Granted Patent B2

US 12,431,014 · App. 18/142,982 · Granted Sep 30, 2025

Inventors: Minhua Shao (Shanghai, CN); Yuncheng Zeng (Shanghai, CN); Lijun Sun (Shanghai, CN)

Assignee: TONGJI UNIVERSITY

G08G1/0145G06N3/0464

Abstract

Claims (39)

1. A method for controlling road network traffic congestion, comprising:

acquiring, by traffic information detection facilities, positions of detectors in a road network and a traffic state of a cross-section where each of the detectors is located at each moment within a set time period, wherein the traffic state comprises a vehicle speed and traffic flow;

constructing, by a computer, a directed graph according to the positions of the detectors in the road network, wherein nodes in the directed graph are cross-sections where the detectors are located in the road network, and edges in the directed graph are road sections between the cross-sections where the detectors are located;

determining, by the computer, a free-flow reachability matrix of the directed graph and a plurality of neighborhood matrices with different orders according to a free-flow vehicle speed between the cross-sections where the detectors are located and the directed graph, wherein when the order is k, the neighborhood matrices indicate whether a vehicle reaches j th detector from an i th detector after passing across k road sections, and the free-flow reachability matrix indicates whether the vehicle reaches the j th detector from the i th detector at the free-flow vehicle speed within a set time;

calculating, by the computer, a convolution operator of the directed graph at each moment within the set time period according to the neighborhood matrices of all the orders, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period;

constructing, by the computer, a long short-term memory neural network model, based on historical traffic data of the road network, to learn spatial-temporal characteristics of the road network:

inputting the convolution operator of the directed graph at each moment within the set time period into the long short-term memory neural network model to obtain a traffic state of a cross-section where each of the detectors is located at each moment in a predicted time period;

determining, according to the traffic state of the cross-section where each of the detectors is located at each moment within the set time period and the traffic state of the cross-section where the detector is located at each moment in the predicted time period, whether the cross-section where the detector is located is a congestion source or single-point congestion;

controlling, when the cross-section where the detector is located is single-point congestion, the traffic flow of the cross-section by using a control method for single-point control; and

controlling, when the cross-section where the detector is located is a congestion source, the cross-section where the detector is located by using a control method for circle layer control, and determining, in real-time, whether to stop circle layer control according to first congestion duration, second congestion duration, a vehicle speed within the first congestion duration before control and a vehicle speed within the first congestion duration after the control, wherein the first congestion duration is congestion duration before control; and the second congestion duration is congestion duration after the control: wherein

the circle layer control is to control, according to a distance between the cross-section where the detector is located and a congestion source, the traffic flow of the cross-section where the detector is located.

2. The method for controlling road network traffic congestion according to claim 1 , wherein determining the free-flow reachability matrix of the directed graph and the plurality of neighborhood matrices with different orders according to the free-flow vehicle speed between the cross-sections where the detectors are located and the directed graph comprises:

determining a shortest distance matrix of the directed graph and the plurality of neighborhood matrices with different orders according to the directed graph; and

calculating the free-flow reachability matrix according to the shortest distance matrix of the directed graph and the free-flow vehicle speed between the cross-sections where the detectors are located.

3. The method for controlling road network traffic congestion according to claim 1 , wherein calculating the convolution operator of the directed graph at each moment within the set time period according to the neighborhood matrices of all the orders, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period comprises:

calculating, for any order, a convolution operator of the directed graph corresponding to the order at each moment within the set time period according to the neighborhood matrices of the order, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period; and

calculating the convolution operator of the directed graph at each moment within the set time period according to the convolution operators corresponding to all the orders of the directed graph at each moment within the set time period.

4. The method for controlling road network traffic congestion according to claim 3 , wherein calculating the convolution operator of the directed graph corresponding to the order at each moment within the set time period according to the neighborhood matrices of the order, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period comprises:

calculating a convolution operator of the k-hop directed graph at a moment t according to formula GC t k =(W gc_k ⊙Ã k ⊙FFR), wherein GC t k represents a convolution operator of the k-hop directed graph at the moment t W gc_k represents a weight matrix of a k-hop neighborhood matrix, Ã k represents a k-hop neighborhood matrix, FFR represents the free-flow reachability matrix and X t represents the traffic state of the cross-section where each of the detectors is located at the moment t.

5. A system for controlling road network traffic congestion, comprising: a computer and traffic information detection facilities;

wherein the traffic information detection facilities are configured to acquire positions of detectors in a road network and a traffic state of a cross-section where each of the detectors is located at each moment within a set time period, wherein the traffic state comprises a vehicle speed and traffic flow; and

the computer is configured to:

construct a directed graph according to the positions of the detectors in the road network, wherein nodes in the directed graph are cross-sections where the detectors are located in the road network, and edges in the directed graph are road sections between the cross-sections where the detectors are located;

determine a free-flow reachability matrix of the directed graph and a plurality of neighborhood matrices with different orders according to a free-flow vehicle speed between the cross-sections where the detectors are located and the directed graph, wherein when the order is k, the neighborhood matrices indicate whether a vehicle reaches a j th detector from an i th detector after passing across k road sections, and the free-flow reachability matrix indicates whether the vehicle reaches the j th detector from the i th detector at the free-flow vehicle speed within a set time;

calculate a convolution operator of the directed graph at each moment within the set time period according to the neighborhood matrices of all the orders, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period;

construct a long short-term memory neural network model, based on historical traffic data of the road network, to learn spatial-temporal characteristics of the road network;

input the convolution operator of the directed graph at each moment within the set time period into the long short-term memory neural network model to obtain a traffic state of a cross-section where each of the detectors is located at each moment in a predicted time period;

determine, according to the traffic state of the cross-section where each of the detectors is located at each moment within the set time period and the traffic state of the cross-section where the detector is located at each moment in the predicted time period, whether the cross-section where the detector is located is a congestion source or single-point congestion;

control, when the cross-section where the detector is located is single-point congestion, the traffic flow of the cross-section by using a control method for single-point control; and

control, when the cross-section where the detector is located is a congestion source, the cross-section where the detector is located by using a control method for circle layer control, and determine, in real-time, whether to stop circle layer control according to first congestion duration, second congestion duration, a vehicle speed within the first congestion duration before control and a vehicle speed within the first congestion duration after the control, wherein the first congestion duration is congestion duration before control; and the second congestion duration is congestion duration after the control;

wherein the circle layer control is to control, according to a distance between the cross-section where the detector is located and a congestion source, the traffic flow of the cross-section where the detector is located.

6. The system for controlling road network traffic congestion according to claim 5 , wherein determining the free-flow reachability matrix of the directed graph and the plurality of neighborhood matrices with different orders according to the free-flow vehicle speed between the cross-sections where the detectors are located and the directed graph comprises:

determining a shortest distance matrix of the directed graph and the plurality of neighborhood matrices with different orders according to the directed graph; and

calculating the free-flow reachability matrix according to the shortest distance matrix of the directed graph and the free-flow vehicle speed between the cross-sections where the detectors are located.

7. The system for controlling road network traffic congestion according to claim 5 , wherein calculating the convolution operator of the directed graph at each moment within the set time period according to the neighborhood matrices of all the orders, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period comprises:

calculating, for any order, a convolution operator of the directed graph corresponding to the order at each moment within the set time period according to the neighborhood matrices of the order, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period; and

calculating the convolution operator of the directed graph at each moment within the set time period according to the convolution operators corresponding to all the orders of the directed graph at each moment within the set time period.

8. The system for controlling road network traffic congestion according to claim 7 , wherein calculating the convolution operator of the directed graph corresponding to the order at each moment within the set time period according to the neighborhood matrices of the order, the free-flow reachability matrix and the traffic state of the cross-section where each of the detectors is located at each moment within the set time period comprises:

calculating a convolution operator of the k-hop directed graph at a moment t according to formula GC t k =(W gc_k ⊙Ã k ⊙FFR), wherein GC t k represents a convolution operator of the k-hop directed graph at the moment t W gc_k represents a weight matrix of a k-hop neighborhood matrix, Ã k represents a k-hop neighborhood matrix, FFR represents the free-flow reachability matrix and X t represents the traffic state of the cross-section where each of the detectors is located at the moment t.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (17)