CN107103754B - Road traffic condition prediction method and system - Google Patents
Road traffic condition prediction method and system Download PDFInfo
- Publication number
- CN107103754B CN107103754B CN201710326733.8A CN201710326733A CN107103754B CN 107103754 B CN107103754 B CN 107103754B CN 201710326733 A CN201710326733 A CN 201710326733A CN 107103754 B CN107103754 B CN 107103754B
- Authority
- CN
- China
- Prior art keywords
- road
- data
- traffic
- neural network
- track
- 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
- G08G1/0108—Measuring and analyzing of parameters relative to traffic conditions based on the source of data
- G08G1/0112—Measuring and analyzing of parameters relative to traffic conditions based on the source of data from the vehicle, e.g. floating car data [FCD]
-
- 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
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- 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
- G08G1/0133—Traffic data processing for classifying traffic situation
Landscapes
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a road traffic condition prediction method and a system, wherein the method comprises the following steps: acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network; the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point. The method adopts the cyclic convolution neural network for prediction, has high prediction accuracy and high stability, and can be widely applied to intelligent cities.
Description
Technical Field
The invention relates to the field of intelligent cities, in particular to a road traffic condition prediction method and a road traffic condition prediction system.
Background
The noun explains:
a recurrent neural network: the English language is called Current Neural Networks, and is abbreviated as RNNs. Has been used in a great number of natural language processing fields with great success and wide application. The concrete expression is that the network memorizes the previous information and applies the previous information to the calculation of the current output, namely, the nodes between the hidden layers are not connected any more but connected, and the input of the hidden layer comprises not only the output of the input layer but also the output of the hidden layer at the last moment.
A convolutional neural network: the English is called the volumetric neural networks, abbreviated as CNNs. An important feature of the convolutional neural network is that the convolution operation can enhance the characteristics of the original signal and reduce noise, and the principle of local image correlation is used to sub-sample the image, so that the data processing amount can be reduced while retaining useful information.
In intelligent traffic systems, road traffic condition prediction is one of the biggest challenges facing today's intelligent cities. Accurate traffic state prediction is the basis of an intelligent traffic system. Map service providers typically use general network traffic flow for road condition prediction, which is implemented based on conventional segmentation methods and road traffic assessment applications. Although the method can realize the prediction of the traffic state to a certain extent, the prediction precision is general, and the development requirement of the intelligent city is difficult to meet.
Disclosure of Invention
In order to solve the above technical problems, an object of the present invention is to provide a road traffic condition prediction method, and an object of the present invention is to provide a road traffic condition prediction system.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a road traffic condition prediction method includes the steps:
acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network;
the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point.
Further, the method also comprises the following steps:
a data acquisition step: acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices;
a data preprocessing step: after data cleaning is carried out on the acquired traffic data, clustering and grouping the traffic data after the data cleaning based on flow similarity;
a feature vector construction step: for each road, constructing a feature vector of the road based on the distance measurement of the public track and the historical traffic condition of the road;
and a deep learning framework construction step: constructing a cyclic convolution neural network based on a deep learning technology, and performing deep feature extraction on the feature vectors;
training and optimizing: and training and optimizing the cyclic convolution neural network based on a momentum random gradient descent algorithm.
Further, in the data preprocessing step, the step of cleaning the acquired traffic data includes:
and sequentially performing abnormal data processing, invalid data deleting processing, data filling processing and data smoothing processing on the acquired traffic data.
Further, in the data preprocessing step, the step of clustering and grouping the traffic data after data cleaning based on the traffic similarity includes:
segmenting each traffic data according to a preset period;
clustering the segmentation result based on Euclidean metric;
and constructing a representative track of each cluster as a mode of the cluster, wherein the representative track is the average value of the length and the angle of the whole track segment belonging to the cluster.
Further, the feature vector constructing step specifically includes: and for each road to be analyzed, acquiring a representative track of the road based on the public track distance measurement, acquiring the first K representative tracks closest to the representative track as K adjacent modes of the road, and further constructing the feature vector of the road according to the K adjacent modes and the historical traffic condition of the road.
Further, the deep learning framework constructing step includes:
calculating and obtaining left and right context information of adjacent modes of each road as a training set, and associating the left and right context information with the traffic condition corresponding to the road;
after a cyclic convolution neural network is constructed, the obtained left and right context information is input into a cyclic convolution layer of the neural network;
after the output of the circulating convolution layer is subjected to linear transformation and tanh activation function calculation, the average value calculation is carried out on the calculation result;
and taking the average value obtained by calculation as the input of the exit layer, and finally obtaining the output of the output layer.
Further, the training optimization step specifically includes:
based on a momentum random gradient descent algorithm, gradient descent calculation is carried out on the training parameters of the cyclic convolution neural network, and the log likelihood value of the training parameters is maximized, so that the output of the cyclic convolution neural network is closest to the actual traffic condition of the road.
The other technical scheme adopted by the invention for solving the technical problem is as follows:
a road traffic condition prediction system comprising a processor and a storage device, the storage device storing a plurality of instructions, the instructions being loaded by the processor and performing the steps of:
acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network;
the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point.
Further, the processor load instruction also performs the steps of:
a data acquisition step: acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices;
a data preprocessing step: after data cleaning is carried out on the acquired traffic data, clustering and grouping the traffic data after the data cleaning based on flow similarity;
a feature vector construction step: for each road, constructing a feature vector of the road based on the distance measurement of the public track and the historical traffic condition of the road;
and a deep learning framework construction step: constructing a cyclic convolution neural network based on a deep learning technology, and performing deep feature extraction on the feature vectors;
training and optimizing: and training and optimizing the cyclic convolution neural network based on a momentum random gradient descent algorithm.
The invention has the beneficial effects that: the road traffic condition prediction method of the present invention includes the steps of: acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network; the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point. The method adopts the cyclic convolution neural network for prediction, and has high prediction accuracy and high stability.
The invention has another effect that: the invention relates to a road traffic condition prediction system, which comprises a processor and a storage device, wherein the storage device stores a plurality of instructions, and the instructions are loaded by the processor and execute the following steps: acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network; the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point. The system adopts the cyclic convolution neural network for prediction, and has high prediction accuracy and high stability.
Drawings
The invention is further illustrated by the following figures and examples.
FIG. 1 is a sample road illustration of an embodiment of the road traffic condition prediction method of the present invention;
FIG. 2 is a schematic structural diagram of a cyclic convolution neural network employed in the road traffic condition prediction method of the present invention;
fig. 3 is a schematic diagram of the result of performing prediction performance verification on the cyclic convolution neural network of the method.
Detailed Description
The technical solution of the present invention is further described with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a road traffic condition prediction method, which comprises the following steps:
acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network;
the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point.
Further as a preferred embodiment, the method further comprises the following steps:
a data acquisition step: acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices;
a data preprocessing step: after data cleaning is carried out on the acquired traffic data, clustering and grouping the traffic data after the data cleaning based on flow similarity;
a feature vector construction step: for each road, constructing a feature vector of the road based on the distance measurement of the public track and the historical traffic condition of the road;
and a deep learning framework construction step: constructing a cyclic convolution neural network based on a deep learning technology, and performing deep feature extraction on the feature vectors;
training and optimizing: and training and optimizing the cyclic convolution neural network based on a momentum random gradient descent algorithm.
Specifically, in the prediction process, a data preprocessing step and a feature vector construction step are also required to be performed on the acquired traffic data, the traffic data are input into a cyclic convolution neural network for prediction, and finally, the accuracy rate is predicted by using a Softmax function.
Further as a preferred embodiment, in the data preprocessing step, the step of performing data cleaning on the acquired traffic data specifically includes:
and sequentially performing abnormal data processing, invalid data deleting processing, data filling processing and data smoothing processing on the acquired traffic data.
Further preferably, in the data preprocessing step, the step of clustering and grouping the traffic data after data cleaning based on the traffic similarity includes:
segmenting each traffic data according to a preset period;
clustering the segmentation result based on Euclidean metric;
and constructing a representative track of each cluster as a mode of the cluster, wherein the representative track is the average value of the length and the angle of the whole track segment belonging to the cluster.
Further preferably, the feature vector constructing step specifically includes: and for each road to be analyzed, acquiring a representative track of the road based on the public track distance measurement, acquiring the first K representative tracks closest to the representative track as K adjacent modes of the road, and further constructing the feature vector of the road according to the K adjacent modes and the historical traffic condition of the road. The historical traffic conditions of the road specifically include: traffic conditions at t on the last week, traffic conditions at t +15 on the same day on the last week, traffic conditions at t on yesterday, traffic conditions at t +15 on yesterday, etc.
Further preferably, the deep learning framework constructing step includes:
calculating and obtaining left and right context information of adjacent modes of each road as a training set, and associating the left and right context information with the traffic condition corresponding to the road;
after a cyclic convolution neural network is constructed, the obtained left and right context information is input into a cyclic convolution layer of the neural network;
after the output of the circulating convolution layer is subjected to linear transformation and tanh activation function calculation, the average value calculation is carried out on the calculation result;
and taking the average value obtained by calculation as the input of the exit layer, and finally obtaining the output of the output layer.
Further as a preferred embodiment, the training optimization step specifically includes:
based on a momentum random gradient descent algorithm, gradient descent calculation is carried out on the training parameters of the cyclic convolution neural network, and the log likelihood value of the training parameters is maximized, so that the output of the cyclic convolution neural network is closest to the actual traffic condition of the road.
The invention also provides a road traffic condition prediction system, which comprises a processor and a storage device, wherein the storage device stores a plurality of instructions, and the instructions are loaded by the processor and execute the following steps:
acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network;
the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point.
Further as a preferred embodiment, the processor load instruction further performs the steps of:
a data acquisition step: acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices;
a data preprocessing step: after data cleaning is carried out on the acquired traffic data, clustering and grouping the traffic data after the data cleaning based on flow similarity;
a feature vector construction step: for each road, constructing a feature vector of the road based on the distance measurement of the public track and the historical traffic condition of the road;
and a deep learning framework construction step: constructing a cyclic convolution neural network based on a deep learning technology, and performing deep feature extraction on the feature vectors;
training and optimizing: and training and optimizing the cyclic convolution neural network based on a momentum random gradient descent algorithm.
On one hand, the method combines similar track data together as a mode of each road for data preprocessing, and adopts a concept based on context to construct a characteristic vector for the modes as an input of an in-depth learning architecture. On the other hand, a deep learning architecture is designed, and a cyclic neural network and a convolutional neural network are combined to learn hidden information from road traffic condition prediction characteristics, so that the prediction performance is further improved, and further, the upcoming traffic jam can be conveniently subjected to pre-treatment and dispersion, and the upcoming traffic jam road can be avoided. The method can accurately predict the traffic state and has high accuracy.
In more detail, the invention mainly comprises the following main steps:
the first step, data acquisition step: the method comprises the steps of acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices, wherein the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the coordinates and the time stamp of the point. The data mainly originates from a GPS device on the taxi.
And secondly, preprocessing data, namely performing data cleaning on the acquired traffic data, and clustering and grouping the traffic data after the data cleaning based on flow similarity.
Data cleaning is firstly carried out through the following four steps:
1. and (3) exception data processing: if the speed of the GPS recording point is more than 50% faster than the average speed of all the GPS recording points in the same period, the data is abnormal data, and the average speed in the same period is used to replace the abnormal data.
2. Invalid data deletion processing: if a certain traffic data has not recorded any data for three hours in a day, the data of the whole day is deleted.
3. And (3) data filling processing: if the data in a certain time stamp is lost, taking the average value of the data before and after the time stamp as a filling value, and performing data filling processing on the time stamp.
4. And (3) data smoothing treatment: and calculating the average value of the speeds of the GPS recording points on every three time stamps, and smoothing the data.
Second, the rest of the trajectory data is grouped according to traffic similarity to obtain the pattern of each road. The present application defines a section as a road. Since one road has different traffic conditions every day. Thus, a clustering method is used to generate one pattern for each portion of the road at each time period of each day. The steps of the clustering method are as follows:
1. each piece of traffic data is segmented according to a preset period, for example, the preset period is 5 minutes, and the traffic data is segmented into a group of line segments according to an interval of 5 minutes.
2. The track segments that are close to each other according to a distance metric are grouped into a cluster. In this step, the distance between two traces is measured using a euclidean metric. Definition of
Is traffic data P1…nAnd Q1…nThe sum of the distances of the points at each sampling time t, then,
wherein the euclidean metric between two points is defined as:
for this clustering step, the DBSCAN algorithm is used because it is based on density clustering and allows the clustering trajectories to form any shape and size.
3. And for each cluster, constructing a representative track of the cluster as a mode of the cluster, wherein the representative track is an average value of the length and the angle of the whole track segment belonging to the cluster, namely the mode represents the cluster characteristics including the traffic condition in a certain time, namely the road condition.
And thirdly, constructing a feature vector. In a road network of a modern city, the traffic condition of one road is always influenced by other roads in the vicinity. When the traffic of the nearby road is busy, the traffic of the road is relatively stable. Therefore, when predicting the traffic conditions of the roads, it is necessary to consider not only the historical traffic conditions of the roads but also the traffic conditions of the roads in the vicinity thereof. Therefore, a set of links is to be selected to provide contextual information of the surroundings for the link for which the traffic prediction is to be made.
If a road is taken as a center point, a common track distance metric derived based on Minimum Bounding Rectangles (MBRs) can be used to capture the overall similarity of the representative track of the road and the representative tracks around the representative track in a certain range. Firstly, B is caused to be1And B2Are respectively a pattern P1And P2The MBR of (a). Distance Dmin(B1,B2) Is represented by B1And B2The minimum distance between any pair of points. Then, according to Dmin(B1,B2) All the modes are sorted, and K adjacent modes of the road in the first K modes closest to the road are selected.
Next, the relevant feature vectors are constructed using the neighboring patterns and the historical traffic conditions for this road. Table 1 below shows a sample of feature vectors if K is set to 5 and it is desired to predict road traffic conditions 15 minutes after time t, the flag indicating the location of the feature in terms of context information. "C" indicates that the function contains information of the link itself, and different subscripts of "C" are used to distinguish different link information. Please note that, since obtaining real-time traffic consumes very much resources, in the method of the present application, the traffic condition of r is not considered, r represents a section of road, r is dynamic, and may be 100 meters, or 300 meters, or 1 km, and r is a section of road of 3 km in the test data of the present embodiment. "L" or "R" indicates a feature on the left or right side of the road context information depending on the geographical position where the adjacent route is located, and subscripts of "L" and "R" are used to distinguish different context information. Fig. 1 shows a case of a road sample, showing a road represented by each row. If a circle is drawn with "AB" as the center point and K is set to 5, then: "BH", "BG" and "AE" can be considered as characteristic of the "L" designation, while "AC" and "CD" can be considered as characteristic of the "R" designationSo that they are all within a circle. However, although a portion of "EF" is also within a circle, the distance from "AB" is not at DminThe first 5 bits.
Table 1: feature vector samples
| Sign (sign) | Feature(s) |
| C1 | Traffic condition at t on the day of the last week |
| C2 | Traffic condition of r at t +15 on the same day of the last week |
| C3 | Traffic condition at t yesterday |
| C4 | R traffic condition at t +15 yesterday |
| C5 | Sign indicating that one day is public holiday/weekend |
| C6 | A value representing the day's working day |
| L1 | Traffic condition of |
| L2 | Traffic condition of neighboring mode 2 at t +15 |
| L3 | Adjacent mode 3 traffic conditions at t +15 |
| R1 | Traffic condition at t +15 for neighboring mode 4 |
| R2 | Traffic condition of |
It should be noted that the present embodiment supports more information to construct the feature vector, for example, more historical traffic conditions of adjacent roads, but considering the calculation cost, the present embodiment preferably uses six landmark features "C" and K adjacent modes to construct the feature vector.
And fourthly, constructing a deep learning framework. In the foregoing step, the feature vector is constructed based on the common trajectory data only. Therefore, it is necessary to apply a deep learning technique to further extract deep features of the feature vector, thereby improving the prediction performance. Will be provided with
And
defined as context information about the road R. In the present embodiment, it is preferred that,
and
is in addition to
And
the traffic conditions of the corresponding neighbor mode of the out-conversion are used as the value of the traffic conditions for direct use. R is also a carrier in which the characteristics of the road itself are contained. Calculated using the following equation
And
where f is a non-linear activation function, W(l)And W(r)Is a matrix that converts the hidden layer (context) to the next hidden layer.
Fig. 2 shows a schematic diagram of the structure of the cyclic convolutional neural network of the present invention, for a cyclic structure, a context is first captured using a directional cyclic neural network. Taking the example of the above-mentioned table 1 as an example,
is context information of the proximity pattern, located on the left side of the nearest road.
And
is context information of a proximity pattern having a second close distance and a third closest distance on the left side of the road. In a similar manner to that described above,
and
contextual information being a proximity pattern with a first close distance and a second close distance to the right of the roadAnd (4) information. Based on the above equation, the context vector captures all left and right context information. Next, the definition of x, represented by R, is shown below, where x is all left context information
All right context information
And the embedded characteristic information of R. Using this context information, the present recurrent neural network can learn the hidden information of R better than a conventional neural model using only fixed windows:
in a forward scan of the road context information, the loop structure may obtain all of the roads in a left-to-right scan
And all of the roads in a right-to-left scan
In this process, the temporal complexity is O (n). After x is obtained, it is subjected to linear transformation and tanh activation function calculation, and the calculation result y is sent to the next layer, where y is the output of a matrix containing various information, also called a cyclic convolution layer, and the formula of y is as follows:
y＝tanh(WX+b)
here, the convolutional neural network in the architecture proposed in the present application is defined as a representative road, and thus, from the perspective of the convolutional neural network, the aforementioned cyclic structure is a convolutional layer. An average pooling layer is then applied to consolidate the features learned and expressed in the previous layer using the following equation:
y*＝average(y)
the average function is an element function, the average value of each element of y is used as the input value of the next layer to promote feature representation, and overfitting to training data is reduced through a model. Maximum pooling is not used here in the present application, since average pooling is more suitable for capturing information in the case of only one convolutional layer. Pooling utilizes the output of the loop structure as input, with a temporal complexity of O (n). The overall model is a concatenation of the cyclic structure and the average pooling layer, so the overall time complexity is still O (n).
An exit layer and a fully connected layer are added to reduce overfitting to create a combination with features and activation functions for later prediction over the network. The last part of the architecture of the present application is the output layer. It is defined as y, similar to a conventional neural network**＝Wy*+b。
In verification, a prediction probability can be calculated for the output of the output layer by using a Softmax function, namely, the prediction accuracy rate is verified.
Fifthly, training and optimizing: based on a momentum random gradient descent algorithm, training and optimizing a cyclic convolution neural network:
if all the parameters to be trained are defined as θ, then the training objectives of the network are used to maximize the log-likelihood value of θ:
wherein
Representing a training set, D representing a training set
Training sample of (1), classDIs the correct category of road traffic conditions, and optimizes the training target by using a momentum random gradient descent algorithm. In each step, an example (D, class) is randomly selectedD) And performing gradient descent processing:
where ∈ is a learning rate, and 100 memory units are provided as a loop layer and 256 hidden units for the fully-connected layer in consideration of actual computing resources that the server can afford. And 10,000 training iterations were performed. After several adjustments, the initial value of ∞ is set to 0.2, and the decrement per 1000 iterations is performed by γ ═ 0.1. In addition, the momentum factor in the momentum random gradient descent algorithm is set to be 0.9, the exit rate is set to be 0.2, the pooling length is set to be 2, and after iterative training, the loss values of the training set and the verification set are basically close to 0, so that the method is high in prediction accuracy.
In addition, after the cyclic convolution neural network is constructed, the training and verification steps of specific details include ways of selecting one part of input data as a training set for training, and the other part of the input data as a verification set for verification, and the like, which are similar to other neural network training steps and are not repeated in the invention.
Road traffic condition predictions may provide advice to drivers, so it is desirable to ensure that the prediction results are as close as possible to the actual road traffic conditions. The method was verified using a data set containing 44580 trace data samples for 30 roads in 30 road segments of beijing, 5 months in 2013. And in the training process, the network is trained by adopting 100 periods, and the final accuracy prediction is carried out by using a Softmax function. Finally, 10-fold cross-validation was used for evaluation. Finally, the prediction accuracy of the method obtained by calculation can reach nearly 90%, and is improved by 5% compared with the traditional convolutional neural network. This is because the convolutional layer can select more information that distinguishes the feature context by averaging pooling and capturing.
In addition, in order to verify the capability of the method for capturing context information, the convolutional neural network and the method are respectively tested by setting K to be 1, 5, 10, 15 and 20, the test result is shown in FIG. 3, it can be seen that the method is superior to the traditional convolutional neural network in all K values, in addition, when K is 10, the two models achieve the best performance, and the performance change of the method along with the change of K is not as obvious as that of the convolutional neural network, therefore, the method does not depend on K very much, the working stability is better, because the loop structure can retain longer context information and reduce noise.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (7)
1. A road traffic condition prediction method is characterized by comprising the following steps:
a data acquisition step: acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices;
a data preprocessing step: after data cleaning is carried out on the acquired traffic data, clustering and grouping the traffic data after the data cleaning based on flow similarity;
a feature vector construction step: for each road, constructing a feature vector of the road based on the distance measurement of the public track and the historical traffic condition of the road; the public track distance measurement comprises a track measurement method derived based on a minimum circumscribed rectangle and used for calculating the overall similarity of the representative track after the current road is clustered and the representative track after the surrounding roads of the current road are clustered in a preset range;
and a deep learning framework construction step: constructing a cyclic convolution neural network based on a deep learning technology, and performing deep feature extraction on the feature vectors;
training and optimizing: training and optimizing a cyclic convolution neural network based on a momentum random gradient descent algorithm;
acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network;
the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point.
2. The method according to claim 1, wherein in the data preprocessing step, the step of performing data cleansing on the acquired traffic data comprises:
and sequentially performing abnormal data processing, invalid data deleting processing, data filling processing and data smoothing processing on the acquired traffic data.
3. The method of claim 1, wherein in the step of preprocessing the data, the step of clustering and grouping the traffic data after data cleaning based on the traffic similarity comprises:
segmenting each traffic data according to a preset period;
clustering the segmentation result based on Euclidean metric;
for each cluster, a representative trajectory for the cluster is constructed as a pattern for the cluster.
4. The method according to claim 1, wherein the feature vector constructing step is specifically: and for each road to be analyzed, acquiring a representative track of the road based on the public track distance measurement, acquiring the first K representative tracks closest to the representative track as K adjacent modes of the road, and further constructing the feature vector of the road according to the K adjacent modes and the historical traffic condition of the road.
5. The method of claim 1, wherein the deep learning framework construction step comprises:
calculating and obtaining left and right context information of adjacent modes of each road as a training set, and associating the left and right context information with the traffic condition corresponding to the road;
after a cyclic convolution neural network is constructed, the obtained left and right context information is input into a cyclic convolution layer of the neural network;
after the output of the circulating convolution layer is subjected to linear transformation and tanh activation function calculation, the average value calculation is carried out on the calculation result;
and taking the average value obtained by calculation as the input of the exit layer, and finally obtaining the output of the output layer.
6. The method according to claim 1, wherein the training optimization step is specifically:
based on a momentum random gradient descent algorithm, gradient descent calculation is carried out on the training parameters of the cyclic convolution neural network, and the log likelihood value of the training parameters is maximized, so that the output of the cyclic convolution neural network is closest to the actual traffic condition of the road.
7. A road traffic condition prediction system comprising a processor and a memory device, said memory device storing a plurality of instructions, said instructions being loaded by said processor and performing the steps of:
a data acquisition step: acquiring traffic data recorded by a plurality of vehicle-mounted GPS devices;
a data preprocessing step: after data cleaning is carried out on the acquired traffic data, clustering and grouping the traffic data after the data cleaning based on flow similarity;
a feature vector construction step: for each road, constructing a feature vector of the road based on the distance measurement of the public track and the historical traffic condition of the road; the public track distance measurement comprises a track measurement method derived based on a minimum circumscribed rectangle and used for calculating the overall similarity of the representative track after the current road is clustered and the representative track after the surrounding roads of the current road are clustered in a preset range;
and a deep learning framework construction step: constructing a cyclic convolution neural network based on a deep learning technology, and performing deep feature extraction on the feature vectors;
training and optimizing: training and optimizing a cyclic convolution neural network based on a momentum random gradient descent algorithm;
acquiring traffic data recorded by vehicle-mounted GPS equipment of a vehicle running on each road to be analyzed, and predicting the road traffic condition of the road by adopting a trained cyclic convolution neural network;
the traffic data comprises a plurality of GPS recording points, and each GPS recording point comprises the current speed, the current coordinate and the current time stamp of the point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710326733.8A CN107103754B (en) | 2017-05-10 | 2017-05-10 | Road traffic condition prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710326733.8A CN107103754B (en) | 2017-05-10 | 2017-05-10 | Road traffic condition prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107103754A CN107103754A (en) | 2017-08-29 |
| CN107103754B true CN107103754B (en) | 2020-05-22 |
Family
ID=59669105
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710326733.8A Active CN107103754B (en) | 2017-05-10 | 2017-05-10 | Road traffic condition prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107103754B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4016412A1 (en) * | 2020-12-18 | 2022-06-22 | Imec VZW | Traffic prediction |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10733877B2 (en) * | 2017-11-30 | 2020-08-04 | Volkswagen Ag | System and method for predicting and maximizing traffic flow |
| CN108205889B (en) * | 2017-12-29 | 2021-04-27 | 长春理工大学 | Method for predicting highway traffic flow based on convolutional neural network |
| CN108986453A (en) * | 2018-06-15 | 2018-12-11 | 华南师范大学 | A kind of traffic movement prediction method based on contextual information, system and device |
| CN108648457B (en) * | 2018-06-28 | 2021-07-13 | 苏州大学 | Method, device and computer readable storage medium for speed prediction |
| CN108830430A (en) * | 2018-08-01 | 2018-11-16 | 苏州大学张家港工业技术研究院 | Convolutional neural networks multiple spot regressive prediction model for traffic flow forecasting |
| CN109068174B (en) * | 2018-09-12 | 2019-12-27 | 上海交通大学 | Video frame rate up-conversion method and system based on cyclic convolution neural network |
| CN109064748B (en) * | 2018-09-18 | 2020-12-25 | 浙江工业大学 | Traffic average speed prediction method based on time cluster analysis and variable convolutional neural network |
| CN109767622B (en) * | 2019-02-14 | 2021-12-17 | 重庆邮电大学 | Traffic flow prediction method for bayonet similarity division and recurrent neural network |
| CN109887279B (en) * | 2019-02-26 | 2021-01-05 | 北京隆普智能科技有限公司 | Traffic jam prediction method and system |
| CN110232820A (en) * | 2019-05-20 | 2019-09-13 | 北京世纪高通科技有限公司 | A kind of method for building up and device of road condition predicting model |
| CN110309304A (en) * | 2019-06-04 | 2019-10-08 | 平安科技（深圳）有限公司 | A kind of file classification method, device, equipment and storage medium |
| CN110570658B (en) * | 2019-10-23 | 2022-02-01 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
| CN110942637B (en) * | 2019-12-17 | 2021-04-06 | 浙江工业大学 | SCATS system road traffic flow prediction method based on airspace map convolutional neural network |
| CN111524348A (en) * | 2020-04-14 | 2020-08-11 | 长安大学 | Long-short term traffic flow prediction model and method |
| CN111986490A (en) * | 2020-09-18 | 2020-11-24 | 北京百度网讯科技有限公司 | Road condition prediction method and device, electronic equipment and storage medium |
| CN112330953A (en) * | 2020-10-10 | 2021-02-05 | 杭州翔毅科技有限公司 | Traffic control method, device and equipment based on cloud data and storage medium |
| CN113379754A (en) * | 2020-12-02 | 2021-09-10 | 哈尔滨理工大学 | Road center line extraction method based on vehicle-mounted GPS data and neural network |
| CN113450562B (en) * | 2021-05-08 | 2022-05-24 | 浙江工业大学 | Road network traffic state discrimination method based on clustering and graph convolution network |
| CN114764977B (en) * | 2022-04-06 | 2023-07-28 | 成都亿盟恒信科技有限公司 | Road congestion calculation method and system based on driving data |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104572892B (en) * | 2014-12-24 | 2017-10-03 | 中国科学院自动化研究所 | A kind of file classification method based on cyclic convolution network |
| CN106250103A (en) * | 2016-08-04 | 2016-12-21 | 东南大学 | A kind of convolutional neural networks cyclic convolution calculates the system of data reusing |
| CN106205126B (en) * | 2016-08-12 | 2019-01-15 | 北京航空航天大学 | Large-scale Traffic Network congestion prediction technique and device based on convolutional neural networks |
| CN106503055B (en) * | 2016-09-27 | 2019-06-04 | 天津大学 | A kind of generation method from structured text to iamge description |
| CN106611162B (en) * | 2016-12-20 | 2019-06-18 | 西安电子科技大学 | Road vehicle real-time detection method based on deep learning SSD frame |
-
2017
- 2017-05-10 CN CN201710326733.8A patent/CN107103754B/en active Active
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4016412A1 (en) * | 2020-12-18 | 2022-06-22 | Imec VZW | Traffic prediction |
| WO2022129421A1 (en) * | 2020-12-18 | 2022-06-23 | Imec Vzw | Traffic prediction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107103754A (en) | 2017-08-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107103754B (en) | Road traffic condition prediction method and system | |
| CN110991311B (en) | Target detection method based on dense connection deep network | |
| CN109978893B (en) | Training method, device, equipment and storage medium of image semantic segmentation network | |
| CN109344736B (en) | Static image crowd counting method based on joint learning | |
| CN110969250B (en) | Neural network training method and device | |
| CN107038478B (en) | Road condition prediction method and device, computer equipment and readable medium | |
| CN110738247B (en) | Fine-grained image classification method based on selective sparse sampling | |
| CN104462190B (en) | A kind of online position predicting method excavated based on magnanimity space tracking | |
| CN107247961B (en) | Track prediction method applying fuzzy track sequence | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| US20170200274A1 (en) | Human-Shape Image Segmentation Method | |
| CN106714110A (en) | Auto building method and system of Wi-Fi position fingerprint map | |
| CN106919957B (en) | Method and device for processing data | |
| CN108986453A (en) | A kind of traffic movement prediction method based on contextual information, system and device | |
| CN109671102A (en) | A kind of composite type method for tracking target based on depth characteristic fusion convolutional neural networks | |
| Markos et al. | Unsupervised deep learning for GPS-based transportation mode identification | |
| CN110443287B (en) | Crowd moving stream drawing method based on sparse trajectory data | |
| CN110096979B (en) | Model construction method, crowd density estimation method, device, equipment and medium | |
| CN112800915A (en) | Building change detection method, building change detection device, electronic device, and storage medium | |
| CN103136757A (en) | SAR image segmentation method based on manifold distance two-stage clustering algorithm | |
| CN111916144A (en) | Protein classification method based on self-attention neural network and coarsening algorithm | |
| CN110751027A (en) | Pedestrian re-identification method based on deep multi-instance learning | |
| CN111931991A (en) | Weather nowcast method, weather nowcast device, computer equipment and storage medium | |
| CN110659601A (en) | Depth full convolution network remote sensing image dense vehicle detection method based on central point | |
| CN112329830B (en) | Passive positioning track data identification method and system based on convolutional neural network and transfer 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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200820 Address after: 518000 Room 408, Huatong business building, No. 3, Xili Guan Long Village, Nanshan District, Shenzhen, Guangdong Patentee after: SHENZHEN LINGHANGZHE AUTOMOBILE INTELLIGENT TECHNOLOGY DEVELOPMENT Co.,Ltd. Address before: 510631 School of computer science, South China Normal University, 55 West Zhongshan Road, Guangzhou, Guangdong, Tianhe District Patentee before: SOUTH CHINA NORMAL University |
|
| TR01 | Transfer of patent right |