CN102360454A - Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network - Google Patents
Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network Download PDFInfo
- Publication number
- CN102360454A CN102360454A CN2011103081628A CN201110308162A CN102360454A CN 102360454 A CN102360454 A CN 102360454A CN 2011103081628 A CN2011103081628 A CN 2011103081628A CN 201110308162 A CN201110308162 A CN 201110308162A CN 102360454 A CN102360454 A CN 102360454A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- wheel track
- track force
- wheel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention discloses a wheel-track force prediction method based on an NARX (Nonlinear Auto-regressive with Extra Inputs) neural network in the technical field of railway safe operation control, comprising the following steps of: collecting track irregularity data by using a track detection vehicle; emulating the track irregularity data to obtain wheel-track force data; carrying out normalization on the track irregularity data and the wheel-track force data; setting a prediction model of the NARX neural network; selecting a training sample to train the prediction model of the NARX neural network; selecting a test sample to test the trained prediction model of the NARX neural network, and outputting the tested wheel-track force data; and analyzing the wheel-track force data in the test sample and the tested wheel-track force data to evaluate properties of the prediction model of the NARX neural network. The wheel-track force prediction method provided by the invention is used for predicting a wheel-track force by using the actual-measured track irregularity data through adopting the prediction model of the NARX neural network, thereby improving the accuracy of safety evaluation of railway train operation.
Description
Technical field
The invention belongs to safe railway operation control technology field, relate in particular to a kind of wheel track force prediction method based on the NARX neural network.
Background technology
Along with transportation by railroad develops towards high speed, heavy duty, large conveying quantity and highdensity direction; The safety and steady operation that guarantees train becomes a current vital task, so China adopts track inspection vehicle or comprehensive detection train etc. that the railway infrastructure service state is carried out the periodicity detection.In many test items of infrastructure; Wheel track power is the key factor that causes track failure/destruction, derail, vehicle part damage; Also be to be used for evaluation driving stationarity and security; And as the main foundation of speed limit, speed-raising, so obtaining of wheel track power has important theory and practical significance.
Physics dynamometry wheel is a kind of special sensor that is used to measure wheel track power to being the important tool that detects track/Vehicular system running status.Physics dynamometry wheel is to utilizing the elastic body of wheel as transmitting element, and when the time spent of doing that receives wheel track power, wheel produces distortion, through detecting and resolving the corresponding relation between this distortion and the wheel track power, can confirm the value of wheel track interphase interaction power.But physics dynamometry wheel is defectives such as cost is excessive, failure rate height to existing, thereby have limited right the promoting the use of of physics dynamometry wheel.
U.S. John Zolock combines neural network modeling approach with and non-linear time series analysis method theoretical based on the dynamic system phase space reconfiguration of time delay; A kind of time-delay neural network (TDNN) is proposed; The track of input actual measurement is uneven suitable, predicts vertical wheel track power.U.S. Transportation Technology Center; Inc. (TTCI) dynamics of vehicle forecasting techniques of proposing based on neural network; This technology adopts the BP neural network; With parameters such as track irregularity and speed is input, has realized the prediction of kinetic parameters such as wheel track power, and further is applied to the judge of track geometry.Because track irregularity/wheel track power system is the complex nonlinear dynamic system, and the BP neural network is a kind of static neural network, so the BP neural network is difficult to the complicated dynamic relationship between accurate description track irregularity and the wheel track power.Britain Gualano L. has developed a kind of new recurrent neural network structure based on the Jordan neural network, can realize the wheel track power prediction based on the actual measurement track irregularity.
Above-mentioned technology has all realized utilizing the prediction of neural network to wheel track power, and still, precision of prediction is desirable not to the utmost, especially all has weak point aspect the prediction of horizontal wheel track power.
Summary of the invention
The objective of the invention is; In order to improve precision of prediction to wheel track power; Remedy the deficiency of the wheel track force measuring method existence aspect horizontal wheel track power prediction that proposes in the background technology, proposed a kind of wheel track force prediction method based on NARX (non-linear regression) neural network.
To achieve these goals, the technical scheme of the present invention's employing is that a kind of wheel track force prediction method based on the NARX neural network is characterized in that said method comprises:
Step 1: utilize track inspection vehicle acquisition trajectory irregularity data;
Step 2: the track irregularity data are carried out emulation, obtain the wheel track force data;
Step 3: track irregularity data and wheel track force data are carried out the normalization processing;
Step 4: set the NARX neural network prediction model;
Step 5: choose training sample in track irregularity data after normalization is handled and the wheel track force data, training NARX neural network prediction model;
Step 6: choose test sample book in track irregularity data after normalization is handled and the wheel track force data, the NARX neural network prediction model that trains is tested, the wheel track force data after the output test;
Step 7: the wheel track force data to after wheel track force data in the test sample book and the test is analyzed, and estimates the performance of NARX neural network prediction model.
Said track irregularity data comprise that left rail is uneven and are uneven along data and right rail rail to the irregularity data to irregularity data, right rail along data, left rail rail.
Said wheel track force data comprises horizontal wheel track force data and vertical wheel track force data.
Saidly track irregularity data and wheel track force data are carried out normalization handle and specifically utilize formula
Wherein, x
iBe track irregularity data/wheel track force data, x
MinBe the minimum value in track irregularity data/wheel track force data, x
MaxBe the maximal value in track irregularity data/wheel track force data, i is for more than or equal to 1 and the integer of the track irregularity data sum that collects smaller or equal to the track inspection vehicle.
Said setting NARX neural network prediction model specifically is; The middle layer node of setting the NARX neural network adopts the tan-sigmoid function; The output layer node adopts linear function, and the input layer number is 4, and the middle layer node number is 15; The output layer interstitial content is 1, and input delay and output delay all are 45.
Said training NARX neural network prediction model specifically adopts Bayesian regularization algorithm training NARX neural network.
The concrete performance that adopts root-mean-square error method, curve fitting method or linear regression analysis method to estimate the NARX neural network prediction model of said step 7;
Wherein, the root-mean-square error method is specifically utilized formula
Y (j) is the wheel track force data in the test sample book, y
m(j) be the wheel track force data after the test, N is the test sample book number;
Said curve fitting method is that the wheel track force data after wheel track force data in the test sample book and the test is carried out curve fitting;
Said linear regression analysis method is that the wheel track force data after wheel track force data in the test sample book and the test is carried out linear regression analysis.
The present invention utilizes the track irregularity data of surveying, and adopts the NARX neural network model, and prediction wheel track power has improved the accuracy that railway operation safety is estimated.
Description of drawings
Fig. 1 is based on the wheel track force prediction method process flow diagram of NARX neural network;
Fig. 2 is NARX neural network structure figure;
Fig. 3 is the wheel track force data curve fitting comparison diagram after the test of horizontal wheel track force data and the NARX neural network output in the test sample book;
Fig. 4 is the wheel track force data correlation analysis figure after the test of horizontal wheel track force data and the NARX neural network output in the test sample book.
Embodiment
Below in conjunction with accompanying drawing, preferred embodiment is elaborated.Should be emphasized that following explanation only is exemplary, rather than in order to limit scope of the present invention and application thereof.
Fig. 1 is based on the wheel track force prediction method process flow diagram of NARX neural network, among Fig. 1, comprises the following steps: based on the wheel track force prediction method of NARX neural network
Step 101: utilize track inspection vehicle acquisition trajectory irregularity data.The track irregularity data comprise that left rail is uneven and are uneven along data and right rail rail to the irregularity data to irregularity data, right rail along data, left rail rail, and with its input data as neural network model.
Step 102: the track irregularity data are carried out emulation, obtain the wheel track force data.
Through dynamics simulation software ADAMS/RAIL; Set up vehicle/dynamics of orbits model, import 3000 track irregularity data, carry out emulation; Obtain the horizontal wheel track power and the vertical wheel track force data of corresponding 3000 the near front wheels, as the target data of neural network model.
Step 103: track irregularity data and wheel track force data are carried out the normalization processing.
Normalization is handled and is specifically utilized formula
Wherein, x
iBe track irregularity data, horizontal wheel track force data or vertical wheel track force data, x
MinBe the minimum value in track irregularity data, horizontal wheel track force data or the vertical wheel track force data, x
MaxBe track irregularity data, the horizontal maximal value in wheel track force data or the vertical wheel track force data, i is for more than or equal to 1 and the integer of the track irregularity data sum that collects smaller or equal to the track inspection vehicle.
Step 104: set the NARX neural network prediction model.
The middle layer node of NARX neural network adopts the tan-sigmoid function, and the output layer node adopts linear function, and the input layer number is 4, and the middle layer node number is 15, and the output layer interstitial content is 1, and input delay and output delay all are 45.
Step 105: choose training sample in the track irregularity data after normalization is handled, horizontal wheel track force data and the vertical wheel track force data, training NARX neural network prediction model.Fig. 2 is NARX neural network structure figure, and wherein, x (i) is the track irregularity data of normalization after handling, and y (i) is horizontal wheel track force data and vertical wheel track force data.Y (k) is the horizontal wheel track force data and the vertical wheel track force data of output.
Extract track irregularity data and corresponding with it horizontal wheel track force data and vertical wheel track force data after 2500 groups of normalization are handled as training sample.In the NARX neural network, needing the training optimum parameters is the weights and the threshold value of node, adopts Bayesian regularization algorithm training neural network.Bayesian regularization (BR, Bayesian Regularization algorithm) is in order to improve network promotion ability.To set up a property parameter that constitutes by each layer output error, weights and threshold value in the training process, the weights and the threshold value of network adjusted, this parameter is minimized according to L-M (Levenberg-Marguart) optimum theory.
Step 106: extract track irregularity data and corresponding with it horizontal wheel track force data and vertical wheel track force data after 500 groups of normalization are handled as test sample book; NARX neural network prediction model to training is tested, horizontal wheel track force data and vertical wheel track force data after the output test.
Step 107: horizontal wheel track force data and vertical wheel track force data to after the horizontal wheel track force data in the test sample book and vertical wheel track force data and the test are analyzed, and estimate the performance of NARX neural network prediction model.
The performance of estimating the NARX neural network prediction model can adopt root-mean-square error method, curve fitting method or linear regression analysis method.
The root-mean-square error method is specifically utilized formula
Y (j) is the horizontal wheel track force data/vertical wheel track force data in the test sample book, y
m(j) be the horizontal wheel track force data/vertical wheel track force data after the test of output, N is the test sample book number; RMS (y, y
m) performance of more little then NARX neural network prediction model is good more.
Curve fitting method is that the horizontal wheel track force data/vertical wheel track force data after the horizontal wheel track force data in the test sample book/vertical wheel track force data and the test is carried out curve fitting; The result of curve fitting embodies the performance of NARX neural network prediction model.Fig. 3 is the wheel track force data curve fitting comparison diagram after the test of horizontal wheel track force data and the NARX neural network output in the test sample book, can find out among Fig. 3 that more approaching through the curve data of over-fitting, the performance of NARX neural network prediction model is relatively good.
The linear regression analysis method is that the horizontal wheel track force data/vertical wheel track force data after the horizontal wheel track force data in the test sample book/vertical wheel track force data and the test is carried out linear regression analysis, confirms the performance of NARX neural network prediction model through correlativity.Fig. 4 is the horizontal wheel track force data correlation analysis figure after the test of horizontal wheel track force data and the NARX neural network output in the test sample book; Can find out among Fig. 4; Horizontal wheel track force data correlativity after the test of horizontal wheel track force data in the test sample book and the output of NARX neural network is more intense, and the performance of NARX neural network prediction model is relatively good.
Wheel track force prediction method based on the NARX neural network provided by the invention; The track irregularity data that utilization is surveyed adopt the NARX neural network model, accurately predicting wheel track power; Improved the accuracy that railway operation safety is estimated, the track traffic security control has been had important practical significance.
The above; Be merely the preferable embodiment of the present invention, but protection scope of the present invention is not limited thereto, any technician who is familiar with the present technique field is in the technical scope that the present invention discloses; The variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.
Claims (7)
1. wheel track force prediction method based on the NARX neural network is characterized in that said method comprises:
Step 1: utilize track inspection vehicle acquisition trajectory irregularity data;
Step 2: the track irregularity data are carried out emulation, obtain the wheel track force data;
Step 3: track irregularity data and wheel track force data are carried out the normalization processing;
Step 4: set the NARX neural network prediction model;
Step 5: choose training sample in track irregularity data after normalization is handled and the wheel track force data, training NARX neural network prediction model;
Step 6: choose test sample book in track irregularity data after normalization is handled and the wheel track force data, the NARX neural network prediction model that trains is tested, the wheel track force data after the output test;
Step 7: the wheel track force data to after wheel track force data in the test sample book and the test is analyzed, and estimates the performance of NARX neural network prediction model.
2. a kind of wheel track force prediction method based on the NARX neural network according to claim 1 is characterized in that said track irregularity data comprise that left rail is uneven and are uneven along data and right rail rail to the irregularity data to irregularity data, right rail along data, left rail rail.
3. a kind of wheel track force prediction method based on the NARX neural network according to claim 1 is characterized in that said wheel track force data comprises horizontal wheel track force data and vertical wheel track force data.
4. a kind of wheel track force prediction method based on the NARX neural network according to claim 1 is characterized in that saidly track irregularity data and wheel track force data are carried out normalization handling and specifically utilizing formula
Wherein, x
iBe track irregularity data/wheel track force data, x
MinBe the minimum value in track irregularity data/wheel track force data, x
MaxBe the maximal value in track irregularity data/wheel track force data, i is for more than or equal to 1 and the integer of the track irregularity data sum that collects smaller or equal to the track inspection vehicle.
5. a kind of wheel track force prediction method according to claim 1 based on the NARX neural network; It is characterized in that said setting NARX neural network prediction model specifically is, the middle layer node of setting the NARX neural network adopts the tan-sigmoid function, and the output layer node adopts linear function; The input layer number is 4; The middle layer node number is 15, and the output layer interstitial content is 1, and input delay and output delay all are 45.
6. a kind of wheel track force prediction method based on the NARX neural network according to claim 1 is characterized in that said training NARX neural network prediction model specifically adopts Bayesian regularization algorithm training NARX neural network.
7. a kind of wheel track force prediction method based on the NARX neural network according to claim 1 is characterized in that the concrete performance that adopts root-mean-square error method, curve fitting method or linear regression analysis method to estimate the NARX neural network prediction model of said step 7;
Wherein, the root-mean-square error method is specifically utilized formula
Y (j) is the wheel track force data in the test sample book, y
m(j) be the wheel track force data after the test, N is the test sample book number;
Said curve fitting method is that the wheel track force data after wheel track force data in the test sample book and the test is carried out curve fitting;
Said linear regression analysis method is that the wheel track force data after wheel track force data in the test sample book and the test is carried out linear regression analysis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011103081628A CN102360454A (en) | 2011-10-12 | 2011-10-12 | Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011103081628A CN102360454A (en) | 2011-10-12 | 2011-10-12 | Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102360454A true CN102360454A (en) | 2012-02-22 |
Family
ID=45585779
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011103081628A Pending CN102360454A (en) | 2011-10-12 | 2011-10-12 | Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102360454A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622519A (en) * | 2012-03-09 | 2012-08-01 | 北京交通大学 | Method for estimating safety domain of track irregularity amplitude |
| CN104794312A (en) * | 2015-05-15 | 2015-07-22 | 西南交通大学 | Method and device for evaluating track regularity |
| CN106326587A (en) * | 2016-08-30 | 2017-01-11 | 中车株洲电力机车有限公司 | Correction method of wheel-rail force with instrumented wheelset and the system thereof |
| CN108664673A (en) * | 2017-03-30 | 2018-10-16 | 中国铁道科学研究院 | A kind of wheel rail force load identification feature-based data model method for building up and device |
| CN109034192A (en) * | 2018-06-20 | 2018-12-18 | 西南交通大学 | A kind of track based on deep learning-body oscillating trend prediction method |
| CN109145764A (en) * | 2018-07-27 | 2019-01-04 | 中国铁道科学研究院集团有限公司 | The unjustified section recognition methods of the multiple groups detection waveform of synthetic detection vehicle and device |
| CN110826184A (en) * | 2019-10-09 | 2020-02-21 | 北京化工大学 | Variational Bayesian identification method for NARX model structure and parameters under time-varying lag |
| CN113919236A (en) * | 2021-10-29 | 2022-01-11 | 中国国家铁路集团有限公司 | Model generation method for predicting track irregularity and prediction method for track irregularity |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102072789A (en) * | 2010-11-03 | 2011-05-25 | 西南交通大学 | Method for continuously processing wheel-track force of track vehicle by means of ground test |
-
2011
- 2011-10-12 CN CN2011103081628A patent/CN102360454A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102072789A (en) * | 2010-11-03 | 2011-05-25 | 西南交通大学 | Method for continuously processing wheel-track force of track vehicle by means of ground test |
Non-Patent Citations (2)
| Title |
|---|
| XUE-MIAO PANG ET AL.: "The Prediction of Derailment Coefficient Using NARX Neural Networks", 《ICTIS 2011》 * |
| 陈建政等: "在线连续测量轮轨接触点的神经网络方法", 《振动与冲击》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622519A (en) * | 2012-03-09 | 2012-08-01 | 北京交通大学 | Method for estimating safety domain of track irregularity amplitude |
| CN102622519B (en) * | 2012-03-09 | 2015-01-07 | 北京交通大学 | Method for estimating safety domain of track irregularity amplitude |
| CN104794312A (en) * | 2015-05-15 | 2015-07-22 | 西南交通大学 | Method and device for evaluating track regularity |
| CN104794312B (en) * | 2015-05-15 | 2017-11-03 | 西南交通大学 | The assessment method and device of rail smooth state |
| CN106326587A (en) * | 2016-08-30 | 2017-01-11 | 中车株洲电力机车有限公司 | Correction method of wheel-rail force with instrumented wheelset and the system thereof |
| CN108664673A (en) * | 2017-03-30 | 2018-10-16 | 中国铁道科学研究院 | A kind of wheel rail force load identification feature-based data model method for building up and device |
| CN109034192A (en) * | 2018-06-20 | 2018-12-18 | 西南交通大学 | A kind of track based on deep learning-body oscillating trend prediction method |
| CN109034192B (en) * | 2018-06-20 | 2022-03-08 | 西南交通大学 | Track-vehicle body vibration state prediction method based on deep learning |
| CN109145764A (en) * | 2018-07-27 | 2019-01-04 | 中国铁道科学研究院集团有限公司 | The unjustified section recognition methods of the multiple groups detection waveform of synthetic detection vehicle and device |
| CN109145764B (en) * | 2018-07-27 | 2020-10-27 | 中国铁道科学研究院集团有限公司 | Method and device for identifying unaligned sections of multiple groups of detection waveforms of comprehensive detection vehicle |
| CN110826184A (en) * | 2019-10-09 | 2020-02-21 | 北京化工大学 | Variational Bayesian identification method for NARX model structure and parameters under time-varying lag |
| CN110826184B (en) * | 2019-10-09 | 2021-08-17 | 北京化工大学 | Variational Bayesian identification method for NARX model structure and parameters under time-varying lag |
| CN113919236A (en) * | 2021-10-29 | 2022-01-11 | 中国国家铁路集团有限公司 | Model generation method for predicting track irregularity and prediction method for track irregularity |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102360454A (en) | Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network | |
| CN103852269B (en) | Bullet train runs kinetic parameter detection method | |
| Burgelman et al. | A new rolling contact method applied to conformal contact and the train–turnout interaction | |
| CN105303832B (en) | Overpass road section traffic volume congestion index computational methods based on microwave vehicle detector | |
| US20120203402A1 (en) | Intelligent Railway System for Preventing Accidents at Railway Passing Points and Damage to the Rail Track | |
| Spiryagin et al. | Friction measurement and creep force modelling methodology for locomotive track damage studies | |
| CN104598753B (en) | Bridge moving vehicle load recognition method based on Brakhage V method | |
| Wang et al. | Longitudinal train dynamics model for a rail transit simulation system | |
| Ye et al. | A wheel wear prediction model of non-Hertzian wheel-rail contact considering wheelset yaw: comparison between simulated and field test results | |
| CN102567576B (en) | Method for predicting rate of wheel load reduction | |
| CN105923014B (en) | A kind of track transition Amplitude Estimation method based on evidential reasoning rule | |
| CN102072789B (en) | Method for continuously processing wheel-track force of track vehicle by means of ground test | |
| Shrestha et al. | Review of adhesion estimation approaches for rail vehicles | |
| CN104063543A (en) | Wheel-rail combination roughness identification method for rail traffic | |
| CN103207097A (en) | Dynamic detection method of track stiffness | |
| CN104881573B (en) | A kind of methods of risk assessment and system of city track traffic engineering | |
| Banerji et al. | Structural health monitoring of a steel railway bridge for increased axle loads | |
| CN102567786A (en) | Method for predicting derailment coefficients | |
| CN103344395A (en) | Determination method and device of bridge reinforcing target bearing capacity | |
| Zhang et al. | Optimal system design for weigh-in-motion measurements using in-pavement strain sensors | |
| CN105128888B (en) | Track waveform abrasion method for quick based on many midpoint string synchro measures | |
| Silva et al. | Traffic monitoring using a structural health monitoring system | |
| CN105303833B (en) | Overpass accident method of discrimination based on microwave vehicle detector | |
| Costello et al. | Stochastic rail wear model for railroad tracks | |
| Zhang et al. | Tunnel damage identification method based on relative entropy of wavelet packet energy: An experimental verification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20120222 |