CN111882878A - Method for maximizing traffic capacity of key roads based on traffic flow prediction - Google Patents
Method for maximizing traffic capacity of key roads based on traffic flow prediction Download PDFInfo
- Publication number
- CN111882878A CN111882878A CN202010906980.7A CN202010906980A CN111882878A CN 111882878 A CN111882878 A CN 111882878A CN 202010906980 A CN202010906980 A CN 202010906980A CN 111882878 A CN111882878 A CN 111882878A
- Authority
- CN
- China
- Prior art keywords
- data
- traffic
- time
- road
- traffic flow
- 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.)
- Granted
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
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- 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
- G08G1/048—Detecting movement of traffic to be counted or controlled with provision for compensation of environmental or other condition, e.g. snow, vehicle stopped at detector
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/07—Controlling traffic signals
Abstract
The invention relates to a solution scheme based on the Internet of things and a big data information technology, in particular to a deep neural network, which trains a prediction model and predicts the traffic flow states of a road network target road section and a Node 30 minutes after the current time point by collecting historical data of the road network traffic flow in short time and long time and applying deep learning of a Node2vec and graph multi-attention neural network (GMAN) fusion mechanism, and carries out real-time intelligent feedback on a green-traffic ratio of a signal lamp according to the prediction result, thereby achieving the purposes of accurately limiting the current, maximally excavating the traffic capacity of the road network and reducing congestion and traffic paralysis caused by accidents. The invention forms a set of complete method for maximizing the traffic capacity of the key road from the collection and the conversion of traffic data to the adjustment and the control of traffic lights, and performs application verification on the actual engineering, thereby achieving the effects of accurately limiting the current, maximally excavating the traffic capacity of a road network and reducing the congestion and traffic paralysis caused by accidents.
Description
Technical Field
The invention relates to the technical field of intelligent traffic system operation and safety management control, in particular to a method for maximizing the traffic capacity of key roads based on traffic flow prediction.
Background
With the rapid growth of automobile holding capacity and the diversification and enrichment of travel demands, road traffic flow appears to migrate towards the direction of space-time multidimensional dynamic development, and under the multi-factor coupling action of the sharp increase of road traffic travel demands, adverse influences of severe weather and the limitation of traffic facility service functions, the traffic capacity and safe operation of roads are greatly challenged, such as: the contradiction between the demand of high-efficiency and safe operation of roads and the service level of traffic facilities caused by holidays, rush hours on duty and off duty and severe weather conditions such as rain, fog, ice, snow, strong wind and the like is quickly shown and activated, and more traffic accidents are caused by key control road sections such as urban main roads, tunnels, bridges and the like, so that traffic jam is caused, and even the road network is seriously paralyzed. Therefore, it has been difficult to adapt to the development of modern road traffic in this context to support traffic facilities and road operations, management and control by relying solely on traditional traffic flow theory. In the road network traffic organization and operation management control process, a scientific and reasonable method is adopted to evaluate various special states and catastrophe events in real time, such as rain, fog, ice and snow and other disastrous weather or traffic events, influence on road network operation and safety risks of the road network operation are evaluated, and the method has an important role in timely making road network safety management decisions, reducing the risks by adopting proper measures or intervening traffic flow operation to prevent traffic accidents.
The research and mining of traffic flow big data aim at scientifically evaluating the use effect of various traffic facilities such as roads, signal lamps and the like, providing improvement measures and scientifically guiding traffic management departments to provide solutions for improving the operation efficiency and traffic safety of traffic systems.
At present, a fixed signal lamp or a manual current limiting device is adopted for current limiting control of key roads in a road network system, the control means is single, advance prediction and real-time active regulation cannot be realized, the traffic capacity of a road network is not fully exerted, even the key roads are blocked to form traffic bottlenecks, the key roads are often irreplaceable traffic essential roads such as bridges, tunnels, viaducts, large intersections and the like, and the road network is paralyzed in a large area. The conventional road network traffic organization management strategy or measure is not integrated enough in the aspects of systematicness and completeness, the decision criterion of part of the traffic organization management strategy is not clear enough, and even a corresponding theoretical basis is lacked, for example, a road network entrance current limiting strategy is lacked with clear basis and method. For the traditional road network traffic organization management under the normal state, the smoothness and the efficiency of a road network are generally taken as main targets, and the corresponding decision basis of the traffic organization management is also taken as the basis. For the traffic organization management of the expressway network under the disaster event, because the environmental conditions cause great security threat to the normal operation of the road network, the road network is very easy to have traffic accidents or other adverse traffic events in the operation process, the realization of the smooth and high-efficiency target of the road network is influenced, and meanwhile, the occurrence of the traffic accidents or other adverse traffic events can cause serious casualties or property loss, so the traffic organization management of the expressway network under the disaster event can meet the requirements of safety, smoothness and efficiency, and the safety, smoothness and efficiency are reasonably balanced.
The existing solution is to open the variable lane, but basically only to relieve a part of the pressure; the flow is then limited at the entrance with a signal light, such as turning on the signal light during peak hours or in rainy or snowy weather. However, this management scheme is relatively extensive, and on one hand, limiting the flow too early or canceling the flow too late will limit the maximum traffic rate; on the other hand, if the current limiting is cancelled too late or too early, the problems of congestion and paralysis of the key roads cannot be solved at all, meanwhile, the time length setting of the signals does not consider the traffic flow change of the key roads, and information technologies such as big data, artificial intelligence, internet of things, internet, 5G and the like become important resources and solving means under the background of 'new infrastructure'.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a solution based on the Internet of things and a big data information technology, in particular to a deep neural network, by collecting historical data of a road network traffic flow in short time and long time, and applying deep learning of a Node2vec and map multi-attention neural network (GMAN) fusion mechanism, a prediction model is trained, the traffic flow state of a target road section and a Node (such as a bridge, a tunnel and the like) of the road network 30 minutes after the current time point (instant) is predicted, and real-time intelligent feedback is carried out on the green-to-green ratio of a signal lamp according to the prediction result, so that the aims of accurately limiting the current, maximally excavating the traffic capacity of the road network and reducing congestion and traffic paralysis caused by accidents are achieved.
Specifically, the technical scheme of the invention is as follows:
1. data acquisition and storage
(1) Road network main data
The collected data includes traffic monitoring sensor locations, connection road information (lane and width, variable lane, alternate lane, signal light settings, design speed, road maintenance conditions, etc.).
(2) Traffic flow data of road network
The traffic flow data of the road network is embodied by collecting the number of vehicles which continuously pass through the cross section within minute time.
(3) Weather, holiday, and major event information
The temperature (hot/cold/good), weather conditions (sunny/rainy/cloudy/snow/fog/ice), etc. are collected through the weather interface of the local weather department. Whether the information is holiday or not is obtained in advance through a government website. The start time and location (kilometers away from this critical section) of a significant event (sporting event, exhibition, literary and artistic activity, etc.) are obtained from a government website. The data are stored in a Hadoop distributed file storage system of a big data system deployed in a public cloud or a local traffic data center, and preparation is made for next processing.
2. Deep learning model training
(1) Road network graph adjacency matrix
The road network map adjacency matrix is a map data structure that describes the relative positions and relationships of traffic flow sampling sensors, and each row of the adjacency matrix represents the start point, end point, and weight of each edge, where the weight represents the distance of a connecting link between two vertices.
