CN-116798227-B - Parameter self-learning signal control method based on cloud edge cooperation

CN116798227BCN 116798227 BCN116798227 BCN 116798227BCN-116798227-B

Abstract

The invention provides a parameter self-learning signal control method based on cloud edge coordination, which solves the problems that traffic signal algorithm control is seriously dependent on traffic parameter configuration, is too dependent on training samples, is difficult to realize real-time all-weather full-scene application and the like. The method mainly comprises the steps of collecting traffic holographic track data, carrying out edge induction signal control, extracting holographic induction control parameters from the holographic data through a parameter self-learning method, establishing a cloud edge cooperative control framework, carrying out parameter self-learning data interaction, and controlling traffic signals in real time. The traffic state of each phase is judged through holographic data, a parameter self-learning method is adopted, a cloud-edge cooperative control system is established, all-weather full-scene application of traffic control is realized, and the method has the advantages of being strong in real-time control capability, adapting to various traffic flow conditions and traffic scenes, avoiding calculation force and trial-and-error cost required by parameter learning and the like.

Inventors

ZHOU JUNJIE
CUI XIA
HU LINGLONG
XU MENG
WU HAO

Assignees

浙江中控信息产业股份有限公司

Dates

Publication Date: 20260505
Application Date: 20230621

Claims (8)

1. A parameter self-learning signal control method based on cloud edge cooperation is characterized by comprising the following steps: S1, acquiring traffic holographic track data, judging traffic states of all phases, and controlling signal lamp duration according to the relation among selected traffic parameters; when the traffic density of other phases is equal to the theoretical blocking density, if the current traffic density of the phase is smaller than the optimal density, switching the phases, otherwise, prolonging the green light time of each phase one by one; S2, extracting a holographic induction control parameter definition parameter self-learning action space from traffic holographic track data, constructing a state matrix definition parameter self-learning state space by adopting an intersection cell structure construction method, and training a parameter self-learning model; constructing a vehicle state matrix by adopting an intersection cell structure according to intersection traffic holographic track data, representing different states of signal lamps of each lane by different values according to the actual condition of the current intersection traffic signal lamp, constructing an intersection signal state matrix with the same size as the state matrix, extracting a characteristic vector at the current moment by adopting a CNN algorithm for a superposition two-matrix result, integrating a characteristic vector sequence at each moment into a characteristic vector of a state space by adopting a transducer algorithm, learning the green light limit time of each phase equilibrium under a saturated scene, taking a sequence formed by each limit time as an action space, generating learning parameters, and training; and S3, establishing a cloud edge cooperative control framework, performing data interaction and controlling traffic signals in real time.
2. The method according to claim 1, wherein in step S1, the traffic hologram trajectory data includes traffic state information of each phase, the traffic state information includes traffic density and headway, and the controlling signal duration according to the relationship between the selected traffic parameters includes: s101, judging the traffic state of an intersection according to the traffic density of each phase based on a traffic flow theory; S102, setting a headway threshold value, and calculating the headway of the current phase; s103, judging whether the green light time needs to be prolonged or not according to the relation between the headway and the headway threshold value and the traffic density by combining a holographic induction algorithm.
3. The method of claim 2, wherein in step S1, the combined holographic induction algorithm determines that when the traffic density of other phases is less than the theoretical blocking density, if the headway of the current phase exceeds the headway threshold, the green time is not prolonged, and if the headway of the current phase does not exceed the headway threshold, the green time is prolonged.
4. The method for controlling a self-learning signal of a parameter based on cloud edge coordination according to claim 1, wherein in step S2, the defining a self-learning action space and a state space of the parameter includes: s201, constructing a vehicle state matrix by adopting an intersection cellular structure construction method according to intersection traffic holographic track data; S202, constructing an intersection signal state matrix according to the actual condition of the current intersection traffic signal lamp; S203, extracting feature vectors at the current moment from the result of overlapping the two matrixes by adopting a convolutional neural network algorithm, serializing the feature vectors at each moment to serve as signal period features, integrating the signal period features into a feature vector by adopting a Transformer algorithm, and defining the feature vector as a state space; S204, learning green light limit extension time for ensuring queuing equalization of each phase under traffic saturation situations, and defining a sequence formed by the green light limit extension time of each phase as an action space; And S205, generating learning parameters by adopting DDPG algorithm, and training a parameter model based on the defined state space and action space.
5. The method for controlling a self-learning signal based on a cloud edge cooperative parameter according to claim 1 or 4, wherein in step S2, the method for constructing a vehicle state matrix by using an intersection cellular structure comprises: S20101, dividing the maximum detection range of a certain lane in an intersection into a plurality of fixed-length cells, wherein the cells contain the position information and the speed value of a vehicle; s20102, setting the cell state of each lane as a row and setting each intersection lane as a column; S20103, carrying out normalization processing of the highest speed limit on the vehicle speed value in each cell as the value of the position, and constructing a vehicle state matrix according to whether the parking speed exists or not and representing the vehicle speed value by different fixed values.
6. The method for controlling a parameter self-learning signal based on cloud edge coordination according to claim 4, wherein in step S2, the indicator of queuing balance is inter-queuing length variance in all directions between intersections.
7. The method for controlling a parameter self-learning signal based on cloud edge cooperative as claimed in claim 1, wherein in step S3, the establishing a cloud edge cooperative control architecture includes: S301, carrying out data interaction through a message transmission channel from a cloud to an edge computing node, wherein the message transmission channel interaction data comprises traffic state information and a parameter model for completing training by the cloud; S302, establishing a control drive synchronization mechanism, wherein the control drive comprises a program for converting a signal scheme into a control command which can interact with control equipment through a communication protocol, and the synchronization is the synchronous determination of the control mode and the priority of an edge generation scheme and an original central lamp control command after the edge calculation node is established; s303, cloud side link management and abnormal degradation control are applied, unified management information of the cloud center comprises edge equipment information, and abnormal degradation conditions comprise edge equipment abnormality and intersection and cloud center communication abnormality.
8. The method for controlling a parameter self-learning signal based on cloud-edge coordination according to claim 7, wherein in step S3, the traffic state information is in the same data format after preliminary preprocessing, and the initial state input of the parameter model for completing training by the cloud comprises traffic state information.

