Movatterモバイル変換

[0]ホーム
URL:

CN101901547A - A variable lane adaptive control method - Google Patents

A variable lane adaptive control methodDownload PDF

Info

Publication number
CN101901547A
CN101901547ACN2010102323131ACN201010232313ACN101901547ACN 101901547 ACN101901547 ACN 101901547ACN 2010102323131 ACN2010102323131 ACN 2010102323131ACN 201010232313 ACN201010232313 ACN 201010232313ACN 101901547 ACN101901547 ACN 101901547A
Authority
CN
China
Prior art keywords
output
layer
variable lane
attribute
network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN2010102323131A
Other languages
Chinese (zh)
Inventor
董红召
郭海锋
傅立骏
陈宁
郭明飞
凌越
曹福灵
马帅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUTfiledCriticalZhejiang University of Technology ZJUT
Priority to CN2010102323131ApriorityCriticalpatent/CN101901547A/en
Publication of CN101901547ApublicationCriticalpatent/CN101901547A/en
Pendinglegal-statusCriticalCurrent

Links

Images

Landscapes

Abstract

Translated fromChinese

一种可变车道自适应控制方法，包括以下步骤：1)、将通过设置在道路的检测器所检测到的时间占有率作为BP神经网络的输入，BP神经网络的输出向量，输出向量为0或1；2)可变车道控制过程如下：2.1)若当前可变车道属性为直行，如果输出向量为1，则在下一个直行相位到来时，将可变车道属性由直行变为左转；如果输出向量为0，则当前可变车道属性不变；2.2)若当前可变车道属性为左转，如果输出向量为1，则在下一个左转相位到来时将可变车道属性由左转变为直行；如果输出向量为0，则可变车道保持左转属性不变。本发明能够有效提高可变车道利用率。

Figure 201010232313

A variable lane adaptive control method, comprising the following steps: 1), with the time occupancy rate detected by the detector that is arranged on road as the input of BP neural network, the output vector of BP neural network , the output vector is 0 or 1; 2) The variable lane control process is as follows: 2.1) If the current variable lane attribute is straight, if the output vector is 1, then when the next straight phase arrives, the variable lane attribute will be changed from straight to If the output vector is 0, the attribute of the current variable lane remains unchanged; 2.2) If the attribute of the current variable lane is turning left, if the output vector is 1, the attribute of the variable lane will be changed when the next left-turn phase arrives. Change from left to straight; if the output vector is 0, the variable lane keeps the left-turn property unchanged. The invention can effectively improve the utilization rate of variable lanes.

Figure 201010232313

Description

Translated fromChinese
一种可变车道自适应控制方法A variable lane adaptive control method

技术领域technical field

本发明涉及交通控制方法，尤其是一种可变车道自适应控制方法。The invention relates to a traffic control method, in particular to a variable lane adaptive control method.

背景技术Background technique

随着经济的发展，尤其是城市经济的快速发展，城市机动车的保有量快速上升，虽然城市道路及交通设施也有了相应的增加，但是道路设施的增长速度远远低于机动车的增长速度，由此造成了交通拥堵、环境污染等一系列问题。特别在上下班高峰期间，“潮汐式”交通引起的交通拥堵问题更加严重。从本质上讲，产生这种现象的原因是交通需求的动态变化与静态的道路设施之间的矛盾。充分挖掘、合理利用城市道路资源是缓解城市交通拥堵状况的有效手段。为此，国内很多大中型城市应用了可变车道技术。With the development of the economy, especially the rapid development of the urban economy, the number of urban motor vehicles has increased rapidly. Although urban roads and traffic facilities have also increased correspondingly, the growth rate of road facilities is far lower than that of motor vehicles. , resulting in a series of problems such as traffic congestion and environmental pollution. Especially during rush hour, the problem of traffic congestion caused by "tidal" traffic is even more serious. Essentially, the reason for this phenomenon is the contradiction between the dynamic change of traffic demand and the static road facilities. Fully excavating and rationally utilizing urban road resources is an effective means to alleviate urban traffic congestion. For this reason, many large and medium-sized cities in China have applied variable lane technology.

目前可变车道诱导方法主要就是利用原有的车道诱导标志，经过加工改装，将标志上固定不变的指向箭头改为可变的箭头，根据交叉口的流量变化情况，由值班交警或者指挥中心调度控制，改变箭头的指向。车道的改变由值班交警根据观测交通流的大小来遥控信号标志牌的方式实现的。这样，车道的切换时间会经常提前或滞后于最佳切换时间，可变车道没有得到充分的使用。At present, the variable lane guidance method is mainly to use the original lane guidance signs, after processing and modification, change the fixed pointing arrows on the signs into variable arrows, and according to the traffic changes at the intersection, the traffic police on duty or the command center Dispatch controls to change the direction of the arrow. Lane changes are realized by the traffic police on duty by remote control of signal signs according to the size of the observed traffic flow. In this way, the lane switching time will often advance or lag behind the optimal switching time, and the variable lanes are not fully used.

发明内容Contents of the invention

为了解决上述可变车道利用率不高的问题，本发明提供一种能够有效提高可变车道利用率的可变车道自适应控制方法。In order to solve the above-mentioned problem of low utilization rate of variable lanes, the present invention provides an adaptive control method for variable lanes that can effectively improve the utilization rate of variable lanes.