(2) Spatial data vectorization
Node2vec is an algorithm for generating a vector of nodes in the network, the input is the adjacency matrix of the network structure diagram obtained in the previous step, and the output is a vector representation of each Node. And sampling nodes in the graph in a depth-first walking mode, and generating a corresponding sequence for each point. And then, taking the sequences as text, importing the text into a skip-gram model in the word2vec, and training to obtain a vector of each node (corresponding to the vector of each word in the word2 vec).
(3) Time series data vectorization
Coding the time of sampling, and dividing into 24 different classes of 0-23 hours; encoding weather according to: sunny, rainy, cloudy, snowy, foggy and icy respectively represent different categories; the temperature is coded and divided into different categories of hot, cold, proper and the like; coding the holiday information, and classifying the holiday information into different categories such as yes, no holiday and the like; the important events nearby are coded and classified into different categories such as existence and nonexistence, and one dimension is added to represent the number of kilometers of the moving site leaving the road network. And respectively carrying out One-Hot (One-Hot) coding on the categories, and finally splicing all the One-Hot vectors and the continuous vectors into a high-dimensional vector (Ut) containing all the information.
(4) Training data construction
The raw data sampled from the sensor is each formatted as follows:
time t, sensor number sensorN, traffic flow (.)
Grouping according to the same time stamp, enabling each piece of data to fuse the data of all the sensors at the time, and then converting the data of the time series into a data format required by supervised learning, wherein for example, data of 30 minutes is required as input, and data of the next 30 minutes is required as output;
assuming that Xt0 is the sensor data vector at time t0 with values of (Xsensor1, Xsensor2, … Xsensor n), then the following format of input/output data may be constructed:
inputting [ X ]t0,Xt1,…,Xt29]
Output [ X ]t30,Xt31,…,Xt59]
The next piece of data is then:
inputting:[Xt1,Xt2,…,Xt30]
output [ X ]t31,Xt32,…,Xt59]
By analogy, a time-series supervised learning training set can be constructed. The input of the data is the data of each sensor in the last 30 minutes, and the output of the data is the data of each sensor in the next 30 minutes;
for each minute time, there is a corresponding high-dimensional vector (U)t) To express the information of time, weather, holidays, major events and the like at the current moment, and convert the data into time series supervised learning data:
inputting [ U ]t0,Ut1,…,Ut29]
Output [ U ]t30,Ut31,…,Ut59]
Training data required for deep learning of the neural network of the graph is prepared according to a required format.
(5) Graph multi-attention neural network (GMAN) training model
GMAN is the architecture of an Encoder-decoder (Encoder-decoder) that extracts features and the decoder predicts, with a translation attention module in between to translate the historical features of the Encoder into a feature representation. Both the encoder and decoder here consist of space-time attention blocks (ST-attention blocks). The spatiotemporal complex correlation problem is then solved by using a spatiotemporal embedding (STE) module to combine spatial and temporal information together and then input into a spatiotemporal attention block.
The inputs to the model were the training data (including sensor data and time information data) constructed in (4) and the spatial data in (2), using the mean absolute error (RME) as a loss function, using Adam optimizer as optimizer, with an initial learning rate of 0.001. The specific method uses GMAN, A Graph Multi-attachment Network for traffic Prediction (https:// arxiv.org/pdf/1911.08415v2. pdf).
(6) Traffic flow prediction and calculation
And (4) converting the input format in the step (4) to obtain a traffic flow input data sequence, wherein the traffic flow data acquired by the sensor N minutes before the current time (N is 30 in the invention).
And (3) vectorizing the time, weather and important activity related data N minutes before the current time, and then converting the input format in the step (4) to obtain a time-related input data sequence.
Although the format conversion in (4) is used, no data is output. And operating the GMAN prediction model and outputting the traffic flow N minutes after the current time.
(7) Traffic control
According to the road traffic capacity handbook: version 2000 (HCM2000), the maximum traffic capacity of a one-way roadway is calculated as CD＝MSVi·N·fw·fHV·fP
Wherein N is the number of unidirectional lanes;
fw-lane width and lateral net width correction factor;
fHV-oversize vehicle correction factor;
fp-a driver condition correction factor;
Msvi-maximum service traffic (pcu/(h x ln)) for level i service level;
according to the calculation model, the maximum traffic capacity in rainy days for a certain river-crossing bridge can be calculated as follows:
C＝1000×2×1×0.8×0.8＝1208(pcu/h)
when the maximum traffic capacity is considered, the maximum traffic flow of the road section needs to be calculated, if the traffic flow of the target road section continues to increase, the traffic density is increased, the traffic capacity is reduced at the same time, the speed of the vehicle is at the critical speed, and the density of the vehicles on the road corresponds to the optimal density.
At this time, the predicted traffic flow data within 30 minutes, particularly the predicted traffic flow of the key road, is extracted from (6) and compared with the maximum traffic capacity of the key road in real time. If the traffic flow is predicted to basically not reach the maximum traffic capacity within 30 minutes, the signal lamp is not required to be adjusted; if the traffic flow is predicted to approach the maximum traffic capacity (which may be set to reach 90% of the maximum traffic capacity value) within the first 10 minutes, the signal light control system is immediately activated.
First, the effect of traffic light control (split ratio) of the directly adjacent access road of the key road segment on the flow of the key node road segment is calculated. The traffic capacity of the entrance road of each directly adjacent road section constitutes the traffic flow of the key road section: call＝∑Ci
For each entry road CiThe traffic capacity calculation model under the control of the signal lamp is as follows:
wherein:
t-signal cycle time(s);
tg-green time(s) of the traffic direction per period of the signal;
to-the time(s) when the green light is on and the first vehicle passes the stop line;
ti-the average time(s) for the vehicle to pass the stop-line;
And adjusting the split green ratio of each entrance road to enable the sum of the traffic capacity of the entrance roads to be equal to the maximum traffic capacity of the key road. The equal ratio adjustment of the traffic capacity can be achieved by adjusting the green ratio of each entrance road in an equal ratio manner, and different weights can be reserved for different entrances, for example, a more important entrance is set with a weight with a larger price ratio.
Thus, the green lights at each entrance are similarly adjusted, and the adjustment can be stopped when the sum is close to the traffic capacity of the key road.
(8) Model update and prediction and signal lamp adjustment
Since the data collection of traffic flow is a continuous process, new data is always added to the data set, and the model should be updated to reflect the latest traffic flow condition of the road network. It can be set that every 24 hours, newly collected data is added into the original data set, and the steps (3), (4) and (5) are repeated to train a new model.
In order to respond to the latest road conditions quickly and in time, the trained model is loaded every 10 minutes, and prediction and signal lamp adjustment are carried out once, namely, the steps (6) and (7) are carried out.
(9) Signal lamp adjustment
The signal lamp control part of the invention consists of a Micro Control Unit (MCU), a 4G data transparent transmission module and a plurality of relays;
the 4G data transparent transmission module is connected to the micro control unit through a serial port and used for receiving signal lamp time length adjusting data from a traffic command center;
the relay is connected with an output port of the micro control unit and is used for receiving a control signal of the micro control unit;
each relay is respectively connected with a corresponding signal lamp circuit, such as an east-west green lamp circuit, an east-west yellow lamp circuit, an east-west red lamp circuit, a north-south green lamp circuit, a north-south yellow lamp circuit and a north-south red lamp circuit, and when a certain relay receives output data from the micro control unit, the corresponding signal lamp connected with the relay can be lightened or extinguished.
