WO2022188387A1 - 基于多榜样学习粒子群的智慧城市信号灯配时优化方法 - Google Patents

基于多榜样学习粒子群的智慧城市信号灯配时优化方法 Download PDF

Info

Publication number
WO2022188387A1
WO2022188387A1 PCT/CN2021/119200 CN2021119200W WO2022188387A1 WO 2022188387 A1 WO2022188387 A1 WO 2022188387A1 CN 2021119200 W CN2021119200 W CN 2021119200W WO 2022188387 A1 WO2022188387 A1 WO 2022188387A1
Authority
WO
WIPO (PCT)
Prior art keywords
particle
signal light
dimension
particle swarm
timing
Prior art date
Application number
PCT/CN2021/119200
Other languages
English (en)
French (fr)
Inventor
詹志辉
黎建宇
Original Assignee
华南理工大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 华南理工大学 filed Critical 华南理工大学
Publication of WO2022188387A1 publication Critical patent/WO2022188387A1/zh

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
    • G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
    • G08G1/0145—Measuring and analyzing of parameters relative to traffic conditions for specific applications for active traffic flow control
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/07—Controlling traffic signals
    • G08G1/08—Controlling traffic signals according to detected number or speed of vehicles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G06Q50/40—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0108—Measuring and analyzing of parameters relative to traffic conditions based on the source of data
    • G08G1/0116—Measuring and analyzing of parameters relative to traffic conditions based on the source of data from roadside infrastructure, e.g. beacons
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A30/00—Adapting or protecting infrastructure or their operation
    • Y02A30/60—Planning or developing urban green infrastructure

Definitions

  • the invention relates to the fields of intelligent traffic control and intelligent optimization algorithms, in particular to a method for optimizing the timing of signal lights in a smart city based on multi-model learning particle swarms.
  • the traditional signal light timing scheme such as the Webster timing scheme, establishes a traffic mathematical model based on the goal of minimizing the intersection delay, and calculates the optimal signal light cycle based on this model, and then divides the corresponding green light time.
  • the Webster model is only applicable to the traffic model under the condition of unsaturated traffic flow. Therefore, when the road traffic volume is oversaturated, the Webster timing scheme is no longer applicable, and the traditional signal light timing scheme has become increasingly unsatisfactory. The needs of smart city transportation.
  • particle swarm optimization has shown excellent performance in various fields such as power system, medical image registration, multi-objective optimization, and machine learning.
  • the particle swarm algorithm has strong global search ability and convergence ability, and is suitable for the search optimization of the best signal timing scheme under different saturation conditions. Therefore, many researchers also apply the particle swarm algorithm to the signal timing optimization of smart traffic. field.
  • the traditional particle swarm optimization algorithm is completely random initialization when initializing the population, but in the field of traffic signal timing optimization, if we obtain the distribution characteristics of traffic flow input, we can use this information to purposefully generate more It may cover the initial population of the region where the optimal solution is located, thereby enhancing the algorithm's search ability. Therefore, based on the above analysis, the traditional particle swarm optimization algorithm needs to be improved and integrated with the inherent characteristic knowledge in the field of traffic signal timing, so as to enhance the optimization ability of the algorithm in the field of traffic signal timing.
  • the purpose of the present invention is to solve the above-mentioned defects in the prior art, and to provide a smart city signal light timing optimization method based on multi-model learning particle swarm.
  • the performance is enhanced to help avoid particles getting stuck in local optimal positions.
  • the traffic flow distribution characteristics input at the intersection are used as pre-knowledge, so that the initial population distribution is more likely to cover the area where the optimal solution is located.
  • a smart city signal light timing optimization method based on multi-model learning particle swarm includes the following steps:
  • the size and proportion of the traffic flow at each entrance in the intersection are the biggest factors affecting the setting of traffic lights.
  • the optimization of timing of traffic lights is regarded as a random optimization problem. This important factor is not taken into account.
  • an initial seed solution is generated according to the traffic flow distribution characteristics of each entrance at the intersection, and then an initial particle swarm population is generated according to the initial seed solution. The process is as follows:
  • the crossing allocates relatively more green light traffic time, so it is necessary to define an average lane flow ratio indicator to measure the distribution ratio of traffic flow.
  • the average lane flow ratio is defined as follows:
  • ALF j represents the average lane flow of the j-th entry
  • VI j represents the input traffic flow of the j-th entry
  • LN j represents the number of lanes of the j-th entry
  • R j represents the j-th entry.
  • g′ j represents the green light time of the seed solution at the jth crossing
  • G′ represents the total green light time of the seed solution
  • G represents the total green light time under the maximum allowable signal light cycle period
  • VI max represents the maximum allowed input at the intersection traffic flow, Indicates that the integer operation symbol is evaluated downward
  • the initial seed solution represents the area where the potential optimal timing scheme may exist. Based on each dimension value of the initial seed solution, an appropriate Gaussian disturbance is added to generate an initial particle population covering the vicinity of the seed solution. Specifically, the initial population is generated according to the following formula:
  • N(0,5) represents a Gaussian distribution with a mean of 0 and a standard deviation of 5;
  • the speed and position of the population can be updated according to the iterative steps of the particle swarm optimization algorithm.
  • the traditional particle swarm algorithm has the disadvantage that the particles are easy to fall into the local optimal solution.
  • the present invention is in the particle speed update link.
  • a multi-example learning strategy is added to overcome this shortcoming.
  • the particles of the population are updated according to the multi-example learning strategy.
  • the update formula is as follows:
  • ⁇ represents the velocity weight
  • gBest j represents the global optimal position of the jth dimension
  • the jth dimension value is used as the example value of the ith particle in this dimension, c 1 , c 2 and c 3 are three update coefficients, and is three different random numbers between the jth dimension [0, 1], the particle position value update is the original position plus the updated new velocity in formula (6) as the new particle position;
  • steps S3 to S4 are continued until the maximum number of iterations set in advance by the algorithm is met and terminated.
  • a knowledge ratio threshold KT is set, and when generating a population individual, a value between [0, 1] is firstly selected. Random number, if the random number is less than KT, use the knowledge-assisted strategy to generate the individual, otherwise generate the individual according to the random method.
  • a model update interval threshold RT is set. If the global optimal position of the algorithm is not improved after the RT generation, the model update is performed, otherwise the model of the previous generation is continued.
  • update coefficients c 1 , c 2 and c 3 are set to 0.75, 0.75 and 1.50, respectively.
  • the knowledge ratio threshold KT is set to 0.8.
  • example update interval threshold RT is set to 7.
  • the maximum number of iterations is set to 40 generations.
  • the present invention has the following advantages and beneficial effects:
  • the present invention assists the initialization of the particle population by using the traffic flow distribution characteristics of the intersection as the pre-knowledge, which helps the initial population to be distributed in the area where the optimal solution is more likely to be covered.
  • the multi-model learning strategy proposed by the present invention helps particles in different dimensions to learn from different models respectively, thereby enhancing the diversity of algorithms and preventing particles from falling into local optimal positions during the search process.
  • a knowledge ratio threshold is used to control the ratio of the knowledge-assisted generation solution and the random solution, which helps to integrate the respective advantages of the knowledge-embedded-assisted strategy and the random search strategy.
  • a model update threshold is used to control the update frequency of the model.
  • this method helps to avoid the problem that the particle search is difficult to converge due to the frequent updating of the learning model of the particle, and on the other hand, it helps to avoid the problem that the particle search is difficult to converge.
  • this method helps to avoid the problem that the particle search is difficult to converge due to the frequent updating of the learning model of the particle, and on the other hand, it helps to avoid the problem that the particle search is difficult to converge.
  • Fig. 1 is the multi-model learning particle swarm algorithm flow chart in the embodiment of the present invention
  • FIG. 2 is a geometric plan view of the intersection of Xinhu Road-Yu'an 1st Road, Bao'an District, Shenzhen City, China according to an embodiment of the present invention.
  • ALF j represents the average lane flow of the j-th entry
  • VI j represents the input traffic flow of the j-th entry
  • LN j represents the number of lanes of the j-th entry
  • R j represents the j-th entry.
  • the initial seed solution is generated according to the following formula, and the green light time of each dimension of the initial seed solution is calculated by the following formula:
  • g'j represents the green light time of the seed solution at the jth entrance
  • G' represents the total green light time of the seed solution
  • G represents the total green light time under the maximum allowable signal light cycle period
  • VI max represents the maximum allowed time at the intersection Enter traffic flow
  • each dimension value is generated according to the following formula:
  • N(0,5) represents a Gaussian distribution with a mean of 0 and a standard deviation of 5;
  • a knowledge ratio threshold KT is set, which is set to 0.8 in this embodiment. Every time the initial particle is generated by formula (11), a random number in the range of [0, 1] is generated first. If If the random number is less than KT, use formula (11) to generate the initial particle, otherwise initialize the particle randomly;
  • step S2 Most of the particles of the initial population in step S1 are generated by Gaussian sampling based on the seed solution, so it may happen that each dimension value does not meet the requirements of the feasible solution, so it is necessary to constrain the generated particle swarm individuals to make them Satisfy the constraints of feasible solutions;
  • an exemplary update threshold RT is set, which is set to 7 in this embodiment. If the global optimal position of the algorithm is not improved after the RT generation, the model must be updated before the particle velocity and position update.
  • the update process of the example is that, for the jth dimension of the example individual, different particles in the other two populations are selected, and the jth dimension of the historical optimal position of the particle with good fitness value is taken as the jth dimension of the example individual. If the global optimal position of the algorithm is improved in the RT generation, it means that the model individual can continue to guide the evolution of particles, and there is no need to update the model, but directly update the speed and position of the particles.
  • the particle adopts a multi-example learning strategy for speed update, and the update formula is as follows:
  • ⁇ represents the velocity weight
  • gBest j represents the global optimal position of the jth dimension
  • c 1 , c 2 and c 3 are three update coefficients, and are three different random numbers between the jth dimension [0, 1], the particle position value update is the original position plus the updated new velocity in formula (12) as the new particle position;
  • this method is easy to fall into the local optimal problem when the traditional particle swarm algorithm is applied to the signal light optimization timing scene, and a new multi-example learning strategy is adopted, which makes the particles move towards their own optimal position and the global situation. While learning the optimal position, learning from different dimensional examples of other particles helps to enhance the diversity of the algorithm and avoid falling into the local optimal position.
  • this method adopts the knowledge embedding auxiliary strategy when generating the initial particle population, and uses the distribution characteristics of the input traffic flow at the intersection as the pre-knowledge to assist the generation of the initial population.
  • the optimization results of traffic flow at a single intersection with different saturation levels show that compared with other timing optimization methods, this method has better diversity under the premise of ensuring the convergence speed, and the optimized timing scheme has better comprehensive performance.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Biophysics (AREA)
  • Economics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Educational Administration (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Primary Health Care (AREA)
  • Game Theory and Decision Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Quality & Reliability (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Traffic Control Systems (AREA)

Abstract

本发明公开了一种基于多榜样学习粒子群的智慧城市信号灯配时优化方法，主要涉及智慧交通控制和智能优化算法领域。本方法针对传统粒子群算法应用于信号灯优化配时场景时容易陷入局部最优的问题，采用一种新的多榜样学习策略，该策略使粒子在向自身最优位置和全局最优位置学习的同时向其他粒子的不同维度榜样学习，有助于增强算法多样性避免陷入局部最优位置。此外，本方法在生成初始粒子种群时采用了知识嵌入辅助策略，将交叉路口输入车流量的分布特征作为前置知识，用于辅助初始种群的生成。在单交叉路口不同饱和度车流量的优化结果显示，本方法相对于其他配时优化方法在保证收敛速度前提下多样性更好，优化得到的配时方案综合表现能力更好。

Description

基于多榜样学习粒子群的智慧城市信号灯配时优化方法 技术领域
本发明涉及智慧交通控制和智能优化算法领域，具体涉及一种基于多榜样学习粒子群的智慧城市信号灯配时优化方法。
背景技术
随着城市化的发展，城市居民的汽车保有量在逐年提高，城市道路尤其是交叉路口面临的交通压力也在逐渐增大。设置交通信号灯能加强道路交通管理，有效地解决车流冲突的问题，但随着道路车流量饱和度的增加，道路通行情况变得日益复杂。传统信号灯配时方案如韦伯斯特配时方案，是基于最小化路口延迟的目标建立交通数学模型，并以此模型计算出最优的信号灯周期，再划分相应的绿灯时间。但韦伯斯特模型仅适用于不饱和车流量情况下的交通模型，因此，当道路车流量达到过饱和时，韦伯斯特配时方案就不再适用，传统信号灯配时方案已经越来越不能满足智慧城市交通的需求。
作为一种重要的进化优化算法，粒子群算法在电力系统、医学图像配准、多目标优化、机器学习等各个领域都显示出优异的性能。粒子群算法拥有较强的全局搜索能力以及收敛能力，适用于不同饱和度情况下的最佳信号灯配时方案的搜索优化，因此许多研究者也将粒子群算法应用于智慧交通的信号配时优化领域。
现有的用于交通信号优化的粒子群算法大多存在容易陷入局部最优解的缺陷。这是由于粒子群是在连续空间里面搜索最优值，而最 终用于配时的信号灯方案是离散的整数值，前者与后者存在取整关系，因此，当粒子的某个维度在局部最优解处收敛到一定程度的时候，便很难在这一维度跳出局部最优位置。
此外，现有研究大多没有将路口输入的流量分布特征用于辅助粒子群算法进行信号灯优化。传统的粒子群优化算法在初始化种群时是完全随机初始化，但在交通信号配时优化领域，如果我们获取了车流量输入的分布特征，我们利用上这些信息，便可以有目的性地生成更有可能覆盖最优解所在区域的初始种群，从而增强算法搜索能力。因此，基于以上分析，传统粒子群优化算法亟待加以改进并整合交通信号配时领域固有特征知识，以增强算法用于交通信号配时领域的优化能力。
发明内容
本发明的目的是为了解决现有技术中的上述缺陷，提供一种基于多榜样学习粒子群的智慧城市信号灯配时优化方法，该方法采用多榜样学习的粒子速度更新策略，使得粒子群算法多样性得到增强，有助于避免粒子陷入局部最优位置。此外，在初始化粒子种群时将交叉路口输入的车流量分布特征作为前置知识使初始种群分布更有可能覆盖最优解所在的区域。
本发明的目的可以通过采取如下技术方案达到：
一种基于多榜样学习粒子群的智慧城市信号灯配时优化方法，所述的智慧城市信号灯配时优化方法包括下列步骤：
S1、交叉路口中各个进道口的车流量大小和比例是影响交通信号灯设置的最大因素，现有的利用智能算法优化交通信号灯配时问题的方法中，将信号灯配时优化作为一个随机优化问题，并没有将这一重 要因素考虑进来，在本发明中，根据交叉路口各个进道口车流量分布特征生成初始种子解，再根据初始种子解生成初始粒子群种群，过程如下：
S101、由于各个进道口的车道数不一定相同，即对于两个车道数不同但输入车流量相同的进道口而言，车道数量少的进道口将会更加“拥堵”，理论上需要为此进道口分配相对更多的绿灯通行时间，因此需要定义一个平均车道流量比率指标来衡量车流量的分布比例，平均车道流量比率定义如下式：
Figure PCTCN2021119200-appb-000001
Figure PCTCN2021119200-appb-000002
其中，ALF j表示第j个进道口的平均车道流量大小，VI j表示第j个进道口的输入车流量大小，LN j表示第j个进道口的车道数，R j表示第j个进道口的平均车道流量比率；
S102、根据整个交叉路口的总输入车流量大小与交叉口设计所能容纳的最大车流量的占比关系计算一个初始总绿灯时长，再基于平均车道流量比率为各个进道口分配绿灯时长生成初始种子解，初始种子解各维度的绿灯时间由以下公式计算：
Figure PCTCN2021119200-appb-000003
Figure PCTCN2021119200-appb-000004
其中g′ j表示第j个进道口的种子解的绿灯时间，G′表示种子解的总绿灯时长，G表示允许的最大信号灯循环周期下的总绿灯时长，VI max表示交叉路口允许的最大输入车流量，
Figure PCTCN2021119200-appb-000005
表示向下求整运算符号；
S103、初始种子解代表潜在最佳配时方案可能存在的区域，基于初始种子解各维度值，增加合适的高斯扰动，生成覆盖在种子解附近的初始粒子种群，具体按照下式生成初始种群：
Figure PCTCN2021119200-appb-000006
其中
Figure PCTCN2021119200-appb-000007
表示种群中第i个粒子的第j维度，N(0,5)表示均值为0，标准差为5的高斯分布；
S2、对生成的粒子群个体进行约束修正，使其满足可行解的约束需求，包括最大、最小绿灯时间约束，最大、最小周期时间约束；
S3、生成初始种群后，即可按照粒子群优化算法的迭代步骤对种群的速度和位置进行更新，然而传统的粒子群算法存在粒子容易陷入局部最优解的缺点，本发明在粒子速度更新环节中增加一个多榜样学习策略用于克服这一缺点，种群的粒子根据多榜样学习策略进行速度更新，更新公式如下：
Figure PCTCN2021119200-appb-000008
其中
Figure PCTCN2021119200-appb-000009
表示第t+1代中第i个粒子第j维的速度，
Figure PCTCN2021119200-appb-000010
表示第t代中第i个粒子第j维的速度，
Figure PCTCN2021119200-appb-000011
表示第t代中第i个粒子第j维的位置值，ω表示速度权重，
Figure PCTCN2021119200-appb-000012
表示第i个粒子第j维的历史最优位置，gBest j表示第j维度的全局最优位置，
Figure PCTCN2021119200-appb-000013
表示第i个粒子第j维的榜样个体值，更具体地，对于第i个粒子的第j维榜样值，是通过随机挑选另外两个粒子，选取适应值较优粒子的历史最佳位置的第j维度值作为第i个粒子在该维度下的榜样值，c 1，c 2和c 3是三个更新系数，
Figure PCTCN2021119200-appb-000014
和
Figure PCTCN2021119200-appb-000015
是第j维度[0,1]之间的三个不同随机数，粒子位 置值更新是原位置加上公式(6)中更新后的新速度作为粒子新位置；
S4、通过微观仿真软件VISSIM对粒子位置值向下取整所代表的交通信号灯配时方案进行仿真评估，将评估结果作为该粒子的适应值；
S5、当算法迭代到规定的最大迭代次数时，将此时的全局最优方案作为最终配时方案，否则继续执行步骤S3～S4直到满足算法事先设置的最大迭代次数终止。
进一步地，所述的步骤S1、根据交叉路口各个进道口车流量分布特征生成初始种子解中，设定一个知识占比阈值KT，在生成种群个体时，先取一个[0,1]之间的随机数，若该随机数小于KT则利用知识辅助策略生成个体，否则按照随机方法生成个体。
进一步地，所述的步骤S3中，设定一个榜样更新间隔阈值RT，若算法全局最优位置在经过RT代都没有改进，则进行榜样更新，否则继续沿用前一代的榜样。
进一步地，所述的更新系数c 1、c 2和c 3分别设置为0.75、0.75和1.50。
进一步地，所述的知识占比阈值KT设置为0.8。
进一步地，所述的榜样更新间隔阈值RT设置为7。
进一步地，所述的最大迭代次数设置为40代。
本发明相对于现有技术具有如下的优点及有益效果：
1、本发明通过采用交叉路口车流量分布特征作为前置知识辅助粒子种群的初始化，有助于初始种群分布在更有可能覆盖最优解所在的区域。
2、本发明提出的多榜样学习策略有助于粒子的不同维度分别向不同榜样学习，进而增强算法多样性，避免粒子在搜索过程中陷入局部最优位置。
3、本发明提出的初始粒子种群生成过程中，使用一个知识占比阈值控制知识辅助生成解和随机解的比例，该方式有助于整合知识嵌入辅助策略和随机搜索策略各自的优势。
4、本发明提出的多榜样学习策略中，使用一个榜样更新阈值控制榜样的更新频率，该方式一方面有助于避免频繁更新粒子的学习榜样而导致粒子搜索难以收敛问题，另一方面有助于及时更新不能继续引导粒子进化的老旧榜样个体。
附图说明
图1是本发明实施例中多榜样学习粒子群算法流程图；
图2是本发明实施例中中国深圳市宝安区新湖路-裕安一路交叉路口几何平面图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
实施例
参见图1，该基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其操作步骤如下：
S1、选取如图2所示的深圳新湖路-裕安一路交叉路口信号灯用于配时优化。该交叉路口是四向十字交叉路口，采用四相位信号灯控制，每个相位的绿灯时长作为粒子的维度值，根据交叉路口的四个进道口输入车流量，定义平均车道流量比率如下式：
Figure PCTCN2021119200-appb-000016
Figure PCTCN2021119200-appb-000017
其中，ALF j表示第j个进道口的平均车道流量大小，VI j表示第j个进道口的输入车流量大小，LN j表示第j个进道口的车道数，R j表示第j个进道口的平均车道流量比率；
然后基于平均车道流量比率按照下式生成初始种子解，初始种子解各维度的绿灯时间由以下公式计算：
Figure PCTCN2021119200-appb-000018
Figure PCTCN2021119200-appb-000019
其中，g′ j表示第j个进道口的种子解的绿灯时间，G′表示种子解的总绿灯时长，G表示允许的最大信号灯循环周期下的总绿灯时长，VI max表示交叉路口允许的最大输入车流量，
Figure PCTCN2021119200-appb-000020
表示向下求整运算符号；
最后基于初始种子解各维度值按照下式生成初始种群：
Figure PCTCN2021119200-appb-000021
其中
Figure PCTCN2021119200-appb-000022
表示种群中第i个粒子的第j维度，N(0,5)表示均值为0，标准差为5的高斯分布；
初始化种群时设置了一个知识占比阈值KT，在此实施例中设置 为0.8，在每次采用公式(11)生成初始粒子时，先生成一个在[0,1]范围内的随机数，若该随机数小于KT则用公式(11)生成初始粒子，否则随机初始化该粒子；
S2、步骤S1中初始种群的大部分粒子是通过基于种子解进行高斯采样生成，因此有可能出现各个维度值不满足可行解要求的情况，因此需要对生成的粒子群个体进行约束修正，使其满足可行解的约束需求；
S3、在粒子进行速度更新前，先判断粒子的榜样个体需不需要进行更新。在此设定一个榜样更新阈值RT，在此实施例中设置为7。若算法的全局最优位置在经过RT代都没有改进，则要先更新榜样再进行粒子的速度和位置更新。榜样的更新过程是，对于榜样个体的第j维，挑选另外两个种群中的不同粒子，取其中适应值好的粒子的历史最优位置的第j维作为榜样个体的第j维。若算法的全局最优位置在在RT代内得到改进，说明该榜样个体还能继续指导粒子进化，则无需进行榜样更新，直接进行粒子的速度和位置更新，
粒子采用多榜样学习策略进行速度更新，更新公式如下：
Figure PCTCN2021119200-appb-000023
其中
Figure PCTCN2021119200-appb-000024
表示第t+1代中第i个粒子第j维的速度，
Figure PCTCN2021119200-appb-000025
表示第t代中第i个粒子第j维的速度，
Figure PCTCN2021119200-appb-000026
表示第t代中第i个粒子第j维的位置值，ω表示速度权重，
Figure PCTCN2021119200-appb-000027
表示第i个粒子第j维的历史最优位置，gBest j表示第j维度的全局最优位置，
Figure PCTCN2021119200-appb-000028
表示第i个粒子第j维的榜样个体值，c 1，c 2和c 3是三个更新系数，
Figure PCTCN2021119200-appb-000029
和
Figure PCTCN2021119200-appb-000030
是第j维 度[0,1]之间的三个不同随机数，粒子位置值更新是原位置加上公式(12)中更新后的新速度作为粒子新位置；
S4、完成粒子速度位置更新后，通过微观仿真软件VISSIM对粒子位置值向下取整所代表的交通信号灯配时方案进行仿真评估，将评估结果作为该粒子的适应值；
S5、当算法迭代到规定的最大迭代次数时，在此实施例中设置为40代，将此时的全局最优解向下取整作为最终配时方案，否则继续执行步骤S3～S4直到满足算法终止条件，即最大迭代次数40代；
综上所述，本方法针对传统粒子群算法应用于信号灯优化配时场景时容易陷入局部最优的问题，采用一种新的多榜样学习策略，该策略使粒子在向自身最优位置和全局最优位置学习的同时向其他粒子的不同维度榜样学习，有助于增强算法多样性避免陷入局部最优位置。此外，本方法在生成初始粒子种群时采用了知识嵌入辅助策略，将交叉路口输入车流量的分布特征作为前置知识，用于辅助初始种群的生成。在单交叉路口不同饱和度车流量的优化结果显示，本方法相对于其他配时优化方法在保证收敛速度前提下多样性更好，优化得到的配时方案综合表现能力更好。
上述实施例为本发明较佳的实施方式，但本发明的实施方式并不受上述实施例的限制，其他的任何未背离本发明的精神实质与原理下所作的改变、修饰、替代、组合、简化，均应为等效的置换方式，都包含在本发明的保护范围之内。

Claims (7)

  1. 一种基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的智慧城市信号灯配时优化方法包括下列步骤：
    S1、根据交叉路口各个进道口车流量分布特征生成初始种子解，再根据初始种子解生成初始粒子群种群，过程如下：
    S101、平均车道流量比率定义如下式：
    Figure PCTCN2021119200-appb-100001
    Figure PCTCN2021119200-appb-100002
    其中，ALF j表示第j个进道口的平均车道流量大小，VI j表示第j个进道口的输入车流量大小，LN j表示第j个进道口的车道数，R j表示第j个进道口的平均车道流量比率；
    S102、基于平均车道流量比率按照下式生成初始种子解，初始种子解各维度的绿灯时间由以下公式计算：
    Figure PCTCN2021119200-appb-100003
    Figure PCTCN2021119200-appb-100004
    其中g′ j表示第j个进道口的种子解的绿灯时间，G′表示种子解的总绿灯时长，G表示允许的最大信号灯循环周期下的总绿灯时长，VI max表示交叉路口允许的最大输入车流量，
    Figure PCTCN2021119200-appb-100005
    表示向下求整运算符号；
    S103、基于初始种子解各维度值按照下式生成初始种群：
    Figure PCTCN2021119200-appb-100006
    其中
    Figure PCTCN2021119200-appb-100007
    表示种群中第i个粒子的第j维度，N(0,5)表示均值为0，标准差为5的高斯分布；
    S2、对生成的粒子群个体进行约束修正，使其满足可行解的约束需求；
    S3、种群的粒子根据多榜样学习策略进行速度更新，更新公式如下：
    Figure PCTCN2021119200-appb-100008
    其中
    Figure PCTCN2021119200-appb-100009
    表示第t+1代中第i个粒子第j维的速度，
    Figure PCTCN2021119200-appb-100010
    表示第t代中第i个粒子第j维的速度，
    Figure PCTCN2021119200-appb-100011
    表示第t代中第i个粒子第j维的位置值，ω表示速度权重，
    Figure PCTCN2021119200-appb-100012
    表示第i个粒子第j维的历史最优位置，gBest j表示第j维度的全局最优位置，
    Figure PCTCN2021119200-appb-100013
    表示第i个粒子第j维的榜样个体值，c 1，c 2和c 3是三个更新系数，
    Figure PCTCN2021119200-appb-100014
    和
    Figure PCTCN2021119200-appb-100015
    是第j维度[0,1]之间的三个不同随机数；
    S4、通过微观仿真软件VISSIM对粒子位置值向下取整所代表的交通信号灯配时方案进行仿真评估，将评估结果作为该粒子的适应值；
    S5、当算法迭代到规定的最大迭代次数时，将此时的全局最优方案作为最终配时方案，否则继续执行步骤S3～S4直到满足算法事先设置的最大迭代次数终止。
  2. 根据权利要求1所述的基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的步骤S1、根据交叉路口各个进道口车流量分布特征生成初始种子解中，设定一个知识占比阈值KT，在生成种群个体时，先取一个[0,1]之间的随机数，若该随机数小 于KT则利用知识辅助策略生成个体，否则按照随机方法生成个体。
  3. 根据权利要求1所述的基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的步骤S3中，设定一个榜样更新间隔阈值RT，若算法全局最优位置在经过RT代都没有改进，则进行榜样更新，否则继续沿用前一代的榜样。
  4. 根据权利要求1所述的基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的更新系数c 1、c 2和c 3分别设置为0.75、0.75和1.50。
  5. 根据权利要求2所述的基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的知识占比阈值KT设置为0.8。
  6. 根据权利要求3所述的基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的榜样更新间隔阈值RT设置为7。
  7. 根据权利要求1所述的基于多榜样学习粒子群的智慧城市信号灯配时优化方法，其特征在于，所述的最大迭代次数设置为40代。
PCT/CN2021/119200 2021-03-11 2021-09-18 基于多榜样学习粒子群的智慧城市信号灯配时优化方法 WO2022188387A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110271046.7A CN113012449B (zh) 2021-03-11 2021-03-11 基于多榜样学习粒子群的智慧城市信号灯配时优化方法
CN202110271046.7 2021-03-11

Publications (1)

Publication Number Publication Date
WO2022188387A1 true WO2022188387A1 (zh) 2022-09-15

Family

ID=76406286

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/119200 WO2022188387A1 (zh) 2021-03-11 2021-09-18 基于多榜样学习粒子群的智慧城市信号灯配时优化方法

Country Status (3)

Country Link
CN (1) CN113012449B (zh)
LU (1) LU503042B1 (zh)
WO (1) WO2022188387A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116434576A (zh) * 2022-12-12 2023-07-14 中电信数字城市科技有限公司 一种交通灯配时方案确定方法、装置、系统及设备
CN117077041A (zh) * 2023-10-16 2023-11-17 社区魔方(湖南)数字科技有限公司 基于物联网的智慧社区管理方法及系统
CN117523823A (zh) * 2023-10-11 2024-02-06 吉林师范大学 一种基于量子遗传算法的区域交通信号控制优化方法

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113012449B (zh) * 2021-03-11 2022-03-29 华南理工大学 基于多榜样学习粒子群的智慧城市信号灯配时优化方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1889114A (zh) * 2006-07-17 2007-01-03 中国科学院地理科学与资源研究所 基于粒子群算法的交通信号离线配时优化方法
CN104134356A (zh) * 2014-06-30 2014-11-05 南京航空航天大学 城市交叉口模型参考自适应信号的控制方法
CN106384521A (zh) * 2016-09-18 2017-02-08 广东工业大学 一种基于公交优先的单交叉口交通信号优化控制方法
CN109035811A (zh) * 2018-08-28 2018-12-18 大连理工大学 一种基于数字信息素的智能交通信号灯实时调控方法
CN113012449A (zh) * 2021-03-11 2021-06-22 华南理工大学 基于多榜样学习粒子群的智慧城市信号灯配时优化方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106022463B (zh) * 2016-05-13 2018-11-13 安徽教育网络出版有限公司 基于改进粒子群算法的个性化学习路径优化方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1889114A (zh) * 2006-07-17 2007-01-03 中国科学院地理科学与资源研究所 基于粒子群算法的交通信号离线配时优化方法
CN104134356A (zh) * 2014-06-30 2014-11-05 南京航空航天大学 城市交叉口模型参考自适应信号的控制方法
CN106384521A (zh) * 2016-09-18 2017-02-08 广东工业大学 一种基于公交优先的单交叉口交通信号优化控制方法
CN109035811A (zh) * 2018-08-28 2018-12-18 大连理工大学 一种基于数字信息素的智能交通信号灯实时调控方法
CN113012449A (zh) * 2021-03-11 2021-06-22 华南理工大学 基于多榜样学习粒子群的智慧城市信号灯配时优化方法

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116434576A (zh) * 2022-12-12 2023-07-14 中电信数字城市科技有限公司 一种交通灯配时方案确定方法、装置、系统及设备
CN116434576B (zh) * 2022-12-12 2024-03-12 中电信数字城市科技有限公司 一种交通灯配时方案确定方法、装置、系统及设备
CN117523823A (zh) * 2023-10-11 2024-02-06 吉林师范大学 一种基于量子遗传算法的区域交通信号控制优化方法
CN117077041A (zh) * 2023-10-16 2023-11-17 社区魔方(湖南)数字科技有限公司 基于物联网的智慧社区管理方法及系统
CN117077041B (zh) * 2023-10-16 2023-12-26 社区魔方(湖南)数字科技有限公司 基于物联网的智慧社区管理方法及系统

Also Published As

Publication number Publication date
LU503042B1 (en) 2023-03-13
CN113012449A (zh) 2021-06-22
CN113012449B (zh) 2022-03-29

Similar Documents

Publication Publication Date Title
WO2022188387A1 (zh) 基于多榜样学习粒子群的智慧城市信号灯配时优化方法
CN111696370B (zh) 基于启发式深度q网络的交通灯控制方法
WO2022121510A1 (zh) 基于随机策略梯度的交通信号控制方法、系统及电子设备
CN111260937A (zh) 一种基于强化学习的十字路口交通信号灯控制方法
CN112700664A (zh) 一种基于深度强化学习的交通信号配时优化方法
CN110097755A (zh) 基于深度神经网络的高速公路交通流量状态识别方法
WO2021189668A1 (zh) 基于排队消散时间的路口交通信号方案优化方法
WO2021051870A1 (zh) 基于强化学习模型的信息控制方法、装置和计算机设备
CN113223305B (zh) 基于强化学习的多路口交通灯控制方法、系统及存储介质
CN112216127B (zh) 一种基于近端策略优化的小型路网交通信号优化方法
CN111814079B (zh) 结合聚类与cmab的群智感知车联网用户筛选方法
CN107393319B (zh) 一种防止单交叉口排队溢出的信号优化控制方法
CN113963553A (zh) 一种道路交叉口信号灯绿信比控制方法、装置和设备
CN108898858A (zh) 一种过饱和交通状态下连续交叉口的信号协调控制方法
CN113780624B (zh) 一种基于博弈均衡理论的城市路网信号协调控制方法
CN106384521A (zh) 一种基于公交优先的单交叉口交通信号优化控制方法
CN115578870B (zh) 一种基于近端策略优化的交通信号控制方法
CN112071062A (zh) 一种基于图卷积网络和图注意力网络的行车时间估计方法
CN111126687A (zh) 一种交通信号的单点离线优化系统及方法
CN114613169B (zh) 一种基于双经验池dqn的交通信号灯控制方法
CN115903485A (zh) 一种具有适应性的dqn模型优化方法及应用
CN117351729A (zh) 黑塞矩阵协助梯度优化的城市交叉口交通信号实时控制方法
CN110021168B (zh) 一种车联网下实现实时智能交通管理的分级决策方法
CN115472023B (zh) 一种基于深度强化学习的智能交通灯控制方法及装置
Wei et al. Intersection signal control approach based on pso and simulation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21929852

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE