CN112687102A - Metropolitan area traffic flow prediction method based on knowledge graph and deep space-time convolution - Google Patents
Metropolitan area traffic flow prediction method based on knowledge graph and deep space-time convolution Download PDFInfo
- Publication number
- CN112687102A CN112687102A CN202011537038.4A CN202011537038A CN112687102A CN 112687102 A CN112687102 A CN 112687102A CN 202011537038 A CN202011537038 A CN 202011537038A CN 112687102 A CN112687102 A CN 112687102A
- Authority
- CN
- China
- Prior art keywords
- time
- data
- stgcn
- space
- network model
- 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 relates to a metropolitan area traffic flow prediction method based on a knowledge graph and deep space-time convolution. According to the invention, the space-time graph convolutional neural network is adopted for traffic flow prediction, and compared with a transmission prediction method, time and space are strongly correlated, so that the accuracy of traffic flow prediction is improved; the invention does not adopt conventional convolution and recursion units, but establishes problems on the graph and establishes a model with a complete convolution structure, so that faster training speed can be brought by fewer parameters, and the computational demand of traffic flow prediction is greatly reduced; the invention adds new data generated by the road network into the training set, continuously performs iterative optimization on the model, and improves the performance of the model.
Description
Technical Field
The invention relates to the field of traffic prediction, in particular to a metropolitan area traffic flow prediction method based on a knowledge graph and deep space-time convolution
Background
The prediction of the traffic flow of a road is very difficult due to the high randomness of traffic, the large influence of various events and the like, and particularly, the problem of the prediction difficulty on a main road of an urban area is particularly prominent. These all bring difficulties for the early deployment, early warning preparation and the like of relevant departments.
At present, in practical level, the rough level management is mainly carried out through the peak in the morning and at the evening, the major events and the like, and the quantitative standards and methods are lacked, so that the reaction is too slow when some sudden traffic jams and the like are faced.
In the conventional method, traffic flow prediction is usually only considered from a time level, spatial correlation among roads is ignored, and although some methods apply the space-time information of a road network to the traffic flow prediction, a large amount of space-time data information also brings a large amount of calculation power consumption problems.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a metropolitan area traffic flow prediction method based on a knowledge graph and deep space-time convolution.
The technical scheme adopted by the invention for solving the technical problem is as follows:
the invention utilizes the complex data collected by the buried coil to carry out arrangement and fusion, and then carries out feature learning through a space-time diagram convolutional neural network, thereby utilizing the prior traffic condition to predict the future traffic volume change, and the method comprises the following steps:
A) data cleansing
1. Extracting valid information
Data helpful for subsequent prediction, including road section ID, road section name, lane direction, lane number, acquisition time, average speed and number of vehicles passing through, are extracted.
The road section ID, the road section name and the lane direction are used for establishing the mutual corresponding relation of the road so as to be convenient for timely corresponding to the road after the prediction is finished.
The acquisition time provides time point reference for lane data fusion and data cleaning.
The average speed and the number of passing vehicles are data input for training the space-time graph convolutional neural network.
2. Lane fusion
Fusing different lane data of the same road section in the same direction according to the following modes:
1) determining a current time point;
2) determining the current road section ID;
3) if the current time point and the road section ID only correspond to one record, skipping the fusion; otherwise, combining the records into one record, and deleting the attribute of the lane number;
4) all time points and link IDs are traversed.
3. Time interpolation fusion and cleaning are specifically as follows:
1) generating time stamps from 6:00 to 23:00 every day, wherein each time point in seconds corresponds to an integer greater than 0, and the time stamps of any two time points with a time difference of 1 second have a time difference of 1;
2) taking out data according to 5-minute intervals, collecting time stamps in time, and recording the time stamps as a standard time stamp sequence;
3) for each data, if the time stamp corresponding to the collection time of the data is in the standard time stamp sequence or the difference is within 5 seconds, the data is kept, otherwise, the data is moved to the area to be deleted;
4) and for each road section and the standard timestamp, if no corresponding record is stored, finding out all the road sections in the area to be deleted, collecting data within 5 minutes of the time difference between the time and the timestamp, and returning the data to the data after lane fusion.
5) And deleting the data of all the areas to be deleted.
B) Traffic state prediction using space-time graph convolutional neural network
The space-time graph convolutional neural network comprises
Data preprocessing,
Constructing an STGCN network model,
STGCN network model training
The STGCN network model is optimized by four parts
The network model is used for modeling
The training is performed until the loss value output by the training is lower than a preset threshold value. After the reliable STGCN network model is obtained, the current traffic data is input, and the traffic state to be predicted in a future period of time can be obtained. Meanwhile, the current traffic data is updated to a training set, and the STGCN model is continuously optimized.
And merging the cleaned data into a data matrix, wherein the data matrix is 204 in total in 6:00-23:00 time periods every day by taking every 5 minutes as a time interval, an adjacent matrix of one intersection is constructed according to the connection relation of n intersections, and the data of one and a half months in succession is merged into a group of data matrices which are used as a data set and a test set for training the STGCN network.
The STGCN network model includes two space-time volume blocks and one full connectivity layer after that.
The space-time convolution block includes two time gate convolution layers and one space gate convolution layer. The input of the STGCN network model is uniformly processed by the spatio-temporal convolution block for spatio-temporal continuity. The synthesized features are integrated by a final fully connected layer to generate the final prediction.
The middle space gate convolution layer bridges two time gate convolution layers, and high-speed space state transmission from graph convolution to time convolution is achieved. Furthermore, normalization is used in each space-time convolution block to prevent overfitting.
Will be provided with
And the obtained data is reintegrated into a large matrix data, the large matrix data is used as a data set of the STGCN network model and is put into the STGCN network model for training, the large matrix data is used as a test set for testing, and finally the STGCN network model which is trained is obtained.
Will be provided with
The STGCN network model after the training is applied to an actual road network for traffic flow prediction, and meanwhile, new data generated by the road network is used according to the traffic flow prediction
And the new data is processed and added into the training set in a medium data preprocessing mode, and the model is continuously subjected to iterative optimization, so that the performance of the model is improved.
The invention has the beneficial effects that: according to the invention, the space-time graph convolutional neural network is adopted for traffic flow prediction, and compared with a transmission prediction method, time and space are strongly correlated, so that the accuracy of traffic flow prediction is improved; the invention does not adopt conventional convolution and recursion units, but establishes problems on the graph and establishes a model with a complete convolution structure, so that faster training speed can be brought by fewer parameters, and the computational demand of traffic flow prediction is greatly reduced; the invention adds new data generated by the road network into the training set, continuously performs iterative optimization on the model, and improves the performance of the model.
Drawings
FIG. 1 is a flow chart of data cleansing;
FIG. 2 is a lane fusion flow chart;
FIG. 3 is a flow chart of temporal interpolation;
FIG. 4 is a flow chart of temporal interpolation;
FIG. 5 is a time series prediction graph;
fig. 6 is a diagram of the structure of the STGCN network.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The invention mainly utilizes a method for predicting traffic by data fusion and a space-time graph convolutional neural network (STGCN network), essentially utilizes complex data collected by buried coils to carry out necessary arrangement and fusion and then carries out feature learning through the STGCN network, thereby predicting future traffic volume change by utilizing the prior traffic condition, and specifically comprises the following steps:
A) data cleaning, see FIG. 1
Data collected by the buried coil is stored in a csv form, and the vertical axis is time and takes 1 minute as a unit; the horizontal axis is traffic volume and attributes, and the specifically collected parameters include:
1. extracting valid information, see FIG. 2
Data helpful for subsequent prediction, including road section ID, road section name, lane direction, lane number, acquisition time, average speed and number of vehicles passing through, are extracted.
The road section ID, the road section name and the lane direction are used for establishing the mutual corresponding relation of the road so as to be convenient for timely corresponding to the road after the prediction is finished.
The acquisition time provides time point reference for lane data fusion and data cleaning.
Wherein the average speed and the number of passing vehicles are data input for the STGCN network model training.
This is followed by a preliminary cleaning of the data, since there may be cases of false recordings of buried coils. The specific method is to scan all records and delete the records in which the SPEED (SPEED) is 0 and the OCCUPANCY (OCCUPANCY) exceeds 1, which are obviously unreasonable.
2. Lane merging, see fig. 3
Each lane of each road in the original data is subjected to independent statistics, and different lane data of the same road section in the same direction are fused for simplifying subsequent operation. The fusion was performed as follows:
1) determining a current point in time (COLLECTIONTIME);
2) determining a current segment id (sectionid);
3) if the current time point and the road section only correspond to one record, skipping the fusion step; otherwise, combining the records into one record and deleting the attribute of the lane number. Other attribute fusion methods are as follows:
a. SPEED (SPEED), OCCUPANCY (OCCUPANCY), flow (VOLUME) were averaged over all records.
b. The vehicles passed (COMMCOUNT) take the sum of all records.
4) All time points and link IDs are traversed.
3. Temporal interpolation fusion and washing, see fig. 3 and 4.
In real-world applications, there is often no need for either a very time-refined prediction or a prediction that is not performed late at night, and therefore, the 1 minute interval data throughout the day provided by the coils may cause data redundancy problems. Only five minute intervals of data of 6:00-23:00 a day in the morning need to be considered. But redundant data can be used to fill in erroneous data that may be generated by the coil, i.e., the data deleted in step 1. The specific method comprises the following steps:
1) generating time stamps from 6:00 to 23:00 of each day, wherein each time point (in seconds) corresponds to an integer larger than 0, and the time stamps of any two time points which are different by 1 second are different by 1;
2) taking out the time stamps in the data collection time according to the 5-minute interval, and recording the time stamps as a standard time stamp sequence;
3) for each data, if the time stamp corresponding to the collection time (COLLECTIONTIME) of the data is in the standard time stamp sequence or the difference is within 5 seconds, the data is retained, otherwise, the data is moved to the area to be deleted;
4) for each road Section (SECTIONID) and standard timestamp, if no corresponding record is stored, finding out that all road Sections (SECTIONID) are the same in the area to be deleted, collecting data with the difference between the time (COLLECTINTIME) and the timestamp within 5 minutes, fusing according to the method in the step 2, and returning the data to the data;
5) and deleting the data of all the areas to be deleted.
B) Data prediction using STGCN network
Aiming at the problem of state prediction on the time series in the traffic field, the invention uses a novel deep learning framework STGCN network. The invention does not adopt the conventional convolution and recursion units, but establishes problems on the graph and establishes a model with a complete convolution structure.
The STGCN network comprises
Data preprocessing,
Constructing an STGCN network model,
STGCN network model training
The STGCN network model is optimized by four parts
The network model is used for modeling
And training the medium data set until the loss value output by the medium data set is lower than a preset threshold value, and stopping training. After the reliable STGCN network model is obtained, the current traffic data is input, and the traffic state to be predicted in a future period of time can be obtained. Meanwhile, the current traffic data is updated to a training set, and the STGCN model is continuously optimized.
Traffic prediction is typically a time series prediction problem, see fig. 5, that is, the most likely traffic measure (e.g., speed or traffic flow) at the next H time step is predicted based on previous M observations of the traffic network, i.e.
Each one of which is
All represent the current traffic state at time step t and are recorded in a data matrix.
In the problem, the invention uses the data obtained in the step A, and combines the data into a data matrix, wherein the data matrix is 204 in total in 6:00-23:00 time periods of every day with every 5 minutes as a time interval, an adjacent matrix of an intersection is constructed according to the connection relation of n intersections, and the data of continuous half month is combined into a group of data matrices as a data set and a test set for training the STGCN network.
The framework of the STGCN network comprises two space-time convolutional layers (ST-Conv) followed by a fully-connected layer. Each ST-Conv includes two time gate convolutional layers and one space gate convolutional layer. The inputs are processed uniformly by ST-Conv for spatial-temporal continuity, and the composite features are integrated by a final fully-connected layer to generate the final prediction.
As shown in fig. 6, the intermediate space-gate convolutional layer bridges the two time-gate convolutional layers, and fast space-state propagation from graph convolution to time convolution can be achieved. Furthermore, normalization is used in each ST-Conv block to prevent overfitting.
The inputs and outputs of ST-Conv are all vectors in 3D. Input for point in time l
Wherein M represents the frame number of the road map, C represents the feature dimension, and the output of the next time point l +1 is
Wherein the invention uses a width KlIs convolved to extract KlA temporal characteristic of 1, then vl+1Calculated by the following equation:
wherein
And
respectively, a front time kernel and a back time kernel in the kernel; thetalIs the spectral kernel of the graph convolution; RELU (·) represents a linear unit function.
After stacking 2 ST-Conv blocks, an extra temporal convolutional layer with a fully connected layer is added at the end as output layer. The time convolution layer maps the output of the last ST-Conv block to single step prediction. Then, Z ∈ R can be obtained from the modeln×CAnd for speed prediction of n intersections by applying a linear transformation on the c-channel, the formula is as follows:
v＝Zw+b
where w is a weight vector and b is a bias.
The performance of the model described above was measured using the L2 norm loss. Therefore, the loss function of the STGCN network to traffic prediction can be expressed as:
wherein, WθAre the parameters of the training in the model,
for model prediction, vt+1Is the true value at the next time instant.
And (3) reintegrating the data obtained in the first step into a large matrix data, putting the large matrix data into the STGCN as a data set of the STGCN for training, and testing by using the large matrix data as a test set to finally obtain the STGCN after training.
Will be provided with
The STGCN network model after the training is applied to an actual road network for traffic flow prediction, and meanwhile, new data generated by the road network is used according to the traffic flow prediction
The method for preprocessing the intermediate data adds the processed new data into a training set, and continuously performs iterative optimization on the model to improve the performance of the model.
Claims (4)
1. A metropolitan area traffic flow prediction method based on a knowledge map and deep space-time convolution utilizes complex data collected by buried coils to carry out sorting and fusion, and then carries out feature learning through a space-time map convolution neural network, thereby predicting future traffic volume change by utilizing the prior traffic condition, and the method is characterized by comprising the following steps:
A) data cleansing
1. Extracting valid information
Extracting data which is beneficial to subsequent prediction and comprises a road section ID, a road section name, a lane direction, a lane number, acquisition time, an average speed and a vehicle passing number;
2. lane fusion
Fusing different lane data of the same road section in the same direction according to the following modes:
1) determining a current time point;
2) determining the current road section ID;
3) if the current time point and the road section ID only correspond to one record, skipping the fusion; otherwise, combining the records into one record, and deleting the attribute of the lane number;
4) traversing all time points and section IDs;
3. time interpolation fusion and cleaning are specifically as follows:
1) generating time stamps from 6:00 to 23:00 every day, wherein each time point in seconds corresponds to an integer greater than 0, and the time stamps of any two time points with a time difference of 1 second have a time difference of 1;
2) taking out data according to 5-minute intervals, collecting time stamps in time, and recording the time stamps as a standard time stamp sequence;
3) for each data, if the time stamp corresponding to the collection time of the data is in the standard time stamp sequence or the difference is within 5 seconds, the data is kept, otherwise, the data is moved to the area to be deleted;
4) for each road section and the standard timestamp, if no corresponding record is stored, finding out all the road sections in the area to be deleted, collecting data within 5 minutes of the time difference between the time and the timestamp, and returning the data to the data after lane fusion;
5) deleting the data of all the areas to be deleted;
B) traffic state prediction using space-time graph convolutional neural network
The space-time graph convolutional neural network comprises
Data preprocessing,
Constructing an STGCN network model,
STGCN network model training
The STGCN network model is optimized by four parts
The network model is used for modeling
Training the data set until the loss value output by the data set is lower than a preset threshold value; after the reliable STGCN network model is obtained, the current traffic data is input, and the traffic state to be predicted in a future period of time can be obtained; at the same time will be presentUpdating traffic data into a training set, and continuously optimizing the STGCN model;
Merging the cleaned data into a data matrix, wherein the data matrix takes every 5 minutes as a time interval, 204 intersections are constructed in a period of 6:00-23:00 every day, an adjacent matrix of one intersection is constructed according to the connection relation of n intersections, and the data of one and a half months are merged into a group of data matrix which is used as a data set and a test set for training the STGCN network;
The STGCN network model comprises two space-time volume blocks and a full connection layer after the two space-time volume blocks;
the space-time convolution block comprises two time gate convolution layers and a space gate convolution layer; the input of the STGCN network model is uniformly processed by a space-time convolution block so as to facilitate the continuity of space and time; the synthesis features are integrated by a final fully connected layer to generate a final prediction;
Will be provided with
The obtained data is reintegrated into a large matrix data, the large matrix data is used as a data set of the STGCN network model and is put into the STGCN network model for training, the large matrix data is used as a test set for testing, and finally the STGCN network model which is trained is obtained;
Will be provided with
The STGCN network model after the training is applied to an actual road network for traffic flow prediction, and meanwhile, new data generated by the road network is used according to the traffic flow prediction
And the new data is processed and added into the training set in a medium data preprocessing mode, and the model is continuously subjected to iterative optimization, so that the performance of the model is improved.
2. The metro traffic flow prediction method based on the knowledge graph and the deep space-time convolution according to claim 1, characterized in that: the extracted effective information comprises:
the road section ID, the road section name and the lane direction are used for establishing the mutual corresponding relation of the road so as to be convenient for timely corresponding to the road after the prediction is finished;
the acquisition time provides time point reference for lane data fusion and data cleaning;
the average speed and the number of passing vehicles are data input for training the space-time graph convolutional neural network.
3. The metro traffic flow prediction method based on the knowledge graph and the deep space-time convolution according to claim 1, characterized in that: in the STGCN network model: the middle space gate convolution layer bridges the two time gate convolution layers, and high-speed space state transmission from graph convolution to time convolution is achieved.
4. The metro traffic flow prediction method based on the knowledge graph and the deep space-time convolution according to claim 1, characterized in that: in the STGCN network model: normalization is used in each space-time convolution block to prevent overfitting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011537038.4A CN112687102A (en) | 2020-12-23 | 2020-12-23 | Metropolitan area traffic flow prediction method based on knowledge graph and deep space-time convolution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011537038.4A CN112687102A (en) | 2020-12-23 | 2020-12-23 | Metropolitan area traffic flow prediction method based on knowledge graph and deep space-time convolution |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112687102A true CN112687102A (en) | 2021-04-20 |
Family
ID=75451001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011537038.4A Pending CN112687102A (en) | 2020-12-23 | 2020-12-23 | Metropolitan area traffic flow prediction method based on knowledge graph and deep space-time convolution |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112687102A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111932026A (en) * | 2020-08-27 | 2020-11-13 | 西南交通大学 | Urban traffic pattern mining method based on data fusion and knowledge graph embedding |
| CN113379156A (en) * | 2021-06-30 | 2021-09-10 | 南方科技大学 | Speed prediction method, device, equipment and storage medium |
| CN113570859A (en) * | 2021-07-23 | 2021-10-29 | 江南大学 | Traffic flow prediction method based on asynchronous space-time expansion graph convolution network |
| CN113792929A (en) * | 2021-04-26 | 2021-12-14 | 青岛大学 | Traffic flow prediction method, electronic device, and storage medium |
| CN114360254A (en) * | 2022-03-17 | 2022-04-15 | 湖南工商大学 | Space-time-based traffic flow prediction method and related equipment |
| CN114360239A (en) * | 2021-12-03 | 2022-04-15 | 武汉工程大学 | Traffic prediction method and system for multilayer space-time traffic knowledge map reconstruction |
| CN116153084A (en) * | 2023-04-20 | 2023-05-23 | 智慧互通科技股份有限公司 | Vehicle flow direction prediction method, prediction system and urban traffic signal control method |
| CN116774086A (en) * | 2023-06-09 | 2023-09-19 | 淮阴工学院 | Lithium battery health state estimation method based on multi-sensor data fusion |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180530A (en) * | 2017-05-22 | 2017-09-19 | 北京航空航天大学 | A kind of road network trend prediction method based on depth space-time convolution loop network |
| US10074271B1 (en) * | 2017-06-02 | 2018-09-11 | Yandex Europe Ag | Methods and systems for generating a traffic prediction |
| US20190116505A1 (en) * | 2017-10-13 | 2019-04-18 | At&T Intellectual Property I, L.P. | Network traffic forecasting for non-ticketed events |
| CN110347668A (en) * | 2019-07-04 | 2019-10-18 | 南京航空航天大学 | The cleaning of ADS-B track and calibrating installation |
| CN110362559A (en) * | 2019-07-04 | 2019-10-22 | 南京航空航天大学 | The cleaning of ADS-B track and calibration method based on part traversal Density Clustering |
| CN110827546A (en) * | 2019-11-21 | 2020-02-21 | 银江股份有限公司 | Signalized intersection road section short-term speed prediction method |
| CN111968375A (en) * | 2020-08-27 | 2020-11-20 | 北京嘀嘀无限科技发展有限公司 | Traffic flow prediction method and device, readable storage medium and electronic equipment |
-
2020
- 2020-12-23 CN CN202011537038.4A patent/CN112687102A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180530A (en) * | 2017-05-22 | 2017-09-19 | 北京航空航天大学 | A kind of road network trend prediction method based on depth space-time convolution loop network |
| US10074271B1 (en) * | 2017-06-02 | 2018-09-11 | Yandex Europe Ag | Methods and systems for generating a traffic prediction |
| US20190116505A1 (en) * | 2017-10-13 | 2019-04-18 | At&T Intellectual Property I, L.P. | Network traffic forecasting for non-ticketed events |
| CN110347668A (en) * | 2019-07-04 | 2019-10-18 | 南京航空航天大学 | The cleaning of ADS-B track and calibrating installation |
| CN110362559A (en) * | 2019-07-04 | 2019-10-22 | 南京航空航天大学 | The cleaning of ADS-B track and calibration method based on part traversal Density Clustering |
| CN110827546A (en) * | 2019-11-21 | 2020-02-21 | 银江股份有限公司 | Signalized intersection road section short-term speed prediction method |
| CN111968375A (en) * | 2020-08-27 | 2020-11-20 | 北京嘀嘀无限科技发展有限公司 | Traffic flow prediction method and device, readable storage medium and electronic equipment |
Non-Patent Citations (2)
| Title |
|---|
| BINGYU等: "Spatio-Temporal Graph Convolutional Networks_ A Deep Learning Framework for Traffic", 《IJCAI-2018》 * |
| 邹易呈: "基于多源数据融合的交通知识图谱构建及应用", 《中国万方学位论文库》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111932026B (en) * | 2020-08-27 | 2022-03-04 | 西南交通大学 | Urban traffic pattern mining method based on data fusion and knowledge graph embedding |
| CN111932026A (en) * | 2020-08-27 | 2020-11-13 | 西南交通大学 | Urban traffic pattern mining method based on data fusion and knowledge graph embedding |
| CN113792929A (en) * | 2021-04-26 | 2021-12-14 | 青岛大学 | Traffic flow prediction method, electronic device, and storage medium |
| CN113792929B (en) * | 2021-04-26 | 2023-09-22 | 青岛大学 | Traffic flow prediction method, electronic device and storage medium |
| CN113379156A (en) * | 2021-06-30 | 2021-09-10 | 南方科技大学 | Speed prediction method, device, equipment and storage medium |
| CN113570859B (en) * | 2021-07-23 | 2022-07-22 | 江南大学 | Traffic flow prediction method based on asynchronous space-time expansion graph convolution network |
| CN113570859A (en) * | 2021-07-23 | 2021-10-29 | 江南大学 | Traffic flow prediction method based on asynchronous space-time expansion graph convolution network |
| CN114360239A (en) * | 2021-12-03 | 2022-04-15 | 武汉工程大学 | Traffic prediction method and system for multilayer space-time traffic knowledge map reconstruction |
| CN114360254A (en) * | 2022-03-17 | 2022-04-15 | 湖南工商大学 | Space-time-based traffic flow prediction method and related equipment |
| CN116153084A (en) * | 2023-04-20 | 2023-05-23 | 智慧互通科技股份有限公司 | Vehicle flow direction prediction method, prediction system and urban traffic signal control method |
| CN116153084B (en) * | 2023-04-20 | 2023-09-08 | 智慧互通科技股份有限公司 | Vehicle flow direction prediction method, prediction system and urban traffic signal control method |
| CN116774086A (en) * | 2023-06-09 | 2023-09-19 | 淮阴工学院 | Lithium battery health state estimation method based on multi-sensor data fusion |
| CN116774086B (en) * | 2023-06-09 | 2024-03-26 | 淮阴工学院 | Lithium battery health state estimation method based on multi-sensor data fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112687102A (en) | Metropolitan area traffic flow prediction method based on knowledge graph and deep space-time convolution | |
| CN111612206B (en) | Neighborhood people stream prediction method and system based on space-time diagram convolution neural network | |
| CN109697852B (en) | Urban road congestion degree prediction method based on time sequence traffic events | |
| CN109285346B (en) | Urban road network traffic state prediction method based on key road sections | |
| CN113487066B (en) | Long-time-sequence freight volume prediction method based on multi-attribute enhanced graph convolution-Informer model | |
| Fang et al. | FTPG: A fine-grained traffic prediction method with graph attention network using big trace data | |
| CN112257934A (en) | Urban people flow prediction method based on space-time dynamic neural network | |
| CN113326974B (en) | Multi-source traffic flow prediction method based on multi-task hypergraph convolutional network | |
| CN109166309A (en) | A kind of missing data on flows restoration methods towards complicated urban traffic network | |
| CN112215427B (en) | Vehicle driving track reconstruction method and system under condition of bayonet data loss | |
| CN106447119A (en) | Short-term traffic flow prediction method and system based on convolutional neural network | |
| CN111489013A (en) | Traffic station flow prediction method based on space-time multi-graph convolution network | |
| CN112150207A (en) | Online taxi appointment order demand prediction method based on space-time context attention network | |
| CN113283581B (en) | Multi-fusion graph network collaborative multi-channel attention model and application method thereof | |
| CN112489426A (en) | Urban traffic flow space-time prediction scheme based on graph convolution neural network | |
| CN112863182B (en) | Cross-modal data prediction method based on transfer learning | |
| CN111047078B (en) | Traffic characteristic prediction method, system and storage medium | |
| CN112633602B (en) | Traffic congestion index prediction method and device based on GIS map information | |
| CN116913088A (en) | Intelligent flow prediction method for expressway | |
| CN111445700B (en) | Regional running and stopping vehicle monitoring method based on license plate identification data | |
| Wu et al. | Spatiotemporal scenario generation of traffic flow based on lstm-gan | |
| CN111461426A (en) | High-precision travel time length prediction method based on deep learning | |
| CN115985104A (en) | Traffic flow prediction device, prediction method and prediction model construction method | |
| CN114913689A (en) | Traffic flow prediction method based on priori knowledge time-varying graph convolution network | |
| CN112309118B (en) | Vehicle trajectory calculation method based on space-time similarity |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210420 |
|
| WD01 | Invention patent application deemed withdrawn after publication |