本发明解决其技术问题所采用的技术方案是： The technical solution adopted by the present invention to solve its technical problems is:

一种可变车道自适应控制方法，所述可变车道自适应控制方法包括以下步骤：A variable lane adaptive control method, the variable lane adaptive control method comprises the following steps:

1）、将通过设置在道路的检测器所检测到的时间占有率作为BP神经网络的输入，BP神经网络包括输入层、隐含层和输出层，其训练过程为在网络各节点的连接权值固定不变的前提下，从输入层开始逐层逐个节点地计算每个节点的输出，再保持输出层各节点的输出不变，从输出层开始反向逐层逐个节点计算连接权值的修改量；如果输出层的网络输出与期望输出相差超于预设值，根据网络输出与期望输出的信号误差进行权值调整，最终使网络输出层的输出值与期望值趋于一致。权值调整公式为：1) The time occupancy rate detected by the detector set on the road is used as the input of the BP neural network. The BP neural network includes an input layer, a hidden layer and an output layer. The training process is the connection weight of each node in the network. Under the premise that the value is fixed, the output of each node is calculated layer by node from the input layer, and then the output of each node in the output layer remains unchanged, and the connection weight is calculated layer by layer from the output layer in reverse. Modification amount; if the difference between the network output of the output layer and the expected output exceeds the preset value, the weight is adjusted according to the signal error between the network output and the expected output, and finally the output value of the network output layer tends to be consistent with the expected value. The weight adjustment formula is:

Figure 2010102323131100002DEST_PATH_IMAGE001
 ，
Figure 519903DEST_PATH_IMAGE002
Figure 2010102323131100002DEST_PATH_IMAGE001
,
Figure 519903DEST_PATH_IMAGE002

其中，

Figure 2010102323131100002DEST_PATH_IMAGE003
为当前某层权值调整量矩阵；为前一次权值调整量矩阵；X代表某层输入向量；
Figure 2010102323131100002DEST_PATH_IMAGE005
为动量系数；为比例系数，
Figure 2010102323131100002DEST_PATH_IMAGE007
；
Figure 779349DEST_PATH_IMAGE008
为误差信号，；in,
Figure 2010102323131100002DEST_PATH_IMAGE003
It is the weight adjustment matrix of a certain layer at present; is the previous weight adjustment matrix; X represents the input vector of a certain layer;
Figure 2010102323131100002DEST_PATH_IMAGE005
is the momentum coefficient; is the proportional coefficient,
Figure 2010102323131100002DEST_PATH_IMAGE007
;
Figure 779349DEST_PATH_IMAGE008
is the error signal, ;

所述BP神经网络的输出向量O=

Figure 146614DEST_PATH_IMAGE010
，所述输出向量为0或1；The output vectorO of the BP neural network =
Figure 146614DEST_PATH_IMAGE010
, the output vector is 0 or 1;

2）、可变车道控制过程如下：2) The variable lane control process is as follows:

目前神经网络的实现仍以软件编程为主，编写相应的算法程序，神经网络训练好之后开始工作，采集相应道路的道路交通信息（时间占有率），交由算法进行计算，如果满足触发条件，则改变可变车道的属性；如果不满足触发条件，则保持可变车道的属性不变。At present, the realization of the neural network is still mainly based on software programming, and the corresponding algorithm program is written. After the neural network is trained, it starts to work, collects the road traffic information (time occupancy rate) of the corresponding road, and submits it to the algorithm for calculation. If the trigger condition is met, Then change the attribute of the variable lane; if the trigger condition is not met, keep the attribute of the variable lane unchanged.

2.1）若当前可变车道属性为直行，将采集到的时间占有率数据输入BP神经网络中，如果输出向量为1，则在下一个直行相位到来时，将可变车道属性由直行变为左转；如果输出向量为0，则当前可变车道属性不变。2.1) If the current variable lane attribute is going straight, input the collected time occupancy data into the BP neural network, and if the output vector is 1, then change the variable lane attribute from straight to left turn when the next straight phase arrives ; If the output vector is 0, the attributes of the current variable lane remain unchanged.

2.2）若当前可变车道属性为左转，将采集到的时间占有率数据输入BP神经网络中，如果输出向量为1，则在下一个左转相位到来时将可变车道属性由左转变为直行；如果输出向量为0，则可变车道保持左转属性不变。2.2) If the current variable lane attribute is left turn, input the collected time occupancy data into the BP neural network, if the output vector is 1, then change the variable lane attribute from left to straight when the next left turn phase arrives ; If the output vector is 0, the variable lane keeps the left-turn property unchanged.

本发明的技术构思为：本发明所述的自适应控制方法采用改进的BP神经网络，BP神经网络是一个3层或3层以上阶层神经网络，网络模型如图1所示。

Figure 2010102323131100002DEST_PATH_IMAGE011
，，…,
Figure 2010102323131100002DEST_PATH_IMAGE013
为网络的输入，
Figure 575639DEST_PATH_IMAGE014
，…,
Figure 2010102323131100002DEST_PATH_IMAGE015
为网络的输出。上下层之间各种神经元实行权连接，即下层的每个单元与上层的每个单元都实现权连接，每层各种神经元之间无连接。最基本的BP网络是3层前馈网络，包括输入层、隐层和输出层。The technical idea of the present invention is: the adaptive control method described in the present invention adopts the improved BP neural network, and the BP neural network is a hierarchical neural network with 3 or more layers, and the network model is as shown in FIG. 1 .
Figure 2010102323131100002DEST_PATH_IMAGE011
, ,…,
Figure 2010102323131100002DEST_PATH_IMAGE013
is the input of the network,
Figure 575639DEST_PATH_IMAGE014
,…,
Figure 2010102323131100002DEST_PATH_IMAGE015
is the output of the network. Various neurons in the upper and lower layers are connected by weight, that is, each unit in the lower layer is connected with each unit in the upper layer, and there is no connection between various neurons in each layer. The most basic BP network is a 3-layer feedforward network, including an input layer, a hidden layer and an output layer.

1.1变量的选择及原始数据的搜集1.1 Selection of variables and collection of raw data

影响可变车道属性的因素主要有车速及时间占有率，本文将相应的左转车道及直行车道的时间占有率作为输入，将触发阈值作为网络的输出。训练样本来自有可变车道的某一实际交叉口的经过分析处理过的有经验的交通警察的现场指挥数据。时间占有率能够反映出左转和直行车道的车流量信息以及左转和直行车道的溢出情况，从而决定是否改变车道的属性。The main factors affecting the attributes of variable lanes are vehicle speed and time occupancy. In this paper, the time occupancy of the corresponding left-turn lane and through lane is taken as input, and the trigger threshold is taken as the output of the network. The training samples come from the field command data of an experienced traffic policeman who has been analyzed and processed at an actual intersection with variable lanes. The time occupancy rate can reflect the traffic flow information of the left-turn and through lanes and the overflow situation of the left-turn and through lanes, so as to decide whether to change the attributes of the lanes.

1.2确定网络的学习算法1.2 Determine the learning algorithm of the network

传统的BP算法具有收敛速度慢，容易陷入局部极小值等缺点，为此，学界提出了很多改进办法，例如，增加动量项法、自适应调节学习率法和引入陡度因子法。本文采用附加动量项的BP神经网络，可以有效地抑制网络陷入局部极小值。权值调整公式为：The traditional BP algorithm has disadvantages such as slow convergence speed and easy to fall into local minimum. For this reason, the academic circle has proposed many improvement methods, such as adding momentum items, adaptively adjusting learning rate methods, and introducing steepness factor methods. This paper adopts the BP neural network with additional momentum term, which can effectively restrain the network from falling into the local minimum. The weight adjustment formula is:

Figure 472925DEST_PATH_IMAGE001
 (
Figure 715819DEST_PATH_IMAGE002
)
Figure 472925DEST_PATH_IMAGE001
(
Figure 715819DEST_PATH_IMAGE002
)

其中：

Figure 19761DEST_PATH_IMAGE003
为当前某层权值调整量矩阵；in:
Figure 19761DEST_PATH_IMAGE003
It is the weight adjustment matrix of a certain layer at present;

    

Figure 600610DEST_PATH_IMAGE004
为前一次权值调整量矩阵；
Figure 600610DEST_PATH_IMAGE004
is the previous weight adjustment matrix;

X代表某层输入向量；X represents the input vector of a certain layer;

Figure 306398DEST_PATH_IMAGE005
为动量系数；
Figure 306398DEST_PATH_IMAGE005
is the momentum coefficient;

为比例系数，

Figure 245852DEST_PATH_IMAGE007
； is the proportional coefficient,
Figure 245852DEST_PATH_IMAGE007
;

Figure 636251DEST_PATH_IMAGE008
为误差信号，
Figure 134228DEST_PATH_IMAGE009
。
Figure 636251DEST_PATH_IMAGE008
is the error signal,
Figure 134228DEST_PATH_IMAGE009
.

1.3确定网络的结构1.3 Determining the structure of the network

在本文中，将相应的直行车道和左转车道的时间占有率作为网络的输入，有可变车道的交叉口入口道如图3所示，触发阈值作为网络的输出，这样，网络的输入层节点数为8，输出层节点数为1，网络含有一个隐含层，隐含层的个数通过经验公式

Figure 968192DEST_PATH_IMAGE016
，确定为17个。对于输入层与隐含层，以及隐含层与输出层之间的传递函数均选为单极性Sigmoid函数，即： In this paper, the time occupancy of the corresponding through lane and left-turn lane is taken as the input of the network, and the intersection entrance road with variable lanes is shown in Figure 3, and the trigger threshold is taken as the output of the network. In this way, the input layer of the network The number of nodes is 8, the number of nodes in the output layer is 1, the network contains a hidden layer, and the number of hidden layers is determined by the empirical formula
Figure 968192DEST_PATH_IMAGE016
, determined to be 17. For the transfer function between the input layer and the hidden layer, as well as the hidden layer and the output layer, the unipolar Sigmoid function is selected, namely:

Figure 2010102323131100002DEST_PATH_IMAGE017
Figure 2010102323131100002DEST_PATH_IMAGE017