Description

Parameter self-learning signal control method based on cloud edge cooperation Technical Field The invention relates to the technical field of traffic signal control, in particular to a parameter self-learning signal control method based on cloud edge cooperation. Background Aiming at the problems of increasingly serious urban traffic jams and green light time waste, along with the continuous iterative upgrading of detection sensing means, an intelligent signal control method is more important. The current traffic control method mainly comprises two types, namely traffic control algorithms based on operation optimization and control theory, such as full induction control, half induction control, self-adaptive control and the like, wherein the basic principle is that the input traffic flow, queuing length and other data are used for generating control parameters such as signal period, green-signal ratio and the like through a control model, and the method achieves a certain effect at a plurality of intersections, but the method is seriously dependent on various traffic parameter configurations, requires a great deal of professional manual experience of traffic practitioners, and different roads need repeated parameter adjustment, is difficult to generalize and popularize, and when traffic flow does not develop according to history, the parameter setting is fixed, so that the control scheme is unreasonable. The other is a control strategy self-generation algorithm based on machine learning and deep learning. The algorithm directly generates a signal scheme by extracting intersection operation characteristics and setting an optimization target. Depending too much on the training sample, the effects achieved on the test set tend to be difficult to achieve in a real environment. The patent publication No. CN113643553A discloses a multi-intersection intelligent traffic signal lamp control method and system based on federal reinforcement learning, which is characterized in that the actual intersection is modeled, the traffic simulation software Cityflow is used for simulating urban traffic and traffic flow, a reinforcement learning algorithm is used for each reinforcement learning agent, the traffic signal lamp is controlled in real time according to the traffic flow condition of the intersection, a federal reinforcement learning framework similar to federal learning is combined, a gradient sharing and parameter transmission process similar to federal learning is introduced, and a better control effect is achieved in the aspect of average running time of vehicles, but the acquisition of the relevant information of the actual road conditions of the patent basically depends on the traffic simulation software, real-time full-scene application is difficult to achieve, and the pressure on each directional road is mainly calculated through observation when the reinforcement learning agent, and a more proper intersection observation method and a traffic parameter self-learning process are not adopted, so that the system is adapted to various traffic flow conditions and traffic saturation scenes. Disclosure of Invention The invention aims to solve the problems that the traffic signal algorithm control is seriously dependent on traffic parameter configuration, is too dependent on training samples, is difficult to realize real-time all-weather full-scene application and the like. The technical problem is solved by the following technical scheme that the parameter self-learning signal control method based on cloud edge cooperation comprises the following steps: s1, acquiring traffic holographic track data, judging traffic states of all phases, and controlling signal lamp duration according to the relation among selected traffic parameters; s2, extracting a holographic induction control parameter definition parameter self-learning action space from traffic holographic track data, constructing a state matrix definition parameter self-learning state space by adopting an intersection cell structure construction method, and training a parameter self-learning model; and S3, establishing a cloud edge cooperative control framework, performing data interaction and controlling traffic signals in real time. The all-round moving track of crossing vehicles can be gathered through detection equipment such as radar, thunder all-in-one, including vehicle track position, vehicle speed, pass detection zone time etc. data to draw traffic related parameters such as traffic flow, vehicle speed, queuing length, traffic density etc. from vehicle track position. The traditional induction control carries out phase extension according to the time interval of passing vehicles, and the method is added to judge the traffic state of each phase, so that the real-time performance of acquired data is stronger, and the data analysis and calculation result is more accurate. The control algorithm based on operation optimization can improve the accuracy of the r