A signalized intersection vehicle delay distribution estimation method

Abstract

The present application relates to a kind of signal intersection vehicle delay distribution estimation method, the method utilizes vehicle delay distribution model based on connected vehicle trajectory data to carry out signal intersection vehicle delay distribution estimation, the vehicle delay distribution model includes convolutional neural network and mixed density network, the method includes: based on the connected vehicle trajectory data construction connected vehicle position matrix, the input feature is obtained by stacking the connected vehicle position matrix;The input feature is input into the convolutional neural network to extract feature data;Based on the feature data, utilize the mixed density network to output the distribution of vehicle delay label;Based on the distribution of vehicle delay label, obtain vehicle delay distribution.Compared with prior art, the present application has the advantages of high estimation accuracy, strong traffic scene adaptability and low data quality requirement.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traffic signal control and traffic parameter estimation, in particular to a signal intersection vehicle delay distribution estimation method. BACKGROUND

[0002] Adaptive control of signal intersection is considered as the most basic means to improve traffic efficiency. Vehicle delay distribution is an important basis for evaluating and optimizing intersection signal control scheme. Connected vehicle trajectory provides detailed driving information of vehicles near the intersection, and mining the required traffic information from it is a kind of traffic technology with great application prospect. The information richness of connected vehicle trajectory also brings complexity of data structure, so the method of extracting information from it needs careful design. Chinese patent application CN116226592A provides a kind of intersection queue length estimation method based on sparse connected vehicle trajectory data, which estimates the queue length at the intersection by using sparse connected vehicle trajectory data, constructs the speed matrix under the red light state, the speed matrix under the green light state and the standard deviation matrix of the speed based on the connected vehicle trajectory at the intersection, and estimates the queue length at the intersection based on the above-mentioned matrix. Although it can effectively infer the queue length in the intersection, the calculation complexity involved in the process of constructing the matrix is high, which is not conducive to real-time estimation, and although it considers various data, it ignores the influence of the penetration rate of connected vehicles on the queue length estimation.

[0003] Therefore, it is a technical problem to provide a method that can simplify the calculation process and take the penetration rate of connected vehicles into account. SUMMARY

[0004] The purpose of the present application is to overcome the defects of the prior art and provide a signal intersection vehicle delay distribution estimation method based on convolutional neural network feature information mining and hybrid density network signal intersection vehicle delay distribution estimation, which has the advantages of high estimation accuracy and low data quality requirement.

[0005] The purpose of the present application can be achieved by the following technical solutions:

[0006] A signal intersection vehicle delay distribution estimation method, the method uses a vehicle delay distribution model to estimate the signal intersection vehicle delay distribution based on connected vehicle trajectory data, the vehicle delay distribution model includes a convolutional neural network and a hybrid density network, and the method includes:

[0007] Based on the connected vehicle trajectory data, a connected vehicle position matrix is constructed, and the connected vehicle position matrix is stacked to obtain input features;

[0008] Input the input features into the convolutional neural network to extract feature data;

[0009] Based on the feature data, output a distribution of vehicle delay labels using the hybrid density network;

[0010] Obtain a vehicle delay distribution based on the distribution of vehicle delay labels.

[0011] As a preferred technical solution, the method for constructing a connected vehicle position matrix comprises:

[0012] Obtain time information of the connected vehicle trajectory data, and perform grid division on all entry roads of the intersection based on the time information and the connected vehicle trajectory data;

[0013] According to the grid division result, determine whether there is a vehicle in each grid, and if there is, set the corresponding grid element to 1, and if there is not, set the corresponding grid element to 0, to generate the connected vehicle position matrix.

[0014] As a preferred technical solution, the method for stacking processing comprises:

[0015] Divide the time information into multiple signal periods;

[0016] Obtain all connected vehicle position matrices in the same signal period, and perform time dimension stacking to generate a three-dimensional tensor of time-lane-distance;

[0017] Obtain three-dimensional tensors of all different signal periods, and perform time dimension splicing to generate the input features.

[0018] As a preferred technical solution, the method for outputting a distribution of vehicle delay labels using the hybrid density network comprises:

[0019] Generate a hybrid distribution component of the hybrid density network, and the expression is:

[0020]

[0021] wherein z represents a vehicle delay label; x is feature data; μ k (x) is an expectation vector of the Gaussian component distribution; Σ k (x) is a diagonal covariance matrix of the Gaussian component distribution wherein σ k.m is the standard deviation of the mth dimension of the kth Gaussian component distribution, and C is the dimension of the Gaussian component distribution;

[0022] Estimate the distribution of the vehicle delay label based on the hybrid distribution component, and the expression is:

[0023]

[0024] where p(z|x) is the distribution of vehicle delay label conditioned on the feature data x, K is the number of components of Gaussian mixture distribution, π k (x) is the mixing coefficient of the kth Gaussian component distribution, p k (z|x) is the mixing distribution component of the kth mixture density network.

[0025] As a preferred technical solution, the training method of the vehicle delay distribution model comprises the following steps:

[0026] Obtaining historical connected vehicle trajectory data, and obtaining historical input features based on the historical connected vehicle trajectory data;

[0027] Calculating vehicle delay values based on the historical connected vehicle trajectory data, and generating vehicle delay labels based on the vehicle delay values;

[0028] Iteratively performing the following steps until the training stop condition is met:

[0029] Randomly selecting a batch of historical input features, inputting the vehicle delay distribution model to be trained, and outputting corresponding historical vehicle delay probability density;

[0030] Calculating a loss value based on the historical vehicle delay probability density and the vehicle delay labels;

[0031] Performing backpropagation to calculate model parameter gradients;

[0032] Optimizing and updating the model parameters of the vehicle delay distribution model based on the model parameter gradients.

[0033] As a preferred technical solution, the method for generating vehicle delay labels comprises the following steps:

[0034] Obtaining historical time information of the historical connected vehicle trajectory data, and obtaining vehicle arrival and departure times of intersections based on the historical time information;

[0035] Calculating vehicle travel times based on the vehicle arrival and departure times of intersections;

[0036] Obtaining expected speeds and travel distances of vehicles, and calculating expected travel times of vehicles;

[0037] Calculating vehicle delay values based on the vehicle travel times and the expected travel times of vehicles;

[0038] Transforming the vehicle delay values to generate the vehicle delay labels.

[0039] As a preferred technical solution, the transformation comprises power transformation and batch normalization, and the expression is:

[0040] z = sigma -1 (y λ -mu),

[0041] wherein z represents a vehicle delay label, lambda represents a parameter of power transformation, sigma represents a standard deviation of batch normalization transformation estimation, mu represents an expectation of batch normalization transformation, and y represents a vehicle delay value.

[0042] As a preferred technical solution, the method for obtaining the vehicle delay distribution is inverse transformation of the distribution of the vehicle delay label.

[0043] As a preferred technical solution, the training stop condition comprises: a percentage of decrease of the loss function in the current iteration compared with the loss function before the current iteration is less than a preset value; or, the number of iterations reaches a maximum number of iterations.

[0044] As a preferred technical solution, the method for calculating the loss value is:

[0045]

[0046] wherein x' represents a historical input feature, x represents a feature data, z represents a vehicle delay label, y represents a vehicle delay value, D represents a data set composed of the historical input feature and the vehicle delay label, p(z|x) is a distribution of the vehicle delay label under the condition of the feature data x, lambda is a parameter of power transformation, sigma is a standard deviation of batch normalization transformation estimation, and mu is an expectation of batch normalization transformation.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1) The present application constructs a connected vehicle position matrix through grid division, stacks the position matrix according to a period, obtains input features containing time and space information, maximally maintains rich information contained in sparse connected vehicle trajectory data, and can automatically extract information required for vehicle delay distribution estimation, compared with existing traffic parameter estimation methods based on connected vehicle trajectory, the data processing process is simpler and faster, and has higher estimation accuracy.

[0049] 2) The present application realizes modeling of vehicle delay distribution in a data-driven manner based on a convolutional neural network and a hybrid density network in a deep learning method, can be applied to complex and variable traffic conditions, and has lower requirements for data quality compared with existing analytical estimation methods of vehicle delay distribution. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The method flowchart of the present application;

[0051] Figure 2A trajectory data processing flowchart of the present application;

[0052] Figure 3 An input feature acquisition flowchart of the present application;

[0053] Figure 4 A vehicle delay distribution model training flowchart of the present application. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0055] Deep learning uses one or more neural networks to approximate the mapping relationship between traffic states and corresponding vehicle delay distributions. Through a large amount of data-driven learning, the neural network fits this mapping relationship so that the output estimation result is close to the true result. Since deep learning has the advantages of fast calculation and description of complex problems, the present application proposes a signal intersection vehicle delay distribution estimation method based on a hybrid density network, the flow of which is shown in Figure 1 As shown in the figure, based on the collected network-connected vehicle trajectory data, the input features and output labels constituting the data set are processed, secondly, the vehicle delay distribution estimation model is constructed based on the convolutional neural network and the hybrid density network, the result obtained by inputting data into the model is used to calculate the negative log-likelihood loss function, then the gradient of the model parameters is calculated by executing the back propagation process, and finally the model parameters are updated using the Adam algorithm based on the gradient. The learning process is repeated to guide the convergence. The mapping relationship between the traffic state reflected by the network-connected vehicle trajectory and the signal intersection vehicle delay distribution is learned by using the hybrid density network, and the advantages of strong universality and high calculation efficiency are highlighted by introducing the network-connected vehicle trajectory processing, delay transformation and hybrid density network and other improvement mechanisms. The detailed steps are as follows:

[0056] S1, data processing:

[0057] The processing of the acquired network-connected vehicle trajectory data of the present application includes two parts as shown in Figure 2 The first part is to acquire input features based on network-connected vehicle trajectory data, and the second part is to acquire vehicle delay labels based on network-connected vehicle trajectory data.

[0058] S11, acquiring input features:

[0059] S111, obtain time information of the connected vehicle trajectory data, divide all entry roads of the intersection into grids based on the time information and the connected vehicle trajectory data, as shown in Figure 3

[0060] S112, according to the grid division result, determine whether there is a vehicle in each grid, if yes, set the corresponding grid element to 1, if no, set the corresponding grid element to 0, and generate a connected vehicle position matrix.

[0061] S113, divide the time information into multiple signal periods.

[0062] S114, obtain all connected vehicle position matrices in the same signal period, and stack in the time dimension to generate a three-dimensional tensor of time-lane-distance.

[0063] S115, obtain three-dimensional tensors of all different signal periods, and splice in the time dimension to generate input features.

[0064] S12, obtain vehicle delay labels:

[0065] S121, obtain historical time information of historical connected vehicle trajectory data, and obtain the time of vehicle arrival and departure from the intersection based on the historical time information.

[0066] S122, calculate the vehicle travel time based on the time of vehicle arrival and departure from the intersection.

[0067] S123, obtain the expected speed and travel distance of the vehicle, and calculate the expected travel time of the vehicle.

[0068] S124, subtract the vehicle travel time from the vehicle expected travel time to obtain the vehicle delay value.

[0069] S125, perform power transformation and batch normalization on the vehicle delay value to generate the vehicle delay label, the expression is:

[0070] z=σ -1 (y λ -μ),

[0071] wherein z represents the vehicle delay label, λ represents the power transformation parameter, σ represents the standard deviation estimated by batch normalization transformation, μ represents the expectation of batch normalization transformation, and y represents the vehicle delay value.

[0072] S2, input the input features into a convolutional neural network to extract feature data, and input the feature data as input of the hybrid density network.

[0073] S3, output the distribution of vehicle delay labels by using the hybrid density network:

[0074] ​S31, generating a mixed distribution component of the mixed density network, expressed as:

[0075]

[0076] wherein z represents a vehicle delay label; x is feature data; μ k (x) is an expectation vector of the Gaussian component distribution; and k (x) is a diagonal covariance matrix of the Gaussian component distribution wherein σ k.m is a standard deviation of the mth dimension of the kth Gaussian component distribution, wherein C is the dimension of the Gaussian component distribution;

[0077] S32, estimating a distribution of the vehicle delay label based on the mixed distribution component, expressed as:

[0078]

[0079] wherein p(z|x) is a distribution of the vehicle delay label under the condition of the feature data x, K is the number of components of the Gaussian mixed distribution, π k (x) is a mixing coefficient of the kth Gaussian component distribution, and p k (z|x) is a mixed distribution component of the kth mixed density network.

[0080] S4, inversely transforming the distribution of the vehicle delay label to obtain a vehicle delay distribution.

[0081] In the present application, the training process of the vehicle delay distribution model is as shown in Figure 4 , comprising the following steps:

[0082] A1, obtaining historical connected vehicle trajectory data, and obtaining historical input features based on the historical connected vehicle trajectory data.

[0083] A2, constructing a vehicle delay label according to the process of step S12.

[0084] Steps A3-A6 are iterated until the percentage of decrease of the loss function in the present iteration compared with the loss function before the present iteration is less than a preset value, or the number of iterations reaches a maximum number of iterations:

[0085] A3, randomly selecting a batch of historical input features, inputting the vehicle delay distribution model to be trained, and outputting corresponding historical vehicle delay probability density.

[0086] A4, calculating a loss value based on the historical vehicle delay probability density and the vehicle delay label, expressed as:

[0087]

[0088] wherein x' represents historical input features, x represents feature data, z represents a vehicle delay label, y represents a vehicle delay value, D represents a data set composed of historical input features and vehicle delay labels, p(z|x) is a distribution of a vehicle delay label under the condition of feature data x, λ is a parameter of a power transformation, σ is a standard deviation of a batch normalization transformation estimation, and μ is an expectation of a batch normalization transformation.

[0089] A5, performing a back propagation calculation model parameter gradient.

[0090] A6, based on the model parameter gradient, using the Adam algorithm to optimize and update the model parameters of the vehicle delay distribution model, and the specific parameter update form is:

[0091]

[0092] In the formula, θ is a model parameter, L is a loss value, and α is a learning rate.

[0093] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for estimating vehicle delay distribution at a signalized intersection, characterized in that, The method utilizes a vehicle delay distribution model to estimate vehicle delay distribution at a signalized intersection based on connected vehicle trajectory data, the vehicle delay distribution model comprising a convolutional neural network and a mixture density network, and the method comprises: constructing a connected vehicle position matrix based on the connected vehicle trajectory data, and stacking the connected vehicle position matrix to obtain input features; inputting the input features into the convolutional neural network to extract feature data; outputting a distribution of vehicle delay labels using the mixture density network based on the feature data; obtaining a vehicle delay distribution based on the distribution of vehicle delay labels; wherein the method of outputting the distribution of vehicle delay labels using the mixture density network comprises: generating a mixture distribution component of the mixture density network, the expression of which is: ， wherein z represents a vehicle delay label; x is feature data; is a mean vector of the Gaussian component distribution; is a diagonal covariance matrix of the Gaussian component distribution wherein is a standard deviation of the mth dimension of the kth Gaussian component distribution, wherein C is a dimension of the Gaussian component distribution; estimating the distribution of vehicle delay labels based on the mixture distribution component, the expression of which is: wherein, is a distribution of vehicle delay labels under feature data x, K is the number of components of the Gaussian mixture distribution, is a mixing coefficient of the kth Gaussian component distribution, is a mixture distribution component of the kth mixture density network.

2. The method of estimating vehicle delay distribution at a signalized intersection according to claim 1, wherein, The method of constructing the connected vehicle position matrix comprises: obtaining time information of the connected vehicle trajectory data, and performing grid division on all entry roads of the intersection based on the time information and the connected vehicle trajectory data; determining whether there is a vehicle in each grid according to the grid division result, and setting the corresponding grid element to 1 if there is a vehicle, or setting the corresponding grid element to 0 if there is no vehicle, to generate the connected vehicle position matrix.

3. The method of estimating vehicle delay distribution at a signalized intersection according to claim 2, wherein, The method of stacking processing comprises: dividing the time information into multiple signal periods; obtaining all connected vehicle position matrices in the same signal period, and stacking them in the time dimension to generate a three-dimensional tensor of time-lane-distance; obtaining three-dimensional tensors of all different signal periods, and splicing them in the time dimension to generate the input features.

4. The method of estimating vehicle delay distribution at a signalized intersection of claim 1, wherein, The training method of the vehicle delay distribution model comprises: obtaining historical connected vehicle trajectory data, and obtaining historical input features based on the historical connected vehicle trajectory data; calculating vehicle delay values based on the historical connected vehicle trajectory data, and generating vehicle delay labels based on the vehicle delay values; iteratively performing the following steps until the training stop condition is met: randomly selecting a batch of historical input features, inputting them into the vehicle delay distribution model to be trained, and outputting corresponding historical vehicle delay probability densities; calculating a loss value based on the historical vehicle delay probability densities and the vehicle delay labels; performing back propagation to calculate model parameter gradients; optimizing and updating model parameters of the vehicle delay distribution model based on the model parameter gradients.

5. The method of estimating vehicle delay distribution at a signalized intersection according to claim 4, wherein, The method of generating vehicle delay labels comprises: obtaining historical time information of the historical connected vehicle trajectory data, and obtaining vehicle arrival and departure times of the intersection based on the historical time information; calculating vehicle travel times based on the vehicle arrival and departure times of the intersection; obtaining expected speed and travel distance of the vehicle, and calculating expected travel time of the vehicle; calculating vehicle delay values based on the vehicle travel times and the expected travel time of the vehicle; transforming the vehicle delay values to generate the vehicle delay labels.

6. The method of estimating vehicle delay distribution at a signalized intersection according to claim 5, wherein, The transformation comprises power transformation and batch normalization, and the expression is: ， where z denotes the vehicle delay label, denotes a parameter of the power transformation, denotes the standard deviation of the batch normalization transformation estimate, denotes the expectation of the batch normalization transformation, y denotes the vehicle delay value.

7. The method of estimating vehicle delay distribution at a signalized intersection according to claim 6, wherein, The method of obtaining the vehicle delay distribution comprises inversely transforming the distribution of vehicle delay labels.

8. The method of estimating vehicle delay distribution at a signalized intersection of claim 4, wherein, The training stopping condition comprises: a percentage of decrease of the loss function in the current iteration compared with the loss function before the current iteration is less than a preset value; or, the number of iterations reaches a maximum number of iterations.

9. The method of estimating vehicle delay distribution at a signalized intersection of claim 4, wherein, The calculation method of the loss value is: ， wherein, represents historical input features, x represents feature data, z represents vehicle delay labels, y represents vehicle delay values, D represents a dataset consisting of historical input features and vehicle delay labels, is a distribution of vehicle delay labels conditioned on feature data x, is a parameter of the power transformation, is a standard deviation of the batch normalization transformation estimate, is an expectation of the batch normalization transformation.

Images