1.4 网络模型与BP学习算法1.4 Network model and BP learning algorithm

本发明所对应的三层网络结构图如图2所示，设输入向量为X=，图中

Figure 2010102323131100002DEST_PATH_IMAGE019
=-1是为隐含层神经元引入阈值而设置的；隐含层输出向量为Y=
Figure 50603DEST_PATH_IMAGE020
，图中
Figure 2010102323131100002DEST_PATH_IMAGE021
=-1是为输出层神经元引入阈值而设置的；输出层输出向量为O=
Figure 278454DEST_PATH_IMAGE010
，期望输出向量为d=。输入层到隐含层之间的权值矩阵用V表示，V=
Figure 2010102323131100002DEST_PATH_IMAGE023
,其中列向量
Figure 845887DEST_PATH_IMAGE024
为隐含层第j个神经元对应的权向量；隐含层到输出层之间的权值矩阵用W表示。W=
Figure 2010102323131100002DEST_PATH_IMAGE025
，W1表示输出层的神经元对应的权向量。The corresponding three-layer network structure diagram of the present invention is as shown in Figure 2, assuming that the input vector isX = , in the figure
Figure 2010102323131100002DEST_PATH_IMAGE019
=-1 is set for the introduction of the threshold for hidden layer neurons; the hidden layer output vector isY =
Figure 50603DEST_PATH_IMAGE020
, in the figure
Figure 2010102323131100002DEST_PATH_IMAGE021
=-1 is set for the introduction of the threshold for the output layer neurons; the output vector of the output layer isO =
Figure 278454DEST_PATH_IMAGE010
, the desired output vector isd = . The weight matrix between the input layer and the hidden layer is represented byV ,V =
Figure 2010102323131100002DEST_PATH_IMAGE023
, where the column vector
Figure 845887DEST_PATH_IMAGE024
is the weight vector corresponding to the jth neuron in the hidden layer; the weight matrix between the hidden layer and the output layer is represented byW.W =
Figure 2010102323131100002DEST_PATH_IMAGE025
, W1 represents the weight vector corresponding to the neurons of the output layer.

对于输出层，有：For the output layer, there are:

Figure 2010102323131100002DEST_PATH_IMAGE027
Figure 2010102323131100002DEST_PATH_IMAGE027

对于隐含层，有：For hidden layers, there are:

Figure 181151DEST_PATH_IMAGE028
  j=1,2,…,17
Figure 181151DEST_PATH_IMAGE028
j=1,2,…,17

Figure 2010102323131100002DEST_PATH_IMAGE029
  j=1,2,…,17
Figure 2010102323131100002DEST_PATH_IMAGE029
j=1,2,…,17

对于输入层与隐含层，以及隐含层与输出层之间的传递函数均选为单极性Sigmoid函数，即：For the transfer function between the input layer and the hidden layer, as well as the hidden layer and the output layer, the unipolar Sigmoid function is selected, namely:

Figure 169967DEST_PATH_IMAGE030
Figure 169967DEST_PATH_IMAGE030

当网络输出与实际输出不等时，存在输出误差E，计算式如下：When the network output is not equal to the actual output, there is an output error E, and the calculation formula is as follows:

Figure 2010102323131100002DEST_PATH_IMAGE031
Figure 2010102323131100002DEST_PATH_IMAGE031

由上式可以看出，网络误差是各层权值

Figure 469099DEST_PATH_IMAGE032
、
Figure 2010102323131100002DEST_PATH_IMAGE033
的函数，因此调整权值可以改变误差E。权值调整公式为：It can be seen from the above formula that the network error is the weight of each layer
Figure 469099DEST_PATH_IMAGE032
,
Figure 2010102323131100002DEST_PATH_IMAGE033
function, so adjusting the weight can change the error E. The weight adjustment formula is:

Figure 568773DEST_PATH_IMAGE001
 (
Figure 941855DEST_PATH_IMAGE002
)
Figure 568773DEST_PATH_IMAGE001
(
Figure 941855DEST_PATH_IMAGE002
)

其中：

Figure 288522DEST_PATH_IMAGE003
为当前某层权值调整量矩阵；in:
Figure 288522DEST_PATH_IMAGE003
It is the weight adjustment matrix of a certain layer at present;

     

Figure 514098DEST_PATH_IMAGE004
为前一次权值调整量矩阵；
Figure 514098DEST_PATH_IMAGE004
is the previous weight adjustment matrix;

X代表某层输入向量；X represents the input vector of a certain layer;

Figure 604414DEST_PATH_IMAGE005
为动量系数；
Figure 604414DEST_PATH_IMAGE005
is the momentum coefficient;

Figure 97581DEST_PATH_IMAGE006
为比例系数，
Figure 349571DEST_PATH_IMAGE007
；
Figure 97581DEST_PATH_IMAGE006
is the proportional coefficient,
Figure 349571DEST_PATH_IMAGE007
;

Figure 62443DEST_PATH_IMAGE008
为误差信号，
Figure 690871DEST_PATH_IMAGE009
。
Figure 62443DEST_PATH_IMAGE008
is the error signal,
Figure 690871DEST_PATH_IMAGE009
.

增加动量项即从前一次权值调整量中取出一部分迭加到本次权值调整量中，

Figure 321702DEST_PATH_IMAGE005
为动量系数，一般有
Figure 682276DEST_PATH_IMAGE002
。动量项反映了以前积累的调整经验，对于t时刻的调整起阻尼作用。当误差曲面出现骤然起伏时，可以减小振荡趋势，提高训练速度。Adding the momentum item means taking a part of the previous weight adjustment and adding it to the current weight adjustment.
Figure 321702DEST_PATH_IMAGE005
is the momentum coefficient, generally
Figure 682276DEST_PATH_IMAGE002
. The momentum term reflects the previous accumulated adjustment experience and plays a damping role for the adjustment at time t. When there are sudden fluctuations in the error surface, the oscillation tendency can be reduced and the training speed can be improved.