Specifically, STM32F103ZET6 is used by the micro control unit, TAS-IT-168 is used by the 4G data transparent transmission module, and CHNT JZX-22F 2Z is used by the relay.
(10) Signal lamp cycle process
In a working period, the micro control unit lights the green lamps in the east-west direction and simultaneously lights the red lamps in the north-south direction through the relay; then, a timer arranged in the micro control unit starts to time, after the duration of the green light in the east-west direction reaches seconds, the east-west green light is turned off through a relay signal, the yellow light in the east-west direction is turned on, and at the moment, the duration of the red light in the north-south direction does not reach yet.
After the east-west yellow lamp is turned on and waits for a corresponding time, the micro control unit turns off the east-west yellow lamp and turns on the east-west red lamp; at the moment, the time length of the timer of the red light in the south-north direction is just reached in seconds, the red light in the south-north direction is extinguished by the microcontroller, and the green light in the south-north direction is lightened.
Similarly, after the green lamps in the south-north direction are turned on for a certain green lamp duration, the microcontroller turns off the green lamps in the south-north direction and turns on the yellow lamps in the south-north direction, and at the moment, the waiting duration of the red lamps in the east-west direction is not reached yet. After the yellow lamps in the north-south directions are lighted for a certain time, the waiting time of just the red lamps in the east-west directions is just reached in seconds.
And then, after one period is finished, the micro control unit checks whether the data with the adjusted time length is received from the 4G data transparent transmission module, and if the data with the adjusted time length exists, the adjusted time length is read and used as the waiting time of the timer corresponding to the next period. Then, the green light in the east-west direction and the red light in the north-south direction are lightened through the relay, and the timer is started to wait for timing.
Preferably, when the micro control unit receives the signal duration adjustment data of the 4G data transparent transmission module, the adjustment is not performed immediately, but is performed uniformly after the current signal lamp period is finished, so as to avoid confusion of the signal lamps at the intersection.
Compared with the prior art, the invention has the advantages that:
the invention adopts the neural network based on the depth map to predict the traffic flow, and correspondingly adjusts the green-signal ratio of the signal lamp according to the prediction result, thereby achieving the purposes of accurately limiting the current and maximizing the traffic capacity of the key road, dynamically predicting the space-time (30 minutes in the future, in the monitoring range of the road network) traffic operation situation of the key road section of the road network, and maximally excavating the potential of the road network for carrying traffic. The method fully considers factors influencing traffic jam, simultaneously acquires road network main data, traffic flow data and data information influencing traffic, such as weather, holidays, major events and the like during data acquisition, vectorizes the acquired information and constructs training data, outputs predicted traffic flow data after training through a GMAN (multi-attention neural network) training model, and automatically optimizes and adjusts the split green ratio of a road intersection through the output predicted traffic flow data; the method of the invention is a method for continuously acquiring data, continuously training and continuously updating and adjusting, so that timely and rapid response can be made through the latest road network traffic flow condition, and the best flow regulation effect is achieved; in the specific adjusting method, the invention uses a micro-control unit to simultaneously control the red, yellow and green lights in different directions of a single intersection, and when the adjusting data is received, the adjusting data is not immediately adjusted, but the adjusting data is uniformly adjusted after the current signal lamp period is finished, thereby effectively avoiding the confusion of the signal lamps at the intersection. The technical scheme of the invention can effectively reduce the congestion and paralysis caused by traffic accidents, and can help reduce the problems of energy consumption and environmental pollution caused by congestion.
The invention forms a set of complete method for maximizing the traffic capacity of the key road from the collection and the conversion of traffic data to the adjustment and the control of traffic lights, and performs application verification on the actual engineering, thereby achieving the effects of accurately limiting the current, maximally excavating the traffic capacity of a road network and reducing the congestion and traffic paralysis caused by accidents.
Drawings
FIG. 1 is a general flow diagram of the invention;
FIG. 2 is a data acquisition and integration diagram;
FIG. 3 is a road network graph vectorization flow chart;
FIG. 4 is a diagram illustrating weather status, time, holidays, and major event codes;
FIG. 5 is a flow chart of the conversion of time series data into supervised learning training data;
FIG. 6 is a diagram of a multi-attention-seeking neural network model component;
FIG. 7 is a flow chart of traffic flow prediction;
FIG. 8 is a signal lamp control decision diagram;
FIG. 9 is a key road and section traffic flow diagram;
FIG. 10 is a general configuration diagram of a self-control signal lamp;
fig. 11 is a control flow chart of self-control signal lamps.
The specific implementation mode is as follows:
the technical solution of the present invention is further explained below with reference to fig. 1 to 11.
A. Data acquisition and storage
As shown in fig. 2:
(1) road network main data
The main data of the road network is relatively objective and fixed within a certain time unless the structure of the current road network is adjusted, such as expansion and the like. The data to be collected include traffic flow monitoring sensor positions, connection road information (lane and width, variable lane, spare lane, signal light settings, design speed, road maintenance conditions, etc.).
(2) Traffic flow data of road network
The traffic flow data of the road network is embodied by collecting the number of vehicles which continuously pass through the cross section within minute time. Typically acquired using buried coils or infrared, radar, etc. detection systems, each time series-based datum including time (minutes), location (sensor number), traffic flow (cumulative number of passes of the vehicle over the minute).
(3) Weather, holiday, and major event information
The influence of weather on the state of traffic flow is significant, and temperature (hot/cold/suitable), weather conditions (clear/rain/cloudy/snow/fog/ice), and the like need to be collected through a weather interface of a local weather department. The time information is reflected through traffic flow data, whether the holidays are equally important or not is judged, and the holiday information can be obtained in advance through a government website. The occurrence of major events (sporting events, exhibitions, literary activities, etc.) can have a significant impact on regional traffic flow, and the starting time and location (miles from this critical road segment) of the major event can be obtained from government websites.
The data provides richer feature support for a traffic flow prediction model, so the data can be stored in a Hadoop distributed file storage system of a big data system deployed in a public cloud or a local traffic data center to prepare for the next processing.
B. Deep learning model training
(1) Road network graph adjacency matrix
As shown in fig. 3, the road network map adjacency matrix is a map data structure that describes the relative positions and relationships of traffic flow sampling sensors. Graph G represents the road network under study containing critical roads with 7 fixed points representing 7 data samplers, then each row of the adjacency matrix represents the start point, end point and weight of each edge, where the weight represents the distance of the connecting road between the two vertices.
(2) Spatial data vectorization
Node2vec is an algorithm for generating a vector of nodes in the network, the input is the adjacency matrix of the network structure diagram obtained in the previous step, and the output is a vector representation of each Node. As shown in fig. 3, the nodes in the graph are sampled in a depth-first walking manner, and a corresponding sequence is generated for each point. And then, taking the sequences as text, importing the text into a skip-gram model in the word2vec, and training to obtain a vector of each node (corresponding to the vector of each word in the word2 vec).
(3) Time series data vectorization
As shown in fig. 4, the time of sampling is encoded into 24 different categories of 0 to 23 hours; encoding weather according to: sunny, rainy, cloudy, snowy, foggy and icy respectively represent different categories; the temperature is coded and divided into different categories of hot, cold, proper and the like; coding the holiday information, and classifying the holiday information into different categories such as yes, no holiday and the like; the important events nearby are coded and classified into different categories such as existence and nonexistence, and one dimension is added to represent the number of kilometers of the moving site leaving the road network. With this amount of class data, One-Hot (One-Hot) encoding can be performed on the classes, which is actually to encode N states with an N-bit state register, each state having an independent register bit, and only One of the register bits being valid. Finally, all the one-hot vectors and the continuous vectors are spliced into a high-dimensional vector (Ut) containing all the information.
(4) Training data construction
The raw data sampled from the sensor is each formatted as follows:
time (20: 05), Sensor number (Sensor1), traffic flow (13)
As shown in fig. 5, first, grouping is performed according to the same time stamp, so that each piece of data fuses the data of all sensors at that time, and the format is as follows:
time (20: 05), Sensor1 data (13), Sensor2 data (24), Sensor3 data (23) … … … Sensor data (.)
Then, we need to convert the time series data into the data format (input/output format) required by supervised learning, for example, 30 minutes of data is required as input, and the next 30 minutes of data is required as output.
Assuming that Xt0 is the sensor data vector at time t0 with values of (Xsensor1, Xsensor2, … Xsensor n), then the following format of input/output data may be constructed:
inputting [ X ]t0,Xt1,…,Xt29]
Output [ X ]t30,Xt31,…,Xt59]
The next piece of data is then:
inputting [ X ]t1,Xt2,…,Xt30]
Output [ X ]t31,Xt32,…,Xt59]
By analogy, a time-series supervised learning training set can be constructed. The input is the data of the last 30 minutes of each sensor, and the output is the data of the next 30 minutes of each sensor.
From above (3), for each minute time, there is a corresponding high latitude vector (Ut) to represent information about the time, weather, holidays, major activities, etc. at the current time. Therefore, similarly, these data can also be converted into time-series supervised learning data:
inputting [ U ]t0,Ut1,…,Ut29]
Output [ U ]t30,Ut31,…,Ut59]
So far, training data required by deep learning of the neural network of the graph is prepared according to a required format.
(5) Graph multi-attention neural network (GMAN) training model
As shown in fig. 6, GMAN is an Encoder-decoder (Encoder-decoder) architecture, where the Encoder is used to extract features and the decoder is used to predict, and there is a translation attention module in between to translate historical features of the Encoder into a feature representation. Both the encoder and decoder here consist of space-time attention blocks (ST-attention blocks). The spatiotemporal complex correlation problem is then solved by using a spatiotemporal embedding (STE) module to combine spatial and temporal information together and then input into a spatiotemporal attention block.
Space-time Embedding, the space-time Embedding is divided into two parts of space and time, the space part is the vector of the graph to be embedded, and the time part is the high-dimensional vector of the time.
A spatiotemporal attention block (ST-attention block) comprising three parts: a spatial attention mechanism (spatial attention), a temporal attention mechanism (temporal attention), and a fusion mechanism (gate fusion). Through the attention mechanism, attention features of other nodes on the graph in the space dimension are extracted, and attention features of the current node in the historical time step in the time dimension are extracted.
Transition attention (transition attention) to address the problem of each step error being amplified over long prediction time, a transition attention layer between the encoding and decoding layers is used to model the direct relationship between each future time step and the historical step, converting the encoded traffic flow characteristics into a representation that produces future decoding input.
The inputs to the model were the training data (including sensor data and time information data) constructed in (4) and the spatial data in (2), using the mean absolute error (RME) as a loss function, using Adam optimizer as optimizer, with an initial learning rate of 0.001.
(6) Traffic flow prediction and calculation
As shown in fig. 7, traffic flow data acquired by a sensor 30 minutes before the current time is subjected to the input format conversion in (4) to obtain a traffic flow input data sequence.
And (3) vectorizing the time, weather and important activity related data 30 minutes before the current time, and then converting the input format in the step (4) to obtain a time-related input data sequence.
Although the format conversion in (4) is used, no data is output. And operating the GMAN prediction model and outputting the traffic flow 30 minutes after the current time.
Through a prediction test of traffic flow of a cross-river bridge in a city, traffic flow data of 30 minutes is extracted, the predicted traffic flow of the later 30 minutes has different average absolute errors (RME) from 10 to 30 under different step lengths from 1 minute to 30 minutes, and the accuracy is lower as time goes on (as the accuracy is the most in the former 10 minutes).
(7) Traffic control
Firstly, the maximum traffic capacity of the key road is calculated, variable lanes or standby lanes are considered, and the influence of weather on the traffic capacity is also considered.
According to the road traffic capacity handbook: 2000 edition (HCM2000), the maximum capacity of a one-way roadway is calculated as:
CD＝MSVi·N·fw·fHV·fP
wherein N is the number of unidirectional lanes;
fw-lane width and lateral net width correction factor;
fHV-oversize vehicle correction factor;
fp-a driver condition correction factor;
Msv-maximum service traffic (pcu/(h x ln)) for level i service level;
according to the calculation model, the maximum traffic capacity in rainy days for a certain river-crossing bridge can be calculated as follows:
C＝1000×2×1×0.8×0.8＝1208(pcu/h)
when the maximum traffic capacity is considered, the maximum traffic flow of the road section needs to be calculated, if the traffic flow of the target road section continues to increase, the traffic density is increased, the traffic capacity is reduced at the same time, the speed of the vehicle is at the critical speed, and the density of the vehicles on the road corresponds to the optimal density.
At this time, as shown in fig. 8, predicted traffic flow data within 30 minutes, particularly predicted traffic flow of the key road, is extracted from (6), and is compared with the maximum traffic capacity of the key road in real time. If the traffic flow is predicted to basically not reach the maximum traffic capacity within 30 minutes, the signal lamp is not required to be adjusted; if the traffic flow is predicted to approach the maximum traffic capacity within the first 10 minutes (which can be set to reach 80% of the maximum traffic capacity value), then feedback is given to the platform and the signal lamp control system needs to be started immediately.
First, as shown in fig. 9, the influence of the traffic light control (split ratio) of the road directly adjacent to the entrance of the key link on the flow rate of the key node link is calculated. The traffic capacity of the entrance road of each directly adjacent road section constitutes the traffic flow of the key road section:
Call＝∑Ci
for each entry road CiThe traffic capacity calculation model under the control of the signal lamp is as follows:
wherein:
t-signal cycle time(s);
tg-green time(s) of the traffic direction per period of the signal;
to-the time(s) when the green light is on and the first vehicle passes the stop line;
ti-the average time(s) for the vehicle to pass the stop-line;
And adjusting the split green ratio of each entrance road to enable the sum of the traffic capacity of the entrance roads to be equal to the maximum traffic capacity of the key road. The equal ratio adjustment of the traffic capacity can be achieved by adjusting the green ratio of each entrance road in an equal ratio manner, and different weights can be reserved for different entrances, for example, a more important entrance is set with a weight with a larger price ratio.
If a crossing of a river-crossing bridge is connected, the traffic capacity under the control of the signal lamp is as follows:
C＝3600/60×[(40-2)/5+1]×0.9＝464(pcu/h)
if the green time is adjusted to 30s, the traffic capacity is:
C＝3600/60×[(30-2)/5+1]×0.9＝356(pcu/h)
thus, the green lights at each entrance are similarly adjusted, and the adjustment can be stopped when the sum is close to the traffic capacity of the key road.
(8) Model updating and prediction
Since the data collection of traffic flow is a continuous process, new data is always added to the data set, and the model should be updated to reflect the latest traffic flow condition of the road network. It can be set that every 24 hours, newly collected data is added into the original data set, and the steps (3), (4) and (5) are repeated to train a new model.
In order to respond to the latest road conditions quickly and in time, the trained model is loaded every 10 minutes, and prediction and signal lamp adjustment are carried out once, namely, the steps (6) and (7) are carried out.
(9) Signal lamp adjustment
As shown in fig. 10, the present invention provides a signal lamp adjusting scheme, specifically, a signal lamp control portion is composed of a Micro Control Unit (MCU), a 4G data transparent transmission module, and a plurality of relays;
the 4G data transparent transmission module is connected to the micro control unit through a serial port and used for receiving signal lamp time length adjusting data from a traffic command center;
the relay is connected with an output port of the micro control unit and is used for receiving a control signal of the micro control unit;
each relay is respectively connected with a corresponding signal lamp circuit, such as an east-west green lamp circuit, an east-west yellow lamp circuit, an east-west red lamp circuit, a north-south green lamp circuit, a north-south yellow lamp circuit and a north-south red lamp circuit (as shown in the figure, the south-north circuit is omitted), and when a certain relay receives output data from the micro control unit, the corresponding signal lamp connected with the relay can be lightened or extinguished.
Specifically, STM32F103ZET6 is used by the micro control unit, TAS-IT-168 is used by the 4G data transparent transmission module, and CHNT JZX-22F 2Z is used by the relay.
(10) Signal lamp cycle process
In a working cycle, as shown in fig. 11, the micro control unit lights the green light in the east-west direction and lights the red light in the north-south direction through the relay; then, a timer arranged in the micro control unit starts to time, after the duration of the green light in the east-west direction reaches seconds, the east-west green light is turned off through a relay signal, the yellow light in the east-west direction is turned on, and at the moment, the duration of the red light in the north-south direction does not reach yet.
After the east-west yellow lamp is turned on and waits for a corresponding time, the micro control unit turns off the east-west yellow lamp and turns on the east-west red lamp; at the moment, the time length of the timer of the red light in the south-north direction is just reached in seconds, the red light in the south-north direction is extinguished by the microcontroller, and the green light in the south-north direction is lightened.
Similarly, after the green lamps in the south-north direction are turned on for a certain green lamp duration, the microcontroller turns off the green lamps in the south-north direction and turns on the yellow lamps in the south-north direction, and at the moment, the waiting duration of the red lamps in the east-west direction is not reached yet. After the yellow lamps in the north-south directions are lighted for a certain time, the waiting time of just the red lamps in the east-west directions is just reached in seconds.
And then, after one period is finished, the micro control unit checks whether the data with the adjusted time length is received from the 4G data transparent transmission module, and if the data with the adjusted time length exists, the adjusted time length is read and used as the waiting time of the timer corresponding to the next period. Then, the green light in the east-west direction and the red light in the north-south direction are lightened through the relay, and the timer is started to wait for timing.
Particularly, when the micro control unit receives the signal duration adjustment data of the 4G data transparent transmission module, the adjustment is not performed immediately, but the unified adjustment is performed after the current signal lamp period is finished, so that the signal lamp confusion of the intersection is avoided.
The protocol was run through the following tests:
a multi-attention-graph neural network model based on PyTorch is constructed, the traffic flow of the first half year of 2019 collected by 120 sensors of a road network which is 5 kilometers near a bridge across a river in a certain city is used as a training set, the absolute value error (MAE) is about 15, and accurate prediction effect is obtained on the training set. Forecasting the traffic flow after 30 minutes, and simulating and controlling the traffic signal lamp by using VISSIM (virtual visual identification system) traffic simulation software when the 80% maximum traffic capacity of the river-crossing bridge is reached, so that the traffic capacity is increased by 20% compared with the simple and manual control, and the jam probability of the river-crossing bridge is reduced by 30%; if not controlled, the possibility of road network paralysis caused by the blockage of the bridge crossing the river reaches 95 percent.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (9)
1. A method for maximizing the traffic capacity of key roads based on traffic flow prediction comprises the following steps:
A. data acquisition and storage
Collecting road network traffic flow data in a minute time period by a traffic flow sensor arranged through a road network, wherein each piece of data based on time series comprises time (minutes), places (sensor numbers) and traffic flow (the accumulated passing number of vehicles in the minute); simultaneously collecting main road network data and weather, holiday and major event information, and storing the collected information;
B. deep learning model training
(1) Describing the relative positions and relationships of the traffic flow sampling sensors by utilizing a road network diagram adjacency matrix, wherein each row of the adjacency matrix represents the starting point, the end point and the weight of each edge, and the weight represents the distance of a connecting road between two vertexes;
(2) utilizing the Node2vec to generate Node vectors in a network, sampling the nodes in the graph in a depth-first random walk mode, generating corresponding sequences for each point, and training the sequences as a skip-gram model with texts imported into the word2vec to obtain the vector of each Node;
(3) respectively carrying out One-Hot (One-Hot) coding on the categories of sampling time, weather, holidays, major event information and the like, and finally splicing all the One-Hot vectors and continuous vectors into a high-dimensionality vector Ut;
(4) converting the time series data obtained by the sensor and constructing training data;
(5) combining the data, training a traffic flow prediction model of a deep map neural network by using a map multi-attention neural network (GMAN);
(6) outputting the traffic flow which is trained in the step (5) and is 30 minutes after the current time in the road network by using the data acquired 30 minutes before the current time;
(7) comparing the predicted traffic flow of the key road obtained in the step (6) with the maximum traffic capacity of the key road to determine whether signal light control needs to be started to limit the traffic flow pressure of the connecting road on the key road;
(8) signal lamp adjustment
The signal lamp control system comprises a Micro Control Unit (MCU), a 4G data transparent transmission module and a plurality of relays;
the 4G data transparent transmission module is connected to the micro control unit through a serial port and used for receiving signal lamp time length adjusting data from a traffic command center;
the relay is connected with an output port of the micro control unit and is used for receiving a control signal of the micro control unit;
each relay is respectively connected with a corresponding signal lamp circuit, and after a certain relay receives output data from the micro control unit, the corresponding signal lamp connected with the relay can be lightened or extinguished;
the specific rule is as follows: in a working period, the micro control unit lights the green lamps in the east-west direction and simultaneously lights the red lamps in the north-south direction through the relay; then, a timer arranged in the micro control unit starts to time, after the duration of the green light in the east-west direction reaches seconds, the east-west green light is turned off through a relay signal, the yellow light in the east-west direction is turned on, and at the moment, the duration of the red light in the north-south direction does not reach;
after the east-west yellow lamp is turned on and waits for a corresponding time, the micro control unit turns off the east-west yellow lamp and turns on the east-west red lamp; at the moment, the time length of the timer of the red light in the south-north direction is just up to seconds, the red light in the south-north direction is extinguished by the microcontroller, and the green light in the south-north direction is lightened;
similarly, after the green lamps in the south-north direction are lighted for a certain green lamp duration, the microcontroller extinguishes the green lamps in the south-north direction and lights the yellow lamps in the south-north direction, and at the moment, the waiting duration of the red lamps in the east-west direction is not reached; after the yellow lamps in the north-south directions are lighted for a certain time, the waiting time of just the red lamps in the east-west directions is just reached in seconds.
2. The method for maximizing the traffic flow prediction-based key road traffic capacity according to claim 1, wherein in the step A, the road network main data to be collected comprise the positions of traffic monitoring sensors, connection road information (comprising lane and width, variable lane, standby lane, signal lamp setting, design speed, road maintenance conditions and the like).
3. The traffic flow prediction-based method for maximizing the traffic capacity of the key roads according to claim 1, wherein the collecting of weather, holiday and major event information specifically comprises: collecting temperature (hot/cold/suitable), weather conditions (sunny/rainy/cloudy/snow/fog/ice), etc. through a weather interface of a local meteorological department; holiday information can be obtained in advance through a government website; the start time and location (miles from this critical section) of a significant event (sporting event, exhibitions, literary activities, etc.) are obtained from a government website.
4. The method for maximizing the traffic flow prediction-based key road traffic capacity according to any one of claims 1 to 3, wherein the data collected in the step A is stored in a big data system Hadoop distributed file storage system deployed in a public cloud or a local traffic data center.
5. The method of maximizing critical road traffic capacity based on traffic flow prediction according to claim 4, characterized by the steps ofThe specific method of B (3) is as follows: coding the sampling time, and dividing the sampling time into 24 different categories from 0 to 23 hours; coding weather, and respectively representing different categories according to sunny, rainy, cloudy, snowy, fog and ice; the temperature is coded and divided into different categories of hot, cold, proper and the like; coding holiday information, and dividing the holiday information into two categories of yes and no holidays; coding nearby major events, classifying the major events into two categories, namely the presence category and the absence category, and adding a dimension to represent the number of kilometers of an activity site leaving a road network; respectively carrying out One-Hot (One-Hot) coding on the categories, and finally splicing all the One-Hot vectors and continuous vectors into a high-dimensional vector (U)t) And contains all the information.
6. The method for maximizing the traffic flow prediction-based key road traffic capacity according to claim 5, wherein the specific method of the step B (4) is as follows:
the raw data sampled from the sensor is each formatted as follows:
time t, sensor number sensorN, traffic flow (.)
Grouping according to the same time stamp, enabling each piece of data to fuse the data of all the sensors at the time, and then converting the data of the time series into a data format required by supervised learning, for example, data of 30 minutes is required as input, and data of the next 30 minutes is required as output, where the 30 minutes is only an exemplary way for explaining the present invention, and when in actual use, the method can be adjusted to other times, such as 28, 29, 31 and 32 minutes;
let Xt0For the sensor data vector at time t0, with values of (Xsensor1, Xsensor2, … Xsensor n), then the following format of input/output data may be constructed:
inputting [ X ]t0,Xt1,…,Xt29]
Output [ X ]t30,Xt31,…,Xt59]
The next piece of data is then:
inputting [ X ]t1,Xt2,…,Xt30]
Output of:[Xt31,Xt32,…,Xt59]
By analogy, a time-series supervised learning training set can be constructed; the input of the data is the data of each sensor in the last 30 minutes, and the output of the data is the data of each sensor in the next 30 minutes;
for each minute time, there is a corresponding high-dimensional vector (U)t) To express the information of time, weather, holidays, major events and the like at the current moment, and convert the data into time series supervised learning data:
inputting [ U ]t0,Ut1,…,Ut29]
Output [ U ]t30,Ut31,…,Ut59]
Training data required for deep learning of the neural network of the graph is prepared.
7. The method for maximizing the traffic flow prediction-based key road traffic capacity according to claim 5, wherein the specific method of the step B (7) is as follows:
comparing the predicted traffic flow of the key road obtained in the step (6) with the maximum traffic capacity of the key road, wherein if the predicted traffic flow does not reach the maximum traffic capacity basically within 30 minutes, the signal lamp is not required to be adjusted; if the traffic flow is predicted to be close to 80% of the maximum traffic capacity within the first 10 minutes, a signal lamp control system needs to be started immediately;
firstly, calculating the influence of the traffic light control (green signal ratio) of the road directly adjacent to the access port of the key road section on the flow of the key node road section; the traffic capacity of the entrance road of each directly adjacent road section constitutes the traffic flow of the key road section:
Call＝∑Ci
for each entry road CiThe traffic capacity calculation model under the control of the signal lamp is as follows:
wherein:
t-signal cycle time(s);
tg-green time(s) of the traffic direction per period of the signal;
to-the time(s) when the green light is on and the first vehicle passes the stop line;
ti-the average time(s) for the vehicle to pass the stop-line;
And adjusting the split ratio of each entrance road to enable the sum of the traffic capacity of the entrance roads to be close to 80% of the maximum traffic capacity of the key road, and stopping adjusting.
8. The method for maximizing key road traffic capacity based on traffic flow prediction according to claim 1, wherein newly collected traffic flow data is added every 24 hours, steps 3) to 5) are repeated, the model trained by the system using the updated data is loaded every 10 minutes, and steps 6) and 7) are performed once.
9. The traffic flow prediction-based method for maximizing the traffic capacity of key roads according to claim 1, wherein in step B (8): and after one period is finished, the micro control unit checks whether the data with the adjusted time length is received from the 4G data transparent transmission module, and if the data with the adjusted time length exists, the adjusted time length is read and used as the waiting time of the timer corresponding to the next period.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010906980.7A CN111882878B (en) | 2020-09-02 | 2020-09-02 | Method for maximizing traffic capacity of key roads based on traffic flow prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010906980.7A CN111882878B (en) | 2020-09-02 | 2020-09-02 | Method for maximizing traffic capacity of key roads based on traffic flow prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111882878A true CN111882878A (en) | 2020-11-03 |
| CN111882878B CN111882878B (en) | 2021-07-02 |
Family
ID=73198918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010906980.7A Active CN111882878B (en) | 2020-09-02 | 2020-09-02 | Method for maximizing traffic capacity of key roads based on traffic flow prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111882878B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112991787A (en) * | 2021-04-15 | 2021-06-18 | 吉林大学 | Method and system for optimizing traffic signals at intersection in ice and snow weather |
| CN113015297A (en) * | 2021-02-22 | 2021-06-22 | 上海工程技术大学 | Road intelligent lighting system based on traffic flow prediction |
| CN113257008A (en) * | 2021-05-12 | 2021-08-13 | 兰州交通大学 | Pedestrian flow dynamic control system and method based on deep learning |
| CN113362605A (en) * | 2021-07-23 | 2021-09-07 | 上海交通大学 | Distributed traffic flow optimization system and method based on potential homogeneous region identification |
| CN113689721A (en) * | 2021-07-30 | 2021-11-23 | 深圳先进技术研究院 | Automatic driving vehicle speed control method, system, terminal and storage medium |
| CN114462313A (en) * | 2022-01-25 | 2022-05-10 | 青岛海信网络科技股份有限公司 | Traffic accident risk prediction method, device, equipment, medium and product |
| WO2023004775A1 (en) * | 2021-07-30 | 2023-02-02 | 深圳先进技术研究院 | Autonomous vehicle speed control method and system, terminal, and storage medium |
| CN116401944A (en) * | 2023-04-04 | 2023-07-07 | 山东锐纳交通科技有限公司 | Intelligent traffic management system based on big data |
Citations (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1998008207A1 (en) * | 1996-08-09 | 1998-02-26 | Dinbis Ab | Method and equipment for motorway control |
| CN1434946A (en) * | 2000-07-21 | 2003-08-06 | 交通预测.Com公司 | Method of providing travel time predictions |
| CN101038700A (en) * | 2007-04-20 | 2007-09-19 | 东南大学 | Mixed controlling method of single dot signal controlling crossing |
| CN101894477A (en) * | 2010-07-08 | 2010-11-24 | 苏州大学 | Self-locking control method for controlling road network traffic through urban signal lamps |
| CN204256963U (en) * | 2014-12-10 | 2015-04-08 | 辽宁省交通高等专科学校 | A kind of intersection traffic signal lamp real-time control apparatus based on GPS |
| CN105139667A (en) * | 2015-09-28 | 2015-12-09 | 大连理工大学 | Reversible lane control method at intersection under influence of left-turn short lane |
| CN105788236A (en) * | 2014-12-26 | 2016-07-20 | 浙江大华技术股份有限公司 | Traffic control method and traffic control device |
| CN106056907A (en) * | 2016-06-28 | 2016-10-26 | 福州大学 | Subway construction area road traffic control method |
| CN106530763A (en) * | 2016-12-28 | 2017-03-22 | 东南大学 | Supersaturation traffic adaptive signal control method with coupling of inlet flow and outlet flow |
| US9633560B1 (en) * | 2016-03-30 | 2017-04-25 | Jason Hao Gao | Traffic prediction and control system for vehicle traffic flows at traffic intersections |
| CN107591011A (en) * | 2017-10-31 | 2018-01-16 | 吉林大学 | Consider the intersection traffic signal adaptive control method of supply side constraint |
| US20180060665A1 (en) * | 2016-08-29 | 2018-03-01 | Nec Laboratories America, Inc. | Dual Stage Attention Based Recurrent Neural Network for Time Series Prediction |
| CN108447291A (en) * | 2018-04-03 | 2018-08-24 | 南京锦和佳鑫信息科技有限公司 | A kind of Intelligent road facility system and control method |
| CN108447278A (en) * | 2018-05-11 | 2018-08-24 | 银江股份有限公司 | A kind of intersection space and time optimization method and system towards tide flow |
| CN108665707A (en) * | 2018-05-24 | 2018-10-16 | 金陵科技学院 | It is a kind of prevent local traffic congestion expressway access shunting and flow cooperative control method |
| CN109919358A (en) * | 2019-01-31 | 2019-06-21 | 中国科学院软件研究所 | A kind of real-time site traffic prediction technique based on neural network space-time attention mechanism |
| CN110211396A (en) * | 2019-05-30 | 2019-09-06 | 华南理工大学 | A kind of dynamic regulation method of freeway toll station and periphery intersection group |
| CN110223510A (en) * | 2019-04-24 | 2019-09-10 | 长安大学 | A kind of multifactor short-term vehicle flowrate prediction technique based on neural network LSTM |
| CN110517481A (en) * | 2019-07-26 | 2019-11-29 | 厦门卫星定位应用股份有限公司 | Vehicle flowrate prediction technique, medium, equipment and device |
| US10503174B1 (en) * | 2019-01-31 | 2019-12-10 | StradVision, Inc. | Method and device for optimized resource allocation in autonomous driving on the basis of reinforcement learning using data from lidar, radar, and camera sensor |
| EP3650297A1 (en) * | 2018-11-08 | 2020-05-13 | Bayerische Motoren Werke Aktiengesellschaft | Method and apparatus for determining information related to a lane change of a target vehicle, method and apparatus for determining a vehicle comfort metric for a prediction of a driving maneuver of a target vehicle and computer program |
| CN111161535A (en) * | 2019-12-23 | 2020-05-15 | 山东大学 | Attention mechanism-based graph neural network traffic flow prediction method and system |
| CN111260919A (en) * | 2020-01-15 | 2020-06-09 | 厦门大学 | Traffic flow prediction method |
| CN111554107A (en) * | 2020-03-27 | 2020-08-18 | 北京星云互联科技有限公司 | Traffic control method, management platform, road side equipment and system |
-
2020
- 2020-09-02 CN CN202010906980.7A patent/CN111882878B/en active Active
Patent Citations (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1998008207A1 (en) * | 1996-08-09 | 1998-02-26 | Dinbis Ab | Method and equipment for motorway control |
| CN1434946A (en) * | 2000-07-21 | 2003-08-06 | 交通预测.Com公司 | Method of providing travel time predictions |
| CN101038700A (en) * | 2007-04-20 | 2007-09-19 | 东南大学 | Mixed controlling method of single dot signal controlling crossing |
| CN101894477A (en) * | 2010-07-08 | 2010-11-24 | 苏州大学 | Self-locking control method for controlling road network traffic through urban signal lamps |
| CN204256963U (en) * | 2014-12-10 | 2015-04-08 | 辽宁省交通高等专科学校 | A kind of intersection traffic signal lamp real-time control apparatus based on GPS |
| CN105788236A (en) * | 2014-12-26 | 2016-07-20 | 浙江大华技术股份有限公司 | Traffic control method and traffic control device |
| CN105139667A (en) * | 2015-09-28 | 2015-12-09 | 大连理工大学 | Reversible lane control method at intersection under influence of left-turn short lane |
| US9633560B1 (en) * | 2016-03-30 | 2017-04-25 | Jason Hao Gao | Traffic prediction and control system for vehicle traffic flows at traffic intersections |
| CN106056907A (en) * | 2016-06-28 | 2016-10-26 | 福州大学 | Subway construction area road traffic control method |
| US20180060665A1 (en) * | 2016-08-29 | 2018-03-01 | Nec Laboratories America, Inc. | Dual Stage Attention Based Recurrent Neural Network for Time Series Prediction |
| CN106530763A (en) * | 2016-12-28 | 2017-03-22 | 东南大学 | Supersaturation traffic adaptive signal control method with coupling of inlet flow and outlet flow |
| CN107591011A (en) * | 2017-10-31 | 2018-01-16 | 吉林大学 | Consider the intersection traffic signal adaptive control method of supply side constraint |
| CN108447291A (en) * | 2018-04-03 | 2018-08-24 | 南京锦和佳鑫信息科技有限公司 | A kind of Intelligent road facility system and control method |
| CN108447278A (en) * | 2018-05-11 | 2018-08-24 | 银江股份有限公司 | A kind of intersection space and time optimization method and system towards tide flow |
| CN108665707A (en) * | 2018-05-24 | 2018-10-16 | 金陵科技学院 | It is a kind of prevent local traffic congestion expressway access shunting and flow cooperative control method |
| EP3650297A1 (en) * | 2018-11-08 | 2020-05-13 | Bayerische Motoren Werke Aktiengesellschaft | Method and apparatus for determining information related to a lane change of a target vehicle, method and apparatus for determining a vehicle comfort metric for a prediction of a driving maneuver of a target vehicle and computer program |
| CN109919358A (en) * | 2019-01-31 | 2019-06-21 | 中国科学院软件研究所 | A kind of real-time site traffic prediction technique based on neural network space-time attention mechanism |
| US10503174B1 (en) * | 2019-01-31 | 2019-12-10 | StradVision, Inc. | Method and device for optimized resource allocation in autonomous driving on the basis of reinforcement learning using data from lidar, radar, and camera sensor |
| CN110223510A (en) * | 2019-04-24 | 2019-09-10 | 长安大学 | A kind of multifactor short-term vehicle flowrate prediction technique based on neural network LSTM |
| CN110211396A (en) * | 2019-05-30 | 2019-09-06 | 华南理工大学 | A kind of dynamic regulation method of freeway toll station and periphery intersection group |
| CN110517481A (en) * | 2019-07-26 | 2019-11-29 | 厦门卫星定位应用股份有限公司 | Vehicle flowrate prediction technique, medium, equipment and device |
| CN111161535A (en) * | 2019-12-23 | 2020-05-15 | 山东大学 | Attention mechanism-based graph neural network traffic flow prediction method and system |
| CN111260919A (en) * | 2020-01-15 | 2020-06-09 | 厦门大学 | Traffic flow prediction method |
| CN111554107A (en) * | 2020-03-27 | 2020-08-18 | 北京星云互联科技有限公司 | Traffic control method, management platform, road side equipment and system |
Non-Patent Citations (11)
| Title |
|---|
| CHEONBOK PARK等: ""STGRAT: A Spatio-Temporal Graph Attention Network for Traffic Forecasting"", 《ARXIV》 * |
| CHUANPAN ZHENG等: ""GMAN: A Graph Multi-Attention Network for Traffic Prediction"", 《ARXIV.ORG》 * |
| LAISEN NIE等: ""Network Traffic Prediction Based on Deep Belief Network in Wireless Mesh Backbone Networks"", 《2017 IEEE WIRELESS COMMUNICATIONS AND NETWORKING CONFERENCE (WCNC)》 * |
| NING CHEN等: ""Energy Consumption Model of Urban Transportation Based on Road Network Capacity"", 《2010 ASIA-PACIFIC POWER AND ENERGY ENGINEERING CONFERENCE》 * |
| 刘畅: ""基于VISSIM的城市道路交叉口自适应信号控制仿真技术研究"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 吴科: ""基于流量预测的交通信号配时优化研究"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 张少坤: ""基于深度学习的短时交通流量预测"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 朱勇: ""基于时空关联混合模型的交通流预测方法研究"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 李萌萌: ""预防交叉口排队溢出的交通信号控制方法研究"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 杜圣东等: ""一种基于序列到序列时空注意力学习的交通流预测模型"", 《计算机研究与发展》 * |
| 程晓鹏: ""基于VISSIM交通仿真的城市高架拥堵分析及控制改善"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113015297A (en) * | 2021-02-22 | 2021-06-22 | 上海工程技术大学 | Road intelligent lighting system based on traffic flow prediction |
| CN113015297B (en) * | 2021-02-22 | 2022-08-26 | 上海工程技术大学 | Road intelligent lighting system based on traffic flow prediction |
| CN112991787A (en) * | 2021-04-15 | 2021-06-18 | 吉林大学 | Method and system for optimizing traffic signals at intersection in ice and snow weather |
| CN112991787B (en) * | 2021-04-15 | 2023-04-28 | 吉林大学 | Intersection traffic signal optimization method and system in ice and snow weather |
| CN113257008A (en) * | 2021-05-12 | 2021-08-13 | 兰州交通大学 | Pedestrian flow dynamic control system and method based on deep learning |
| CN113362605A (en) * | 2021-07-23 | 2021-09-07 | 上海交通大学 | Distributed traffic flow optimization system and method based on potential homogeneous region identification |
| CN113689721A (en) * | 2021-07-30 | 2021-11-23 | 深圳先进技术研究院 | Automatic driving vehicle speed control method, system, terminal and storage medium |
| WO2023004775A1 (en) * | 2021-07-30 | 2023-02-02 | 深圳先进技术研究院 | Autonomous vehicle speed control method and system, terminal, and storage medium |
| CN114462313A (en) * | 2022-01-25 | 2022-05-10 | 青岛海信网络科技股份有限公司 | Traffic accident risk prediction method, device, equipment, medium and product |
| CN114462313B (en) * | 2022-01-25 | 2024-04-16 | 青岛海信网络科技股份有限公司 | Traffic accident risk prediction method, device, equipment, medium and product |
| CN116401944A (en) * | 2023-04-04 | 2023-07-07 | 山东锐纳交通科技有限公司 | Intelligent traffic management system based on big data |
| CN116401944B (en) * | 2023-04-04 | 2023-12-22 | 广州华南路桥实业有限公司 | Intelligent traffic management system based on big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111882878B (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111882878B (en) | Method for maximizing traffic capacity of key roads based on traffic flow prediction | |
| CN109670277B (en) | Travel time prediction method based on multi-mode data fusion and multi-model integration | |
| Moosavi et al. | A countrywide traffic accident dataset | |
| Lin et al. | A combined M5P tree and hazard-based duration model for predicting urban freeway traffic accident durations | |
| CN110322695B (en) | Short-term traffic flow prediction method based on deep learning | |
| Barrett et al. | Developing a dynamic traffic management modeling framework for hurricane evacuation | |
| CN106548645B (en) | Vehicle route optimization method and system based on deep learning | |
| CN109697852A (en) | Urban road congestion degree prediction technique based on timing traffic events | |
| CN111866132B (en) | Traffic operation control method based on Internet of things | |
| CN110956807B (en) | Highway flow prediction method based on combination of multi-source data and sliding window | |
| CN111080018A (en) | Intelligent internet automobile speed prediction method based on road traffic environment | |
| CN115063978B (en) | Bus arrival time prediction method based on digital twins | |
| Kubíková et al. | Microscopic simulation of optimal use of communication network | |
| Alam et al. | Mass evacuation of Halifax, Canada: a dynamic traffic microsimulation modeling approach | |
| CN112862240A (en) | Road obstacle risk assessment method and device based on urban big data and readable storage medium | |
| AU2021106420A4 (en) | Time and topology structure based traffic flow prediction with artificial intelligence and neural network | |
| CN111009140A (en) | Intelligent traffic signal control method based on open-source road condition information | |
| Hounsell et al. | Urban network traffic control | |
| CN111340068A (en) | Road traffic state prediction method and device | |
| Lu et al. | Research on the intelligent assignment model of urban traffic planning based on optimal path optimization algorithm | |
| Li et al. | Traffic flow prediction with big data: A learning approach based on SIS-complex networks | |
| CN114091772A (en) | Multi-angle fusion road traffic flow prediction method based on encoder and decoder | |
| Balaji et al. | Distributed multi-agent type-2 fuzzy architecture for urban traffic signal control | |
| CN112785840A (en) | Urban intelligent traffic computer control system and method | |
| Yao et al. | Deep characteristics analysis on travel time of emergency traffic |
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 |