set threshold value, or when the deviation between the measured value of the device and the measured value of the manual is less than the set threshold valueMarking the value y as an effective value, otherwise, marking the value y as an invalid value; when no manual measured value is compared, whether the data is valid or not can be judged manually according to the experience of a detection operator and by combining the previous historical data. Preferably, each geometric measurement value y correspondingly records the driving mileage of the wheel pair of the vehicle during measurement, and can be obtained from a railway information system, such as a railway freight train technical management system HMIS; when the mileage cannot be obtained, the accumulated detected times or time of the wheel set passing through the detection equipment can be recorded; or the vehicle and the wheel position are not distinguished, and all wheel data are only grouped according to the geometric measurement value;
when the number of the data sets is large, one part of the wheel set size data set is randomly selected to serve as a training data set of the prediction model, and the rest of the wheel set size data set is used for training the evaluation model.
In the second step, based on comparison and analysis between the historical effective data and the actual measurement value, an evaluation model is suggested, and the specific steps are as follows:
calculating the deviation between the current actual measurement value and the weighted arithmetic mean value of the historical effective data according to the following formula, and carrying out threshold judgment on the deviation between the current actual measurement value and the historical effective data to determine the effectiveness of the current actual measurement value:
wherein ,ynIs the current actual measured value; q (y)n) For the evaluation results, 1 represents valid, and 0 represents invalid; y isiIs a valid measurement value before the current actual measurement value, i<n and Q (y)i)=1;wiFor weight, M is the number of historical valid data participating in the calculation, T is the evaluation threshold, wiM, T are used as the super parameter of the reference prediction model in whole or in part, and are determined by the super parameter optimization of the subsequent steps.
It should be noted that, the wheel set size change rules of different sites, different vehicle types, different use durations and use conditions of the railway have differences, and the conventional method for manually setting the fixed threshold value is used to directly judge whether the current measurement data is valid or not, which is often unreliable, and it is difficult for a person to find the appropriate set threshold value. Therefore, the invention provides a method for comparing and judging the historical effective data to realize the evaluation of the current measurement data.
In step three, the reference prediction model can be established by adopting the following steps:
pr=Fr(cn,wn,x)
the reference model and the prediction model are defined the same, and the parameters are different after training.
And step seven, in order to solve the problem that the prediction model prediction value of the measured data in the training data set and the actual field application has larger error due to different possible data deviations caused by different vehicle types, operation environments and the like, respectively establishing an online prediction model according to each size parameter for each wheel of each train on each field, selecting effective data in part or all historical data of the measured size parameters from the training data of the current equipment according to the vehicle number and the wheel number, and training the online prediction model on line.
In the first step, a part of data in the wheel set size data set is randomly selected as a training data set of the prediction model, and the rest of data or a part of data is used as a training data set of the evaluation model for training the evaluation model.
Step eight, inputting the driving mileage or the detected times or time of the current measured value into an online prediction model to obtain an online prediction result; inputting the current measured value into an evaluation model, correcting the measured value into a prediction result if the output of the evaluation model is invalid, further comparing the deviation between the prediction result and the measured value if the output of the evaluation model is valid, and correcting the measured value into a predicted value if the output of the evaluation model is larger than a set threshold value. Wherein the threshold value can be reduced appropriately based on the threshold parameter of the evaluation model.
The invention also discloses a wheel set size detection data analysis and correction device, which analyzes and corrects by adopting a wheel set size detection data analysis and correction method, and comprises a wheel set size detection acquisition measurement unit 1, a data storage unit 2, a data analysis and correction unit 3 and a data display and service processing unit 4, wherein the wheel set size detection acquisition measurement unit 1, the data storage unit 2 and the data analysis and correction unit 3 are sequentially connected, the data storage unit 2 is in two-way communication with the data analysis and correction unit 3, the data storage unit 2 is in two-way communication with the data display and service processing unit 4, the wheel set size detection acquisition measurement unit 1 acquires image data, wheel set size, vehicle number and wheel number information by adopting a structured light three-dimensional measurement method, analyzes and processes the information, and calculates to obtain the three-dimensional information of the wheel according to a structured light three-dimensional reconstruction method, and the measured value of each geometric parameter is obtained according to the wheel set size index in the railway standard, and then uploaded to the data storage unit 2 for storage and management, the data in the data storage unit 2 is extracted through the data analysis and correction unit 3, model training and calculation processing are carried out, the processing result is returned to the data storage unit 2, the analyzed and corrected measured data in the data storage unit 2 is obtained through the data display and service processing unit 4, and displayed, and the data display and service processing unit 4 provides data query, statistics and management functions to match and support corresponding service processing of a user, and also provides user service operation functions such as manual review data recording, user management, scheduling management and the like.
The data storage unit 2 stores the data in the form of pieces by the vehicle number, wheel number, measurement value, mileage or number of times of measurement or time of measurement using a database. According to the purpose and source of the data, the data can be separately stored and managed according to the accumulated training data and the current equipment data, wherein the accumulated training data is collected from each equipment in a plurality of fields, and the data are used for training a prediction model and an evaluation model according to the steps from one step to four; the current equipment data is used for recording the measurement result of the wheel pair size detection equipment and the correction result processed by the data analysis and correction unit, the result can be provided for a user to carry out operation and maintenance business operation, and meanwhile, the recorded current equipment historical data is used for creating and updating the online prediction model in the step six; the data storage unit 2 also provides data interfacing functions and interfaces with other information systems (e.g., HMIS systems, etc.) of the user, transmits wheel set size detection data to the other information systems, and obtains other information system data.
The training process from the first step to the fourth step can be processed on an off-site computer, the trained model is deployed on the on-site computer, and the processing process from the eighth step to the eleventh step needs to be deployed on the on-site computer for real-time processing. The data analysis and correction unit 3 returns the processed result to the data storage unit 2.
The parts of the invention not specifically described can be realized by adopting the prior art, and the details are not described herein.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by the present specification, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (8)
1. A method for analyzing and correcting wheel set size detection data, comprising the steps of:
step one, establishing a wheel set dimension measurement value prediction model, and training the prediction model after manually marking historical data of different fields;
establishing an evaluation model for evaluating the effectiveness of the wheel set size detection data and judging whether the error is overlarge or not and whether the error is effective data or not;
step three, establishing a reference prediction model, wherein the reference prediction model is consistent with the prediction model in the step one in principle, and the evaluation model parameters are used as the super-parameters for reference prediction model training;
calculating the effectiveness of each item of data by an evaluation model for historical data sets of different sites, and updating effective attributes in the manual labeling information before updating to obtain an updated training data set;
step five, training the reference prediction model by adopting the training data set updated in the step four, and taking a predicted value obtained by calculation of the prediction model in the training as a data true value;
step six, repeating the step four to the step five, and performing iterative optimization on the super parameters of the reference prediction model, namely the parameters of the evaluation model to obtain the trained evaluation model;
establishing an online prediction model for the current equipment, determining the effectiveness of the current equipment after the field historical data of the current equipment is processed by an evaluation model, taking the current equipment as training data of the online prediction model, and continuously training and updating the online prediction model;
step eight, comparison and correction:
and calculating to obtain a predicted value of the current measurement according to the online prediction model, inputting the current measurement value into the evaluation model, correcting the measurement value into the predicted value if the output of the evaluation model is invalid, comparing the deviation between the predicted value and the measurement value if the output of the evaluation model is valid, and correcting the measurement value into the predicted value if the output of the evaluation model is larger than a set threshold value.
2. The method for analyzing and correcting wheel set dimension detection data of claim 1, wherein in the first step and the fourth step, a part of data in historical data sets of different sites is selected as a training data set of the prediction model and a training set of the reference prediction model.
3. The method for analyzing and correcting wheel set size detection data of claim 1, wherein in the first step, the prediction model is built in the following way:
for each wheel of each train, respectively establishing a prediction model according to each size parameter:
p=Fcn,wn(x)
wherein cn is the vehicle number; wn is the number of the round; x is the driving mileage of the vehicle wheel set, and when the driving mileage cannot be obtained, the detected times or time of the vehicle wheel set passing through the detection equipment is taken by x; p is a wheel set size predicted value;
the prediction model is trained as follows:
in a training data set of the prediction model, selecting an effective geometric measurement value y detected by wheel pair detection equipment of a vehicle and a corresponding running mileage x of a vehicle wheel pair, and training the prediction model to obtain a trained model; and after x is input, calculating by using the trained model to obtain the wheel set size prediction p corresponding to each geometric parameter.
4. A method of analysing and correcting wheel set dimensional test data according to claim 3, wherein a training data set of the predictive model is formed by collecting field wheel set dimensional test data over different time periods and different vehicle types, and labelled by:
the geometric measured value y detected by wheel pair detection equipment of the vehicle and the manually measured geometric parameter value ymOr comparing the empirical values and marking, wherein each measurement data needs to record the affiliated vehicle number and wheel position or wheel number, and when the geometric measurement value y detected by the wheel pair detection equipment of the vehicle and the manually measured geometric parameter value y are usedmWhen the deviation between the geometric measurement values is smaller than a set threshold value, the geometric measurement value y is marked as an effective value, otherwise, the geometric measurement value y is marked as an invalid value; and each geometric measurement value correspondingly records the driving mileage of the wheel set of the vehicle during measurement, and when the driving mileage cannot be acquired, the accumulated detected times or time of the wheel set passing through the detection equipment can be recorded.
5. The method for analyzing and correcting wheel set size detection data according to claim 1, wherein in the second step, an evaluation model is established based on comparison and analysis between historical valid data and actual measurement values, and the specific steps are as follows:
calculating the deviation between the current actual measurement value and the weighted arithmetic mean value of the historical effective data according to the following formula, and carrying out threshold judgment on the deviation between the current actual measurement value and the historical effective data to determine the effectiveness of the current actual measurement value:
wherein ,ynIs the current actual measured value; q (y)n) For the evaluation results, 1 represents valid, and 0 represents invalid; y isiIs a valid measurement value before the current actual measurement value, i<n and Q (y)i)=1;wiFor weight, M is the number of historical valid data participating in the calculation, T is the evaluation threshold, wiM, T are used as the super parameter of the reference prediction model in whole or in part, and are determined by the super parameter optimization of the subsequent steps.
6. The method for analyzing and correcting wheel set size detection data according to claim 1, wherein in the seventh step, the online prediction model is respectively established according to each size parameter for each wheel of each train, and effective data in part or all historical data of the measured size parameters are selected from training data of current equipment according to the number of the vehicle and the number of the wheels to be used as training data, and the online prediction model is trained online.
7. The method for analyzing and correcting wheel set dimension detection data according to claim 1, wherein in the first step, a part of the wheel set dimension data set is randomly selected as a training data set of the predictive model, and all or a part of the remaining data is used as a training data set of the evaluation model for training the evaluation model.
8. An analysis and correction device for wheel set size detection data, characterized in that, the analysis and correction method for wheel set size detection data according to any one of claims 1 to 7 is adopted for analysis and correction, the analysis and correction device comprises a wheel set size detection acquisition and measurement unit, a data storage unit, a data analysis and correction unit and a data display and service processing unit, the wheel set size detection acquisition and measurement unit, the data storage unit and the data analysis and correction unit are sequentially connected, the data storage unit is in bidirectional communication with the data analysis and correction unit, the data storage unit is in bidirectional communication with the data display and service processing unit, the wheel set size, vehicle number and wheel number information are acquired and analyzed and processed through the wheel set size detection acquisition and measurement unit, three-dimensional information of wheels is obtained, and measurement values of various geometric parameters are obtained according to wheel set size indexes in railway standards, and then the data are uploaded to a data storage unit for storage and management, the data in the data storage unit are extracted through a data analysis and correction unit, model training and calculation processing are carried out, the processing result is returned to the data storage unit, and the measurement data analyzed and corrected in the data storage unit are obtained through a data display and service processing unit and displayed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110929385.XA CN113569491B (en) | 2021-08-13 | 2021-08-13 | Wheel set size detection data analysis and correction method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110929385.XA CN113569491B (en) | 2021-08-13 | 2021-08-13 | Wheel set size detection data analysis and correction method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113569491A true CN113569491A (en) | 2021-10-29 |
CN113569491B CN113569491B (en) | 2023-08-18 |
Family
ID=78171592
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110929385.XA Active CN113569491B (en) | 2021-08-13 | 2021-08-13 | Wheel set size detection data analysis and correction method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113569491B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114061454A (en) * | 2021-11-10 | 2022-02-18 | 南京国轩电池有限公司 | Evaluation method for dynamic stability of lithium ion battery coating CCD (Charge coupled device) measuring system |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010262544A (en) * | 2009-05-08 | 2010-11-18 | Fuji Corporation:Kk | Online sales method of tire wheel |
CN103336906A (en) * | 2013-07-15 | 2013-10-02 | 哈尔滨工业大学 | Sampling GPR method of continuous anomaly detection in collecting data flow of environment sensor |
CN107179064A (en) * | 2017-05-27 | 2017-09-19 | 广州地铁集团有限公司 | A kind of determination method of the confidence level of wheelset profile on-line detecting system measured value |
CN207035997U (en) * | 2016-09-13 | 2018-02-23 | 秦皇岛信越智能装备有限公司 | Online aluminum vehicle wheel dimension measuring device |
CN108536652A (en) * | 2018-03-15 | 2018-09-14 | 浙江大学 | A kind of short-term vehicle usage amount prediction technique based on arma modeling |
CN110442991A (en) * | 2019-08-12 | 2019-11-12 | 江南大学 | A kind of dynamic sulfur recovery soft-measuring modeling method based on parametrization FIR model |
CN110688710A (en) * | 2019-09-27 | 2020-01-14 | 南京理工大学 | Turning repairing method based on rail transit vehicle wheel pair service life statistical model |
CN112732960A (en) * | 2020-12-30 | 2021-04-30 | 北京理工大学 | Image classification method based on online federal learning |
CN113239025A (en) * | 2021-04-23 | 2021-08-10 | 四川大学 | Ship track classification method based on feature selection and hyper-parameter optimization |
-
2021
- 2021-08-13 CN CN202110929385.XA patent/CN113569491B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010262544A (en) * | 2009-05-08 | 2010-11-18 | Fuji Corporation:Kk | Online sales method of tire wheel |
CN103336906A (en) * | 2013-07-15 | 2013-10-02 | 哈尔滨工业大学 | Sampling GPR method of continuous anomaly detection in collecting data flow of environment sensor |
CN207035997U (en) * | 2016-09-13 | 2018-02-23 | 秦皇岛信越智能装备有限公司 | Online aluminum vehicle wheel dimension measuring device |
CN107179064A (en) * | 2017-05-27 | 2017-09-19 | 广州地铁集团有限公司 | A kind of determination method of the confidence level of wheelset profile on-line detecting system measured value |
CN108536652A (en) * | 2018-03-15 | 2018-09-14 | 浙江大学 | A kind of short-term vehicle usage amount prediction technique based on arma modeling |
CN110442991A (en) * | 2019-08-12 | 2019-11-12 | 江南大学 | A kind of dynamic sulfur recovery soft-measuring modeling method based on parametrization FIR model |
CN110688710A (en) * | 2019-09-27 | 2020-01-14 | 南京理工大学 | Turning repairing method based on rail transit vehicle wheel pair service life statistical model |
CN112732960A (en) * | 2020-12-30 | 2021-04-30 | 北京理工大学 | Image classification method based on online federal learning |
CN113239025A (en) * | 2021-04-23 | 2021-08-10 | 四川大学 | Ship track classification method based on feature selection and hyper-parameter optimization |
Non-Patent Citations (1)
Title |
---|
王琳琳;郑春波;: "铁路车辆轮对参数实时在线自动检测系统检测方案的实现", 哈尔滨铁道科技 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114061454A (en) * | 2021-11-10 | 2022-02-18 | 南京国轩电池有限公司 | Evaluation method for dynamic stability of lithium ion battery coating CCD (Charge coupled device) measuring system |
Also Published As
Publication number | Publication date |
---|---|
CN113569491B (en) | 2023-08-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020244288A1 (en) | Method and apparatus for evaluating truck driving behaviour based on gps trajectory data | |
CN110376003B (en) | Intelligent train service life prediction method and system based on BIM | |
CN105335692B (en) | A kind of tire x-ray image detection recognition method and system | |
US9981671B2 (en) | Railway inspection system | |
CN104164829A (en) | Pavement smoothness detection method based on mobile terminal and intelligent pavement information real-time monitoring system | |
CN111127906B (en) | Intelligent road surface management system and method based on Internet of things | |
CN108197731B (en) | Motor vehicle exhaust remote measurement and vehicle inspection result consistency method based on co-training | |
CN116227752B (en) | Park facility management system based on Internet of things | |
CN115389385B (en) | Dust intelligent monitoring and early warning system based on operation environment and human occupational health | |
CN112153574B (en) | Method and system for checking accuracy of roadside device clock based on floating vehicle | |
CN113569491A (en) | Analysis and correction method and device for wheel set size detection data | |
CN115439643A (en) | Road disease size analysis and management method based on monocular measurement | |
CN114971166A (en) | Lightweight road pavement service quality inspection system | |
CN109029881B (en) | Track bed state evaluation method based on track rigidity and ground penetrating radar detection | |
CN117119588B (en) | Wi-Fi6 technology-based method for realizing positioning of vehicle in track parking lot | |
CN113657041A (en) | Intelligent sensing and forecasting system for physical and mechanical states of roadbed in alpine region | |
CN111680888B (en) | Method for determining road network capacity based on RFID data | |
CN108920655B (en) | Method and device for quantifying space-time coverage range of road weather information system | |
CN110555188A (en) | analysis method for wheel parameter abrasion trend of motor train unit | |
CN113032907A (en) | Method and system for correcting vehicle shaking disease data deviation based on waveform correlation | |
CN110455370B (en) | Flood-prevention drought-resisting remote measuring display system | |
CN116228839A (en) | Train wheel set size detection big data correction device and method and storage medium thereof | |
CN114169370A (en) | Method and system for constructing road space-time load pedigree based on grating array | |
CN116336939A (en) | Automatic detection-based pavement rut index instant analysis method | |
CN112071073A (en) | Road noise automatic correction method of noise map system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |