WO2022142418A1

WO2022142418A1 - Traffic performance index prediction method and device based on gis map information

Info

Publication number: WO2022142418A1
Application number: PCT/CN2021/114906
Authority: WO
Inventors: 邢玲; 余意
Original assignee: 深圳云天励飞技术股份有限公司
Priority date: 2020-12-31
Filing date: 2021-08-27
Publication date: 2022-07-07
Also published as: CN112633602B; CN112633602A

Abstract

A traffic performance index prediction method and device based on GIS map information. The method comprises the following steps: obtaining a GIS map structured vector graphic sequence and an environment vector sequence of a region to be predicted, each frame of a GIS map structured vector graphic comprising a first preset number of static graphic layers and a second preset number of dynamic graphic layers, and one frame of the GIS map structured vector graphic corresponding to one environment vector sequence; and inputting the GIS map structured vector graphic sequence and the environment vector sequence into a preset convolutional neural network, separately extracting the temporal and spatial features of the GIS map structured vector graphic sequence and the mapping feature of the environment vector sequence, and outputting a traffic performance index result of said region within a preset time period according to the temporal and spatial features and the mapping feature.

Description

A traffic congestion index prediction method and device based on GIS map information

This application claims the priority of the Chinese patent application filed on December 31, 2020, with the application number of 202011640824.7 and the invention titled "A GIS Map Information-based Traffic Congestion Index Prediction Method and Device", the entire contents of which are Incorporated herein by reference.

technical field

The invention relates to the technical field of traffic congestion prediction, in particular to a traffic congestion index prediction method and device based on GIS map information.

Background technique

With the development of the urban economy, the number of urban vehicles has increased sharply, and the subsequent road traffic congestion has become a major problem in urban management. In order to quantify the situation of road traffic congestion, scholars proposed an index to use the data of floating vehicles on the road to determine the level of road congestion, and named it the traffic congestion index. The value range is 0-10. The larger the value, the more serious the congestion is . Through the accurate and real-time prediction of the road traffic congestion index, the traffic management department can allocate the traffic flow, ease the congestion in advance, and improve the traffic capacity of the road network.

Existing traffic congestion index prediction methods have several flaws:

First, the traditional traffic congestion index forecast uses a time series prediction method to predict the traffic congestion index for a period of time in the future based on historical traffic conditions. This method can often only learn the time-varying laws of a few road segments, while ignoring the topology of complex traffic roads, so it cannot model the influence of adjacent road segments/regions. At the same time, traditional models have limited expressive ability and cannot model the impact of weather, holidays, POI information, etc. on traffic. With the maturity of deep learning technology, some scholars have proposed a spatiotemporal neural network model, which can simultaneously model the spatial correlation and temporal regularity of road traffic. . However, the existing spatiotemporal neural network models are often based on static road connection relationships (ie, maps), which cannot well describe the dynamic relationship of traffic flow between different roads.

Second, the traditional traffic congestion index prediction model does not take into account the spatial information of the road, at most only the connection relationship between the roads, and the spatial information of the road, such as whether there are subway stations, bus stations, high-speed railway stations, schools, etc. And the width of the road is a very important factor affecting traffic congestion.

It can be seen that the existing traffic congestion index prediction method has the problem of low prediction accuracy of the traffic congestion index.

SUMMARY OF THE INVENTION

The embodiment of the present invention provides a traffic congestion index prediction method based on GIS map information, which can solve the problem of low prediction accuracy of the traffic congestion index in the existing traffic congestion index prediction method.

In a first aspect, an embodiment of the present invention provides a traffic congestion index prediction method based on GIS map information, and the GIS map information-based traffic congestion index prediction method includes:

Obtain a GIS map structured vector map sequence and an environmental vector sequence of the area to be predicted. Each frame of GIS map structured vector map includes a first preset number of static layers and a second preset number of dynamic layers, and one frame of GIS map includes a first preset number of static layers and a second preset number of dynamic layers. The map structured vector map corresponds to an environment vector sequence;

Inputting the GIS map structured vector diagram sequence and the environment vector sequence into a preset convolutional neural network, extracting the spatiotemporal features of the GIS map structured vector graphics sequence, the mapping features of the environment vector sequence, and According to the spatiotemporal feature and the mapping feature, a traffic congestion index result of the to-be-predicted area within a preset time is output.

Optionally, the dynamic layer includes traffic dynamic data, and the step of obtaining a GIS map structured vector map sequence of the area to be predicted includes:

Obtain the static layer and traffic dynamic data of the GIS map of the area to be predicted at historical time points. The static layer includes one or more of a road network map, a building map and a point of interest map. The traffic dynamic data The data includes traffic congestion index and vehicle trajectory data corresponding to the historical time point;

At the historical time point, the traffic congestion index is mapped to the road network map to obtain a traffic congestion index map;

The vehicle trajectory data is mapped into a two-dimensional network, and the average vehicle speed and the number of vehicles in each network are calculated to obtain a traffic flow speed map and a vehicle number map;

Splicing the road network map, building map, point of interest map, traffic congestion index map, traffic speed map and vehicle quantity map as channels to obtain a GIS map structured vector map corresponding to the historical time point;

The GIS map structured vector graphics corresponding to the continuous historical time points are spliced according to the time sequence of the continuous historical time points to obtain the GIS map structured vector graphics sequence of the to-be-predicted area.

Optionally, the step of obtaining the environment vector sequence of the area to be predicted includes:

Obtain the weather data and date data of the area to be predicted at the historical time point;

encoding the weather data and date data into an environment vector according to a preset encoding rule;

The environment vectors corresponding to the continuous historical time points are spliced according to the time sequence of the continuous historical time points to obtain the environment vector sequence of the to-be-predicted area.

Optionally, the preset convolutional neural network includes a three-dimensional convolutional network, a feature mapping network, a fusion network and a fully connected layer network;

Inputting the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, respectively extracting the spatiotemporal features of the GIS map structured vector graphics sequence, and the mapping features of the environmental vector sequence , and the step of outputting the result of the traffic congestion index of the to-be-predicted area within a preset time according to the spatiotemporal feature and the mapping feature includes:

The 3D convolution calculation is performed on the GIS map structured vector diagram sequence through the 3D convolution network to obtain the spatiotemporal characteristics of the GIS map structured vector diagram sequence;

Perform a mapping calculation on the environment vector sequence through the feature mapping network, and output the mapping feature of the environment vector sequence;

Perform fusion calculation on the spatiotemporal feature and the mapping feature through the fusion network, and output the fusion feature;

The first full-connection calculation is performed on the fusion feature through the fully-connected layer network, and the result of the traffic congestion index of the to-be-predicted area within a preset time is output.

Optionally, the three-dimensional convolution network is constructed according to a residual network, and the three-dimensional convolution calculation is performed on the GIS map structured vector diagram sequence through the three-dimensional convolution network to obtain the GIS map structured vector Steps of spatiotemporal features of graph sequences, including:

In the process of three-dimensional convolution calculation, the convolution calculation is carried out in combination with the residual of the previous convolution calculation layer, and the spatiotemporal feature map is obtained;

A second full connection calculation is performed on the spatiotemporal feature map to obtain the spatiotemporal feature of the GIS map structured vector map sequence.

Optionally, the fusion network is a gated network, and the step of performing fusion calculation on the spatiotemporal feature and the mapping feature through the fusion network, and outputting the fusion feature, includes:

Through the gating network, non-linear processing is performed on the mapping feature to obtain the gating feature;

Calculate the dot product of the gated feature and the spatiotemporal feature to obtain a fusion feature.

Optionally, the method further includes:

Obtain a training sample set, the training sample set includes a plurality of training samples, each training sample includes a sample GIS map structured vector map sequence, a sample environment vector sequence corresponding to the GIS map structured vector map sequence, and The real label of the traffic congestion index at the predicted time point, the sample GIS map structured vector map sequence and the GIS map structured vector map sequence have the same data structure, the sample environment vector sequence and the environment vector sequence have the same data structure data structure;

Through the training sample set, the convolutional neural network is trained, so that the convolutional neural network learns the predicted output of the traffic congestion index at the predicted time point, and the preset convolutional neural network is obtained.

In a second aspect, an embodiment of the present invention also provides a device for predicting a traffic congestion index based on GIS map information, including:

The first acquisition module is used to acquire the GIS map structured vector graphics sequence and the environmental vector sequence of the area to be predicted, and each frame of the GIS map structured vector graphics includes a first preset number of static layers and a second preset number of dynamic layers. layer, and a frame of GIS map structured vector graphics corresponds to an environmental vector sequence;

The prediction module is used to input the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, and extract the spatiotemporal features of the GIS map structured vector diagram sequence, and the environmental vector sequence and output the traffic congestion index result of the to-be-predicted area within a preset time according to the spatiotemporal feature and the mapping feature.

In a third aspect, an embodiment of the present invention further provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program At the same time, the steps in the method for predicting the traffic congestion index based on the GIS map information provided in the above embodiment are implemented.

In a fourth aspect, embodiments of the present invention further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, implements the GIS-based map provided in the foregoing embodiments Information on the steps in the traffic congestion index prediction method.

In the embodiment of the present invention, by obtaining the GIS map structured vector map sequence and the environment vector sequence of the area to be predicted, each frame of the GIS map structured vector map includes a first preset number of static layers and a second preset number of static layers. A dynamic layer, and a frame of GIS map structured vector graphics corresponds to an environmental vector sequence; input the GIS map structured vector graphics sequence and the environmental vector sequence into a preset convolutional neural network, and extract the GIS map respectively The spatiotemporal feature of the structured vector diagram sequence, the mapping feature of the environment vector sequence, and the traffic congestion index result of the to-be-predicted area within a preset time is output according to the spatiotemporal feature and the mapping feature. In this way, multi-dimensional data such as the first preset number of static layers and the second preset number of dynamic layers and the environmental vector sequence in the GIS map structured vector map sequence of the area to be predicted can be input into the preset convolutional neural network. In the network, the temporal and spatial characteristics of the GIS map structured vector diagram sequence and the mapping characteristics of the environmental vector sequence are extracted to predict the traffic congestion index of the area to be predicted within a preset time, thereby improving the prediction accuracy of the traffic congestion index. .

Description of drawings

1 is a flowchart of a method for predicting a traffic congestion index based on GIS map information provided by an embodiment of the present invention;

FIG. 2 is a flowchart of a method provided in step 101 in an embodiment of the present invention;

3 is a flowchart of another method provided in step 101 in an embodiment of the present invention;

4 is a flowchart of a method provided in step 102 in an embodiment of the present invention;

5 is a flowchart of another traffic congestion index prediction method based on GIS map information provided by an embodiment of the present invention;

6 is a schematic structural diagram of a device for predicting traffic congestion index based on GIS map information provided by an embodiment of the present invention;

7 is a schematic structural diagram provided by a first acquisition module in an embodiment of the present invention;

8 is another schematic structural diagram provided by a first acquisition module in an embodiment of the present invention;

9 is a schematic structural diagram provided by a prediction module in an embodiment of the present invention;

10 is a schematic structural diagram of another device for predicting traffic congestion index based on GIS map information provided by an embodiment of the present invention;

FIG. 11 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

Please refer to FIG. 1. FIG. 1 is a flowchart of a method for predicting traffic congestion index based on GIS map information provided by an embodiment of the present invention. As shown in FIG. 1, the method for predicting traffic congestion index based on GIS map information includes the following steps :

Step 101: Obtain a GIS map structured vector diagram sequence and an environment vector sequence of the area to be predicted.

The GIS map structured vector graphics sequence includes multi-frame GIS map structured vector graphics, and each frame of GIS (Geographic Information System, geographic information system) map structured vector graphics includes a first preset number of static layers and a second preset number of static layers. Set a number of dynamic layers, and each frame of GIS map structured vector graphics corresponds to an environmental vector sequence.

The above-mentioned first preset number of static layers may include three static layers, and each static layer includes one static channel. The three static layers correspond to three static channels, which can be called RBP channels, where R represents the road channel, B represents the building channel, and P represents the POI (Point of interest) channel. Of course, for the first preset number of static layers, different numbers and different functions can be set according to actual needs, and the number and functions of the static layers are not limited here. Correspondingly, the number of static channels is also not limited.

The above-mentioned second preset number of dynamic layers may include three dynamic layers, and each dynamic layer includes one dynamic channel. The three dynamic layers correspond to three dynamic channels, which can be called TSC channels, where T represents the TTI (Travel Time Index, traffic congestion index) channel, S represents the speed (speed) channel, and C represents the count (quantity) channel. Of course, for the second preset number of dynamic layers, different numbers of dynamic layers with different functions can be set according to actual needs, and the number and functions of the dynamic layers are not limited here. Of course, the number of dynamic channels is not limited.

For example, each frame of the GIS map structured vector diagram above includes a six-channel RBPTSC layer, wherein the N*N*6 three-dimensional vector of the six-channel RBPTSC layer represents spatial information, plus the number of time points in a certain time period, For example, a slice of 6 time points forms a 6*N*N*6 four-dimensional vector to represent the spatiotemporal information. Each frame of GIS map structured vector graphics contains its own spatiotemporal information. Among them, N is an integer greater than or equal to 1, and the value of N can be set according to actual needs.

The above-mentioned area to be predicted is a target area where the user needs to predict the traffic congestion index, which may be a certain intersection, a certain road, a certain area, and the like. It can also be multiple intersections, multiple roads, multiple areas, and so on.

Specifically, as shown in FIG. 2 , FIG. 2 is a flowchart of a method provided in step 101 in an embodiment of the present invention. The dynamic layer includes traffic dynamic data, and the above-mentioned steps of obtaining a GIS map structured vector map sequence of the area to be predicted include:

Step 201: Obtain the static layer and traffic dynamic data of the GIS map of the area to be predicted at historical time points.

The static layer includes one or more of a road network map, a building map, and a point-of-interest map, and the traffic dynamic data includes traffic congestion index and vehicle trajectory data corresponding to historical time points. Each area to be predicted has its own road network map, building map and point of interest map.

Step 202 , at a historical time point, map the traffic congestion index to a road network map to obtain a traffic congestion index map.

Step 203: Map the vehicle trajectory data into a two-dimensional network, calculate the average speed of vehicles and the number of vehicles in each network, and obtain a vehicle flow speed map and a vehicle number map.

Step 204 , splicing the road network map, the building map, the point of interest map, the traffic congestion index map, the traffic speed map, and the vehicle number map as channels to obtain a GIS map structured vector map corresponding to a historical time point.

Step 205 , splicing the GIS map structured vector graphics corresponding to the consecutive historical time points according to the time sequence of the consecutive historical time points to obtain a GIS map structured vector graphics sequence of the area to be predicted.

The above historical time point can be a certain time point taken in a certain time unit in a certain time period in the past, for example, a certain time point taken in a unit of 10 minutes in the past hour, so that the unit of 10 minutes in the past hour can be used. Take 6 time points (that is, 6 data points), then the historical time point can be one of the 6 time points. Of course, a certain time period in the past can be set as a certain time point in the past two hours, or in units of 10 minutes, that is, 12 time points (ie 12 data points) are obtained, etc. Then the historical time point can be one of the 12 time points. In the embodiment of the present invention, the description is mainly based on a unit of 10 minutes. Of course, the unit time can be set according to actual requirements.

The above-mentioned point-of-interest map may include maps corresponding to some points of interest such as bus stops, subway stations, hotels, schools, hospitals, high-speed railway stations, bus stations, parks, and parking lots. The above point of interest map can be referred to as a POI layer.

The above-mentioned vehicle trajectory data may include vehicle speed, vehicle position (GPS positioning position), and the like. The above-mentioned vehicles may include online car-hailing, taxis, private cars, and the like.

More specifically, after acquiring the road network map, building map, and point-of-interest map of the GIS map of the area to be predicted at historical time points, a grid is drawn based on the GPS latitude and longitude. By constructing three N*N two-dimensional vector layers according to the GPS latitude and longitude from the three layers of the road network map, the building map and the POI map to construct the channel, that is, the corresponding road network map, the building map and the POI map are obtained. The RBP channels are the road channel, building channel, and POI channel.

At the same time, the traffic congestion data of the GIS map of the area to be predicted at historical time points are obtained. Further according to the GIS map of the road, the traffic congestion index of the corresponding road is mapped to the corresponding road network map based on the R channel, and the traffic congestion index map of the road is obtained, that is, a TTI channel is constructed and recorded as the T channel. For example, if the historical time is set to one hour (ie, 6 data points), then the traffic congestion index for the past hour can be recorded as

Among them, M represents the number of roads to be predicted. Then according to the GIS map of the road, using the R channel in the constructed RBP channel, the

The traffic congestion index TTI of the corresponding road is mapped to the corresponding road network map, and the traffic congestion index map of the past hour can be obtained, that is, a TTI channel of the past hour can be constructed and recorded as the T channel.

It should be noted that, in the embodiment of the present invention, the recording of the traffic congestion index mainly takes 10 minutes as a unit time, that is, each road has a value of the traffic congestion index every 10 minutes. If you need to predict the future half an hour (i.e., one hour is divided into 6 time points (that is, 6 data points), that is, 6 traffic congestion indexes are obtained) based on the past hour (a traffic congestion index is obtained every 10 minutes in an hour). 3 data points), the traffic congestion index for the next half hour can be recorded as

M represents the number of roads to be predicted. However, the traffic congestion index for the next half hour is not mapped, but is regarded as the real value of the traffic congestion index for the next half hour.

Further, the vehicle speed and vehicle position (GPS positioning position) in the vehicle trajectory data of the GIS map of the area to be predicted at historical time points are acquired. And according to the GPS latitude and longitude information, the vehicle speed and vehicle position are mapped to the two-dimensional grid, and the average vehicle speed and the total number of vehicles in each two-dimensional grid are counted, and the traffic speed map and the number of vehicles are obtained, that is, the traffic flow is constructed. The traffic speed channel and the vehicle number channel corresponding to the speed map and the number of vehicles map are denoted as the SC channel, where S represents the speed channel and C represents the count channel.

In this way, a GIS map structured vector map sequence corresponding to the area to be predicted can be obtained based on multi-dimensional maps such as road network maps, building maps, points of interest maps, traffic congestion index maps, traffic speed maps, and vehicle quantity maps. The traffic congestion index of the region in the future period of time is predicted, thereby improving the prediction accuracy of the traffic congestion index.

Specifically, as shown in FIG. 3 , FIG. 3 is a flowchart of another method provided in step 101 of the embodiment of the present invention. The above steps of obtaining the environment vector sequence of the area to be predicted include:

Step 301: Obtain weather data and date data of the area to be predicted at a historical time point.

Step 302: Encode weather data and date data into environment vectors according to a preset encoding rule.

Step 303 , splicing the environmental vectors corresponding to the consecutive historical time points according to the time sequence of the consecutive historical time points to obtain the environmental vector sequence of the area to be predicted.

The above weather data may include three indicators of temperature, humidity, and rainfall.

The above-mentioned date data may include holiday information. For date data, a preset encoding rule, such as discrete features, can be used to encode date data. For example, 1 means that the day is a working day, and 0 means that it is a holiday. At the same time, time-related information can also be extracted as features, such as the day of the week, the number of minutes, the number of hours, whether the morning and evening peaks, etc.

Specifically, the obtained weather data and date data of the area to be predicted at historical time points are encoded into environmental vectors of each historical time point according to a preset coding rule, and then the environmental vectors of multiple consecutive historical time points are coded as continuous The sequence splicing of historical time points obtains the environment vector sequence of the area to be predicted. In this way, the actual weather data and date data can be converted into an environment vector sequence that can be recognized by a computer, so as to facilitate the analysis and processing of the environment vector sequence. In this way, multi-source heterogeneous data such as weather data and date data can be combined to predict the traffic congestion index of a certain area to be predicted in the future, and improve the prediction accuracy of the traffic congestion index.

Step 102: Input the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, respectively extract the spatiotemporal features of the GIS map structured vector graphics sequence and the mapping feature of the environmental vector sequence, and extract the spatial and temporal features of the GIS map structured vector map sequence and the environmental vector sequence according to the spatiotemporal features. Output the traffic congestion index result of the area to be predicted within the preset time with the mapping feature.

The above-mentioned preset convolutional neural network includes a three-dimensional convolutional network, a feature mapping network, a fusion network, and a fully connected layer network.

Specifically, as shown in FIG. 4 , FIG. 4 is a flowchart of a method provided in step 102 in an embodiment of the present invention. Step 102 includes the steps:

Step 401 , performing a three-dimensional convolution calculation on the GIS map structured vector graphics sequence through a three-dimensional convolutional network to obtain the spatiotemporal characteristics of the GIS map structured vector graphics sequence.

Wherein, the above three-dimensional convolutional network is constructed according to the residual network. The three-dimensional convolutional network is constructed according to the residual network, which can avoid the loss of information caused by the depth of the neural network.

Specifically, in the three-dimensional convolution calculation process, the convolution calculation is performed in combination with the residuals of the previous convolution calculation layer to obtain a spatiotemporal feature map. The second full connection calculation is performed on the spatiotemporal feature map to obtain the spatiotemporal feature of the structured vector map sequence of the GIS map.

It should be noted that the spatiotemporal feature is represented by a Z-dimensional vector, and the Z-dimensional vector can be expressed as

Taking the four-dimensional space-time vector data of 6*N*N*6 as an example, the Z-dimensional vector can be expressed as

Step 402 , perform mapping calculation on the environment vector sequence through the feature mapping network, and output the mapping feature of the environment vector sequence.

Among them, the above feature mapping network is a two-layer fully connected feedforward neural network (with Relu (Rectified Linear Unit) as a nonlinear activation function). The input to the feature map network is the weather data in the sequence of environment vectors along with the date data.

Specifically, through the feature mapping network, the weather data and date data in the environmental vector sequence are encoded and mapped to a vector space, and then the mapping features are obtained.

Step 403 , perform a fusion calculation on the spatiotemporal feature and the mapping feature through a fusion network, and output the fusion feature.

Among them, the above-mentioned fusion network is a gated network.

Specifically, through the gating network, the mapping feature is nonlinearly processed to obtain the gating feature. Calculate the dot product of the gated feature and the spatiotemporal feature to get the fusion feature.

More specifically, the input of the gating network is the output of the feature mapping network, and the mapping feature is nonlinearly processed through a sigmoid activation function through the gating network, and a gate with a value between (0, 1) is obtained. gating feature (gating parameter vector

). Combined with the Z-dimensional vector of spatiotemporal features

Compute the gating parameter vector with the gating feature

The dot product (quantitative product) between , the final gating network outputs the fusion feature F _out =Feature·V.

From this, it can be seen that the gating parameter vector

The size of the value represents the degree to which the value of the gated network is passed. The closer the value is to 1, the more it passes, and vice versa.

Step 404: Perform a first full-connection calculation on the fusion feature through a fully-connected layer network, and output the result of the traffic congestion index of the area to be predicted within a preset time.

Wherein, the above-mentioned fully connected layer network includes two layers of fully connected layer network. The input of the above fully connected layer network is the output of the gating network (fused features).

Specifically, the fusion feature is input into a two-layer fully-connected layer network to perform a fully-connected calculation, and then the result of the traffic congestion index of the area to be predicted within a preset time is output. For example, when you want to predict the traffic congestion index of each road section in the next half hour, you can actually output the traffic congestion index of each road section in the next half hour according to the fully connected layer network.

and put the

The actual traffic congestion index of each road section in the next half hour with the preset

to compare, if

and

The closer it is, the more accurate the prediction result is.

In the embodiment of the present invention, by obtaining the GIS map structured vector map sequence and the environment vector sequence of the area to be predicted, each frame of the GIS map structured vector map includes a first preset number of static layers and a second preset number of static layers. Dynamic layers, and a frame of GIS map structured vector graphics corresponds to an environmental vector sequence; input the GIS map structured vector graphics sequence and the environmental vector sequence into the preset convolutional neural network, and extract the GIS map structured vector graphics respectively The spatiotemporal features of the sequence, the mapping features of the environmental vector sequence, and the traffic congestion index results of the area to be predicted within the preset time are output according to the spatiotemporal features and the mapping features. In this way, multi-dimensional data such as the first preset number of static layers and the second preset number of dynamic layers and the environmental vector sequence in the GIS map structured vector map sequence of the area to be predicted can be input into the preset convolutional neural network. In the network, the temporal and spatial characteristics of the GIS map structured vector diagram sequence and the mapping characteristics of the environmental vector sequence are extracted to predict the traffic congestion index of the area to be predicted within the preset time, thereby improving the prediction accuracy of the traffic congestion index.

Referring to FIG. 5, FIG. 5 is a flowchart of another traffic congestion index prediction method based on GIS map information provided by an embodiment of the present invention. The traffic congestion index prediction method based on GIS map information further includes the steps:

Step 501: Obtain a training sample set.

The training sample set includes multiple training samples, and each training sample includes a sample GIS map structured vector map sequence, a sample environment vector sequence corresponding to the GIS map structured vector map sequence, and a traffic congestion index at a predicted time point true label.

The sample GIS map structured vector map sequence has the same data structure as the GIS map structured vector map sequence, and the sample environment vector sequence has the same data structure as the environment vector sequence.

Specifically, the training sample set may be a sample set that is pre-collected, processed and stored in a database for training. Of course, it can also be a training sample set collected and processed in real time at the scene of a certain prediction area.

Step 502 , train the convolutional neural network through the training sample set, so that the convolutional neural network learns the predicted output of the traffic congestion index at the predicted time point to obtain a preset convolutional neural network.

Specifically, after the training sample set is obtained, the training sample set can be input into the convolutional neural network for prediction training, so that the convolutional neural network can learn the predicted output of the traffic congestion index at the predicted time point, and then obtain Preset Convolutional Neural Networks. In this way, the preset convolutional neural network can be used to predict the traffic congestion index in a certain forecast area in a future period of time based on the GIS map information, thereby improving the accuracy of the forecast result.

In the embodiment of the present invention, when training the convolutional neural network, the loss function used by the convolutional neural network is the mean square error (MSE, Mean Square error), and can pass SGD (Stochastic Gradient Descent, stochastic gradient descent) The algorithm learns end-to-end, thereby improving the accuracy of the prediction results of the preset convolutional neural network.

In the embodiment of the present invention, the convolutional neural network is trained through the training sample set to obtain a preset convolutional neural network to predict the traffic congestion index of a certain prediction area for a period of time in the future, so as to improve the preset convolutional neural network. The prediction performance of the pre-set convolutional neural network is further improved.

Referring to FIG. 6, FIG. 6 is a schematic structural diagram of a traffic congestion index prediction device based on GIS map information provided by an embodiment of the present invention. The GIS map information-based traffic congestion index prediction device 600 includes:

The first acquisition module 601 is used to acquire a GIS map structured vector map sequence and an environment vector sequence of the area to be predicted. Each frame of the GIS map structured vector map includes a first preset number of static layers and a second preset number of static layers. Dynamic layers, and a frame of GIS map structured vector graphics corresponds to an environmental vector sequence;

The prediction module 602 is used to input the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, respectively extract the spatiotemporal features of the GIS map structured vector graphics sequence, the mapping feature of the environmental vector sequence, and Output the traffic congestion index results of the area to be predicted within the preset time according to the spatiotemporal features and mapping features.

Optionally, as shown in FIG. 7 , FIG. 7 is a schematic structural diagram provided by the first acquisition module in an embodiment of the present invention. The dynamic layer includes traffic dynamic data, and the first acquisition module 601 includes:

The first obtaining unit 6011 is used to obtain the static layer and traffic dynamic data of the GIS map of the area to be predicted at historical time points. The static layer includes one or more of a road network map, a building map and a point of interest map, Traffic dynamic data includes traffic congestion index and vehicle trajectory data corresponding to historical time points;

The first mapping unit 6012 is used to map the traffic congestion index to the road network map at a historical time point to obtain a traffic congestion index map;

The second mapping unit 6013 is used to map the vehicle trajectory data into a two-dimensional network, calculate the average vehicle speed and the number of vehicles in each network, and obtain a traffic flow speed map and a vehicle number map;

The first splicing unit 6014 is used for splicing the road network map, the building map, the point of interest map, the traffic congestion index map, the traffic speed map and the vehicle quantity map as channels to obtain the GIS map structured vector diagram corresponding to the historical time point;

The second splicing unit 6015 is configured to splicing the GIS map structured vector graphics corresponding to the consecutive historical time points according to the time sequence of the consecutive historical time points to obtain the GIS map structured vector graphics sequence of the area to be predicted.

Optionally, as shown in FIG. 8 , FIG. 8 is another schematic structural diagram provided by the first acquisition module in the embodiment of the present invention. The first acquisition module 601 includes:

The second obtaining unit 6016 is used to obtain weather data and date data of the area to be predicted at historical time points;

The encoding unit 6017 is used to encode weather data and date data into environmental vectors according to preset encoding rules;

The third splicing unit 6018 is configured to splicing the environmental vectors corresponding to the consecutive historical time points according to the time sequence of the consecutive historical time points, to obtain the environmental vector sequence of the area to be predicted.

Optionally, the preset convolutional neural network includes a three-dimensional convolutional network, a feature mapping network, a fusion network, and a fully connected layer network;

As shown in FIG. 9 , FIG. 9 is a schematic structural diagram provided by a prediction module in an embodiment of the present invention. Prediction module 602 includes:

The convolution calculation unit 6021 is used to perform three-dimensional convolution calculation on the GIS map structured vector diagram sequence through a three-dimensional convolution network to obtain the spatiotemporal characteristics of the GIS map structured vector diagram sequence;

The mapping calculation unit 6022 is used to perform mapping calculation on the environment vector sequence through the feature mapping network, and output the mapping feature of the environment vector sequence;

The fusion calculation unit 6023 is used to fuse the spatiotemporal feature and the mapping feature through the fusion network, and output the fusion feature;

The fully connected calculation unit 6024 is configured to perform the first fully connected calculation on the fusion feature through the fully connected layer network, and output the result of the traffic congestion index of the area to be predicted within a preset time.

Optionally, the three-dimensional convolution network is constructed according to the residual network, and the convolution calculation unit 6021 includes:

The convolution calculation sub-unit is used to perform convolution calculation in combination with the residuals of the previous convolution calculation layer during the three-dimensional convolution calculation process to obtain a spatiotemporal feature map;

The full connection calculation subunit is used to perform the second full connection calculation on the spatiotemporal feature map to obtain the spatiotemporal features of the structured vector map sequence of the GIS map.

Optionally, the fusion network is a gated network, and the fusion computing unit 6023 includes:

The linearization processing subunit is used to perform nonlinear processing on the mapping feature through the gating network to obtain the gating feature;

The point-pole calculation subunit is used to calculate the dot product of the gated feature and the spatiotemporal feature to obtain the fusion feature.

Optionally, as shown in FIG. 10 , FIG. 10 is a schematic structural diagram of another device for predicting traffic congestion index based on GIS map information provided by an embodiment of the present invention. The device 600 for predicting traffic congestion index based on GIS map information further includes: :

The second obtaining module 603 is configured to obtain a training sample set, the training sample set includes a plurality of training samples, and each training sample includes a sample GIS map structured vector graphics sequence and a sample corresponding to the GIS map structured vector graphics sequence The environment vector sequence and the real label of the traffic congestion index at the predicted time point, the sample GIS map structured vector map sequence and the GIS map structured vector map sequence have the same data structure, and the sample environment vector sequence and the environment vector sequence have the same data structure;

The training module 604 is used for training the convolutional neural network through the training sample set, so that the convolutional neural network learns the predicted output of the traffic congestion index at the predicted time point to obtain a preset convolutional neural network.

The apparatus 600 for predicting a traffic congestion index based on GIS map information provided by the embodiment of the present invention can implement the various implementations in the foregoing method embodiments and the corresponding beneficial effects, which are not repeated here to avoid repetition.

Referring to FIG. 11, FIG. 11 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 700 includes: a memory 702, a processor 701, and a computer program stored in the memory 702 and running on the processor 701 , when the processor 701 executes the computer program, the steps in the method for predicting the traffic congestion index based on GIS map information provided by the above embodiments are implemented, and the processor 701 executes the following steps:

Obtain a GIS map structured vector map sequence and an environment vector sequence of the area to be predicted. Each frame of GIS map structured vector map includes a first preset number of static layers and a second preset number of dynamic layers, and one frame of GIS map includes a first preset number of static layers and a second preset number of dynamic layers. The map structured vector map corresponds to an environment vector sequence;

Input the GIS map structured vector map sequence and the environmental vector sequence into the preset convolutional neural network, extract the spatiotemporal features of the GIS map structured vector map sequence and the mapping feature of the environmental vector sequence, and extract the spatial and temporal features and mapping features according to the spatial and temporal features and mapping features. Output the traffic congestion index results of the area to be predicted within the preset time.

Optionally, the dynamic layer includes traffic dynamic data, and the steps performed by the processor 701 to obtain the GIS map structured vector diagram sequence of the area to be predicted include:

Obtain the static layer and traffic dynamic data of the GIS map of the area to be predicted at historical time points. The static layer includes one or more of the road network map, building map, and point of interest map. The traffic dynamic data includes historical time points. Corresponding traffic congestion index and vehicle trajectory data;

At historical time points, map the traffic congestion index to the road network map to obtain a traffic congestion index map;

Map the vehicle trajectory data into a two-dimensional network, calculate the average vehicle speed and the number of vehicles in each network, and obtain the traffic speed map and the vehicle number map;

The road network map, building map, point of interest map, traffic congestion index map, traffic speed map and vehicle number map are spliced as channels to obtain the GIS map structured vector map corresponding to the historical time point;

The GIS map structured vector graphics corresponding to the consecutive historical time points are spliced according to the time series of the consecutive historical time points, and the GIS map structured vector graphics sequence of the area to be predicted is obtained.

Optionally, the step of acquiring the environment vector sequence of the area to be predicted performed by the processor 701 includes:

Obtain the weather data and date data of the area to be predicted at historical time points;

Encode weather data and date data into environment vectors according to preset coding rules;

The environmental vectors corresponding to the consecutive historical time points are spliced according to the time sequence of the consecutive historical time points to obtain the environmental vector sequence of the area to be predicted.

The processor 701 inputs the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, respectively extracts the spatiotemporal features of the GIS map structured vector graphics sequence, the mapping feature of the environmental vector sequence, and according to the The steps of outputting the result of the traffic congestion index of the area to be predicted within the preset time by the spatiotemporal feature and the mapping feature include:

The 3D convolution calculation is performed on the GIS map structured vector map sequence through the 3D convolution network, and the spatiotemporal characteristics of the GIS map structured vector map sequence are obtained;

Perform the mapping calculation on the environment vector sequence through the feature mapping network, and output the mapping feature of the environment vector sequence;

The spatiotemporal feature and the mapping feature are fused and calculated through the fusion network, and the fusion feature is obtained as the output;

The first full connection calculation is performed on the fusion feature through the fully connected layer network, and the result of the traffic congestion index of the area to be predicted within the preset time is obtained as output.

Optionally, the three-dimensional convolution network is constructed according to the residual network, and the processor 701 performs three-dimensional convolution calculation on the GIS map structured vector diagram sequence through the three-dimensional convolution network to obtain the spatiotemporal characteristics of the GIS map structured vector diagram sequence. steps, including:

The second full connection calculation is performed on the spatiotemporal feature map to obtain the spatiotemporal feature of the structured vector map sequence of the GIS map.

Optionally, the fusion network is a gated network, and the steps performed by the processor 701 to fuse the spatiotemporal feature and the mapping feature through the fusion network, and output the fused feature, include:

Through the gated network, the mapping feature is nonlinearly processed to obtain the gated feature;

Calculate the dot product of the gated feature and the spatiotemporal feature to get the fusion feature.

Optionally, the processor 701 further performs the following steps:

Obtain a training sample set, the training sample set includes multiple training samples, and each training sample includes a sample GIS map structured vector map sequence, a sample environment vector sequence corresponding to the GIS map structured vector map sequence, and a predicted time point. The traffic congestion index real label, the sample GIS map structured vector map sequence and the GIS map structured vector map sequence have the same data structure, and the sample environment vector sequence and the environment vector sequence have the same data structure;

Through the training sample set, the convolutional neural network is trained, so that the convolutional neural network learns the predicted output of the traffic congestion index at the predicted time point, and a preset convolutional neural network is obtained.

The electronic device 700 provided in the embodiment of the present invention can implement each implementation manner in the foregoing method embodiment and the corresponding beneficial effects, which are not repeated here in order to avoid repetition.

Embodiments of the present invention further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, implements the traffic congestion index prediction method based on GIS map information provided by the embodiment of the present invention and can achieve the same technical effect, in order to avoid repetition, it will not be repeated here.

Those of ordinary skill in the art can understand that the realization of all or part of the processes in the methods of the above embodiments can be accomplished by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and the program can be executed when the program is executed. , may include the flow of the above-mentioned method embodiments. The storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM) or a random access memory (Random Access Memory, RAM for short).

Claims

A traffic congestion index prediction method based on GIS map information, characterized in that the method comprises the following steps:

Obtain a GIS map structured vector map sequence and an environment vector sequence of the area to be predicted. Each frame of GIS map structured vector map includes a first preset number of static layers and a second preset number of dynamic layers, and one frame of GIS map includes a first preset number of static layers and a second preset number of dynamic layers. The map structured vector map corresponds to an environment vector sequence;

Inputting the GIS map structured vector diagram sequence and the environment vector sequence into a preset convolutional neural network, extracting the spatiotemporal features of the GIS map structured vector graphics sequence, the mapping features of the environment vector sequence, and According to the spatiotemporal feature and the mapping feature, a traffic congestion index result of the to-be-predicted area within a preset time is output.
The traffic congestion index prediction method based on GIS map information according to claim 1, wherein the dynamic layer includes traffic dynamic data, and the step of obtaining a GIS map structured vector diagram sequence of the area to be predicted comprises:

Obtain the static layer and traffic dynamic data of the GIS map of the area to be predicted at historical time points. The static layer includes one or more of a road network map, a building map and a point of interest map. The traffic dynamic data The data includes traffic congestion index and vehicle trajectory data corresponding to the historical time point;

At the historical time point, the traffic congestion index is mapped to the road network map to obtain a traffic congestion index map;

Map the vehicle trajectory data into a two-dimensional network, calculate the average vehicle speed and the number of vehicles in each network, and obtain a vehicle flow speed map and a vehicle number map;

Splicing the road network map, building map, point of interest map, traffic congestion index map, traffic speed map and vehicle quantity map as channels to obtain a GIS map structured vector map corresponding to the historical time point;

The GIS map structured vector graphics corresponding to the continuous historical time points are spliced according to the time sequence of the continuous historical time points to obtain the GIS map structured vector graphics sequence of the to-be-predicted area.
The method for predicting traffic congestion index based on GIS map information according to claim 2, wherein the step of obtaining the environmental vector sequence of the area to be predicted comprises:

Obtain the weather data and date data of the area to be predicted at the historical time point;

encoding the weather data and date data into an environment vector according to a preset encoding rule;

The environment vectors corresponding to the continuous historical time points are spliced according to the time sequence of the continuous historical time points to obtain the environment vector sequence of the to-be-predicted area.
The traffic congestion index prediction method based on GIS map information according to any one of claims 1 to 3, wherein the preset convolutional neural network comprises a three-dimensional convolutional network, a feature mapping network, a fusion network, and a Fully connected layer network;

Inputting the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, respectively extracting the spatiotemporal features of the GIS map structured vector graphics sequence, and the mapping features of the environmental vector sequence , and the step of outputting the result of the traffic congestion index of the to-be-predicted area within a preset time according to the spatiotemporal feature and the mapping feature includes:

The 3D convolution calculation is performed on the GIS map structured vector diagram sequence through the 3D convolution network to obtain the spatiotemporal characteristics of the GIS map structured vector diagram sequence;

Perform a mapping calculation on the environment vector sequence through the feature mapping network, and output the mapping feature of the environment vector sequence;

Perform fusion calculation on the spatiotemporal feature and the mapping feature through the fusion network, and output the fusion feature;

The first full-connection calculation is performed on the fusion feature through the fully-connected layer network, and the result of the traffic congestion index of the to-be-predicted area within a preset time is output.
The traffic congestion index prediction method based on GIS map information according to claim 4, wherein the three-dimensional convolutional network is constructed according to a residual network, and the GIS map structure is determined by the three-dimensional convolutional network. The steps of obtaining the spatiotemporal features of the GIS map structured vector graphics sequence by performing three-dimensional convolution calculation on the sequence of the GIS map structured vector graphics include:

In the process of three-dimensional convolution calculation, the convolution calculation is carried out in combination with the residual of the previous convolution calculation layer, and the spatiotemporal feature map is obtained;

A second full connection calculation is performed on the spatiotemporal feature map to obtain the spatiotemporal feature of the GIS map structured vector map sequence.
The traffic congestion index prediction method based on GIS map information according to claim 5, wherein the fusion network is a gated network, and the spatiotemporal feature and the mapping feature are fused through the fusion network The steps of calculating and outputting fused features include:

Through the gating network, non-linear processing is performed on the mapping feature to obtain the gating feature;

Calculate the dot product of the gated feature and the spatiotemporal feature to obtain a fusion feature.
The traffic congestion index prediction method based on GIS map information as claimed in claim 6, wherein the method further comprises:

Obtain a training sample set, the training sample set includes a plurality of training samples, each training sample includes a sample GIS map structured vector map sequence, a sample environment vector sequence corresponding to the GIS map structured vector map sequence, and The real label of the traffic congestion index at the predicted time point, the sample GIS map structured vector map sequence and the GIS map structured vector map sequence have the same data structure, the sample environment vector sequence and the environment vector sequence have the same data structure data structure;

Through the training sample set, the convolutional neural network is trained, so that the convolutional neural network learns the predicted output of the traffic congestion index at the predicted time point, and the preset convolutional neural network is obtained.
A traffic congestion index prediction device based on GIS map information, characterized in that it includes:

The first acquisition module is used to acquire the GIS map structured vector graphics sequence and the environmental vector sequence of the area to be predicted, and each frame of the GIS map structured vector graphics includes a first preset number of static layers and a second preset number of dynamic layers. layer, and a frame of GIS map structured vector graphics corresponds to an environmental vector sequence;

The prediction module is used to input the GIS map structured vector diagram sequence and the environmental vector sequence into a preset convolutional neural network, and extract the spatiotemporal features of the GIS map structured vector diagram sequence, and the environmental vector sequence and output the traffic congestion index result of the to-be-predicted area within a preset time according to the spatiotemporal feature and the mapping feature.
An electronic device, characterized by comprising: a memory, a processor, and a computer program stored on the memory and running on the processor, the processor implementing the computer program as claimed in claim 1 when the processor executes the computer program Steps in the traffic congestion index prediction method based on GIS map information described in any one of to 7.
A computer-readable storage medium, characterized in that, a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the GIS-based GIS-based system according to any one of claims 1 to 7 is realized when the computer program is executed by a processor. Steps in a traffic congestion index prediction method for map information.