CN114819083A - Train positioning method based on cross-line composite transfer learning - Google Patents
Train positioning method based on cross-line composite transfer learning Download PDFInfo
- Publication number
- CN114819083A CN114819083A CN202210445242.6A CN202210445242A CN114819083A CN 114819083 A CN114819083 A CN 114819083A CN 202210445242 A CN202210445242 A CN 202210445242A CN 114819083 A CN114819083 A CN 114819083A
- Authority
- CN
- China
- Prior art keywords
- train
- learning model
- deep learning
- line
- target domain
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 239000002131 composite material Substances 0.000 title claims abstract description 10
- 238000013526 transfer learning Methods 0.000 title claims abstract description 10
- 238000013136 deep learning model Methods 0.000 claims abstract description 77
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000012795 verification Methods 0.000 claims description 22
- 238000005516 engineering process Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 14
- 238000010220 Pearson correlation analysis Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 230000008014 freezing Effects 0.000 claims description 5
- 238000007710 freezing Methods 0.000 claims description 5
- 238000012216 screening Methods 0.000 claims description 5
- 238000013508 migration Methods 0.000 abstract description 12
- 230000005012 migration Effects 0.000 abstract description 12
- 102100028952 Drebrin Human genes 0.000 description 4
- 101000838600 Homo sapiens Drebrin Proteins 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0409—Adaptive resonance theory [ART] networks
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61L—GUIDING RAILWAY TRAFFIC; ENSURING THE SAFETY OF RAILWAY TRAFFIC
- B61L25/00—Recording or indicating positions or identities of vehicles or trains or setting of track apparatus
- B61L25/02—Indicating or recording positions or identities of vehicles or trains
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Economics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Mechanical Engineering (AREA)
- Primary Health Care (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The invention provides a train positioning method based on cross-line composite transfer learning, which is characterized by comprising the following steps of: the train positioning method comprises the following steps: the method comprises the steps of training an existing deep learning model by using a mature route A as a source domain, obtaining an available target domain deep learning model of a target domain of a route B after model parameter migration and statistical characteristic variation migration, and positioning a train operated on the route B through the available target domain deep learning model. By adopting the train positioning method, the deep learning model for train positioning can be obtained for training and verifying a new train running line quickly, and the output precision of the model to the train positioning position can be improved by the method, so that the positioning accuracy and the stability of the train are improved.
Description
Technical Field
The invention relates to the technical field of transportation, in particular to a train positioning method based on cross-line composite transfer learning.
Background
The train positioning technology is the basic and key technology of an automatic train operation system (ATO), and the traditional train positioning technology can be summarized into four major aspects: the train positioning technology is based on a speed sensor, satellite navigation, map matching and multi-source information fusion. With the development of artificial intelligence technology, the automatic train positioning technology based on the machine learning mode comes to work, a large amount of running data such as train positions, automatic driving speed curves, gradients, turnouts and the like can be generated and accumulated in the running process of a certain mature line of a train, the running data can be obtained from a train running monitoring device LKJ and a train control and management system TCMS device, a deep learning model for positioning the train of the line can be trained through the big data, and the train running on the line can be positioned and predicted by utilizing the deep learning model.
For a certain mature train running line, because a large amount of running data are accumulated, a deep learning model for train positioning is easy to train, and the train positioning accuracy obtained by the method is high, and the method has effectiveness and reliability. However, for a line which is not long in opening time and insufficient in accumulated driving data volume, due to insufficient data and small labeled data volume, the deep learning model obtained by adopting the small sample training has the problems of poor overfitting and generalization capability, and the deep learning model is used for train positioning, so that the positioning accuracy is low, and even the deep learning model cannot be used at all.
Disclosure of Invention
Aiming at the problems of the background art, the invention provides a train positioning method based on cross-line composite transfer learning, which aims to solve the problem that a deep learning model cannot be adopted to position a train on a line due to insufficient line driving data accumulation in the prior art.
In order to achieve the purpose of the invention, the invention provides a train positioning method based on cross-line composite transfer learning, which is characterized by comprising the following steps: the train positioning method comprises the following steps:
two train operation lines are arranged: line A and line B; the line A accumulates enough driving data to realize train positioning running on the line A by establishing a deep learning model, and the line B accumulates insufficient driving data to realize train positioning running on the line B by establishing the deep learning model; an available depth target domain learning model is established by adopting the following method, and the train operated on the line B is positioned by the available depth target domain learning model:
recording a set of driving data accumulated from a route A as a source domain data set, dividing the source domain data set into a source domain training set and a source domain verification set, extracting train operation dynamic characteristics, route characteristics and train attribute characteristics in the source domain data set as input characteristics, and taking train positioning position characteristics as output characteristics; recording a set of driving data accumulated from the line B as a target domain data set, dividing the target domain data set into a target domain training set and a target domain verification set, extracting train operation dynamic characteristics, line characteristics and train attribute characteristics in the target domain data set as input characteristics, and taking train positioning position characteristics as output characteristics;
1) training the deep learning model by using a source domain training set, and then verifying the deep learning model by using a source domain verification set to obtain a source domain deep learning model;
2) freezing part of shallow submodels of the source domain deep learning model, and then finely adjusting the weight and the threshold of the unfrozen submodels in the source domain deep learning model by using the labeled data in the target domain training set to obtain a target domain deep learning model;
3) performing field self-adaption processing on the target field deep learning model by using MK-MMD, and recording the target field deep learning model after the field self-adaption processing as an effective target field deep learning model;
4) verifying the effective target domain deep learning model by using the sample data in the target domain verification set, and if the effective target domain deep learning model passes the verification, taking the effective target domain deep learning model as an available target domain deep learning model; otherwise, return to step 2).
As optimization, the dynamic characteristics, the line characteristics and the train attribute characteristics of train operation extracted from the source domain data set are screened by using a Pearson correlation analysis technology, and the screened characteristics are used as input characteristics for deep learning model training or verification; and screening the train operation dynamic characteristics, the line characteristics and the train attribute characteristics extracted from the target domain data set by using a Pearson correlation analysis technology, and using the screened characteristics as input characteristics for deep learning model training or verification.
As optimization, the train operation dynamic characteristics comprise the speed of a previous sampling point, the speed of a current sampling point, the average speed of the previous sampling point, current gear information and train running time; the line characteristics comprise the average gradient of the position of a previous sampling point, the average speed of the gradient of the position of the previous sampling point, the gradient of the position of the previous sampling point and the residual length of the gradient of the position of the previous sampling point; the train attribute characteristic includes train weight.
The principle of the invention is as follows:
the deep learning emphasizes the automatic extraction of the depth and the features of the model, the model can extract more abstract features from the input features of a source domain, and the more abstract features are used as a basic model for train positioning based on the excellent feature mapping and data mining capability of a deep learning network; the shallow layer of the deep learning model learns generalized characteristics and belongs to general knowledge, and firstly, the general knowledge learned by a source domain can be transferred to a target domain by freezing parameters such as weight values, threshold values and the like of a shallow sub-network, so that the time required by training a large number of models can be reduced, and the model training efficiency is improved; secondly, a small amount of data with labels in the target domain is used for fine-tuning deep parameters of the target domain, so that high-level abstract specific features of the target domain can be learned, and negative effects caused by data distribution differences are reduced to a certain extent. The fine tuning of the parameters of the deep migration network model has great effect improvement, and the network model after migration has better performance than the original network model. The above process is actually a migration of model parameters.
On the other hand, in a train positioning application scene, parameter values such as route length, speed limit, gradient and the like between a source domain (route a) and a target domain (route B) are different, and data samples between the source domain and the target domain often have domain deviation, so that the migration learning effect is poor, and the application requirement cannot be met. In order to solve the problems, the domain self-adaptation is adopted to solve the domain migration problem among different domains, and the migration effect of the model is improved. The purpose of domain adaptation is mainly to reduce the data distribution difference between different domains through a certain strategy so as to reduce the influence of the data distribution difference on the transfer learning. In order to solve the problem of data distribution difference between the source field and the target field, the deep learning adds an adaptive layer to realize the self-adaptation between the data distribution of the source field and the data distribution of the target field. The self-adaptive technology can solve the problem of distribution difference between the source field and the target field in a targeted manner, so that a network model has better precision and stability.
