WO2020133721A1 - 一种基于非参数贝叶斯框架的信号交叉口状态估计方法 - Google Patents

一种基于非参数贝叶斯框架的信号交叉口状态估计方法 Download PDF

Info

Publication number
WO2020133721A1
WO2020133721A1 PCT/CN2019/078187 CN2019078187W WO2020133721A1 WO 2020133721 A1 WO2020133721 A1 WO 2020133721A1 CN 2019078187 W CN2019078187 W CN 2019078187W WO 2020133721 A1 WO2020133721 A1 WO 2020133721A1
Authority
WO
WIPO (PCT)
Prior art keywords
state
model
data
estimation
measurement
Prior art date
Application number
PCT/CN2019/078187
Other languages
English (en)
French (fr)
Inventor
金峻臣
王辉
杨宪赞
李瑶
周浩敏
郭海锋
Original Assignee
银江股份有限公司
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 银江股份有限公司 filed Critical 银江股份有限公司
Publication of WO2020133721A1 publication Critical patent/WO2020133721A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125Traffic data processing
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0137Measuring and analyzing of parameters relative to traffic conditions for specific applications

Definitions

  • the invention belongs to the field of traffic control, and relates to a signal intersection state estimation method based on a non-parametric Bayesian framework.
  • Traffic State Estimation refers to the process of inferring changes in traffic state using part of traffic observation data with noise. These data are obtained from various monitoring technologies. Signal-controlled intersections are an indispensable component of urban traffic networks. Accurate and practical TSE methods play an important role in the planning and operation of signal-controlled intersections, which can effectively alleviate traffic congestion. Especially for traditional signal control systems, estimating the traffic state is of great significance for measuring the performance of intersections and further optimizing signal control schemes. In addition, for most emerging adaptive traffic signal control systems, their basic idea is to understand the evolution of the traffic state. In general, the more precise and advanced the signal control system, the more accurate and frequent traffic status data is required.
  • TSE methods can be divided into two categories based on different assumptions and input data that they rely on.
  • One is model-driven methods, and the other is data-driven methods.
  • the model-driven TSE method relies on the physical model of the transportation system, which is characterized by the need for empirical relationships and the need for careful model selection and calibration procedures. In specific cases, a large amount of data is required to test the rationality of the model or to calibrate the model.
  • the data-driven TSE method must consider historical data under various traffic conditions, otherwise the method may fail if an emergency occurs. In addition, the cost of training and learning may be relatively high. However, with the continuous development of data and sensing technologies, data-driven models have attracted more and more attention.
  • the present invention is to provide a signal intersection state estimation method based on a non-parametric Bayesian framework, does not require an accurate traffic model, is driven by data, has a wide range of applications, and has an estimated accuracy rate high.
  • a state estimation method of signalized intersection based on non-parametric Bayesian framework the steps are as follows:
  • Intersection state estimation the extended Kalman filter is used to linearize the transfer model and measurement model, and then the traffic state and signal control parameters at the previous moment are input to the transfer model to obtain the predicted state and its covariance, and then the The obtained predicted state, its covariance, and the measured value at the current time are input to the measurement model to predict the best estimated value of the state at the current time.
  • step (4) verification of the state estimation method the introduction of connected car data, calculation of intersection state estimation accuracy at different estimation intervals and connected vehicle penetration rates.
  • state data is traffic flow data
  • state vector is expressed as:
  • n k,t represents the number of vehicles on the kth lane at time t
  • N lane represents the total number of lanes at the intersection
  • the control data is the green signal ratio, and the control vector is expressed as:
  • g k,t refers to the green signal ratio controlling the k-th lane at time t within the defined estimation interval of traffic lights.
  • the transition model is expressed as:
  • g( ⁇ ) represents a mapping between the state-control pair (x t-1 , u t-1 ) at time t-1 and the state x t at time t , and ⁇ follows the covariance matrix of ⁇ tran and the mean is zero White Gaussian distribution process;
  • the state transition probability is as follows:
  • the measurement model is expressed as:
  • h ( ⁇ ) represents the state at time t and the measured values X t mapping between z t, ⁇ subject covariance matrix ⁇ meas, white Gaussian process with zero mean;
  • the measurement probability is as follows:
  • ⁇ meas represents the covariance matrix of the measurement model.
  • the number of elements of the state, control, and measurement vectors are respectively represented by M x , Mu , and M z , and the data points of the three are the same size, then the training data sets of the transfer model and the measurement model are respectively for:
  • X tran and Y tran are the data points of the input and output of the transfer model, X meas and Y meas are the input and output data points of the measurement model;
  • z q,t refers to the q-th element of the measurement vector
  • linearization of the transfer model and the measurement model specifically uses an extended Kalman filter, uses a first-order Taylor expansion to construct the function value, and slopes to approximate the function linearly.
  • predicting the optimal estimate of the state at the current moment also includes: calculating the Kalman gain using the predicted state and the covariance matrix, adding the predicted state to the new state estimate based on the correctness of the measured value, and calculating the optimal State, the degree of accuracy is proportional to the Kalman gain, and proportional to the deviation between the current measured value and the predicted measured value.
  • non-parametric Bayesian framework adds estimated interval and vehicle permeability parameters, specifically:
  • measuring elements can be calculated by the following formula:
  • r k,t refers to the penetration rate of the connected vehicle at time t
  • ⁇ t refers to the estimated interval duration
  • represents the length of time to update the permeability
  • the verification parameters for estimating the accuracy of the intersection state estimation include: mean absolute error (MAE) and weighted mean absolute error (WAPE), the calculation formula is as follows:
  • T is the total time period of the sequence
  • MAE refers to the average absolute error that can be expected in the estimation method. WAPE allows the comparison of estimates under different conditions.
  • the design of the frame is compatible with the traditional signal control system and the adaptive signal control system.
  • the generated lane-based state can be used for the signal controller based on the group or lane; the frame is not limited by the data type, that is, the fixed position detection data and Mobile data can be used; the included model is non-parametric and does not require prior knowledge of setting parameters.
  • Figure 1 is the offline training process of Bayesian filter based on Gaussian process.
  • Figure 2 is the online estimation process of Bayesian filter based on Gaussian process.
  • Figure 3 is the pseudocode for one-step state estimation using the BFGP modeling framework.
  • Fig. 4 is the pseudo code of the detailed steps of the extended Kalman filter based on the Gaussian process.
  • Figure 5 is a typical independent intersection layout.
  • Fig. 6 is a semaphore phase sequence.
  • This embodiment provides a method for estimating the state of a signalized intersection based on a non-parametric Bayesian framework. The steps are as follows:
  • the problem of recursive state estimation is to determine the most likely state in a certain period of time given all the past measurement and control inputs. Therefore, the probability rules describing the evolution of the state are determined by the probability distribution conditioned on the measurement and control signals.
  • the posterior probability of the state variable of the probability distribution is referred to as belief distribution for short. Assuming that the system starts from the initial state x 0 and performs the initial control u 0 , the first measurement vector is defined as z 1 . Then the belief distribution bel(x t ) of the state variable x t at time t is expressed as follows:.
  • the initial belief distribution which refers to the prediction of the state at time t based on the previous state posterior distribution bel(x t-1 ) before combining the measured values.
  • the estimated bel(x t ) is usually called measurement correction or measurement update. Therefore, the realization of the Bayesian filter requires three probability distributions: state transition probability P(x t
  • the state transition model is usually expressed as:
  • h( ⁇ ) refers to the mapping between the state x t and the measured value z t at time t .
  • X is an N ⁇ M matrix
  • the rows of the matrix are a 1 ⁇ M vector, representing the input data.
  • y represents the training output, which is an N ⁇ 1 matrix, and the training data is represented as (12)(13):
  • x i and y i are the column vector and scalar value of the i-th training data set.
  • K represents the covariance matrix determined by the kernel function of the input data, and its elements are expressed as follows:
  • K( ⁇ ) is a kernel function, indicating the degree of similarity between data points. Specifically, if two data points (x i , x j ) are more similar, then their instance values (f i , f j ) are more related. The degree of similarity depends on the differences between the applied variables.
  • the joint distribution of output variables is a multivariate Gaussian distribution, including new variables conditioned on input data points and hyperparameters. which is:
  • c N+1 is an N ⁇ 1 matrix, defined as follows:
  • c N+1 [C(x 1 ,x N+1 , ⁇ , ⁇ 1 ),C(x 2 ,x N+1 , ⁇ , ⁇ 2 ),...,C(x N ,x N+ 1 , ⁇ , ⁇ N+1 )) T (26)
  • ⁇ (x N+1 ,D) and ⁇ (x N+1 ,D) refer to the mean and variance functions, respectively, as shown in (28)(29) below:
  • X tran and Y tran are the input and output data points of the transfer model
  • X meas and Y meas are the input and output data points of the measurement model.
  • Each input data point of the transfer model contains a state and control vector, which is a (M x +M u ) ⁇ 1 column vector, defined as follows:
  • z q,t refers to the q-th element of the measurement vector
  • the extended Kalman filter is used to linearize the transfer model and measurement model, and then the traffic state and signal control parameters at the previous moment are input to the transfer model to obtain the predicted state and its covariance, and then the obtained The predicted state and its covariance, and the measured value at the current time are input to the measurement model to predict the best estimated value of the state at the current time, see Figure 2.
  • Figure 3 shows the one-step estimation process of new observations.
  • the BF one-step estimation process also requires the state, covariance, and control data that were estimated at the previous moment. If the historical data sets D tran and D meas exist, then the historical data set can be used to update the transfer and measurement models.
  • the Kalman filter When performing state estimation and prediction in a state space model, the Kalman filter is a widely used framework. However, equations (34) and (35) show that the transfer and measurement models are nonlinear functions. In order to deal with this nonlinearity, many algorithms are proposed under the BF modeling framework, such as extended Kalman filter, unscented Kalman filter, and particle filter. The present invention chooses to apply an extended Kalman filter and uses the first-stage Taylor expansion to linearize the nonlinear function based on GP.
  • An extended Kalman filter based on a Gaussian process uses the last estimated state x t-1 , covariance ⁇ t-1 , control u t-1 , and current measured value z t-1 to predict the current state x t and its covariance matrix ⁇ t .
  • Figure 4 shows the steps of DPEKF in detail using pseudocode, and then describes the mathematical rules of each step.
  • Lines 1 and 2 represent prediction steps, and lines 3, 4, and 5 represent update steps.
  • the predicted state s t can be composed of the following matrix:
  • a first-order Taylor expansion can be used to construct the function value and the slope to linearly approximate the function.
  • the covariance matrix of the predicted state is calculated by (38) below:
  • the third line in the pseudocode indicates that the predicted state and the covariance matrix are used to calculate the Kalman gain.
  • the specific expression is shown in (39):
  • H t represents the Jacobian determinant of the GP mean function in the measurement model:
  • the current estimated state is (41):
  • the invention utilizes information gain Adjust the covariance matrix of the predicted state to update the covariance matrix of the estimated state, such as (42):
  • the TSE method described in the present invention is verified using connected vehicle traffic data, and the traffic state data comes from a micro simulator.
  • the experiment applies the estimation method to a typical, independent intersection, see Figure 5.
  • the signal controller at this intersection uses phase-based phases, in which the traffic lights operate in a fixed phase sequence, see Figure 6 for the phase sequence.
  • the induction detector includes a short induction detector and a long induction detector, which are placed at 80 meters and 10 meters away from the parking line, respectively.
  • the present invention adopts a "vehicle drive" signal timing method, that is, the green light distribution time varies according to the number of vehicles detected by the loop detector.
  • the elements of the state vector represent the number of vehicles, including queued vehicles, approaching vehicles, and vehicles in the lanes associated with intersections. The manner in which this state is defined has been applied to multiple adaptive signal control systems.
  • the state vector is shown in equation (43):
  • N lane 12
  • Control data is obtained by collecting general information (green light, yellow light, red light) indicated by traffic lights, and any type of signal controller can access these data. In the process of state transition, green and non-green light indicators will have an important impact on the number of vehicles in the lane.
  • the control vector is as follows (44):
  • g k,t refers to the green signal ratio controlling the k-th lane at time t within the defined estimation interval of traffic lights.
  • the framework uses Internet of Vehicles data sources.
  • V2I vehicle-to-vehicle interface
  • the signal controller can access the vehicle location.
  • the number of connected vehicles in each lane can be extracted in real time.
  • each connected vehicle enters an intersection and an ID is given.
  • the signal controller is “responsible” to record the vehicle ID, and the flow of connected vehicles within a certain time interval is equal to the number of unique vehicle IDs.
  • ⁇ t refers to the estimated interval duration
  • r k,t refers to a rough estimate of the penetration rate of the connected vehicle at time t
  • the recursive update equation is as follows:
  • represents the length of time to update the permeability. Represents the number of connected vehicles on lane k at time l, the corresponding measurement vector is defined as follows (47):
  • T is the total time period of the sequence
  • I the observed state vector.
  • MAE refers to the average absolute error that can be expected in the estimation method. WAPE allows the comparison of estimates under different conditions.
  • the traffic model is built on the open source micro simulator SUMO0.19.0. Then connect the developed and designed signal controller software program with the SUMO simulator, and set the traffic light signal changes in the simulation based on "vehicle drive" control.
  • SUMO records the detection information and the number of vehicles on each lane through the programming interface TraCI provided by the application.
  • a commonly used car tracking model-Intelligent Driver Model (IDM) is used. The parameters of the car following model and signal control parameters are shown in Table 1.
  • the training data set refers to a set of data samples used to discover the potential relationship between state, control, and measurement; according to the performance standards described, the verification set is used to compare model performance or estimated accuracy, and the test set is used to evaluate The effectiveness of the proposed estimation method and provide detailed content of the estimation results.
  • L, T, R represent the left turn rate, straight rate, and right turn rate, respectively.
  • the "uniform" scheme in Table 2 assumes that the traffic flow is the same in all directions at the intersection.
  • the "mainline” scheme considers the north-south direction or the east-west direction to be the main road.
  • the corresponding traffic flow is defined as “medium” or “high” level.
  • the “medium” level indicates that the traffic flow at the intersection is normal, and the “high” level indicates that there is a significant increase in traffic flow (about 20%) compared to the “medium” level.
  • the traffic flow is randomly generated, and the selected values are shown in Table 2.
  • Each verification data set contains 600 simulated data points.
  • the traffic flow starts from the "East-West Main Line (Middle)" scheme, and the pattern of the traffic flow scheme is arranged from "East-West Main Line East (Middle)” ⁇ "East-West Main Line (High)” ⁇ "Uniform (High)” ⁇ "North-South Main Line (High)” ⁇ "Uniform (Middle)” ⁇ "North-South Main Line (Middle)”.
  • Simulate each traffic flow scenario scenario for a duration of 600 seconds, for a total of 3600 seconds. Due to the randomness of traffic simulation, the time of vehicle generation is different.
  • Table 3 and Table 4 summarize the estimation results of the four sets of permeability for the validation set under two typical estimation intervals (ie, 1s and 20s).
  • the TSE framework proposed by the present invention can provide a feasible solution for the location information of networked vehicles when the estimation interval is small (such as 1 s) and the penetration rate is low.
  • the estimation interval will increase, which will lead to a decrease in the effectiveness of the estimation model.

Abstract

一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其步骤如下:(1)数据采集:获取交叉口历史交通数据和对应信号控制参数,向量化处理,分别建立状态数据集和控制数据集;(2)建立非参数贝叶斯框架:结合递归状态估计和高斯过程回归模型,利用所述状态数据集和控制数据集训练并优化转移模型和测量模型;(3)交叉口状态估计:采用扩展的卡尔曼滤波器来线性化转移模型和测量模型,再将上一时刻的交通状态、信号控制参数输入至转移模型得到预测状态及其协方差,然后将得到的预测状态及其协方差、当前时刻的测量值输入到测量模型,预测当前时刻状态的最优估计值。本发明不需要精确的交通模型,由数据驱动,适用范围广,估计准确率高。

Description

一种基于非参数贝叶斯框架的信号交叉口状态估计方法 技术领域
本发明属于交通控制领域,涉及一种基于非参数贝叶斯框架的信号交叉口状态估计方法。
背景技术
交通状态估计(TSE)是指利用带有噪声的部分交通观测数据来推断交通状态变化的过程,这些数据是从各类监控技术中获得的。信号控制交叉口是城市交通网络中不可或缺的组成部分,准确并实用的TSE方法在信控交叉口的规划和运营中发挥着重要的作用,可有效缓解交通拥堵。尤其对于传统的信号控制系统,估计交通状态对衡量交叉口性能和进一步优化信号控制方案有重大意义。此外,对于大多数新兴自适应交通信号控制系统,它们的基本思路就是了解交通状态的演变过程。通常情况下,信号控制系统越精确越先进,就越需要更精准更频繁的交通状态数据。
通常,根据假设不同和所依赖的输入数据,TSE方法可以分为两类,一是模型驱动方法,二是数据驱动方法。简单说来,模型驱动的TSE方法依赖交通系统的物理模型,其特点是需要经验关系,需要仔细选择模型和校准过程。在具体的案例中,检验模型的合理性或者对模型进行校准都需要大量的数据。
而数据驱动的TSE方法必须考虑各种交通状态下的历史数据,否则如果出现突发事件,方法可能会失效。另外,训练和学习的成本可能会比较高。不过随着数据和传感技术的不断发展,数据驱动模型引起了越来越多的关注。
发明内容
为了克服现有技术中存在的不足,本发明在于提供了一种基于非参数贝叶斯框架的信号交叉口状态估计方法,不需要精确的交通模型,由数据驱动,适用范围广,估计准确率高。
本发明采用的技术方案是:
一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其步骤如下:
(1)数据采集:获取交叉口历史交通数据和对应信号控制参数,向量化处理,分别建立状态数据集和控制数据集;
(2)建立非参数贝叶斯框架:结合递归状态估计和高斯过程回归模型,利用所述状态数据集和控制数据集训练并优化转移模型和测量模型;
(3)交叉口状态估计:采用扩展的卡尔曼滤波器来线性化转移模型和测量模型,再将上一时刻的交通状态、信号控制参数输入至转移模型得到预测状态及其协方差,然后将得到的预测状态及其协方差、当前时刻的测量值输入到测量模型,预测当前时刻状态的最优估计值。
进一步,还包括步骤(4)状态估计方法验证:引入车联网数据,计算不同估计间隔和联网车辆渗透率下交叉口状态估计精度。
进一步,所述状态数据为交通流量数据,状态向量表示为:
Figure PCTCN2019078187-appb-000001
其中n k,t表示t时刻第k个车道上的车辆数,N lane表示交叉口车道的总数;
所述控制数据为绿信比,控制向量表示为:
Figure PCTCN2019078187-appb-000002
其中g k,t指在交通灯定义估计区间内控制t时刻第k个车道的绿信比。
进一步,基于递归状态估计,所述转移模型表示为:
x t=g(x t-1,u t-1)+ε
其中g(·)表示t-1时刻状态-控制对(x t-1,u t-1)与t时刻状态x t之间的一个映射,ε服从协方差矩阵为Σ tran、均值为零的白高斯分布过程;
状态转移概率如下式所示:
P(x t|x t-1,u t-1)=N(g(x t-1,u t-1),Σ tran)
所述测量模型表示为:
z t=h(x t)+ζ
其中h(·)表示t时刻状态x t与测量值z t之间的映射,ζ服从协方差矩阵为Σ meas、均值为零的白高斯分布过程;
测量概率如下式所示:
P(z t|x t)=N(h(x t),Σ meas)
其中,Σ meas代表测量模型的协方差矩阵。
进一步,结合高斯过程回归模型,状态、控制、测量向量的元素个数分别由M x,M u,M z表示,且三者数据点规模相同,则转移模型和测量模型的训练数据集分别表示为:
D tran=<X tran,Y tran
D meas=<X meas,Y meas
其中X tran,Y tran是转移模型输入输出的数据点,X meas,Y meas是测量模型的输入输出数据点;
对于t时刻状态向量中的任一个元素x p,t(p=1,2,…,M x),状态向量x t的分布如下:
Figure PCTCN2019078187-appb-000003
其中
Figure PCTCN2019078187-appb-000004
Figure PCTCN2019078187-appb-000005
矩阵的第p个行向量,
Figure PCTCN2019078187-appb-000006
是对应的超参数;
测量变量z t的分布如下:
Figure PCTCN2019078187-appb-000007
其中z q,t指测量向量的第q个元素,
Figure PCTCN2019078187-appb-000008
是Y meas的第q个行向量,
Figure PCTCN2019078187-appb-000009
是第q个测量模型的超参数,
Figure PCTCN2019078187-appb-000010
是第q个测量模型的均值和方差函数。
进一步,转移模型和测量模型的线性化,具体为采用扩展的卡尔曼滤波器,使用一级泰勒展开式构建函数值,斜率来对函数进行线性近似。
进一步,预测当前时刻状态的最优估计值还包括:使用被预测状态和协方差矩阵计算卡尔曼增益,根据测量值的正确性程度将被预测的状态加入到新的状态估计中,计算最优的状态,所述正确性程度与卡尔曼增益成正比,与当前测量值和预测测量值之间的偏差成正比。
进一步,所述非参数贝叶斯框架添加估计间隔和车辆渗透率参数,具体为:
假设
Figure PCTCN2019078187-appb-000011
代表前五分钟内车道k上的联网车流量,测量元素
Figure PCTCN2019078187-appb-000012
可由下式计算:
Figure PCTCN2019078187-appb-000013
其中r k,t是指t时刻联网车辆的渗透率,Δt指估计间隔时长;
递归更新等式如下:
Figure PCTCN2019078187-appb-000014
τ表示更新渗透率的时间长度,
Figure PCTCN2019078187-appb-000015
表示l时刻车道k上的联网车流量,则对应的测量向量定义如下:
Figure PCTCN2019078187-appb-000016
进一步,估计交叉口状态估计精度的验证参数包括:平均绝对误差(MAE)和加权平均绝对误差(WAPE),计算公式如下:
Figure PCTCN2019078187-appb-000017
Figure PCTCN2019078187-appb-000018
其中T是序列的总时间周期,
Figure PCTCN2019078187-appb-000019
是观测的状态向量,从计算公式分析,MAE指的是估计方法中可预期的平均绝对误差,WAPE允许在不同条件下对估计进行比较。
本发明的有益效果:
1、框架的设计与传统信号控制系统和自适应信号控制系统均兼容,生成的基于车道的状 态可用于基于组或车道的信号控制器;框架不受数据类型的限制,即固定位置检测数据和移动数据均可被采用;内含的模型是非参数的,不需要知道设置参数的先验知识。
2、能够为短期交通状态估计(比如1秒)提供一个长期有效的方案,这被认为是一项更具挑战性的任务。如果每秒钟都能以近乎精确的方式来估计状态,那么就可以直观地得到相对长期的TSE估计(如逐周期循环)。此外,还利用联网车辆的不同渗透率进行了敏感性分析,该方法不需要布设昂贵的环路检测器,即可估计交通状态,即使整个交叉路口的少数车辆(仅25%)是联网的,也表现出一定的有效性,所以这是一个非常竞争力的候选方案。
附图说明
图1是基于高斯过程的贝叶斯滤波器的离线训练过程。
图2是基于高斯过程的贝叶斯滤波器的在线估计过程。
图3是使用BFGP建模框架进行一步状态估计的伪代码。
图4是基于高斯过程的扩展卡尔曼滤波器详细步骤的伪码。
图5是典型独立交叉口布局。图6是信号灯相位序列。
具体实施方式
下面结合具体实施例来对本发明进行进一步说明,但并不将本发明局限于这些具体实施方式。本领域技术人员应该认识到,本发明涵盖了权利要求书范围内所可能包括的所有备选方案、改进方案和等效方案。
本实施例提供了一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其步骤如下:
建立非参数贝叶斯框架:结合递归状态估计和高斯过程回归模型,利用所述状态数据集和控制数据集训练并优化转移模型和测量模型,参见图1。
具体如下:
1.1递归状态估计
递归状态估计问题就是在给定所有过去的测量和控制输入的条件下确定某个时间段最可能的状态。因此,描述状态演化的概率规则是由以测量和控制信号为条件的概率分布决定的。该概率分布状态变量的后验概率,简称为信念分布。假设系统从初始状态x 0开始,并执行初始控制u 0,第一次的测量向量被定义为z 1。那么t时刻状态变量x t的信念分布bel(x t)表示如下:。
bel(x t)=P(x t|z 1:t,u 0:t-1)  (1)
那么估计的状态,可由公式(2)得到:
Figure PCTCN2019078187-appb-000020
根据贝叶斯滤波器中的马尔可夫假设,下个状态的条件概率分布仅与当前的状态和控制有关,与之前的事件序列无关。这种状态、控制和测量的演化过程被称为马尔可夫过程模型,由等式(3)(4)表示:
P(x t|x t-1,z 1:t-1,u 0:t-1)=P(x t|x t-1,u t-1)    (3)
P(z t|x t,z 1:t-1,u 0:t-1)=P(z t|x t)   (4)
其中,P(x t|x t-1,u t-1)代表状态转移概率,P(z t|x t)代表测量概率。综上,信念函数可由以下等式(5)(6)进行递归计算:
Figure PCTCN2019078187-appb-000021
Figure PCTCN2019078187-appb-000022
其中
Figure PCTCN2019078187-appb-000023
代表初始信念分布,指结合测量值之前,基于前一个状态后验分布bel(x t-1)的情况下,对t时刻状态的预测。根据
Figure PCTCN2019078187-appb-000024
估计bel(x t),通常称为测量校正或测量更新。因此,贝叶斯滤波器的实现需要三个概率分布:状态转移概率P(x t|x t-1,u t-1),测量概率P(z t|x t),和初始信念P(x 0)。
状态转移模型通常表示为:
x t=g(x t-1,u t-1)+ε   (7)
其中g(·)表示t-1时刻状态-控制对(x t-1,u t-1)与t时刻状态x t之间的一个映射,ε服从协方差矩阵为Σ tran、均值为零的白高斯分布过程,因此,当前状态x t的条件分布如等式(8)所示:
P(x t|x t-1,u t-1)=N(g(x t-1,u t-1),Σ tran)   (8)
同样的,假设ζ也服从零均值白高斯噪声过程,测量模型和测量分布分别由等式(9)(10)给出,即:
z t=h(x t)+ζ   (9)
P(z t|x t)=N(h(x t),Σ meas)   (10)
其中Σ meas代表测量模型的协方差矩阵,h(·)指的是t时刻状态x t与测量值z t之间的映射。
1.2高斯过程(GP)回归模型
假设一组规模大小为N的训练集D,数据由(11)表示:
D=<X,y>   (11)
其中X是N×M的矩阵,矩阵的行是个1×M的向量,代表输入数据。y表示训练的输出, 是N×1矩阵,训练数据表示为(12)(13):
X=[x 1,...,x i,...,x N] T   (12)
y=[y 1,...,y i,...,y N] T,i=1,2,...,N   (13)
这里的x i和y i是第i个训练数据集的列向量和标量值。
为将GP模型运用到回归问题,假设每个输出值均从噪声过程中提取,一般表示为等式(14):
y i=f(x i)+ε i=f ii,i=1,2,...,N   (14)
其中f(x i)输入x i和实例值f i之间的映射函数。ε i服从均值为零、方差为η i的高斯分布。因此,输出变量y i的概率分布是:
P(y i|f ii)=N(f ii),i=1,2,3...,N   (15)
假设噪声变量ε 1…ε N对应的数据点都是相互独立的,如(16)所示,其中I N是N×N矩阵,f和η指N×1阶矩阵,分别由等式(17)(18)表示。
Figure PCTCN2019078187-appb-000025
f=[f 1,...,f i,...,f N] T   (17)
η=[η 1,...,η i,...,η N] T,i=1,2,...,N   (18)
假定实例变量的无穷集合服从GP,根据GP的定义,则f的任何子集变量都会服从以输入数据X、核参数θ为条件的联合高斯分布,如等式(19)所示:
P(f|X,θ)=N(0,K)   (19)
其中K代表由输入数据的核函数所确定的协方差矩阵,它的元素如下表示:
K i,j=K(x i,x j,θ),i=1,2,...,N,j=1,2,...,N   (20)
其中K(·)是一个核函数,表示数据点之间的相似程度。具体地,如果两个数据点(x i,x j)越相似,那么它们的实例值(f i,f j)就越相关。相似的程度取决于应用变量之间的差异。
以输入值和超参数(θ,η)为条件,输出变量y的边缘分布如(21)所示:
P(y|X,θ,η)=∫P(y|f,θ)P(f|X,η)df   (21)
更进一步的表示为多元高斯分布,如(22):
P(y|X,θ,η)=N(0,C)   (22)
其中C代表协方差矩阵。C的元素计算如下(23):
C i,j=C(x i,x j,θ,η)=K(x i,x j,θ)+η iδ i,j,i=1,2,...,N,j=1,2,...,N   (23)
其中δ i,j由Kronecker脉冲函数得之。
为了预测新输入下的输出变量,我们把新输入数据x N+1插入到N个采样训练数据中,对应新数据的输出变量表示为y N+1。根据公式(22),输出变量的联合分布是一个多元高斯分布,包括以输入数据点和超参数为条件的新变量。即:
P(y new|X new,θ,η)=N(0,C new)   (24)
其中y new,X new,C new表示为包含新数据的输入变量、输出变量、协方差矩阵:
Figure PCTCN2019078187-appb-000026
c N+1是一个N×1的矩阵,定义如下:
c N+1=[C(x 1,x N+1,θ,η 1),C(x 2,x N+1,θ,η 2),...,C(x N,x N+1,θ,η N+1)] T  (26)
对于这样一个多变量高斯分布,如果两组变量集均服从联合高斯分布,那么其中一组变量集的分布就是以另一组变量集为条件的高斯分布。因此,新输出变量y N+1的条件分布也是高斯分布,如(27)所示:
P(y N+1|X,x N+1,y,θ,η)=N(ν(x N+1,D),Γ(x N+1,D))  (27)
其中ν(x N+1,D),Γ(x N+1,D)分别指均值和方差函数,如下(28)(29)所示:
Figure PCTCN2019078187-appb-000027
Figure PCTCN2019078187-appb-000028
1.3基于GP的转移和测量模型
首先,假设状态、控制、测量向量的元素个数分别由M x,M u,M z表示,且这三个向量的数据点的规模是相同的,都用N表示。联立等式(7)和(9)中的状态转移模型和测量模型,训练数据集由等式(30)(31)给出:
D tran=<X tran,Y tran>   (30)
D meas=<X meas,Y meas>   (31)
其中X tran,Y tran是转移模型输入输出的数据点,X meas,Y meas是测量模型的输入输出数据点。转移模型的每一个输入数据点都包含状态和控制向量,是一个(M x+M u)×1的列向量,定义如下:
Figure PCTCN2019078187-appb-000029
在等式(27)中,可以推导出标量输出的预测分布。因此,对于t时刻状态向量中的任一个元素x p,t(p=1,2,…,M x),它的分布可由t-1时刻状态-控制对(x t-1,u t-1)给出:
Figure PCTCN2019078187-appb-000030
其中
Figure PCTCN2019078187-appb-000031
Figure PCTCN2019078187-appb-000032
矩阵的第p个行向量,
Figure PCTCN2019078187-appb-000033
是对应的超参数。
Figure PCTCN2019078187-appb-000034
是根据等式(28)(29)计算的均值和方差函数。假设状态变量维数和测量变量维数均是相互独立的。那么t时刻状态向量x t的分布如下:
Figure PCTCN2019078187-appb-000035
相似地,t时刻测量变量z t的分布如下:
Figure PCTCN2019078187-appb-000036
其中z q,t指测量向量的第q个元素,
Figure PCTCN2019078187-appb-000037
是Y meas的第q个行向量。
Figure PCTCN2019078187-appb-000038
是第q个测量模型的超参数。
Figure PCTCN2019078187-appb-000039
是第q个测量模型的均值和方差函数。
交叉口状态在线估计:采用扩展的卡尔曼滤波器来线性化转移模型和测量模型,再将上一时刻的交通状态、信号控制参数输入至转移模型得到预测状态及其协方差,然后将得到的预测状态及其协方差、当前时刻的测量值输入到测量模型,预测当前时刻状态的最优估计值,参见图2。
具体如下:
2.1一步估计过程
图3表示新观测值的一步估计过程。除了当前的测量值,BF一步估计过程还需要上一时刻被估计的状态、协方差、和控制数据。如果历史数据集D tran、D meas是存在的,那么可使用历史数据集更新转移和测量模型。
在状态空间模型中进行状态估计和预测时,卡尔曼滤波器是广泛使用的框架。但是,等式(34)(35)表明转移和测量模型是非线性函数。为了应对这个非线性,在BF建模框架下提出许多算法,例如扩展的卡尔曼滤波器、无迹卡尔曼滤波、粒子滤波器。本发明选择应用 扩展的卡尔曼滤波器,使用第一级泰勒展开式来线性化基于GP的非线性函数。基于高斯过程的扩展卡尔曼滤波器,使用上一时刻被估计的状态x t-1、协方差Σ t-1、控制u t-1、和当前的测量值z t-1来预测当前的状态x t和它的协方差矩阵Σ t
2.2基于高斯过程的扩展卡尔曼滤波器(GPEKF)
图4使用伪代码详细展示了DPEKF的步骤,之后描述了每一步的数学规则。1,2行表示预测步骤,3,4,5行表示更新步骤。假设状态是维数独立的,那么被预测状态s t可由如下矩阵组成:
Figure PCTCN2019078187-appb-000040
其中均值矩阵
Figure PCTCN2019078187-appb-000041
的计算等式由(28)给出。
为了线性化非线性函数,可使用一级泰勒展开式构建函数值,斜率来对函数进行线性近似。
假设G t是状态转移模型中GP均值函数的雅可比行列式如下(37):
Figure PCTCN2019078187-appb-000042
被预测状态的协方差矩阵由下(38)计算:
Figure PCTCN2019078187-appb-000043
其中转移模型的方差函数
Figure PCTCN2019078187-appb-000044
由等式(29)计算得出。
伪码中第三行表明,使用被预测状态和协方差矩阵计算卡尔曼增益。具体的表达式如(39)所示:
Figure PCTCN2019078187-appb-000045
H t表示测量模型中GP均值函数的雅可比行列式:
Figure PCTCN2019078187-appb-000046
其中测量模型的均值函数
Figure PCTCN2019078187-appb-000047
可由等式(28)得出。
最后,根据测量值的正确性程度将被预测的状态加入到新的状态估计中,来计算最优的状态。这种正确性程度与卡尔曼增益成正比,与当前测量值和预测测量值之间的偏差成正比。当前被估计的状态是(41):
Figure PCTCN2019078187-appb-000048
本发明利用信息增益
Figure PCTCN2019078187-appb-000049
调整被预测状态的协方差矩阵,来更新被估计状态的协方差矩阵,如(42):
Figure PCTCN2019078187-appb-000050
其中
Figure PCTCN2019078187-appb-000051
是一个M x×M x的单位矩阵。
3、数值实验
进一步的,使用车联网交通数据验证本发明所述的TSE方法,交通状态数据来自微观仿真器。
3.1实验设置
实验将估计方法应用到一个典型的、独立的交叉口中,见图5。该交叉口的信号控制器采用基于阶段的相位,其中交通灯按照固定的相位序列运行,相位序列见图6。感应检测器包括短感应检测器和长感应检测器,分别安置在距离停车线80米和10米的位置处。除此之外,本发明采用了一种“车辆驱动”信号配时方法,即绿灯分配时长根据环路探测器检测到的车辆存在的数量而变化。
状态向量的元素表示车辆数,包括排队的车辆、正在接近的车辆和与交叉口相关的车道上的车辆,该状态定义的方式已经应用于多个自适应信号控制系统。状态向量如式(43)所示:
Figure PCTCN2019078187-appb-000052
其中n k,t表示t时刻第k个车道上的车辆数,N lane表示交叉口车道的总数,本次测试实验中N lane=12。
控制数据通过采集交通灯指示的一般信息(绿灯,黄灯,红灯)得到,任何类型的信号控制器都可以访问这些数据。在状态转移过程中,绿灯和非绿灯指示会对车道上车辆数的变化有着重要的影响。控制向量如下(44)表示:
u t=[g 1,t,…g k,t,…g Nlane,t],k=1,2,...,N lane   (44)
其中g k,t指在交通灯定义估计区间内控制t时刻第k个车道的绿信比。
实验环节,估计框架使用车联网数据源。当车辆与基础设施(V2I)之间的通信启用时,信号控制器可以访问车辆位置。根据交叉口的几何形状,可实时提取每个车道上联网车辆的数量。在实验过程中,每辆联网车辆进入交叉口时都给定一个ID,信号控制器“负责”记录车辆ID,则一定时间间隔内联网车辆的流量等于唯一车辆ID的数量。
假设
Figure PCTCN2019078187-appb-000053
代表前五分钟内车道k上流动的联网车辆的数量,测量元素
Figure PCTCN2019078187-appb-000054
可由如下(45)计算:
Figure PCTCN2019078187-appb-000055
其中Δt指估计间隔时长,r k,t指的是t时刻联网车辆的渗透率的粗略估计,递归更新等式如下:
Figure PCTCN2019078187-appb-000056
τ表示更新渗透率的时间长度。
Figure PCTCN2019078187-appb-000057
表示l时刻车道k上联网车辆的数量,则对应的测量向量定义如下(47):
Figure PCTCN2019078187-appb-000058
根据一个时间序列内观测值与估计状态之间的差异,所有实验的估计均基于两个标准:平均绝对误差(MAE)和加权平均绝对误差(WAPE)。计算公式如下:
Figure PCTCN2019078187-appb-000059
Figure PCTCN2019078187-appb-000060
其中T是序列的总时间周期,
Figure PCTCN2019078187-appb-000061
是观测的状态向量。从计算公式分析,MAE指的是估计方法中可预期的平均绝对误差,WAPE允许在不同条件下对估计进行比较。
3.2数据准备
根据实验测试的交叉口布局,在开源微观仿真器SUMO0.19.0上建立交通模型。然后将开发设计的信号控制器软件程序与SUMO模拟器相连,设置仿真中的交通灯信号变化基于“车 辆驱动”控制。为生成有效的状态、控制和测量数据,SUMO通过应用程序提供的编程接口TraCI,记录每条车道上的检测信息和车辆数量。实验中,采用了一种常用的汽车跟踪模型——智能驾驶员模型(IDM),汽车跟随模型参数和信号控制参数如表1所示。
表1模型参数的IDM和信号控制参数
Figure PCTCN2019078187-appb-000062
对于每次实验,模拟生成训练、验证和测试三个数据集。训练数据集指的是一组数据样例,用于发现状态、控制和测量之间的潜在关系;根据所述的性能标准,验证集用来比较模型性能或者估计的精度,测试集用于评估所提出的估计方法的有效性,并提供估计结果的详细内容。
为了生成模拟数据,根据泊松过程随机抽取车辆,以每秒车辆到达率为单位。表2给出了6个应用的交通流方案,每个方案产生500个数据点。
表2实验十字路口每一个转弯动作的交通量[车辆/小时]
Figure PCTCN2019078187-appb-000063
L,T,R分别代表左转率,直行率,右转率。
表2中的“均匀”方案假设交叉口的所有方向的交通流都是相同的,“主干线”方案认为南北方向或者东西方向是主干道。相应的交通流量被定义为“中”或“高”水平。“中”水 平表示十字路口的交通流状况是正常,“高”水平表示相较于“中”水平,交通流量有一个显著的增加(约20%)。
对于验证数据集,交通流量是随机生成的,所选择值如表2所示。每个验证数据集包含600个模拟数据点。在生成测试集的模拟中,交通流量从“东西主干线(中)”方案开始,交通流方案的模式安排从“东西主干线东(中)”→“东西主干线(高)”→“均匀(高)”→“南北主干线(高)”→“均匀(中)”→“南北主干线(中)”。对每个交通流方案场景进行模拟,时间持续600秒,模拟共进行3600秒。由于交通仿真的随机性,车辆产生的时间是不同的。
3.3结果和讨论
为了评估该方法的有效性,使用车联网数据进行了几次测试。除了观察分析不同估计区间对估计精度的影响外,还对联网车辆的不同渗透率进行了分析。渗透率应用范围从0%—100%四种,即25%、50%、75%和90%,注意到在我们的分析中既不需要0%也不需要100%,因为渗透率为0%,即没有任何信息可提供,TSE将不能运行;同时如果整个交叉口的车辆信息是完全可访问的,即渗透率为100%时,获取状态将是没有意义的。
在对车联网数据使用各种估计间隔进行估计实验后,发现随着估计间隔的增大,估计的精度会有所下降。此外,对渗透率的变化进行了敏感性分析。表3和表4分别总结了验证集在两个典型估计区间(即1s和20s)下,采用四种渗透率所产生的估计结果。
表3估计间隔为1s时四种渗透率下(25%、50%、75%和90%)的车道估计误差
Figure PCTCN2019078187-appb-000064
表4估计间隔为20s时四种渗透率下(25%、50%、75%和90%)的车道估计误差
Figure PCTCN2019078187-appb-000065
Figure PCTCN2019078187-appb-000066
从表(3)(4)的结果分析,在TSE中如果有更多的车辆来提供信息,即渗透率越来越高的情况下,估计准确率会有所提高。在表(3)中,估计间隔为1s时,渗透率达到25%之前,估计精度并不会随着渗透率的降低而显著变化。尽管WAPE的增幅达8.67%(即在L8车道从5.18%增大到13.85%),但估计性能仍然可以接受,因为在仅使用了交叉口25%的车辆信息的情况下,平均绝对误差均在0.3以下。相反,每20s估计一次状态时车辆信息缺乏较多,此时估计模型是不可行的。例如,车道“L12”使用了验证方案的模拟数据,估计模型给出了一个较大的百分比误差,WAPE值为70.83%。
因此,在估计区间较小(如1s),且渗透率较低的情况下,本发明所提出的TSE框架能够为联网车辆的位置信息提供一种可行的方案。但若在两个估计时间点内缺少转移信息和测量信息,估计间隔将会增大,就会导致估计模型的有效性下降。

Claims (9)

  1. 一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其步骤如下:
    (1)数据采集:获取交叉口历史交通数据和对应信号控制参数,向量化处理,分别建立状态数据集和控制数据集;
    (2)建立非参数贝叶斯框架:结合递归状态估计和高斯过程回归模型,利用所述状态数据集和控制数据集训练并优化转移模型和测量模型;
    (3)交叉口状态估计:采用扩展的卡尔曼滤波器来线性化所述转移模型和所述测量模型,再将上一时刻的交通状态、信号控制参数输入至转移模型得到预测状态及其协方差,然后将得到的预测状态及其协方差、当前时刻的测量值输入到测量模型,预测当前时刻状态的最优估计值。
  2. 根据权利要求1所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:还包括步骤(4)状态估计方法验证:引入车联网数据,计算不同估计间隔和联网车辆渗透率下交叉口状态估计精度。
  3. 根据权利要求1或2所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:所述状态数据为交通流量数据,状态向量表示为:
    Figure PCTCN2019078187-appb-100001
    其中n k,t表示t时刻第k个车道上的车辆数,N lane表示交叉口车道的总数;
    所述控制数据为绿信比,控制向量表示为:
    Figure PCTCN2019078187-appb-100002
    其中g k,t指在交通灯定义估计区间内控制t时刻第k个车道的绿信比。
  4. 根据权利要求3所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:基于递归状态估计,所述转移模型表示为:
    x t=g(x t-1,u t-1)+ε
    其中g(·)表示t-1时刻状态-控制对(x t-1,u t-1)与t时刻状态x t之间的一个映射,ε服从协方差矩阵为Σ tran、均值为零的白高斯分布过程;
    状态转移概率如下式所示:
    P(x t|x t-1,u t-1)=N(g(x t-1,u t-1),Σ tran)
    所述测量模型表示为:
    z t=h(x t)+ζ
    其中h(·)表示t时刻状态x t与测量值z t之间的映射,ζ服从协方差矩阵为Σ meas、均值为零的白高斯分布过程;
    测量概率如下式所示:
    P(z t|x t)=N(h(x t),Σ meas)
    其中,Σ meas代表测量模型的协方差矩阵。
  5. 根据权利要求4所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:结合高斯过程回归模型,状态、控制、测量向量的元素个数分别由M x,M u,M z表示,且三者数据点规模相同,则转移模型和测量模型的训练数据集分别表示为:
    D tran=<X tran,Y tran
    D meas=<X meas,Y meas
    其中X tran,Y tran是转移模型输入输出的数据点,X meas,Y meas是测量模型的输入输出数据点;
    对于t时刻状态向量中的任一个元素x p,t(p=1,2,…,M x),状态向量x t的分布如下:
    Figure PCTCN2019078187-appb-100003
    其中
    Figure PCTCN2019078187-appb-100004
    Figure PCTCN2019078187-appb-100005
    矩阵的第p个行向量,
    Figure PCTCN2019078187-appb-100006
    是对应的超参数;
    测量变量z t的分布如下:
    Figure PCTCN2019078187-appb-100007
    其中z q,t指测量向量的第q个元素,
    Figure PCTCN2019078187-appb-100008
    是Y meas的第q个行向量,
    Figure PCTCN2019078187-appb-100009
    是第q个测量模型的超参数,
    Figure PCTCN2019078187-appb-100010
    是第q个测量模型的均值和方差函数。
  6. 根据权利要求5所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:所述转移模型和测量模型的线性化,具体为采用扩展的卡尔曼滤波器,使用一级泰勒展开式构建函数值,斜率来对函数进行线性近似。
  7. 根据权利要求6所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:所述预测当前时刻状态的最优估计值还包括:使用被预测状态和协方差矩阵计算卡尔曼增益,根据测量值的正确性程度将被预测的状态加入到新的状态估计中,计算最优的状态,所述正确性程度与卡尔曼增益成正比,与当前测量值和预测测量值之间的偏差成正比。
  8. 根据权利要求2所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:所述引入车联网数据,具体为:
    假设
    Figure PCTCN2019078187-appb-100011
    代表前五分钟内车道k上的联网车流量,测量元素
    Figure PCTCN2019078187-appb-100012
    k=1,2,...,N lane可由下式计算:
    Figure PCTCN2019078187-appb-100013
    其中r k,t是指t时刻联网车辆的渗透率,Δt指估计间隔时长;
    r k,t递归更新等式如下:
    Figure PCTCN2019078187-appb-100014
    τ表示更新渗透率的时间长度,
    Figure PCTCN2019078187-appb-100015
    表示l时刻车道k上的联网车流量,则对应的测量向量定义如下:
    Figure PCTCN2019078187-appb-100016
  9. 根据权利要求8所述的一种基于非参数贝叶斯框架的信号交叉口状态估计方法,其特征在于:所述交叉口状态估计精度的验证参数包括:平均绝对误差MAE和加权平均绝对误差WAPE,计算公式如下:
    Figure PCTCN2019078187-appb-100017
    Figure PCTCN2019078187-appb-100018
    其中T是序列的总时间周期,
    Figure PCTCN2019078187-appb-100019
    是观测的状态向量。
PCT/CN2019/078187 2018-12-26 2019-03-14 一种基于非参数贝叶斯框架的信号交叉口状态估计方法 WO2020133721A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811596573.XA CN109615860B (zh) 2018-12-26 2018-12-26 一种基于非参数贝叶斯框架的信号交叉口状态估计方法
CN201811596573.X 2018-12-26

Publications (1)

Publication Number Publication Date
WO2020133721A1 true WO2020133721A1 (zh) 2020-07-02

Family

ID=66011373

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/078187 WO2020133721A1 (zh) 2018-12-26 2019-03-14 一种基于非参数贝叶斯框架的信号交叉口状态估计方法

Country Status (2)

Country Link
CN (1) CN109615860B (zh)
WO (1) WO2020133721A1 (zh)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110321917A (zh) * 2019-04-24 2019-10-11 武汉大学 无参数的农田数据同化方法
CN110634288B (zh) * 2019-08-30 2022-06-21 上海电科智能系统股份有限公司 基于三元高斯混合模型的多维城市交通异常事件识别方法
EP3907663B1 (en) * 2020-05-06 2024-02-21 Robert Bosch GmbH Predicting a state of a computer-controlled entity
CN111951571B (zh) * 2020-07-06 2022-05-17 江苏大学 一种交通事故下路段拥堵车辆疏导系统及方法
CN112530177B (zh) * 2020-11-23 2022-03-04 西南交通大学 车联网环境下基于卡尔曼滤波的车辆排队长度估计方法
CN113724741A (zh) * 2021-06-02 2021-11-30 深圳芒果未来教育科技有限公司 一种乐器演奏或演唱测评方法及系统
CN114333335A (zh) * 2022-03-15 2022-04-12 成都交大大数据科技有限公司 基于轨迹数据的车道级交通状态估计方法、装置及系统
CN115208462A (zh) * 2022-07-14 2022-10-18 上海交通大学 光通信系统光模块控制参数优化方法及系统

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102011113019A1 (de) * 2011-07-14 2012-05-10 Daimler Ag Verfahren zur Ermittlung und Bewertung von Gefahren einer Situation zwischen zumindest zwei Verkehrsteilnehmern in einem Straßenkreuzungsbereich und Verfahren zur Unterstützung eines Fahrers beim Führen eines Fahrzeugs
CN103839412A (zh) * 2014-03-27 2014-06-04 北京建筑大学 一种基于贝叶斯加权的路口动态转向比例组合估计方法
CN103927891A (zh) * 2014-04-29 2014-07-16 北京建筑大学 一种基于双贝叶斯的路口动态转向比例两步预测方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103247177B (zh) * 2013-05-21 2016-01-20 清华大学 大规模路网交通流实时动态预测系统
US9489632B2 (en) * 2013-10-29 2016-11-08 Nec Corporation Model estimation device, model estimation method, and information storage medium
CN104809879B (zh) * 2015-05-14 2017-05-03 重庆大学 基于动态贝叶斯网络的高速公路路段交通状态估计方法
US10445576B2 (en) * 2016-09-23 2019-10-15 Cox Automotive, Inc. Automated vehicle recognition systems
CN106781556B (zh) * 2016-12-30 2019-09-10 大唐高鸿信息通信研究院(义乌)有限公司 一种适用于车载短距离通信网络的交通信号灯时长判断方法
CN108269395B (zh) * 2016-12-30 2019-10-25 大唐高鸿信息通信研究院(义乌)有限公司 适用于车载短距离通信网络交通拥塞预测和处理方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102011113019A1 (de) * 2011-07-14 2012-05-10 Daimler Ag Verfahren zur Ermittlung und Bewertung von Gefahren einer Situation zwischen zumindest zwei Verkehrsteilnehmern in einem Straßenkreuzungsbereich und Verfahren zur Unterstützung eines Fahrers beim Führen eines Fahrzeugs
CN103839412A (zh) * 2014-03-27 2014-06-04 北京建筑大学 一种基于贝叶斯加权的路口动态转向比例组合估计方法
CN103927891A (zh) * 2014-04-29 2014-07-16 北京建筑大学 一种基于双贝叶斯的路口动态转向比例两步预测方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JIN, JUNCHEN ET AL.: "A Non-parametric Bayesian Framework for Traffic-state Estimation at Signalized Intersections", DIVA-ACADEMIC ARCHIVE ON-LINE, 1 January 2017 (2017-01-01), XP085702335, DOI: 20190722121803 *

Also Published As

Publication number Publication date
CN109615860A (zh) 2019-04-12
CN109615860B (zh) 2020-10-16

Similar Documents

Publication Publication Date Title
WO2020133721A1 (zh) 一种基于非参数贝叶斯框架的信号交叉口状态估计方法
Michelmore et al. Uncertainty quantification with statistical guarantees in end-to-end autonomous driving control
Altché et al. An LSTM network for highway trajectory prediction
CN108820039B (zh) 一种无人驾驶车辆弯道横向控制系统及方法
CN107293115B (zh) 一种用于微观仿真的交通流量预测方法
Sirmatel et al. Modeling, estimation, and control in large-scale urban road networks with remaining travel distance dynamics
CN106781489B (zh) 一种基于递归神经网络的路网状态预测方法
Sacks et al. Statistically-based validation of computer simulation models in traffic operations and management
Tang et al. Travel time estimation using freeway point detector data based on evolving fuzzy neural inference system
WO2016187746A1 (zh) 提高人工神经网络定位性能的方法和装置
CN108008099A (zh) 一种污染源定位方法
CN111190211B (zh) 一种gps失效位置预测定位方法
Jin et al. A non-parametric Bayesian framework for traffic-state estimation at signalized intersections
CN105424043B (zh) 一种基于判断机动的运动状态估计方法
CN102663495A (zh) 一种用于非线性器件建模的神经网络数据生成方法
CN110991776A (zh) 一种基于gru网络实现水位预测的方法及系统
Wu et al. Connections between classical car following models and artificial neural networks
Hanke et al. Classification of sensor errors for the statistical simulation of environmental perception in automated driving systems
CN115099328A (zh) 基于对抗网络的交通流量预测方法、系统、设备及存储介质
CN103279030B (zh) 基于贝叶斯框架的动态软测量建模方法及装置
Yuan et al. Efficient traffic state estimation and prediction based on the ensemble Kalman filter with a fast implementation and localized deterministic scheme
CN106202926A (zh) 基于多节点协同探测的空间系统偏差配准优化方法
Lu et al. Determining the required probe vehicle size for real-time travel time estimation on signalized arterial
Zhong et al. Understanding Heterogeneity of Automated Vehicles and Its Traffic-level Impact: A Stochastic Behavioral Perspective
Zammit et al. Joint state and parameter estimation for a macro traffic junction model

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: 19904933

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19904933

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19904933

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04.01.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19904933

Country of ref document: EP

Kind code of ref document: A1