BP神经网络的工作过程通常由两个阶段组成。一个阶段是在网络各节点的连接权值固定不变的前提下，从输入层开始逐层逐个节点地计算每个节点的输出；另一阶段是学习阶段，在这一阶段，输出层中各节点的输出保持不变，网络学习从输出层开始，反向逐层逐个节点地计算各连接权值的修改量，以修改各连接的权值，直到输入层为止。这两个阶段分别称为正向传播和反向传播过程。在正向传播中，如果输出层的网络输出与期望输出相差较大，则开始反向传播过程。根据网络输出与所期望输出的信号误差对网络节点间的各连接权值进行修改，以此来减小网络输出信号与所期望输出的误差。BP网络通过不断进行的正向传播和反向传播，最终使网络输出层的输出值与期望值趋于一致。The working process of BP neural network usually consists of two stages. One stage is to calculate the output of each node layer by layer from the input layer on the premise that the connection weights of each node in the network are fixed; the other stage is the learning stage. In this stage, each node in the output layer The output of the node remains unchanged, and the network learning starts from the output layer, and reversely calculates the modification amount of each connection weight one by one node by layer, so as to modify the weight of each connection until the input layer. These two stages are called forward propagation and back propagation process respectively. In forward propagation, if the network output of the output layer is far from the expected output, the backpropagation process starts. According to the signal error between the network output and the expected output, the connection weights between the network nodes are modified to reduce the error between the network output signal and the expected output. The BP network finally makes the output value of the network output layer consistent with the expected value through continuous forward propagation and back propagation.

本发明的有益效果主要表现在：通过实时采集的道路交通数据，由相应的自适应控制算法进行计算，来判断是否切换可变车道。这样，可以根据实时的道路车辆信息来实时地改变可变车道的属性，使得可变车道的利用率能够达到最大化。The beneficial effect of the present invention is mainly manifested in that: the road traffic data collected in real time is calculated by a corresponding adaptive control algorithm to judge whether to switch variable lanes. In this way, the attributes of the variable lane can be changed in real time according to the real-time road vehicle information, so that the utilization rate of the variable lane can be maximized.

附图说明Description of drawings

图1是三层神经网络模型的示意图。Figure 1 is a schematic diagram of a three-layer neural network model.

图2是三层神经网络结构图。Figure 2 is a three-layer neural network structure diagram.

图3是路口施工图。Figure 3 is the construction drawing of the intersection.

图4是可变车道触发流程图。Fig. 4 is a flow chart of variable lane triggering.

具体实施方式Detailed ways

下面结合附图对本发明作进一步描述。The present invention will be further described below in conjunction with the accompanying drawings.

参照图1～图4，一种基于BP神经网络的可变车道自适应控制方法，所述可变车道自适应控制方法包括以下步骤：Referring to Fig. 1～Fig. 4, a kind of variable lane adaptive control method based on BP neural network, described variable lane adaptive control method comprises the following steps:

1）、将所述可变车道处于直行车道或左转车道时的时间占有率作为BP神经网络的输入，BP神经网络包括输入层、隐含层和输出层，其训练过程为在网络各节点的连接权值固定不变的前提下，从输入层开始逐层逐个节点地计算每个节点的输出，再保持输出层各节点的输出不变，从输出层开始反向逐层逐个节点计算连接权值的修改量；如果输出层的网络输出与期望输出相差超于预设值，根据网络输出与期望输出的信号误差进行权值调整，权值调整公式为：1), the time occupancy rate when the variable lane is in the straight lane or the left-turn lane is used as the input of the BP neural network. The BP neural network includes an input layer, a hidden layer and an output layer. The training process is that each node of the network On the premise that the connection weights of the input layer are fixed, the output of each node is calculated layer by layer from the input layer, and then the output of each node in the output layer remains unchanged, and the connection is calculated layer by layer from the output layer in reverse. The modification amount of the weight; if the difference between the network output of the output layer and the expected output exceeds the preset value, the weight is adjusted according to the signal error between the network output and the expected output. The weight adjustment formula is:

Figure 131712DEST_PATH_IMAGE001
 ，
Figure 314563DEST_PATH_IMAGE002
Figure 131712DEST_PATH_IMAGE001
,
Figure 314563DEST_PATH_IMAGE002

其中，

Figure 267475DEST_PATH_IMAGE003
为当前某层权值调整量矩阵；
Figure 376114DEST_PATH_IMAGE004
为前一次权值调整量矩阵；X代表某层输入向量；
Figure 47267DEST_PATH_IMAGE005
为动量系数；为比例系数，
Figure 106807DEST_PATH_IMAGE007
；为误差信号，
Figure 544796DEST_PATH_IMAGE009
；in,
Figure 267475DEST_PATH_IMAGE003
It is the weight adjustment matrix of a certain layer at present;
Figure 376114DEST_PATH_IMAGE004
is the previous weight adjustment matrix; X represents the input vector of a certain layer;
Figure 47267DEST_PATH_IMAGE005
is the momentum coefficient; is the proportional coefficient,
Figure 106807DEST_PATH_IMAGE007
; is the error signal,
Figure 544796DEST_PATH_IMAGE009
;

所述BP神经网络的输出向量O=

Figure 69450DEST_PATH_IMAGE010
，所述输出向量为0或1；The output vectorO of the BP neural network =
Figure 69450DEST_PATH_IMAGE010
, the output vector is 0 or 1;

2）、可变车道控制过程如下：2) The variable lane control process is as follows:

2.1）若当前可变车道属性为直行，将采集到的时间占有率数据输入BP神经网络，如果输出向量为1，则在下一个直行相位到来时，将可变车道属性由直行变为左转；如果输出向量为0，则当前可变车道属性不变。2.1) If the current variable lane attribute is going straight, input the collected time occupancy data into the BP neural network. If the output vector is 1, when the next straight phase comes, change the variable lane attribute from going straight to turning left; If the output vector is 0, the current variable lane attributes remain unchanged.

2.2）若当前可变车道属性为左转，将采集到的时间占有率数据输入BP神经网络，如果输出向量为1，则在下一个左转相位到来时将可变车道属性由左转变为直行；如果输出向量为0，则可变车道保持左转属性不变。2.2) If the current variable lane attribute is left turn, input the collected time occupancy data into the BP neural network, and if the output vector is 1, change the variable lane attribute from left to straight when the next left turn phase arrives; If the output vector is 0, the variable lane keeps the left turn property unchanged.

本实施例中，神经网络的实现仍以软件编程为主，编写相应的算法程序，神经网络训练好之后开始工作，采集相应道路的道路交通信息（时间占有率），交由算法进行计算，如果满足触发条件，则改变可变车道的属性；如果不满足触发条件，则保持可变车道的属性不变。In this embodiment, the realization of the neural network is still based on software programming, and the corresponding algorithm program is written. After the neural network is trained, it starts to work, collects the road traffic information (time occupancy rate) of the corresponding road, and submits it to the algorithm for calculation. If If the trigger condition is met, the attribute of the variable lane is changed; if the trigger condition is not met, the attribute of the variable lane is kept unchanged.

(1)若当前可变车道属性为直行，将采集到的数据交由控制算法计算，得出可变车道触发条件满足，则在下一个直行相位到来时将可变车道属性由直行变为左转。(1) If the attribute of the current variable lane is going straight, the collected data will be calculated by the control algorithm, and the trigger condition of the variable lane is satisfied, then the attribute of the variable lane will be changed from straight to left when the next straight phase arrives .

(2)若当前可变车道属性为直行，将采集到的数据交由控制算法计算，得到可变车道触发条件不满足，这时可变车道保持直行属性不变。(2) If the attribute of the current variable lane is going straight, the collected data is submitted to the control algorithm for calculation, and it is found that the trigger condition of the variable lane is not satisfied, and the attribute of the variable lane remains unchanged.

(3)若当前可变车道属性为左转，将采集到的数据交由控制算法计算，得到可变车道触发条件满足，则在下一个左转相位到来时将可变车道属性由左转变为直行。(3) If the current variable lane attribute is left turn, the collected data will be calculated by the control algorithm, and the variable lane trigger condition is satisfied, then the variable lane attribute will be changed from left to straight when the next left turn phase arrives .

(4)若当前可变车道属性为左转，将采集到的数据交由控制算法计算，得到可变车道触发条件不满足，这时可变车道保持左转属性不变。(4) If the attribute of the current variable lane is left turn, the collected data is submitted to the control algorithm for calculation, and the trigger condition of the variable lane is not satisfied, and the left turn attribute of the variable lane remains unchanged.

路口施工图如图3所示The construction drawing of the intersection is shown in Figure 3

图3所示为一个交叉口的某一进口路段，1为渠化段，2为过渡段，3为不分流向段，4为所设置的能够检测时间占有率的检测器，检测器编号规则为，过渡段自左向右依次为①、②、③、④，不分流向段自左向右依次为⑤、⑥、⑦、⑧。5为可变车道。Figure 3 shows a certain entrance section of an intersection, 1 is the channelization section, 2 is the transition section, 3 is the section regardless of the flow direction, 4 is the set detector that can detect the time occupancy rate, and the detector numbering rules For, the transition section from left to right is ①, ②, ③, ④, and the section regardless of the flow direction is ⑤, ⑥, ⑦, ⑧ from left to right. 5 is a variable lane.

可变车道触发流程图如下图4所示。The flow chart of variable lane triggering is shown in Figure 4 below.

具体的控制实例如下所示：The specific control example is as follows:

路口施工图如图3所示，假如当前可变车道的属性为直行，检测器采集相应道路的时间占有率分别为（0.80，0.75，0.43，0.40，0.79，0.77，0.45，0.42），并传递到交通控制中心或者路边控制机，由相应的BP神经网络来计算，得到的触发值为0.93，0.93

Figure 996954DEST_PATH_IMAGE034
（0.9，1）（在数据样本里，可变车道属性如可以变换，设置触发值为1，不能变换设为0，则触发条件满足，这时在下一个直行相位到来时将可变车道属性由直行变为左转。The construction drawing of the intersection is shown in Figure 3. If the attribute of the current variable lane is going straight, the detector collects the time occupancy of the corresponding road (0.80, 0.75, 0.43, 0.40, 0.79, 0.77, 0.45, 0.42), and transmits To the traffic control center or roadside control machine, calculated by the corresponding BP neural network, the obtained trigger value is 0.93, 0.93
Figure 996954DEST_PATH_IMAGE034
(0.9, 1) (In the data sample, if the attribute of the variable lane can be changed, set the trigger value to 1, if it cannot be changed, set it to 0, then the trigger condition is satisfied. At this time, the attribute of the variable lane will be changed from Go straight and turn left.

Claims (1)

1. A variable lane adaptive control method, characterized by: the variable lane adaptive control method includes:
1) the training process is that under the premise that the connection weight of each node of the network is fixed and unchanged, the output of each node is calculated layer by layer from the input layer, the output of each node of the output layer is kept unchanged, and the modification quantity of the connection weight is calculated layer by layer from the output layer; if the difference between the network output of the output layer and the expected output exceeds a preset value, carrying out weight adjustment according to the signal error between the network output and the expected output, and finally enabling the output value of the network output layer to be consistent with the expected value, wherein the weight adjustment formula is as follows:
Figure 2010102323131100001DEST_PATH_IMAGE001
，
Figure 422177DEST_PATH_IMAGE002
wherein,
Figure 2010102323131100001DEST_PATH_IMAGE003
adjusting the quantity matrix for the current certain layer of weight;
Figure 401634DEST_PATH_IMAGE004
adjusting the quantity matrix for the previous weight; x represents a layer input vector;
Figure 2010102323131100001DEST_PATH_IMAGE005
is the momentum coefficient;
Figure 56737DEST_PATH_IMAGE006
is a coefficient of proportionality that is,；in order to be able to detect the error signal,
Figure 2010102323131100001DEST_PATH_IMAGE009
；
output vector of the BP neural networkO=The output vector is 0 or 1;
2) the lane-changing control process comprises the following steps:
2.1) if the current variable lane attribute is straight, inputting the acquired time occupancy data into a BP (back propagation) neural network, and if the output vector is 1, changing the variable lane attribute from straight to left-turn when the next straight phase arrives; if the output vector is 0, the current variable lane attribute is unchanged;
2.2) if the current variable lane attribute is left turn, inputting the acquired time occupancy data into a BP neural network, and if the output vector is 1, changing the variable lane attribute from left turn to straight when the next left turn phase arrives; if the output vector is 0, the variable lane keeping left turn attribute is unchanged.
CN2010102323131A2010-07-212010-07-21 A variable lane adaptive control methodPendingCN101901547A (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN2010102323131ACN101901547A (en)2010-07-212010-07-21 A variable lane adaptive control method

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN2010102323131ACN101901547A (en)2010-07-212010-07-21 A variable lane adaptive control method

Publications (1)

Publication NumberPublication Date
CN101901547Atrue CN101901547A (en)2010-12-01

Family

ID=43227047

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN2010102323131APendingCN101901547A (en)2010-07-212010-07-21 A variable lane adaptive control method

Country Status (1)

CountryLink
CN (1)CN101901547A (en)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN102280026A (en)*2011-07-202011-12-14浙江工业大学Vehicle fractal cooperation control method under holographic environment
CN102787569A (en)*2012-08-272012-11-21西北工业大学Control method of junction reversible lane marks and lane line lengths
CN103186984A (en)*2013-04-222013-07-03东南大学Method for triggering transformation of steering function of variable guidance lanes at urban intersections
CN103208196A (en)*2013-04-222013-07-17东南大学Lane adjusting method for urban artery turning traffic
CN103426316A (en)*2013-07-192013-12-04安锐Traffic control method and system based on variable lane
CN103541319A (en)*2013-09-252014-01-29浙江吉利控股集团有限公司Variable lane splitting system
CN106960590A (en)*2017-05-272017-07-18深圳市乃斯网络科技有限公司Intelligent transportation track course distribution method and system
CN107161155A (en)*2017-04-272017-09-15大连理工大学 A method and system for cooperative lane changing of vehicles based on artificial neural network
CN107730920A (en)*2017-10-232018-02-23淮阴工学院 A Dynamic Variable Lane Control Method Based on Spike Lights
CN108177653A (en)*2016-12-082018-06-19本田技研工业株式会社Controller of vehicle
WO2018218410A1 (en)*2017-05-272018-12-06深圳市乃斯网络科技有限公司Method and system for allocating travel path among lanes in intelligent traffic
CN111915894A (en)*2020-08-062020-11-10北京航空航天大学Variable lane and traffic signal cooperative control method based on deep reinforcement learning
CN112419726A (en)*2020-11-202021-02-26华南理工大学Urban traffic signal control system based on traffic flow prediction
CN113570855A (en)*2021-06-222021-10-29阿波罗智联(北京)科技有限公司Variable lane control method, device, equipment and storage medium
CN114694382A (en)*2022-04-012022-07-01桂林电子科技大学 A dynamic one-way traffic control system based on the Internet of Vehicles environment

Citations (4)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
JPH07220193A (en)*1994-01-281995-08-18Nagoya Denki Kogyo Kk Traffic situation determination device
US20080010643A1 (en)*2006-07-072008-01-10Nec Electronics CorporationMultiprocessor system and access right setting method in the multiprocessor system
US20080106436A1 (en)*1997-10-222008-05-08Intelligent Technologies International, Inc.In-Vehicle Signage Techniques
CN101650877A (en)*2009-08-312010-02-17吉林大学Method for setting crossing self-adapting changeable driveway

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
JPH07220193A (en)*1994-01-281995-08-18Nagoya Denki Kogyo Kk Traffic situation determination device
US20080106436A1 (en)*1997-10-222008-05-08Intelligent Technologies International, Inc.In-Vehicle Signage Techniques
US20080010643A1 (en)*2006-07-072008-01-10Nec Electronics CorporationMultiprocessor system and access right setting method in the multiprocessor system
CN101650877A (en)*2009-08-312010-02-17吉林大学Method for setting crossing self-adapting changeable driveway

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
樊晓平等: "基于神经网络的交叉口可变相序模糊控制方法", 《交通运输系统工程与信息》, vol. 8, no. 1, 29 February 2008 (2008-02-29), pages 80 - 85*
许伦辉等: "孤立交叉口多相位自适应模糊控制及其神经网络实现", 《中国公路学报》, vol. 18, no. 3, 31 July 2005 (2005-07-31)*

Cited By (21)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN102280026A (en)*2011-07-202011-12-14浙江工业大学Vehicle fractal cooperation control method under holographic environment
CN102787569A (en)*2012-08-272012-11-21西北工业大学Control method of junction reversible lane marks and lane line lengths
CN102787569B (en)*2012-08-272014-09-10西北工业大学Control method of junction reversible lane marks and lane line lengths
CN103186984A (en)*2013-04-222013-07-03东南大学Method for triggering transformation of steering function of variable guidance lanes at urban intersections
CN103208196A (en)*2013-04-222013-07-17东南大学Lane adjusting method for urban artery turning traffic
CN103186984B (en)*2013-04-222015-01-07东南大学Method for triggering transformation of steering function of variable guidance lanes at urban intersections
CN103426316A (en)*2013-07-192013-12-04安锐Traffic control method and system based on variable lane
CN103426316B (en)*2013-07-192015-04-15安锐Traffic control method and system based on variable lane
CN103541319A (en)*2013-09-252014-01-29浙江吉利控股集团有限公司Variable lane splitting system
CN108177653A (en)*2016-12-082018-06-19本田技研工业株式会社Controller of vehicle
CN107161155A (en)*2017-04-272017-09-15大连理工大学 A method and system for cooperative lane changing of vehicles based on artificial neural network
CN106960590A (en)*2017-05-272017-07-18深圳市乃斯网络科技有限公司Intelligent transportation track course distribution method and system
WO2018218410A1 (en)*2017-05-272018-12-06深圳市乃斯网络科技有限公司Method and system for allocating travel path among lanes in intelligent traffic
CN107730920A (en)*2017-10-232018-02-23淮阴工学院 A Dynamic Variable Lane Control Method Based on Spike Lights
CN111915894A (en)*2020-08-062020-11-10北京航空航天大学Variable lane and traffic signal cooperative control method based on deep reinforcement learning
CN111915894B (en)*2020-08-062021-07-27北京航空航天大学 Variable lane and traffic signal cooperative control method based on deep reinforcement learning
CN112419726A (en)*2020-11-202021-02-26华南理工大学Urban traffic signal control system based on traffic flow prediction
CN112419726B (en)*2020-11-202022-09-20华南理工大学Urban traffic signal control system based on traffic flow prediction
CN113570855A (en)*2021-06-222021-10-29阿波罗智联(北京)科技有限公司Variable lane control method, device, equipment and storage medium
CN113570855B (en)*2021-06-222023-04-28阿波罗智联(北京)科技有限公司Variable lane control method, device, equipment and storage medium
CN114694382A (en)*2022-04-012022-07-01桂林电子科技大学 A dynamic one-way traffic control system based on the Internet of Vehicles environment

Similar Documents

PublicationPublication DateTitle
CN101901547A (en) A variable lane adaptive control method
CN108510764B (en)Multi-intersection self-adaptive phase difference coordination control system and method based on Q learning
CN100444210C (en) A hybrid control method for single-point signal-controlled intersections
CN102722989B (en)Expressway microclimate traffic early warning method based on fuzzy neural network
CN110264757B (en)Intelligent networking automobile layered speed planning method based on continuous signal lamp information
CN106710220B (en)A kind of urban road layering Dynamic coordinated control algorithm and control method
CN103280114B (en) An Intelligent Control Method of Traffic Lights Based on BP-PSO Fuzzy Neural Network
CN104240523B (en)The green ripple control method in arterial street
CN106205156B (en)A kind of intersection self-healing combination control method for the mutation of part lane flow
CN103794065B (en)Active urban road area signal timing parameter collaborative optimization method
CN115083174B (en) A traffic light control method based on cooperative multi-agent reinforcement learning
CN106447119A (en)Short-term traffic flow prediction method and system based on convolutional neural network
Chen et al.Real-coded genetic algorithm for signal timing optimization of a single intersection
CN111461500A (en) A tidal phenomenon control method for shared bicycle system based on dynamic electronic fence and reinforcement learning
Zeng et al.Training reinforcement learning agent for traffic signal control under different traffic conditions
CN115359672A (en) A Traffic Area Boundary Control Method Combining Data-Driven and Reinforcement Learning
Zhao et al.Enhancing traffic signal control with composite deep intelligence
CN111951574A (en) Adaptive Iterative Learning Control Method for Traffic Signals Based on Attenuated Memory Removal
Turky et al.Using genetic algorithm for traffic light control system with a pedestrian crossing
Boukerche et al.Feco: An efficient deep reinforcement learning-based fuel-economic traffic signal control scheme
Mahmood et al.A two stage fuzzy logic adaptive traffic signal control for an isolated intersection based on real data using SUMO simulator
CN116895158A (en)Urban road network traffic signal control method based on multi-agent Actor-Critic and GRU
CN117912223A (en) An intelligent traffic signal control method based on the energy consumption of pure electric vehicles
CN1971655A (en)Method for reducing traffic jam using intelligent traffic information
CN113724507B (en)Traffic control and vehicle guidance cooperative method and system based on deep reinforcement learning

Legal Events

DateCodeTitleDescription
C06Publication
PB01Publication
C10Entry into substantive examination
SE01Entry into force of request for substantive examination
C12Rejection of a patent application after its publication
RJ01Rejection of invention patent application after publication

Application publication date:20101201
[8]ページ先頭
©2009-2025 Movatter.jp