By adopting the model parameter migration and statistical characteristic transformation composite type migration learning train positioning model, the problems that target domain data are insufficient and running data need to be collected and accumulated in a long time are solved, a new model does not need to be retrained by consuming a large amount of time, only a source domain (mature line) deep learning model needs to be migrated to other target domains (newly opened lines), shared knowledge migration capacity is high and training speed is high in different train positioning scenes, and the target domain with higher train positioning precision can be obtained only by adjusting the existing source domain positioning model to a certain degree, so that the train positioning information accuracy is improved, and a train operation control system based on machine learning is safer and more stable.
As an optimization scheme, the input features of the source domain and the target domain are screened by using a Pearson correlation analysis technology to remove redundant features, the calculation complexity is reduced, the input features of the model are more reasonable, and the training efficiency of the model is further improved.
Therefore, the invention has the following beneficial effects: by adopting the method, the deep learning model for train positioning can be quickly acquired for a new train running line, and the output precision of the deep learning model to the train positioning position can be improved by the method, so that the positioning accuracy and the stability of the train are improved.
Drawings
The drawings of the present invention are described below.
FIG. 1 is a schematic structural diagram of a source domain deep learning model;
FIG. 2 is a schematic structural diagram of a deep learning model of a target domain;
fig. 3 is a schematic structural diagram of performing domain adaptive processing on a target domain deep learning model.
Detailed Description
The present invention will be further described with reference to the following examples.
Two train operation lines are arranged: line A and line B; the line A accumulates enough driving data to realize train positioning running on the line A by establishing a deep learning model, and the line B accumulates insufficient driving data to realize train positioning running on the line B by establishing the deep learning model;
for example, a line between a Chongqing No. 3 line Zheng yard substation and a Tang yard substation is used as a line A, normal speed limit data is used as a source domain data set, a line between a Guangzhou No. 7 line university southwest station and a cliff station is used as a line B, normal speed limit data is used as a target domain data set, and parameters such as line length, gradient, speed limit and train weight are different between the two data sets, which are respectively shown in the following tables 1 and 2:
TABLE 1 line operation parameter Table between Zheng Hotel substation and Tang Hotel substation
Table 2 table of line operation parameters between south station of Guangzhou No. 7 line university city and cliff station
An available depth target domain learning model is established by adopting the following method, and the train operated on the line B is positioned by the available depth target domain learning model:
recording a set of driving data accumulated from a line A as a source domain data set, wherein the source domain data set contains 647590 position report point data in total of 1440 times of train inter-station operation, dividing the source domain data set into a source domain training set and a source domain verification set according to the ratio of 7:3, extracting train operation dynamic characteristics, line characteristics and train attribute characteristics in the source domain data set as input characteristics for deep learning model training, and taking train positioning position characteristics as output characteristics for deep learning model training; as shown in table 3, for a specific definition of the input features:
TABLE 3 input feature definitions
Screening train operation dynamic features, line features and train attribute features extracted from a source domain data set by using a Pearson correlation analysis technology, removing redundant features F1, F3 and F9, finally retaining the features of F2, F4, F5, F6, F7, F8 and F10, and using the screened features as input features for deep learning model training or verification;
according to the prior art, the pearson correlation coefficient calculation formula is as follows:
in the formula: f. of i And f j Respectively representing the ith and jth features;
and
mean values representing the ith and jth features, respectively; k represents the kth sample; p ij Is the pearson correlation coefficient between feature i and feature j; p ij The value range is [ -1, 1]In general, | P ij |>0.5 consider the two to be linearly related, P ij 0 then indicates complete irrelevancy, P ij >0 represents a positive correlation, P ij <0 is negatively correlated.
Recording a set of driving data accumulated from the line B as a target domain data set, dividing the target domain data set into a target domain training set and a target domain verification set, extracting train operation dynamic characteristics, line characteristics and train attribute characteristics in the target domain data set as input characteristics for deep learning model training, and taking train positioning position characteristics as output characteristics for deep learning model training; and screening the train operation dynamic characteristics, the line characteristics and the train attribute characteristics extracted from the target domain data set by using a Pearson correlation analysis technology, and using the screened characteristics as input characteristics for deep learning model training or verification.
1) Training the deep learning model by using a source domain training set, and then verifying the deep learning model by using a source domain verification set to obtain a source domain deep learning model; as shown in fig. 1, in this embodiment, the source domain deep learning model may adopt an integrated deep learning model integrating two deep belief networks DBN1 and DBN2, and each of DBN1 and DBN2 is composed of a plurality of sub models with different depths of layer, where DBN1 includes 3 RBM sub models and 1 BP sub model, DBN2 includes 2 RBM sub models and 1 BP sub model, and the whole integrated deep learning model further includes 1 fully connected layer sub model;
2) freezing part of shallow submodels of the source domain deep learning model, and then finely adjusting the weight and the threshold of the unfrozen submodels in the source domain deep learning model by using the labeled data in the target domain training set to obtain a target domain deep learning model; as shown in fig. 2, freezing the first layer of RMB submodel of the DBN1 and the first layer of RMB submodel of the DBN2, other submodels can be fine-tuned; the selected frozen shallow submodels mainly refer to those shallow submodels belonging to general knowledge and generalization characteristics;
3) as shown in fig. 3, performing domain adaptive processing on the target domain deep learning model by using MK-MMD (multi-core maximum mean error), and marking the target domain deep learning model after the domain adaptive processing as an effective target domain deep learning model;
4) verifying the effective target domain deep learning model by using the sample data in the target domain verification set, and if the effective target domain deep learning model passes the verification, taking the effective target domain deep learning model as an available target domain deep learning model; otherwise, return to step 2).
Table 4 shows the results of experiments performed according to the protocol described in this example:
TABLE 5 results of the experiment
In the table, MAE AS The average accumulated error between stations, ME the average error, MAE the average absolute error, and MSE the military error. It is shown from experimental data that, in the cross-line scenario, the test precision value MAE after the migration of the scheme proposed in this embodiment AS The method meets the requirement that the maximum measurement accumulated error of the train position is less than 2% of the length of an uncorrected section, and meets the standard requirement that the maximum measurement accumulated error of the train positioning precision is between 0.25 and 6m in the performance and function requirements of a CBTC (communication-based train control system) in the standard IEEE 1474.1. The above verification methods and experimental standards are all the contents of the prior art.
The deep learning model, the Pearson correlation analysis technology and the multi-kernel maximum mean error MK-MMD applied in the invention are common processing means in the prior art, and related contents can be obtained from related documents in the prior art by a person skilled in the art.
Claims (3)
1. A train positioning method based on cross-line composite transfer learning is characterized in that: the train positioning method comprises the following steps:
two train operation lines are arranged: line A and line B; the line A accumulates enough driving data to realize train positioning running on the line A by establishing a deep learning model, and the line B accumulates insufficient driving data to realize train positioning running on the line B by establishing the deep learning model; an available depth target domain learning model is established by adopting the following method, and the train operated on the line B is positioned by the available depth target domain learning model:
recording a set of driving data accumulated from a route A as a source domain data set, dividing the source domain data set into a source domain training set and a source domain verification set, extracting train operation dynamic characteristics, route characteristics and train attribute characteristics in the source domain data set as input characteristics, and taking train positioning position characteristics as output characteristics; recording a set of driving data accumulated from the line B as a target domain data set, dividing the target domain data set into a target domain training set and a target domain verification set, extracting train operation dynamic characteristics, line characteristics and train attribute characteristics in the target domain data set as input characteristics, and taking train positioning position characteristics as output characteristics;
1) training the deep learning model by using a source domain training set, and then verifying the deep learning model by using a source domain verification set to obtain a source domain deep learning model;
2) freezing part of shallow submodels of the source domain deep learning model, and then finely adjusting the weight and the threshold of the unfrozen submodels in the source domain deep learning model by using the labeled data in the target domain training set to obtain a target domain deep learning model;
3) performing field self-adaption processing on the target field deep learning model by using MK-MMD, and recording the target field deep learning model after the field self-adaption processing as an effective target field deep learning model;
4) verifying the effective target domain deep learning model by using the sample data in the target domain verification set, and if the effective target domain deep learning model passes the verification, taking the effective target domain deep learning model as an available target domain deep learning model; otherwise, return to step 2).
2. The train positioning method based on cross-line composite type transfer learning of claim 1, characterized in that: screening train operation dynamic features, line features and train attribute features extracted from a source domain data set by using a Pearson correlation analysis technology, and using the screened features as input features for deep learning model training or verification; and screening the train operation dynamic characteristics, the line characteristics and the train attribute characteristics extracted from the target domain data set by using a Pearson correlation analysis technology, and using the screened characteristics as input characteristics for deep learning model training or verification.
3. The train positioning method based on cross-line composite type transfer learning according to claim 1 or 2, characterized in that: the train operation dynamic characteristics comprise the speed of a previous sampling point, the speed of a current sampling point, the average speed of the previous sampling point, current gear information and train running time; the line characteristics comprise the average gradient of the position of a previous sampling point, the average speed of the gradient of the position of the previous sampling point, the gradient of the position of the previous sampling point and the residual length of the gradient of the position of the previous sampling point; the train attribute characteristic includes train weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210445242.6A CN114819083A (en) | 2022-04-26 | 2022-04-26 | Train positioning method based on cross-line composite transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210445242.6A CN114819083A (en) | 2022-04-26 | 2022-04-26 | Train positioning method based on cross-line composite transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114819083A true CN114819083A (en) | 2022-07-29 |
Family
ID=82508435
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210445242.6A Pending CN114819083A (en) | 2022-04-26 | 2022-04-26 | Train positioning method based on cross-line composite transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114819083A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115855066A (en) * | 2023-02-22 | 2023-03-28 | 湖南迈克森伟电子科技有限公司 | High-speed rail coordinate positioning correction method |
-
2022
- 2022-04-26 CN CN202210445242.6A patent/CN114819083A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115855066A (en) * | 2023-02-22 | 2023-03-28 | 湖南迈克森伟电子科技有限公司 | High-speed rail coordinate positioning correction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114362367B (en) | Cloud-edge-cooperation-oriented power transmission line monitoring system and method, and cloud-edge-cooperation-oriented power transmission line identification system and method | |
| CN110879989A (en) | Ads-b signal target identification method based on small sample local machine learning model | |
| CN114819083A (en) | Train positioning method based on cross-line composite transfer learning | |
| CN109003446B (en) | Urban peak-shifting traffic effect analysis method based on RFID data | |
| CN113327248B (en) | Tunnel traffic flow statistical method based on video | |
| CN114241307B (en) | Self-attention network-based synthetic aperture radar aircraft target identification method | |
| CN110909794A (en) | Target detection system suitable for embedded equipment | |
| CN105046959B (en) | Urban Travel Time extracting method based on Dual-window shiding matching mechanism | |
| CN114862768A (en) | Improved YOLOv5-LITE lightweight-based power distribution assembly defect identification method | |
| CN114926638A (en) | Unsupervised multi-source domain adaptive image semantic segmentation method based on weighted mutual learning | |
| CN117572457A (en) | Cross-scene multispectral point cloud classification method based on pseudo tag learning | |
| CN105553574A (en) | Support-vector-machine-based MAC protocol identification method in cognitive radio | |
| CN111027397A (en) | Method, system, medium and device for detecting comprehensive characteristic target in intelligent monitoring network | |
| CN113781404A (en) | Road disease detection method and system based on self-supervision pre-training | |
| Huang et al. | Train driving data learning with S-CNN model for gear prediction and optimal driving | |
| CN113222109A (en) | Internet of things edge algorithm based on multi-source heterogeneous data aggregation technology | |
| CN117349748A (en) | Active learning fault diagnosis method based on cloud edge cooperation | |
| CN116310328A (en) | Semantic segmentation knowledge distillation method and system based on cross-image similarity relationship | |
| CN115100847B (en) | Queuing service time estimation method for low-permeability network-connected track data | |
| CN111626508B (en) | Track traffic vehicle-mounted data prediction method based on xgboost model | |
| CN115242496A (en) | Tor encrypted traffic application behavior classification method and device based on residual error network | |
| CN113938889A (en) | Small sample Wi-Fi masquerading attack detection method and system based on meta-learning | |
| CN111930960A (en) | Knowledge graph technology-based optical transport network knowledge testing method | |
| CN107992590B (en) | Big data system beneficial to information comparison | |
| CN117576164B (en) | Remote sensing video sea-land movement target tracking method based on feature joint learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |