CN115056829A - Train motion state estimation method for multi-vehicle type continuous learning - Google Patents
Train motion state estimation method for multi-vehicle type continuous learning Download PDFInfo
- Publication number
- CN115056829A CN115056829A CN202210533015.9A CN202210533015A CN115056829A CN 115056829 A CN115056829 A CN 115056829A CN 202210533015 A CN202210533015 A CN 202210533015A CN 115056829 A CN115056829 A CN 115056829A
- Authority
- CN
- China
- Prior art keywords
- train
- data
- motion state
- vehicle type
- feature
- 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 47
- 239000010410 layer Substances 0.000 claims abstract description 19
- 239000002356 single layer Substances 0.000 claims abstract description 16
- 230000001537 neural effect Effects 0.000 claims abstract description 13
- 230000002776 aggregation Effects 0.000 claims abstract description 9
- 238000004220 aggregation Methods 0.000 claims abstract description 9
- 238000000605 extraction Methods 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000006116 polymerization reaction Methods 0.000 claims description 4
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 230000002860 competitive effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 239000011229 interlayer Substances 0.000 claims description 2
- 238000003062 neural network model Methods 0.000 claims 1
- 238000002910 structure generation Methods 0.000 claims 1
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000004088 simulation Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 230000001133 acceleration Effects 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000000513 principal component analysis Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000032683 aging Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000003337 fertilizer Substances 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61L—GUIDING RAILWAY TRAFFIC; ENSURING THE SAFETY OF RAILWAY TRAFFIC
- B61L27/00—Central railway traffic control systems; Trackside control; Communication systems specially adapted therefor
- B61L27/60—Testing or simulation
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Mechanical Engineering (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The invention relates to a train motion state estimation method for multi-vehicle type continuous learning, belonging to the technical field of computers and artificial intelligence. Firstly, determining a basic feature set of train operation data by a feature extraction method, constructing a single-layer graph structure aiming at the sample relation of each feature, and combining the single-layer graphs into a multiple graph according to the feature association relation; secondly, learning a topological structure of a multiple graph based on a neural gas network, and iteratively updating the topological structure by using multi-vehicle type operation data; then, carrying out data aggregation on the relationship between layers and the in-layer relationship of the multiple images, and generating a relationship code of a sample according to the multi-vehicle knowledge topological structure; and finally, combining the basic characteristics through nonlinear transformation, and combining the basic characteristics with the relational coding of the samples to predict the motion state of the train. The method combines single-vehicle operation data with multi-vehicle knowledge topology, and realizes continuous modeling and motion state estimation of multi-vehicle operation data under the condition of limited real operation data.
Description
Technical Field
The invention relates to a train motion state estimation method for multi-vehicle type continuous learning, belonging to the technical field of computers and artificial intelligence.
Background
In the face of a large, complex and large-passenger-capacity rail transit system, how to safely and efficiently control train operation is always a key problem of common attention in the academic world and the industrial world. Meanwhile, the running effect of the train is greatly tested by the running characteristic difference between train workshops and lines and the fluctuation of traction force and braking force caused by train aging, dynamic suite replacement and the like.
The train motion state estimation is the basis of the research and development of a train control system, can estimate the running state based on a proper motion model, is commonly used for a train simulation and automatic driving system, and calculates or predicts the speed and the position to assist in calculating the speed control level. The motion state modeling can support the establishment of a train simulation system and assist in completing analysis and testing of an automatic driving algorithm. The input of the train motion model is train mass, initial position, traction force, braking force, resistance parameters and the like, and the output is train speed and position.
The train motion state estimation mainly adopts a kinematics modeling method, a single mass point or multi-mass point model is selected for trains with different quality, length and running conditions, rail performance parameters are obtained through real train tests, and a general simulation model is established. Along with the development of machine learning technology, data-driven state estimation methods are successively proposed, the method is based on big data of the train, the historical operation data is utilized to accurately identify the motion mode of the train, an operation simulation model taking a control command as input and a motion state as output is established, and more accurate state estimation can be realized for the on-line train.
1. Traditional kinematic modeling method
The existing kinematics modeling method needs a first train dynamic dispatching experiment, physical model parameters are manually measured, and a general simple substance point model facing to a train with the same model on the whole line is established by utilizing performance data. Although the method has good physical foundation and stability, the method does not consider the non-negligible performance difference among vehicles, including acting force, control delay, resistance coefficient and the like generated by the same level. The general model is established based on the first train performance data, and the simulation errors of different trains are positively correlated with the difference between the train performance and the first train performance. Along with rail ageing or equipment replacement, the performance of the train changes along with the rail ageing, and model parameters such as actual traction force and resistance change, so that the model needs to be periodically optimized.
2. Data-driven state estimation method
The data-driven motion state modeling method comprises the steps of firstly selecting a proper model set according to tasks, then utilizing real train operation data, learning model parameters and fitting a specific mapping function between a motion state and influence factors of the motion state. The model inputs characteristics such as train position, speed, track gradient and the like at the current moment and a level sequence (considering delay system influence), and the output is estimated acceleration, so that the train motion state in the next period is calculated (the approximation is carried out by adopting a uniform acceleration hypothesis).
Compared with a kinematic model for measuring actual traction force, braking force and various resistance parameters at different speeds step by step, the data driving method can carry out end-to-end learning, directly establishes a mapping function from a motion state, a step to an acceleration, and can meet the requirement of complex pattern recognition by a model set with high VC dimension (high fitting performance). However, the existing data-driven modeling methods all rely on a large amount of running data, generally train a general simulation model by taking a line as a unit, ignore the difference of workshop performances, and take the average value of the performances of each vehicle as a fitting object so as to achieve the optimal average estimation accuracy.
In summary, the existing modeling method for the train motion state based on the kinematics model or the data driving is designed for a fixed train type and has higher modeling parameter acquisition cost, and is difficult to continuously model different trains one by one under the condition of limited real operation data, the kinematics model can only establish a general model based on individual vehicle dynamic dispatching data, while the data driving method generally needs to introduce a non-target train heterogeneous data training model, neglect the difference of the workshop performance, generate systematic training deviation which is positively influenced by the difference between the target train performance and the average train performance, and has low simulation accuracy.
Disclosure of Invention
The invention aims to solve the problems that the existing method is designed for a fixed vehicle type, has higher modeling parameter acquisition cost, does not fully consider the performance difference of train workshops and is difficult to face the problem that different trains are continuously modeled one by one under the condition of limited real operation data, a characteristic relation diagram of train operation data is constructed through a Neural network (GNN), and a multi-graph structure of multi-vehicle type data is maintained based on Neural Gas network (Neural Gas) topology, so that the train motion state estimation method for multi-vehicle type continuous learning is provided.
The design principle of the invention is as follows: firstly, determining a basic feature set of train operation data by a feature extraction method, constructing a single-layer graph structure aiming at the sample relation of each feature, and combining the single-layer graphs into a multiple graph according to the feature association relation; secondly, learning a topological structure of a multiple graph based on a neural gas network, and iteratively updating the topological structure by using multi-vehicle type operation data; then, carrying out data aggregation on the relationship between layers and the in-layer relationship of the multiple images, and generating a relationship code of a sample according to the multi-vehicle knowledge topological structure; and finally, combining the basic characteristics through nonlinear transformation, and combining the basic characteristics with the relational coding of the samples to predict the motion state of the train.
The technical scheme of the invention is realized by the following steps:
step 1, preprocessing train operation data and dividing a data set.
And 1.1, converting the log type data into form type data, and marking the processed train data according to different train types.
And 2, determining a basic feature set of train operation data through a feature extraction method, constructing a single-layer graph structure aiming at the sample relationship of each feature, combining the single-layer graphs into a multi-layer graph according to the feature association relationship, and initializing the embedded nodes of the multi-layer graph by using vectorization representation of samples.
And 2.1, fitting a random forest model by using table data of the first train type, selecting the top 20% of features as basic features based on a ranking importance method, and constructing a sample relation multiple graph of the current train type according to the selected basic features.
And 2.2, carrying out feature coding on the sample data, processing discrete variables of the sample by using the one-hot coding, and reducing the dimension of the data by using a Principal Component Analysis (PCA) algorithm. And splicing continuous variables and discrete variables of the sample, and processing by using a Convolutional Neural Network (CNN) to obtain a vectorization representation with a fixed length, wherein the vectorization representation is used as embedded node information of the multi-graph.
And 3, learning the topological structure of the multiple graph based on the neural gas network.
And 3.1, classifying according to the characteristic relation to obtain a characteristic vector set corresponding to each single-layer graph in the multi-graph structure.
And 3.2, training the neural gas network by using the feature vector sets in the step 3.1 in sequence by adopting a competitive hebrew learning method to obtain the knowledge topological structure of the feature space at the stage.
And 4, iteratively updating the neural gas network by using the motion data of the rest vehicle types according to the steps 2 to 3 to serve as a knowledge topological structure of the multi-vehicle type train data.
And 5, carrying out data aggregation on the relationship between layers and the in-layer relationship of the multiple images, and generating a relationship code of the sample according to the multi-vehicle knowledge topological structure.
And 5.1, converting the feature code of the sample data through the projection matrix, and acquiring the statistical information of each single-layer graph of the multi-graph structure by using an aggregation function.
And 5.2, generating a relation code of the sample through the multi-vehicle type knowledge topological structure.
And 6, combining the basic characteristics through nonlinear transformation, and combining the basic characteristics with the relational coding of the samples to predict the motion state of the train.
Advantageous effects
Compared with the traditional kinematic modeling method and the data-driven state estimation method, the method has the capability of continuously learning the motion data of the new-model train while maintaining the modeling capability of the historical train model. And combining single-vehicle type running data with multi-vehicle type knowledge topology to realize continuous modeling and motion state estimation of the multi-vehicle type running data under the condition of limited real running data.
Drawings
FIG. 1 is a schematic diagram of a train motion state estimation method for multi-vehicle type continuous learning according to the present invention.
Detailed Description
To better illustrate the objects and advantages of the present invention, embodiments of the method of the present invention are described in further detail below with reference to examples.
The experimental data are collected from West-An subway airport lines and a combined-fertilizer three-line train, which are provided by China Tong-Hao company. The data of the Xian subway airport line is sampled from 12 real trains, and 3,243,718 effective data are total, including 11-dimensional characteristics such as train control period, state machine, position, speed, gradient, stop point position, target speed, control speed, level and the like. The three-line data of the fat-combined subway is sampled from 22 real trains, 648,411 effective sampling data are summed, and the characteristic dimension and format of the data are consistent with those of the airport line of the Xian subway. In addition, a small amount of Chongqing line five operation data is also included. And segmenting the operation station according to actual requirements and scenes in the using process of the data. The number s of operating stations of each line is 2(n-1), and n is the number of stations.
TABLE 1 Source and basic information of train operating data
By considering the test data scale and the acceleration label value range of the motion state modeling experiment, similar effects can be obtained by each index in the model performance sequencing. Therefore, the Mean Square Error (MSE) with higher sensitivity is used as a representative evaluation index in the experiment, and the positive and negative error counteraction is avoided, and the specific calculation method comprises the following steps:
in the formula, y i The true tag value representing the ith test sample,
the estimated label value output by the model is obtained, n is the number of test samples, and the index can effectively evaluate the biased error level of the single-period acceleration estimation.
The experimental hardware environment is an MSI Prestige desktop computer, the CPU model is an Intel Core i 7-10700K processor, the CPU dominant frequency is 3.8GHz, the physical memory is 32G, the memory frequency is 2400MHz, the video card is a GeForce RTX 2080SUPER, and the video card is provided with an 8GB independent video memory.
The specific process of the experiment is as follows:
step 1, preprocessing train operation data and dividing a data set.
Step 1.1, converting the log type data into table type data, marking the processed train data according to different train types, carrying out standardized processing on the row and column names of the data of each train type, checking the integrity and noise of the data, deleting or filling the train data with missing values, and arranging the train types from large to small according to the data volume. And segmenting the operation station according to actual requirements and scenes in the using process of the data.
And 2, determining a basic feature set of train operation data through a feature extraction method, constructing a single-layer graph structure aiming at the sample relationship of each feature, combining the single-layer graphs into a multi-layer graph according to the feature association relationship, and initializing the embedded nodes of the multi-layer graph by using vectorization representation of samples.
Step 2.1, fitting a random forest model by using table data of the first train type, and selecting a weighted average method
As a combined strategy of a learner, where w i Is a single learning machine h i H (x) is a combination of learners, requiring w i ≥0，
And the prediction of the train motion state is realized. And selecting the top 20% of features as basic features based on the ranking importance method, and constructing a sample relation multiple graph of the current vehicle type according to the selected basic features.
And 2.2, carrying out feature coding on the sample data, processing discrete variables of the sample by using the one-hot coding, and reducing the dimension of the data by using a Principal Component Analysis (PCA) algorithm. And splicing continuous variables and discrete variables of the sample, performing sliding window calculation on the variable-length sequence by using a Convolutional Neural Network (CNN), outputting a vector sequence subjected to feature extraction, and changing the vector sequence into a fixed-length vector sequence through a pooling layer (Pooling) to serve as embedded node information of the multi-graph.
Step 3, learning the topological structure of the multiple graph based on the neural gas network, and defining an undirected graph G ═<V，E>Node v j Centroid for position of E V in feature space
Description of e ij E stores the relationship of the neighbor node. For new input vectors, euclidean distance is used to calculate similarity to each node.
And 3.1, classifying according to the characteristic relation to obtain a characteristic vector set corresponding to each single-layer graph in the multi-graph structure.
Step 3.2, training the neural gas network by using the competitive hebrew learning method and the feature vector set in the step 3.1 in sequence to obtain the topological representation G of the feature space at the stage (t) 。
Node r of topological space i Updating the strategy:
wherein, f represents the input feature vector,
is node r i Center of mass, eta represents learning rate, e -i Is a function of the attenuation of the light beam,
is the centroid to be updated, N (t) Is the number of nodes at time t.
Edge e of the topological space rij Updating the strategy:
wherein j represents the newly added node, and the connection relation between the newly added node and the existing node is through a ij And (6) describing variables.
And 4, iteratively updating the neural gas network by using the motion data of the other vehicle types according to the steps 2 to 3 to serve as a knowledge topological structure of the multi-vehicle type train data, so as to realize continuous learning of the multi-vehicle type relation characteristics.
And 5, carrying out data aggregation on the relationship between layers and the in-layer relationship of the multiple images, and generating a relationship code of the sample according to the multi-vehicle knowledge topological structure.
Step 5.1, through projection matrix M r Conversion sample data h x Feature coding of
And obtaining the statistical information of each single-layer graph of the multi-graph structure by using an aggregation function.
The polymerization method in the layer is as follows:
wherein AGG is a data aggregation function, related to the GNN model; w is a group of r Is a trainable weight matrix for sample data sharing, σ is an activation function,
is a relational representation of a single layer.
The interlayer polymerization method comprises the following steps:
wherein R represents the number of layers, and z represents the overall relation of the multiple graphs x 。
Step 5.2, generating a relation code of the sample through the multi-vehicle type knowledge topological structure:
which is composed of
And b o Is a trainable parameter, adjust z x Dimension and size.
Step 6, combining the basic characteristics x of the train operation data through nonlinear transformation to generate high-order representation
And coding in relation to the samples
In combination, a gradient enhancement machine (GBM) is adopted as a table data training model for predicting the motion state of the train.
And (3) testing results: according to the train motion state estimation method for experimental multi-vehicle type continuous learning, the average MSE on train data of a train line of a Siemens subway, a compound fertilizer third line and a Chongqing fifth line is 23.34, the train operation data can be continuously modeled, the vehicle motion mode can be accurately identified, and the train motion state estimation method has a good effect.
The above detailed description is intended to illustrate the objects, aspects and advantages of the present invention, and it should be understood that the above detailed description is only exemplary of the present invention and is not intended to limit the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (5)
1. The train motion state estimation method for multi-vehicle type continuous learning is characterized by comprising the following steps of:
step 1, preprocessing train operation data, dividing a data set, converting log type data into form type data, and marking the processed train data according to different train types;
step 2, determining a basic feature set of train operation data through a feature extraction method, constructing a single-layer graph structure aiming at the sample relation of each feature, combining the single-layer graphs into a multiple graph according to the feature association relation, and initializing an embedded node of the multiple graph by using vectorization representation of a sample;
step 3, learning the topological structure of the multiple graph based on the neural gas network;
step 4, iteratively updating the neural gas network by using the motion data of other vehicle types according to the steps 2 to 3 to serve as a knowledge topological structure of the multi-vehicle type train data;
step 5, carrying out data aggregation on the relationship between layers and the in-layer of the multiple images, and generating a relation code of a sample according to the multi-vehicle knowledge topological structure;
and 6, combining the basic characteristics through nonlinear transformation, and combining the basic characteristics with the relational coding of the samples to predict the motion state of the train.
2. The multi-vehicle type continuously learned train motion state estimation method according to claim 1, characterized in that: and 2, fitting the combination strategy into a random forest model by using train operation data, wherein the random forest model is a weighted average method, selecting the first 20% of characteristics as basic characteristics based on a ranking importance method, and constructing a sample relation multiple graph of the current vehicle type according to the selected basic characteristics.
3. The multi-vehicle type continuously learned train motion state estimation method according to claim 1, characterized in that: step 3, training the node r of the neural gas network and the topological space by adopting a competitive hebrew learning method i Updating the strategy:
wherein, f represents the input feature vector,
is node r i Center of mass, eta represents learning rate, e -i Is a function of the attenuation of the light beam,
is the centroid to be updated, N (t) Is the number of nodes at time t; edges of topological space
Updating the strategy:
wherein j represents the newly added node, and the connection relation between the newly added node and the existing node is through a ij The variables describe.
4. The multi-vehicle type continuously learned train motion state estimation method according to claim 1, characterized in that: the polymerization method in the layer in the step 5 comprises the following steps:
wherein AGG is a data aggregation function associated with the graph neural network model; w r Is a trainable weight matrix for sample data sharing, σ is an activation function,
is the feature encoding of the sample data,
is a relational representation of a single layer; the interlayer polymerization method comprises the following steps:
wherein R represents the number of layers, and z represents the overall relation of the multiple graphs x 。
5. The multi-vehicle type continuously learned train motion state estimation method according to claim 1, characterized in that: step 5 according to multiple vehiclesAnd (3) carrying out relational coding on the type knowledge topological structure generation samples:
wherein
And b o Is a trainable parameter, G (t) Is a relation topology representation of train operation data at the time t, and can adjust z x Dimension and size.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210533015.9A CN115056829A (en) | 2022-05-16 | 2022-05-16 | Train motion state estimation method for multi-vehicle type continuous learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210533015.9A CN115056829A (en) | 2022-05-16 | 2022-05-16 | Train motion state estimation method for multi-vehicle type continuous learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115056829A true CN115056829A (en) | 2022-09-16 |
Family
ID=83198277
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210533015.9A Pending CN115056829A (en) | 2022-05-16 | 2022-05-16 | Train motion state estimation method for multi-vehicle type continuous learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115056829A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11912321B2 (en) * | 2021-10-18 | 2024-02-27 | Tata Consultancy Services Limited | System and method for railway network access planning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112241814A (en) * | 2020-10-20 | 2021-01-19 | 河南大学 | Traffic prediction method based on reinforced space-time diagram neural network |
| WO2021082681A1 (en) * | 2019-10-29 | 2021-05-06 | 支付宝(杭州)信息技术有限公司 | Method and device for multi-party joint training of graph neural network |
| CN113255225A (en) * | 2021-05-28 | 2021-08-13 | 北京理工大学 | Train motion state estimation method for few-sample-element lifting learning |
| CN113285831A (en) * | 2021-05-24 | 2021-08-20 | 广州大学 | Network behavior knowledge intelligent learning method and device, computer equipment and storage medium |
| CN113780002A (en) * | 2021-08-13 | 2021-12-10 | 北京信息科技大学 | Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning |
| CN113806857A (en) * | 2021-08-17 | 2021-12-17 | 西安理工大学 | High-speed train energy-saving braking method based on variational graph self-encoder |
| CN113822419A (en) * | 2021-09-26 | 2021-12-21 | 广东技术师范大学 | Self-supervision graph representation learning operation method based on structural information |
-
2022
- 2022-05-16 CN CN202210533015.9A patent/CN115056829A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021082681A1 (en) * | 2019-10-29 | 2021-05-06 | 支付宝(杭州)信息技术有限公司 | Method and device for multi-party joint training of graph neural network |
| CN112241814A (en) * | 2020-10-20 | 2021-01-19 | 河南大学 | Traffic prediction method based on reinforced space-time diagram neural network |
| CN113285831A (en) * | 2021-05-24 | 2021-08-20 | 广州大学 | Network behavior knowledge intelligent learning method and device, computer equipment and storage medium |
| CN113255225A (en) * | 2021-05-28 | 2021-08-13 | 北京理工大学 | Train motion state estimation method for few-sample-element lifting learning |
| CN113780002A (en) * | 2021-08-13 | 2021-12-10 | 北京信息科技大学 | Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning |
| CN113806857A (en) * | 2021-08-17 | 2021-12-17 | 西安理工大学 | High-speed train energy-saving braking method based on variational graph self-encoder |
| CN113822419A (en) * | 2021-09-26 | 2021-12-21 | 广东技术师范大学 | Self-supervision graph representation learning operation method based on structural information |
Non-Patent Citations (1)
| Title |
|---|
| 钱晓明 等: "基于聚类与改进最小二乘法支持向量机算法的汽车总装输送装备故障预警方法", 计算机集成制造系统, no. 12, 31 December 2019 (2019-12-31), pages 244 - 249 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11912321B2 (en) * | 2021-10-18 | 2024-02-27 | Tata Consultancy Services Limited | System and method for railway network access planning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111382542B (en) | Highway electromechanical device life prediction system facing full life cycle | |
| CN111832814A (en) | Air pollutant concentration prediction method based on graph attention machine mechanism | |
| CN113487061A (en) | Long-time-sequence traffic flow prediction method based on graph convolution-Informer model | |
| CN107293115B (en) | Traffic flow prediction method for microscopic simulation | |
| WO2022021726A1 (en) | Pmu-based power system state estimation performance evaluation method | |
| CN113393032B (en) | Track circulation prediction method based on resampling | |
| CN111523778A (en) | Power grid operation safety assessment method based on particle swarm algorithm and gradient lifting tree | |
| CN109657709A (en) | A kind of equipment fault prediction technique based on particle group optimizing support vector regression | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN111191824B (en) | Power battery capacity attenuation prediction method and system | |
| CN112734094B (en) | Intelligent city intelligent rail vehicle fault gene prediction method and system | |
| CN111652835A (en) | Method for detecting insulator loss of power transmission line based on deep learning and clustering | |
| CN110555551A (en) | air quality big data management method and system for smart city | |
| CN116245406B (en) | Software operation and maintenance quality evaluation method and system based on operation and maintenance quality management database | |
| CN114510778A (en) | Track irregularity prediction method based on hybrid intelligent optimization LSTM | |
| CN113225346A (en) | Network operation and maintenance situation assessment method based on machine learning | |
| CN112884014A (en) | Traffic speed short-time prediction method based on road section topological structure classification | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| CN115542429A (en) | XGboost-based ozone quality prediction method and system | |
| CN115438726A (en) | Device life and fault type prediction method and system based on digital twin technology | |
| CN113341919A (en) | Computing system fault prediction method based on time sequence data length optimization | |
| CN115056829A (en) | Train motion state estimation method for multi-vehicle type continuous learning | |
| CN116663419A (en) | Sensorless equipment fault prediction method based on optimized Elman neural network | |
| CN112749508B (en) | Road feel simulation method based on GMM and BP neural network | |
| CN114048546A (en) | Graph convolution network and unsupervised domain self-adaptive prediction method for residual service life of aircraft engine |
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 |