CN108648457B - Method, device and computer readable storage medium for speed prediction - Google Patents
Method, device and computer readable storage medium for speed prediction Download PDFInfo
- Publication number
- CN108648457B CN108648457B CN201810690697.8A CN201810690697A CN108648457B CN 108648457 B CN108648457 B CN 108648457B CN 201810690697 A CN201810690697 A CN 201810690697A CN 108648457 B CN108648457 B CN 108648457B
- Authority
- CN
- China
- Prior art keywords
- target
- network
- matrix
- speed
- road
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- 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
Abstract
The embodiment of the invention discloses a method, a device and a computer readable storage medium for predicting speed, which are used for calculating an initial speed vector of a target road network according to track data in a preset time period; carrying out convolution processing on the initial speed vector and the adjacent road section matrix by utilizing a pre-trained convolution neural network based on a road network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section; and converting each feature vector into a time sequence, and processing the time sequence by using a pre-trained long-term and short-term memory network to obtain a target speed matrix corresponding to the target road network. Because the feature vector is obtained under the condition of considering the road network topological structure, the influence of wrong space dynamic evolution features on the precision of the feature vector is effectively avoided. The target speed vector predicted according to the characteristic vector is more accurate, and the precision of speed prediction is effectively improved.
Description
Technical Field
The present invention relates to the field of urban traffic technologies, and in particular, to a method, an apparatus, and a computer-readable storage medium for speed prediction.
Background
With the development of social economy, the quantity of private cars is gradually increased year by year, and the road reconstruction and expansion speed of some big cities cannot meet the increasing requirements of motor vehicles, so that the problems of urban traffic jam, traffic accidents and the like caused by the speed are important problems which puzzle the urban development and the travel of residents. For this reason, Intelligent Transport Systems (ITS) has been a hot spot of research as an effective solution. The traffic system is a system which is participated by people, time-varying, huge and complex, and how the traffic speed state of urban roads which is changing at any moment can be accurately predicted is one of the core problems of the ITS.
Various subsystems of the ITS, such as timely adjusting a traffic management control scheme and controlling traffic jam; issuing travel information for travelers and providing an optimal path selection scheme; accurate arrival time estimation is performed, so that reasonable departure time of a recommended user and the like all need accurate urban traffic speed prediction as a basis.
In recent years, traffic speed prediction is widely researched by academia and industry, and related methods are mainly classified into a traditional prediction method and a deep learning method. On the traditional prediction method, time series analysis is the most typical method for simulating time series patterns. Wherein the differential autoregressive moving average model predicts traffic speeds on a single road segment by a linear combination of past traffic speeds. Meanwhile, in order to describe periodic trends such as morning and evening peaks and the like, a seasonal differential autoregressive moving average model is proposed. In addition, some conventional machine learning methods such as linear regression and support vector regression are also used to perform time series pattern learning of individual road segments.
With the development of deep learning in recent years, more and more researchers are beginning to use deep learning technology to predict traffic speed. Among them, Long Short-Term Memory networks (LSTM) are good at capturing the time dependence of longer sequences and thus are used for traffic speed prediction of a single road segment.
However, the method mainly predicts a single road section, ignores the influence of surrounding road sections, and does not consider the spatial evolution relation of urban traffic speed prediction. In this regard, a Convolutional Neural Network (CNN) which is good at capturing spatial relationships is used for urban traffic speed prediction.
In the process of spatial evolution, since a traffic network of a city is a topological structure, the spatial evolution is influenced by the topological structure of the traffic network, that is, a road segment directly influences surrounding adjacent road segments. However, the conventional CNN can only learn the neighboring units in the matrix, but the city-level spatio-temporal velocity matrix cannot guarantee that the neighboring road segments are in the neighboring rows, and the method for predicting the traffic velocity of the city by using the CNN does not consider this point, so that some wrong spatial dynamic evolution characteristics are learned to influence the prediction accuracy.
Therefore, how to improve the accuracy of the speed prediction is an urgent problem to be solved by those skilled in the art.
Disclosure of Invention
The embodiment of the invention aims to provide a speed prediction method, a speed prediction device and a computer readable storage medium, which can improve the precision of speed prediction.
To solve the foregoing technical problem, an embodiment of the present invention provides a method for predicting a speed, including:
calculating initial speeds corresponding to all road segments in a target road network according to the track data in a preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network;
carrying out convolution processing on the initial speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a pre-trained convolutional neural network based on the road network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section;
converting the characteristic vector of the target road section into a time sequence, and processing the time sequence by using a pre-trained long-short term memory network to obtain a target speed vector corresponding to the target road section; the target road section is any one of all road sections in the target road network, and the target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network.
Optionally, the training process of the road network-based convolutional neural network and the long-short term memory network includes:
calculating the historical speed corresponding to each road segment in the target road network according to the training set track data; historical speeds corresponding to all the road sections form a historical speed vector of the target road network;
carrying out convolution processing on the historical speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a convolution neural network based on the road network to obtain a corresponding historical characteristic matrix; the historical feature matrix comprises historical feature vectors corresponding to the road sections respectively;
converting the historical characteristic vector of the target road section into a historical time sequence, and processing the historical time sequence by using a long-term and short-term memory network to obtain a predicted speed vector corresponding to the target road section; the predicted speed vectors corresponding to all the target road sections form a predicted speed matrix of the target road network;
and adjusting the values of the model parameters of the road network-based convolutional neural network and the long-short term memory network according to the predicted speed matrix and the actual speed matrix corresponding to the target road network until the model parameters meet preset requirements.
Optionally, the calculating an initial speed corresponding to each road segment in the target road network according to the trajectory data in the preset time period includes:
calculating the corresponding initial speed of the target road section r in the period time t by using the following formula
Wherein T represents all track sets passing through the target road section r within a cycle time T in the track data; traj represents one track in the track set T; f (traj, t, r) is used to represent the velocity value of the trajectory traj through the target road segment r within a cycle time t; length represents the total length of target road segment r;
startTime represents the start time of the track traj entering the target road section r within the cycle time t;
endTime indicates that the track traj departs within a cycle time tThe departure time of the target road segment r.
Optionally, the method further includes:
extracting a reference matrix corresponding to the target road section from historical track data by using a full connection layer of a convolutional neural network;
and performing fusion processing on the target speed matrix and the reference matrix, and taking an obtained fusion result as the latest target speed matrix of the target road network.
Optionally, the method further includes:
acquiring an actual speed matrix corresponding to the target speed matrix;
and adjusting the value of each model parameter in the network model according to the target speed matrix and the actual speed matrix obtained after the fusion processing.
The embodiment of the invention also provides a speed prediction device, which comprises a calculation unit, a processing unit and a prediction unit;
the calculating unit is used for calculating the initial speed corresponding to each road segment in the target road network according to the track data in the preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network;
the processing unit is used for carrying out convolution processing on the initial speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a pre-trained road network-based convolutional neural network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section;
the prediction unit is used for converting the characteristic vector of the target road section into a time sequence, and processing the time sequence by utilizing a pre-trained long-short term memory network to obtain a target speed vector corresponding to the target road section; the target road section is any one of all road sections in the target road network, and the target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network.
Optionally, the apparatus further includes an adjusting unit for training the road network-based convolutional neural network and the long-short term memory network;
the calculation unit is further used for calculating the historical speed corresponding to each road segment in the target road network according to the training set track data; historical speeds corresponding to all the road sections form a historical speed vector of the target road network;
the processing unit is further used for carrying out convolution processing on the historical speed vector corresponding to the target road network and the adjacent road section matrix by using a convolution neural network based on the road network to obtain a corresponding historical characteristic matrix; the historical feature matrix comprises historical feature vectors corresponding to the road sections respectively;
the prediction unit is further used for converting the historical characteristic vector of the target road section into a historical time sequence, and processing the historical time sequence by using a long-term and short-term memory network to obtain a predicted speed vector corresponding to the target road section; the predicted speed vectors corresponding to all the target road sections form a predicted speed matrix of the target road network;
and the adjusting unit is used for adjusting the values of the model parameters of the road network-based convolutional neural network and the long-short term memory network according to the predicted speed matrix and the actual speed matrix corresponding to the target road network until the model parameters meet preset requirements.
Optionally, the calculating unit is specifically configured to calculate an initial speed of the target road segment r within the cycle time t by using the following formula
Wherein T represents the trajectoryAll track sets in the data passing through the target road section r within a period time t; traj represents one track in the track set T; f (traj, t, r) is used to represent the velocity value of the trajectory traj through the target road segment r within a cycle time t; length represents the total length of target road segment r;
startTime represents the start time of the track traj entering the target road section r within the cycle time t;
endTime represents the departure time of the track traj from the target link r within the cycle time t.
Optionally, the system further comprises an extraction unit and a fusion unit;
the extraction unit is used for extracting a reference matrix corresponding to the target road section from historical track data by using a full connection layer of a convolutional neural network;
and the fusion unit is used for performing fusion processing on the target speed matrix and the reference matrix, and taking an obtained fusion result as the latest target speed matrix of the target road network.
Optionally, the system further comprises an obtaining unit and an adjusting unit;
the acquiring unit is used for acquiring an actual speed matrix corresponding to the target speed matrix;
and the adjusting unit is used for adjusting the values of the model parameters in the network model according to the target speed matrix and the actual speed matrix obtained after the fusion processing.
The embodiment of the invention also provides a speed prediction device, which comprises:
a memory for storing a computer program;
a processor for executing the computer program to implement the steps of the method of velocity prediction as described above.
Embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the method for velocity prediction as described above.
According to the technical scheme, the initial speed corresponding to each road segment in the target road network is calculated according to the track data in the preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network; carrying out convolution processing on the initial speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a pre-trained convolutional neural network based on the road network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section; converting the characteristic vector into a time sequence, and processing the time sequence by using a pre-trained long-short term memory network to obtain a target speed vector corresponding to the target road section; the target road section is any one of all road sections in a target road network, and the target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network. Because the feature vector is obtained under the condition of considering the road network topological structure, the influence of wrong space dynamic evolution features on the precision of the feature vector is effectively avoided. The target speed vector predicted according to the characteristic vector is more accurate, and the precision of speed prediction is effectively improved.
Drawings
In order to illustrate the embodiments of the present invention more clearly, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings can be obtained by those skilled in the art without inventive effort.
FIG. 1 is a flow chart of a method for speed prediction according to an embodiment of the present invention;
fig. 2a is a schematic structural diagram of a simple road network according to an embodiment of the present invention;
fig. 2b is a schematic diagram of initial speeds of road segments in the road network shown in fig. 2a in a preset time period according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a velocity prediction model according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of an apparatus for speed prediction according to an embodiment of the present invention;
fig. 5 is a schematic hardware structure diagram of an apparatus for speed prediction according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present invention without any creative work belong to the protection scope of the present invention.
In order that those skilled in the art will better understand the disclosure, the invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
Next, a method for speed prediction according to an embodiment of the present invention is described in detail. Fig. 1 is a flowchart of a method for speed prediction according to an embodiment of the present invention, where the method includes:
s101: calculating initial speeds corresponding to all road segments in a target road network according to the track data in a preset time period; and the initial speeds corresponding to all the road sections form an initial speed vector of the target road network.
With the rapid development of global positioning systems, wireless communication and mobile internet technologies, trajectory data are easier to obtain, and a large amount of trajectory data can be mapped to the traffic speed conditions of the whole city. In the embodiment of the present invention, the speed value of each road segment may be calculated using the trajectory data.
The target road network may be a road segment set for which road segment speed prediction is required, that is, the target road network includes a plurality of road segments.
In the embodiment of the present invention, the cycle time may be preset. For example, the cycle time may be set to 20 minutes. And acquiring the track data of the target road network once every cycle time.
In practical application, when the speed of each road segment in the target road network in the next cycle time needs to be predicted, the track data of the first several cycle times immediately adjacent to the cycle time can be acquired as a reference value, that is, the preset time period may include one or more cycle times.
In the embodiment of the present invention, specific values of the preset time period are not limited. For convenience of subsequent description, the method takes the trajectory data in 3 cycle times as a reference value to predict the speed value of each road segment in the target road network in the next cycle time as an example for description. Wherein the 3 cycle times are the first 3 cycle times immediately adjacent to the next cycle time to be predicted.
For example, if the preset time period includes 3 cycle times, assuming that the cycle time is 20 minutes, and the speed value of the target road network is predicted from 10 am to 10 am and 20 minutes, the trajectory data in 3 cycle times from 9 am to 20 am, from 9 am to 9 am and from 9 am to 40 am, and from 9 am to 40 am may be used as the reference value.
The initial speed of each road segment in the target road network in each cycle time is calculated in a similar manner, taking one cycle time as an example, in a specific implementation, the following formula can be used to calculate the corresponding initial speed of the target road segment r in the cycle time t
In the formula, T represents all track sets passing through the target road section r within the cycle time T in the track data; traj denotes a railOne track in the track set T; f (traj, t, r) is used to represent the velocity value of the trajectory traj through the target link r within the cycle time t; length represents the total length of target road segment r;
startTime represents the start time of the track traj entering the target road section r within the cycle time t;
endTime represents the departure time of the track traj from the target link r within the cycle time t.
With reference to the calculation process of the initial speed of the target road segment, the initial speed of all the road segments of the whole target road network in the period time t can be calculated, wherein the initial speed vector X is usedtIt is shown that,
where | E | represents the number of road segments in the target road network.
Referring to fig. 2a, a simple road network is shown, wherein the road network shown in fig. 2a comprises 7 road segments, which are r in sequence0-r6Assuming that the preset time period is 20 minutes, the initial speed of the road network from 8 to 8 points 20 minutes in the morning is shown in fig. 2b, the unit of the initial speed corresponding to each road segment is km/h, and correspondingly, the initial speed vector X corresponding to the road networkt＝[40,38.6,23.5,11.6,21.7,35.7,25.2]。
S102: and carrying out convolution processing on the initial speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a pre-trained convolutional neural network based on the road network to obtain a corresponding characteristic matrix.
The feature matrix comprises feature vectors corresponding to each road section.
Since the conventional CNN cannot be directly applied to the data having the topology structure, such as the road network, in the embodiment of the present invention, the topology structure of the road network is embedded into the convolution operation by using a Look-up convolution Layer (LC), and the Convolutional neural network introduced into the LC layer may be referred to as a road network-based Convolutional neural network.
In a specific implementation, a matrix M of contiguous links is first defined to record contiguous links of all links.
For a road segment r, the adjacency set S may be usedrAn adjacent link representing the link, specifically:
Sr＝{r,r'∈E|r'.s＝r.e or r'.e＝r.s}；
where r.s denotes a start point of the link r, r.e denotes an end point of the link r, r '. s. r.e denotes all links having the end point of r as a start point, and r'. e. r.s denotes all links having the start point of r as an end point. Thus SrThe link r itself is recorded as well as all links with the end point of r as the starting point and all links with the start point of r as the end point.
For all segments we can get a maximum number of adjacencies, denoted by a, so we can construct a matrix M of a x E. Column i of M records a road section riAll adjacent sections of the road, i.e.
The road segment (1).
For the schematic diagram of the road network shown in fig. 2a, the third road segment r3Is S ═ {0, 2, 3, 6}, where 0 denotes the link r0And 2 denotes a link r2And 3 denotes a road section r3And 6 denotes a link r6. Thus M [: 3 [ ]]＝[0,2,3,6]T，M[:,3]Representing a road section r3All the contiguous road segment matrices.
Since the contiguous road segments of each road segment are different, in an embodiment of the present invention, the missing values up to a may be filled with identification information of the road segment itself, such as id.
After the topology of the road network is expressed by M, it needs to be embedded into convolution operation to extract the features of spatial dynamic evolution.
Taking any 1 layer as an example, the input of M and the previous layerOut of Xl-1As input, to obtain the output X of the current layerlWhere l represents the number of LC layers.
Wherein the input of the first layer may consist of the velocity vector of the current cycle time and the velocity vectors of the previous p cycle times, i.e. X0＝[Xt,Xt-1,…,Xt-p]T。
The value of p can be set according to actual requirements, and the value of p is only required to be matched with the number of the cycle time included in each piece of training data during model training. Assuming that, during model training, each piece of training data predicts the speed of the target road network at the 4 th cycle time according to the speed of the target road network at the previous 3 cycles, when speed prediction is performed, a speed vector corresponding to the current cycle time is subtracted, and the value of p is 3-1-2.
As with conventional convolution operations, to obtain more diverse features, multiple convolution kernels may be used for convolution.
Taking the kth convolution kernel as an example, the kth feature matrix can be obtained after the search convolution operation, and the formula is as follows:
wherein the k-th feature matrix is actually composed of the | E | feature vectors of the road segments, and the i-th feature vector of the road segment can be obtained by the following formula,
in this formula, L (P, Q) is used to represent a look-up operation, which will return a vector or matrix based on the values in Q as an index to the second dimension in P.
Specifically, L (P, Q) ═ P [: Q [, ]]Thus, in this formula, L (M, i) ═ M [: i [, ] i]All the adjoining links, L (X), representing the acquisition link il-1,M[:,i]) That is, all the adjacent road sections are passed to obtain the sub-featuresSign matrix, here we use
And (4) showing.
Next, the submatrix is convolved by a conventional convolution operation, denoted the convolution operation, Wl,kIs a convolution kernel of h × A, h is usually set to 1 or 2, and the sub-matrix is scanned by the convolution kernel to obtain
I.e. can be expressed by the following formula:
wherein relu is a modified linear activation function, and takes the maximum value of the input value and 0, specifically relu (x) max (0, x).
Furthermore, to prevent gradient disappearance and a fast training process, in embodiments of the invention, each LC may be followed by a Batch Normalization (BN).
The BN layer is a layer that is commonly used in a convolutional neural network in the prior art to accelerate the network processing speed, and the working principle thereof is not described in detail.
S103: and converting the characteristic vector into a time sequence, and processing the time sequence by using a pre-trained long-short term memory network to obtain a target speed vector corresponding to the target road section.
The target road section is any one of all road sections included in the target road network, and the target speed vectors corresponding to all the target road sections can form a target speed matrix of the target road network.
After passing through a plurality of LC layers, we can obtain the space dynamic characteristic map on the whole road network, and the output of the last LC layer is assumed to be
Wherein k isnThe number of convolution kernels used for the last LC layer.
Before processing the feature vectors, the feature vectors corresponding to each road segment need to be transformed into a time series
Wherein each feature vector can be obtained by the following formula,
and taking the time sequence corresponding to each road section as the input of the LSTM. The hidden sequence [ h ] can be obtained iteratively using LSTM0,…,ht,…,hp]。
Wherein the content of the first and second substances,
based on the last hidden state, we can use z units to predict the traffic speed z cycle times later, and the specific calculation is as follows:
yi＝[yi,t+1,yi,t+2,…,yi,t+z]；
finally, the predicted outputs of all road sections are connected to form a target speed matrix YST，
YST＝[y0,y1,…,y|E|-1]T。
In the embodiment of the invention, the speed of the target road network is predicted by using the convolutional neural network based on the road network and the long-short term memory network, so that the combination of the time-space characteristics is realized. And an LC layer is introduced into the convolutional neural network based on the road network, so that the topological structure of the road network is embedded into the convolution operation, and the predicted speed value is more accurate.
Before the speed of the target road network is predicted by using the road network-based convolutional neural network and the long-short term memory network, model parameters in the road network-based convolutional neural network and the long-short term memory network need to be adjusted, wherein the model parameters are parameters used when corresponding functions are executed.
For convenience of introduction, the combination of the road network-based convolutional neural network and the long-short term memory network can be regarded as a network model. In the embodiment of the invention, the network model can be trained by using the historical track data corresponding to each road segment in the target road network within a period of time, so as to adjust the corresponding model parameters.
Specifically, the historical speed corresponding to each road segment in the target road network can be calculated according to the training set track data; and the historical speeds corresponding to all the road sections form a historical speed vector of the target road network. Then, carrying out convolution processing on the historical speed vector corresponding to the target road network and the adjacent road section matrix by using a convolution neural network based on the road network to obtain a corresponding historical characteristic matrix; the historical feature matrix comprises historical feature vectors corresponding to all road sections; converting the historical characteristic vector into a historical time sequence, and processing the historical time sequence by using a long-term and short-term memory network to obtain a predicted speed vector corresponding to the target road section; and the historical speed vectors corresponding to all the target road sections form a predicted speed matrix of the target road network.
The working process of determining the predicted speed matrix of the target road network by processing the training set trajectory data by using the convolutional neural network and the long-term and short-term memory network based on the road network is similar to the steps of S101 to S103, and is not described herein again.
In the embodiment of the present invention, trajectory data of a day, a week, or a month before the current time may be selected as the corpus.
Taking a day as an example, the trajectory data of the target road network in the time of day before the current time may be used as the corpus, and if the period time is assumed to be 20 minutes in the above description, 72 trajectory data may be obtained as the corpus.
Similar to the way of adjusting model parameters by using historical data in the prior art, when the model is trained, the training corpus can be divided into training set track data and verification set track data. And adjusting the model parameters of the network model by using the training set track data. And verifying whether each model parameter in the network model meets the requirement or not according to the verification set track data.
Because the actual speed value of each road section in the training corpus is a known quantity, in the embodiment of the invention, the values of each model parameter of the convolutional neural network and the long-short term memory network based on the road network can be adjusted according to the predicted speed matrix and the actual speed matrix of the target road network until each model parameter meets the preset requirement.
When adjusting the model parameters, the predicted speed value and the actual speed value can be subjected to mean square error calculation, and then the value of the model parameters is adjusted through a back propagation algorithm.
When the model parameters are verified, the mean square error of the predicted speed matrix and the actual speed matrix can be calculated according to the verification set track data, and the mean square error is used as a basis for judging whether the model parameters of the network model meet preset requirements or not.
When the mean square error value is smaller than the last mean square error value, the model parameters of the network model are indicated to have a further adjustment space, and the model parameters can be further adjusted according to the training set trajectory data.
When the mean square error value is compared with the last mean square error value and the mean square error is unchanged, the model parameters of the network model are adjusted to be optimal, and the training of the network model can be ended.
When the mean square error value is larger than the last mean square error value, it indicates that the model parameter after the last adjustment has reached the optimum, and the model parameter after the last adjustment can be used as the model parameter of the network model, and the training of the network model is finished.
According to the technical scheme, the initial speed corresponding to each road segment in the target road network is calculated according to the track data in the preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network; carrying out convolution processing on an initial speed vector corresponding to a target road network and an adjacent road section matrix by using a pre-trained road network-based convolutional neural network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section; converting the characteristic vector into a time sequence, and processing the time sequence by using a pre-trained long-short term memory network to obtain a target speed vector corresponding to a target road section; the target road section is any one of all road sections included in the target road network, and the target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network. Because the feature vector is obtained under the condition of considering the road network topological structure, the influence of wrong space dynamic evolution features on the precision of the feature vector is effectively avoided. The target speed vector predicted according to the characteristic vector is more accurate, and the precision of speed prediction is effectively improved.
In practical application, when the speed of the road network is predicted, besides the change modes of spatial evolution and time sequence, the periodic change rule of the speed of each road section and the traffic situation have certain influence on the prediction of the speed value of the road section. In the embodiment of the invention, in order to further improve the accuracy of the speed prediction of the road section, a reference matrix corresponding to the target road section can be extracted from historical track data by utilizing a full connection layer of a convolutional neural network; and performing fusion processing on the target speed vector and the reference matrix, and taking the obtained fusion result as the latest target speed vector of the target road section.
The reference matrix can include a speed vector extracted based on the traffic speed periodic variation rule and a speed vector extracted based on the influence of different traffic scene factors.
On some road sections of the target road network, the traffic speed is the same time every dayThe intervals will exhibit similarities and will also exhibit trends at the same time intervals weekly. To extract this information, in the embodiment of the present invention, one fully-connected layer may be used to extract the average velocity of the previous d days and another fully-connected layer may be used to extract the trend information of the previous w weeks, and finally they are combined to obtain the velocity vector YP。
The specific values of d and w can be determined according to the periodic variation rule of the speed of each road section in the target road network.
Traffic scenario data may include holidays, weather conditions, and some metadata such as weekends, non-weekends, days of the week, hours of the day, rush hours, off-peak hours, etc. In the embodiment of the invention, one full-connection layer can be used for extracting the information, and then another full-connection layer is used for mapping the low latitude characteristics to the high dimension to obtain the velocity vector YC。
In a specific implementation, the two speeds Y obtained can be usedPAnd YCAdding to obtain a reference matrix YEThen, the reference matrix and the target speed vector are fused by using the parameter matrix, and the obtained fusion result is obtained
As the latest target velocity vector, a specific formula is as follows,
wherein, WSTIs YSTCorresponding model parameter, WEIs YEThe corresponding model parameters.
In the embodiment of the present invention, a combination of a convolutional neural network based on a road network, a long-short term memory network, and a fully connected layer of the convolutional neural network may be used as a complete network model, a schematic structural diagram of the network model is shown in fig. 3, the convolutional neural network based on the road network may include a plurality of LC layers, in order to prevent gradient disappearance and accelerate a training process, a BN layer may be connected to each LC layer, and the BN layer belongs to a BN layerThe specific working principle of the common technology in the prior art is not described in detail. After the processing of a plurality of LC layers and BN layers, a characteristic matrix corresponding to the target road network can be obtained, all characteristic vectors contained in the characteristic matrix are converted into time sequences, and each time sequence can be processed by utilizing a long-short term memory network to obtain a target speed matrix Y corresponding to the target road networkST. By full connection layer FC S1 can extract the velocity vector YP(ii) a By full connection layer FC S2 the velocity vector Y can be extractedc(ii) a For YPAnd YcSumming to obtain a reference matrix YE. The Fusion of FIG. 3 includes the corresponding model parameters WSTAnd WEUsing hyperbolic function tanh vs YSTAnd YEPerforming fusion processing to obtain a final target speed matrix
According to the network model shown in fig. 3, the speed value of each road segment in the target road network can be predicted. In the model training stage, an actual speed matrix Y corresponding to the target speed matrix can be obtainedt(ii) a According to the target speed matrix obtained after the fusion processing
And the actual velocity matrix YtAnd adjusting the values of the model parameters in the network model. Specifically, the target speed matrix may be calculated by a Loss function (Loss)
And the actual velocity matrix YtMean square error of, i.e.
And updating each model parameter in the network model through a back propagation algorithm according to the mean square error.
Fig. 4 is a schematic structural diagram of an apparatus for speed prediction according to an embodiment of the present invention, which includes a calculating unit 41, a processing unit 42, and a predicting unit 43;
the calculating unit 41 is configured to calculate an initial speed corresponding to each road segment in the target road network according to the trajectory data in the preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network;
the processing unit 42 is configured to perform convolution processing on an initial speed vector corresponding to the target road network and an adjacent road matrix by using a pre-trained road network-based convolutional neural network to obtain a corresponding feature matrix; the feature matrix comprises a feature vector corresponding to each road section;
the prediction unit 43 is configured to convert the feature vector into a time sequence, and process the time sequence by using a pre-trained long-term and short-term memory network to obtain a target speed vector corresponding to the target road segment; the target road section is any one of all road sections included in the target road network, and the target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network.
Optionally, the device further includes an adjusting unit for training the convolutional neural network and the long-term and short-term memory network based on the road network;
the calculation unit is also used for calculating the historical speed corresponding to each road segment in the target road network according to the training set track data; historical speeds corresponding to all road sections form a historical speed vector of the target road network;
the processing unit is also used for carrying out convolution processing on the historical speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a convolution neural network based on the road network to obtain a corresponding historical characteristic matrix; the historical feature matrix comprises historical feature vectors corresponding to all road sections;
the prediction unit is also used for converting the historical characteristic vector into a historical time sequence, and processing the historical time sequence by using a long-term and short-term memory network to obtain a predicted speed vector corresponding to the target road section; historical speed vectors corresponding to all target road sections form a predicted speed matrix of the target road network;
and the adjusting unit is used for adjusting the values of the model parameters of the convolutional neural network and the long-term and short-term memory network based on the road network according to the predicted speed matrix and the actual speed matrix of the target road network until the model parameters meet the preset requirements.
Optionally, the calculating unit is specifically configured to calculate an initial speed of the target road segment r within the cycle time t by using the following formula
In the formula, T represents all track sets passing through the target road section r within the cycle time T in the track data; traj represents one track in the track set T; f (traj, t, r) is used to represent the velocity value of the trajectory traj through the target link r within the cycle time t; length represents the total length of target road segment r;
startTime represents the start time of the track traj entering the target road section r within the cycle time t;
endTime represents the departure time of the track traj from the target link r within the cycle time t.
Optionally, the system further comprises an extraction unit and a fusion unit;
the extracting unit is used for extracting a reference matrix corresponding to the target road section from the historical track data by utilizing a full connection layer of the convolutional neural network;
and the fusion unit is used for carrying out fusion processing on the target speed vector and the reference matrix and taking the obtained fusion result as the latest target speed vector of the target road section.
Optionally, the system further comprises an obtaining unit and an adjusting unit;
an acquisition unit configured to acquire an actual speed matrix corresponding to the target speed matrix;
and the adjusting unit is used for adjusting the values of the model parameters in the network model according to the target speed matrix and the actual speed matrix obtained after the fusion processing.
The description of the features in the embodiment corresponding to fig. 4 may refer to the related description of the embodiment corresponding to fig. 1, and is not repeated here.
According to the technical scheme, the initial speed corresponding to each road segment in the target road network is calculated according to the track data in the preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network; carrying out convolution processing on an initial speed vector corresponding to a target road network and an adjacent road section matrix by using a pre-trained road network-based convolutional neural network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section; converting the characteristic vector into a time sequence, and processing the time sequence by using a pre-trained long-short term memory network to obtain a target speed vector corresponding to a target road section; the target road section is any one of all road sections included in the target road network, and the target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network. Because the feature vector is obtained under the condition of considering the road network topological structure, the influence of wrong space dynamic evolution features on the precision of the feature vector is effectively avoided. The target speed vector predicted according to the characteristic vector is more accurate, and the precision of speed prediction is effectively improved.
Fig. 5 is a schematic structural diagram of a speed predicting apparatus 50 according to an embodiment of the present invention, including:
a memory 51 for storing a computer program;
a processor 52 for executing a computer program to implement the steps of the method for velocity prediction as described above.
Embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the method for predicting speed as described above.
The method, the apparatus and the computer-readable storage medium for speed prediction according to the embodiments of the present invention are described in detail above. The embodiments are described in a progressive manner in the specification, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description. It should be noted that, for those skilled in the art, it is possible to make various improvements and modifications to the present invention without departing from the principle of the present invention, and those improvements and modifications also fall within the scope of the claims of the present invention.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
Claims (10)
1. A method of velocity prediction, comprising:
calculating initial speeds corresponding to all road segments in a target road network according to the track data in a preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network;
carrying out convolution processing on the initial speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a pre-trained convolutional neural network based on the road network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section;
converting the characteristic vector of the target road section into a time sequence, and processing the time sequence by using a pre-trained long-short term memory network to obtain a target speed vector corresponding to the target road section; the target road section is any one of all road sections in the target road network, and target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network;
embedding the topological structure of the road network into convolution operation by utilizing a search convolution layer, and calling the convolution neural network introduced with the search convolution layer as a road network-based convolution neural network; the adjacent road section matrix records adjacent road sections of all road sections;
the convolutional neural network based on the road network comprises a plurality of search convolutional layers, and each search convolutional layer is connected with a batch normalization layer; and obtaining a characteristic matrix corresponding to the target road network after the processing of the plurality of search convolution layers and the batch normalization layer.
2. The method of claim 1, wherein the training process of the road network-based convolutional neural network and the long-short term memory network comprises:
calculating the historical speed corresponding to each road segment in the target road network according to the training set track data; historical speeds corresponding to all the road sections form a historical speed vector of the target road network;
carrying out convolution processing on the historical speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a convolution neural network based on the road network to obtain a corresponding historical characteristic matrix; the historical feature matrix comprises historical feature vectors corresponding to the road sections respectively;
converting the historical characteristic vector of the target road section into a historical time sequence, and processing the historical time sequence by using a long-term and short-term memory network to obtain a predicted speed vector corresponding to the target road section; the predicted speed vectors corresponding to all the target road sections form a predicted speed matrix of the target road network;
and adjusting the values of the model parameters of the road network-based convolutional neural network and the long-short term memory network according to the predicted speed matrix and the actual speed matrix corresponding to the target road network until the model parameters meet preset requirements.
3. The method of claim 1, wherein the calculating an initial speed corresponding to each road segment in the target road network according to the trajectory data within the preset time period comprises:
calculating the corresponding initial speed of the target road section r in the period time t by using the following formula
Wherein T represents all track sets passing through the target road section r within a cycle time T in the track data; traj represents one track in the track set T; f (traj, t, r) is used to represent the velocity value of the trajectory traj through the target road segment r within a cycle time t; length represents the total length of target road segment r;
represents the start time of the track traj entering the target road section r within the cycle time t;
represents the departure time of the trajectory traj from the target road segment r within the cycle time t.
4. The method of any one of claims 1-3, further comprising:
extracting a reference matrix corresponding to the target road section from historical track data by using a full connection layer of a convolutional neural network;
and performing fusion processing on the target speed matrix and the reference matrix, and taking an obtained fusion result as the latest target speed matrix of the target road network.
5. The method of claim 4, further comprising:
acquiring an actual speed matrix corresponding to the target speed matrix;
and adjusting the value of each model parameter in the network model according to the target speed matrix and the actual speed matrix obtained after the fusion processing.
6. The device for predicting the speed is characterized by comprising a computing unit, a processing unit and a predicting unit;
the calculating unit is used for calculating the initial speed corresponding to each road segment in the target road network according to the track data in the preset time period; the initial speeds corresponding to all the road sections form an initial speed vector of the target road network;
the processing unit is used for carrying out convolution processing on the initial speed vector corresponding to the target road network and the adjacent road section matrix by utilizing a pre-trained road network-based convolutional neural network to obtain a corresponding characteristic matrix; the feature matrix comprises a feature vector corresponding to each road section;
the prediction unit is used for converting the characteristic vector of the target road section into a time sequence, and processing the time sequence by utilizing a pre-trained long-short term memory network to obtain a target speed vector corresponding to the target road section; the target road section is any one of all road sections in the target road network, and target speed vectors corresponding to all the target road sections form a target speed matrix of the target road network;
embedding the topological structure of the road network into convolution operation by utilizing a search convolution layer, and calling the convolution neural network introduced with the search convolution layer as a road network-based convolution neural network; the adjacent road section matrix records adjacent road sections of all road sections;
the convolutional neural network based on the road network comprises a plurality of search convolutional layers, and each search convolutional layer is connected with a batch normalization layer; and obtaining a characteristic matrix corresponding to the target road network after the processing of the plurality of search convolution layers and the batch normalization layer.
7. The apparatus according to claim 6, wherein the apparatus further comprises an adjusting unit for training process of the road network based convolutional neural network and the long-short term memory network;
the calculation unit is further used for calculating the historical speed corresponding to each road segment in the target road network according to the training set track data; historical speeds corresponding to all the road sections form a historical speed vector of the target road network;
the processing unit is further used for carrying out convolution processing on the historical speed vector corresponding to the target road network and the adjacent road section matrix by using a convolution neural network based on the road network to obtain a corresponding historical characteristic matrix; the historical feature matrix comprises historical feature vectors corresponding to the road sections respectively;
the prediction unit is further used for converting the historical characteristic vector of the target road section into a historical time sequence, and processing the historical time sequence by using a long-term and short-term memory network to obtain a predicted speed vector corresponding to the target road section; the predicted speed vectors corresponding to all the target road sections form a predicted speed matrix of the target road network;
and the adjusting unit is used for adjusting the values of the model parameters of the road network-based convolutional neural network and the long-short term memory network according to the predicted speed matrix and the actual speed matrix corresponding to the target road network until the model parameters meet preset requirements.
8. The apparatus according to claim 6 or 7, further comprising an extraction unit and a fusion unit;
the extraction unit is used for extracting a reference matrix corresponding to the target road section from historical track data by using a full connection layer of a convolutional neural network;
and the fusion unit is used for performing fusion processing on the target speed matrix and the reference matrix, and taking an obtained fusion result as the latest target speed matrix of the target road network.
9. An apparatus for velocity prediction, comprising:
a memory for storing a computer program;
a processor for executing the computer program to carry out the steps of the method of velocity prediction according to any of claims 1 to 5.
10. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, which computer program, when being executed by a processor, carries out the steps of the method of speed prediction according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810690697.8A CN108648457B (en) | 2018-06-28 | 2018-06-28 | Method, device and computer readable storage medium for speed prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810690697.8A CN108648457B (en) | 2018-06-28 | 2018-06-28 | Method, device and computer readable storage medium for speed prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108648457A CN108648457A (en) | 2018-10-12 |
| CN108648457B true CN108648457B (en) | 2021-07-13 |
Family
ID=63749776
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810690697.8A Active CN108648457B (en) | 2018-06-28 | 2018-06-28 | Method, device and computer readable storage medium for speed prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108648457B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109410575B (en) * | 2018-10-29 | 2020-05-01 | 北京航空航天大学 | Road network state prediction method based on capsule network and nested long-time memory neural network |
| CN109544911B (en) * | 2018-10-30 | 2021-10-01 | 中山大学 | Urban road network traffic state prediction method based on LSTM-CNN |
| CN109740811A (en) * | 2018-12-28 | 2019-05-10 | 斑马网络技术有限公司 | Passage speed prediction technique, device and storage medium |
| CN111739283B (en) * | 2019-10-30 | 2022-05-20 | 腾讯科技（深圳）有限公司 | Road condition calculation method, device, equipment and medium based on clustering |
| CN111152796B (en) * | 2020-04-07 | 2020-08-07 | 北京三快在线科技有限公司 | Vehicle motion state prediction method and device |
| CN111862595B (en) * | 2020-06-08 | 2021-12-31 | 同济大学 | Speed prediction method, system, medium and device based on road network topological relation |
| CN111833600B (en) * | 2020-06-10 | 2022-07-08 | 北京嘀嘀无限科技发展有限公司 | Method and device for predicting transit time and data processing equipment |
| CN113971885B (en) * | 2020-07-06 | 2023-03-03 | 华为技术有限公司 | Vehicle speed prediction method, device and system |
| CN111950810B (en) * | 2020-08-27 | 2023-12-15 | 南京大学 | Multi-variable time sequence prediction method and equipment based on self-evolution pre-training |
| CN112265546B (en) * | 2020-10-26 | 2021-11-02 | 吉林大学 | Networked automobile speed prediction method based on time-space sequence information |
| CN112863180B (en) * | 2021-01-11 | 2022-05-06 | 腾讯大地通途（北京）科技有限公司 | Traffic speed prediction method, device, electronic equipment and computer readable medium |
| CN113160570A (en) * | 2021-05-27 | 2021-07-23 | 长春理工大学 | Traffic jam prediction method and system |
| CN115410386B (en) * | 2022-09-05 | 2024-02-06 | 同盾科技有限公司 | Short-time speed prediction method and device, computer storage medium and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106886023A (en) * | 2017-02-27 | 2017-06-23 | 中国人民解放军理工大学 | A kind of Radar Echo Extrapolation method based on dynamic convolutional neural networks |
| CN106981198A (en) * | 2017-05-24 | 2017-07-25 | 北京航空航天大学 | Deep learning network model and its method for building up for predicting travel time |
| CN107273782A (en) * | 2016-04-08 | 2017-10-20 | 微软技术许可有限责任公司 | Detected using the online actions of recurrent neural network |
| CN108196535A (en) * | 2017-12-12 | 2018-06-22 | 清华大学苏州汽车研究院（吴江） | Automated driving system based on enhancing study and Multi-sensor Fusion |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654729B (en) * | 2016-03-28 | 2018-01-02 | 南京邮电大学 | A kind of short-term traffic flow forecast method based on convolutional neural networks |
| US10289936B2 (en) * | 2016-11-08 | 2019-05-14 | Nec Corporation | Surveillance system with landmark localization on objects in images using convolutional neural networks |
| CN106779050A (en) * | 2016-11-24 | 2017-05-31 | 厦门中控生物识别信息技术有限公司 | The optimization method and device of a kind of convolutional neural networks |
| CN107103754B (en) * | 2017-05-10 | 2020-05-22 | 华南师范大学 | Road traffic condition prediction method and system |
| CN107180530B (en) * | 2017-05-22 | 2019-09-06 | 北京航空航天大学 | A kind of road network trend prediction method based on depth space-time convolution loop network |
-
2018
- 2018-06-28 CN CN201810690697.8A patent/CN108648457B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273782A (en) * | 2016-04-08 | 2017-10-20 | 微软技术许可有限责任公司 | Detected using the online actions of recurrent neural network |
| CN106886023A (en) * | 2017-02-27 | 2017-06-23 | 中国人民解放军理工大学 | A kind of Radar Echo Extrapolation method based on dynamic convolutional neural networks |
| CN106981198A (en) * | 2017-05-24 | 2017-07-25 | 北京航空航天大学 | Deep learning network model and its method for building up for predicting travel time |
| CN108196535A (en) * | 2017-12-12 | 2018-06-22 | 清华大学苏州汽车研究院（吴江） | Automated driving system based on enhancing study and Multi-sensor Fusion |
Non-Patent Citations (2)
| Title |
|---|
| Michaël Defferrard.Convolutional Neural Networks on Graphs.《NIPS》.2016, * |
| Traffic Flow Prediction With Big Data:;Yisheng Lv;《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS》;20150430;第865-872页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108648457A (en) | 2018-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648457B (en) | Method, device and computer readable storage medium for speed prediction | |
| CN108197739B (en) | Urban rail transit passenger flow prediction method | |
| CN104462190B (en) | A kind of online position predicting method excavated based on magnanimity space tracking | |
| CN103106280B (en) | A kind of range query method of uncertain space-time trajectory data under road network environment | |
| CN111145541B (en) | Traffic flow data prediction method, storage medium, and computer device | |
| CN112289034A (en) | Deep neural network robust traffic prediction method based on multi-mode space-time data | |
| CN110390349A (en) | Bus passenger flow volume based on XGBoost model predicts modeling method | |
| CN112071062B (en) | Driving time estimation method based on graph convolution network and graph attention network | |
| CN110956807B (en) | Highway flow prediction method based on combination of multi-source data and sliding window | |
| CN110570672B (en) | Regional traffic signal lamp control method based on graph neural network | |
| CN108986453A (en) | A kind of traffic movement prediction method based on contextual information, system and device | |
| CN107977734A (en) | A kind of Forecasting Methodology based on mobile Markov model under space-time big data | |
| CN112633602B (en) | Traffic congestion index prediction method and device based on GIS map information | |
| CN115206092B (en) | Traffic prediction method of BiLSTM and LightGBM models based on attention mechanism | |
| US20180192245A1 (en) | Extraction and Representation method of State Vector of Sensing Data of Internet of Things | |
| CN113887704A (en) | Traffic information prediction method, device, equipment and storage medium | |
| CN108985488A (en) | The method predicted to individual trip purpose | |
| CN115691167A (en) | Single-point traffic signal control method based on intersection holographic data | |
| CN114694382A (en) | Dynamic one-way traffic control system based on Internet of vehicles environment | |
| CN113240219A (en) | Land utilization simulation and prediction method | |
| CN113327417A (en) | Traffic flow prediction method based on 3D dynamic space-time residual convolution associated network | |
| CN117237475A (en) | Vehicle traffic track generation method and device based on diffusion generation model | |
| CN116824851A (en) | Path-based urban expressway corridor traffic jam tracing method | |
| CN115640756A (en) | Parking demand prediction model and method based on multi-source data and application thereof | |
| CN115472023A (en) | Intelligent traffic light control method and device based on deep reinforcement 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |