Highway traffic flow prediction method and electronic device

Abstract

The present disclosure relates to a highway traffic flow prediction method, which receives highway monitoring equipment collected highway traffic data, forms a fixed interval time highway traffic sequence data; through a variational gated recurrent unit auto-encoding network, the correlation between the traffic sequence data and multiple dimensions is mapped to a hidden space, which is used to extract the long and short term time sequence characteristics of the traffic sequence data; the current real-time collected data is input into the variational gated recurrent unit auto-encoding network, and the traffic prediction information at the future time is obtained, which has a certain robustness to the noise caused by device abnormalities, human operations and other factors, and further shows more reliable and accurate traffic prediction performance.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent transportation, and in particular to a method and electronic device for predicting highway traffic flow. Background Technology

[0002] With the increasing number of highways and the growing transportation demand from travelers, the imbalance between highway supply and demand is intensifying. This, coupled with highway congestion and frequent accidents, has become an obstacle to the stable development of highways. Therefore, acquiring historical traffic flow data through detection equipment on highways and conducting statistical analyses to predict future traffic flow changes is of paramount importance for highway safety management.

[0003] Reliable and accurate traffic flow forecasting is fundamental to the intelligent management of smart highways. Traffic flow forecasting methods can be broadly categorized into two types: one is based on statistical mathematical models, such as Kalman filtering and time series forecasting methods, which are generally suitable for linear and stationary time series data; the other is based on artificial intelligence models, including artificial neural networks and support vector machine regression, which can effectively mine nonlinear time series data.

[0004] In the process of predicting highway traffic flow, the first step is to acquire highway traffic flow data. Currently, road traffic flow data collection mainly relies on integrated detectors, cameras, loop detectors, and geomagnetic sensors. Improving the accuracy of traffic prediction models requires statistical analysis of large amounts of data, making continuous and accurate traffic flow information collection essential. However, in real-world environments, data loss can occur due to equipment malfunctions or improper human operation, directly leading to erroneous data mining and predictions by data-driven prediction models. Furthermore, convolutional neural networks, leveraging their translation and scale invariance characteristics, have been used to mine the temporal and spatial dependencies of traffic flow, improving the performance of large-scale traffic flow prediction models through temporal dual-dimensional feature modeling. However, these discriminative deep learning models rely on large amounts of data for neural network training, and they cannot provide robust predictive performance for traffic flow prediction modeling with missing or outlier data. Summary of the Invention

[0005] This disclosure provides an exemplary embodiment of a highway traffic flow prediction method and electronic device, which makes the reconstructed traffic flow more similar to the actual traffic flow, thereby exhibiting a higher robust traffic flow prediction capability.

[0006] A first aspect of this disclosure provides a traffic flow prediction method, the method comprising:

[0007] It receives highway vehicle passage data collected from highway monitoring equipment and forms highway traffic flow sequence data at fixed intervals.

[0008] The highway traffic flow sequence data is Z-score standardized to obtain normalized traffic flow sequence data;

[0009] The traffic flow sequence data and the correlation between multiple dimensions are mapped to the latent space through a variational gated recursive unit autoencoder network to extract the long and short-term time series features of the traffic flow sequence data. The variational gated recursive unit autoencoder network models the probability distribution of the highway traffic flow sequence data through variational inference, and the probability distribution of the variables in the latent space is continuous.

[0010] The currently collected data is input into the variational gated recursive unit autoencoder network to obtain traffic prediction information for future time moments.

[0011] In one embodiment, the variational gated recurrent unit autoencoder network includes an input layer for accepting the normalized flow sequence data x = {x1, x2, ..., x}. T Let p be the joint distribution of the latent variable z and the true variable. θ (z; x), the latent vector z is derived from the prior distribution p. θ The data points x are sampled from (z), and the reconstructed data points are obtained from the distribution p conditioned on z. θ It is generated in (x|z), where p θ (x|z) represents the distribution of the normalized flow sequence data x given the latent variable z.

[0012] In one embodiment, the variational gated recurrent unit autoencoder network includes an encoding layer for encoding input traffic sequence data x to obtain the feature distribution of the latent variable z. The traffic sequence data x is input to the GRU recurrent neural network, and the output of the GRU recurrent neural network is an encoding vector h. enc The Gaussian distribution parameters of the latent variable z, including the mean μ and variance σ, are obtained through linear transformation. 2 The approximate posterior distribution q of the latent variable z is obtained. φ (z|x).

[0013] In one embodiment, the variational gated recurrent unit autoencoder network includes a sampling layer, which is used to sample the encoded latent variable z to obtain the feature representation vector of the flow sequence. Samples generate feature representation z, where from ε is obtained by upsampling, and z is calculated using reparameterization.

[0014]

[0015]

[0016] z=μ+σ×ε

[0017] Where k represents the dimension of the latent variable z.

[0018] In one embodiment, the variational gated recurrent unit autoencoder network includes an output layer. This output layer maps the latent variable features to the target output through a fully connected layer. The latent variable features z are mapped to the final prediction result y. The number of hidden layer units is n1, and the specific mapping is as follows:

[0019] y = W3*z + b3

[0020] in, These are the weights and bias parameters of the fully connected layer.

[0021] In one embodiment, the loss function of the variational gated recurrent unit autoencoder network consists of a first loss function and a second loss function. The first loss function is a negative log-likelihood function, which measures the reconstruction error of the input data. The second loss function is the KL divergence (Kullback–Leibler divergence) between the true distribution and the approximate posterior distribution, which measures the similarity between the two distributions.

[0022] In one embodiment, the variational gated recurrent unit autoencoder network employs adaptive moment estimation (Adam), which has an RMSprop of momentum term. The learning rate of each parameter is dynamically adjusted using the first-order moment estimation and second-order moment estimation of the gradient. After bias correction, the learning rate has a defined range for each iteration.

[0023] This invention combines short-term data collected on actual highways with a variational autoencoder to map the original monitored traffic flow data and the correlations between multiple dimensions into a latent space, enabling deeper extraction of temporal features. Simultaneously, the variational autoencoder can model the probability distribution of highway traffic flow data through variational inference. Since the probability distribution of variables learned in the latent space is continuous, it exhibits robustness to noise caused by factors such as equipment malfunctions and human error. Based on reliable traffic flow data, a fully connected neural network is used to perform supervised prediction model learning, thereby achieving an accurate and robust highway traffic flow prediction model. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating a traffic flow prediction method according to one embodiment of the present disclosure;

[0026] Figure 2 This is a schematic diagram of a traffic flow prediction method according to an embodiment of the present disclosure;

[0027] Figure 3 This is a structural diagram of a gated recursive unit according to an embodiment of the present disclosure;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0030] In this disclosure, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0031] The application scenarios described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure, and do not constitute a limitation on the technical solutions provided in this disclosure. As those skilled in the art will know, with the emergence of new application scenarios, the technical solutions provided in this disclosure are also applicable to similar technical problems. In the description of this disclosure, unless otherwise stated, "multiple" means two or more.

[0032] This disclosure utilizes short-term data collected from actual highways to learn and construct a potential generation distribution that conforms to the distribution of real highway traffic flow data by combining variational autoencoders and gated recurrent unit (GRU) recurrent neural networks. During the encoding and decoding process of the variational autoencoder, the time-series variation characteristics of traffic flow data are considered. The GRU recurrent neural network is used to represent and predict future traffic flow sequences. On the one hand, by combining the GRU recurrent neural network, the long-term and short-term time-series features of the data can be effectively extracted. On the other hand, by combining the variational autoencoder, the original monitored traffic flow data and the correlations between multiple dimensions can be mapped into the latent space, achieving deeper extraction of time-series features. Simultaneously, the variational autoencoder can model the probability distribution of highway traffic flow data through variational inference, and the probability distribution of variables learned in the latent space is continuous, thus providing robustness to noise caused by factors such as equipment malfunctions and human operation.

[0033] This disclosure provides a method for predicting highway traffic flow, including:

[0034] It receives highway vehicle passage data collected from highway monitoring equipment and forms highway traffic flow sequence data at fixed intervals.

[0035] The highway traffic flow sequence data is Z-score standardized to obtain normalized traffic flow sequence data;

[0036] The traffic flow sequence data and the correlation between multiple dimensions are mapped to the latent space through a variational gated recursive unit autoencoder network to extract the long and short-term time series features of the traffic flow sequence data. The variational gated recursive unit autoencoder network models the probability distribution of the highway traffic flow sequence data through variational inference, and the probability distribution of the variables in the latent space is continuous.

[0037] The currently collected data is input into the variational gated recursive unit autoencoder network to obtain traffic prediction information for future time moments.

[0038] This embodiment uses a fully connected neural network to perform supervised prediction model learning based on reliable traffic data, thereby achieving an accurate and robust highway traffic prediction model.

[0039] Another embodiment, such as Figure 1 As shown, the traffic data acquisition module S1 acquires vehicle data through highway monitoring equipment such as checkpoints and electronic police systems, and stores it in a time-series database according to a standard format for subsequent processing and statistical analysis.

[0040] The traffic flow data preprocessing module S2 first preprocesses the vehicle passage data records and integrates them into traffic flow data within fixed time intervals, thus forming highway traffic flow sequence data. The vehicle passage data recorded by the monitoring equipment is converted into traffic flow sequence data S = {s1,…,s...} at specified time intervals. t ,…,s T}, where T represents the total length of the time series, and the time interval between adjacent elements is 5 minutes. The original flow sequence data is Z-score standardized to obtain the normalized sequence data X. The standardization calculation formula is:

[0041]

[0042] Where μ represents the mean of the original sequence S, and σ represents the standard deviation of the original sequence S.

[0043] The traffic prediction model training module S3 specifically includes:

[0044] S31: The input layer is used to receive the traffic sequence after data preprocessing;

[0045] S32: The encoding layer is used to encode the input traffic sequence to obtain the latent variable feature distribution;

[0046] S33: The sampling layer is used to sample the latent variable feature distribution after encoding to obtain the encoded feature vector;

[0047] S34: The decoding layer is used to decode the encoded feature vector to reconstruct the input data;

[0048] S35: The output layer is used to map the learned latent variable features to the target output, and a fully connected layer based on the sigmoid activation function is used to map the high-dimensional temporal feature vector to the final target prediction sequence.

[0049] The traffic flow prediction model training module S3, comprising a variational inference submodule and a prediction output submodule, first processes the raw traffic flow data using a variational autoencoder network, then accurately learns the latent distribution patterns of the traffic flow data through variational time-series derivation, and finally maps continuous latent variable features to the final traffic flow prediction output. The specific steps of the traffic flow prediction model training module S3 are as follows:

[0050] S31: The input layer receives the traffic flow sequence after data preprocessing. Assume the actual input traffic flow is x = {x1, x2, ..., x...} T Let p be the joint distribution of the latent variable z and the true variable. θ (z;x).

[0051] The latent vector z can be obtained from the prior distribution p θData points x are sampled from (z), and then reconstructed from the distribution p conditioned on z. θ It is generated in (x|z).

[0052] The variational autoencoder model defines the joint distribution p of the real data and the latent variables. θ (z;x)=p θ (x|z)p θ (z).

[0053] Variational inference is equivalent to calculating the posterior probability of z given the observed variable x, which can be defined according to Bayes' theorem as follows:

[0054]

[0055] Where, p θ (z|x) represents the posterior distribution of the latent variable z given the actual traffic flow x, p θ (x|z) represents the distribution of the actual flow x given the latent variable z. However, directly calculating p... θ (x) is very time-consuming and difficult to process, and an approximate posterior distribution q that can be assumed is usually used. φ (z|x) is used to approximate the true posterior distribution p. θ (z|x).

[0056] S32: The coding layer is used to encode the input flow sequence x to obtain the feature distribution of the latent variable z.

[0057] like Figure 2 The diagram shows a model for a traffic flow prediction method, based on a variational gated recursive unit autoencoder network.

[0058] First, let the encoder build a multivariate Gaussian model q with a diagonal covariance structure. φ (z|x)=N(z|μ,diag(σ 2 The network is trained on real traffic volume data using a GRU recurrent neural network. The network input is traffic flow data x, and the network output is an encoded vector. Then, the Gaussian distribution parameters of the latent variable z, namely the mean μ and variance σ, are obtained through linear transformation. 2 Next, by sampling this normal distribution, we can obtain the approximate posterior distribution q of the latent variable z. φ (z|x).

[0059] The input x is encoded by the GRU network and calculated as follows.

[0060]

[0061] in This represents the hidden state output of the GRU coding network at time t. Specifically, the GRU unit node contains two key sigmoid gates: the reset gate r t and update gate z t Used to control data flow.

[0062] Specifically, reset gate r t Determine whether it is necessary to combine the current state with previous historical information and update gate z. t Define the amount of information to retain.

[0063] Figure 4 The diagram shows the structure of a GRU gated recursive unit, where the data transfer process is calculated as follows:

[0064]

[0065] Where, x t Let {W} represent the input flow sequence data at time t. r W z ,Wh} and {U r U z ,Uh} represents the weight matrix parameters, {b r ,b z ,bh} represents the bias vector, h t-1 This represents the hidden state at the previous time step, and σ and ⊙ represent the sigmoid activation function and the element-wise multiplication operator, respectively.

[0066] S33: The sampling layer is used to sample the encoded latent variable z to obtain the feature representation vector of the flow sequence; however, directly from... The sampling operation performed by the distribution on μ and σ 2 Since the network is not differentiable, gradient descent cannot be used for optimization during network training. Therefore, reparameterization from variational autoencoder training is employed, starting from... ε is obtained by upsampling, and then z is calculated using reparameterization, thus the result is differentiable.

[0067]

[0068]

[0069] z=μ+σ×ε (7)

[0070] Where k represents the dimension of the latent variable z. In this operation, the process from encoder output to z involves only linear operations and can be optimized using gradient descent.

[0071] S34: The decoding layer is used to decode the sampled latent variable feature vectors to reconstruct the input data x.

[0072] This embodiment employs a GRU network with the same length as the corresponding encoder, where each GRU has the same number of hidden units as the encoder, which is m. The decoder aims to output an input that is as identical as possible to the encoder's input, defined as follows.

[0073]

[0074]

[0075] in, Let be the hidden state output of the GRU network at time t. Represents the initial cell state value in the decoder network, where

[0076] The message passing process in the GRU unit refers to formula (4) above, and the decoder output is uniformly expressed as...

[0077] The goal of the variational inference submodule described above is to minimize q while ensuring that the reconstruction error of the input data is sufficiently small. φ (z|x) and p θ The KL divergence between the (z|x) distributions is used to model traffic flow distribution in a continuous latent space, achieving data imputation for missing traffic data, and thus enabling further application in robust traffic flow prediction. The composite loss function of the variational inference submodule is defined as follows:

[0078]

[0079] The first term represents the expected reconstruction loss, or the expected negative log-likelihood function of the i-th data point, which can obtain the feature distribution on the model encoder, thereby enabling the model's decoder to accurately reconstruct the data.

[0080] The second term represents the KL divergence of the regularization term, which is used to measure the divergence of q. φ (z|x) approximates p θ The amount or degree of information loss is (z|x), so by minimizing the KL divergence, we can obtain an approximate value for p. θ (z|x), where x represents the input vector and z represents the latent variables in the latent space.

[0081] For the observed variable x = {x 1 ,x 2 ,x 3 ,…,x Z} and latent variable z = {z 1 ,z 2 ,z 3 ,…,zM The joint probability distribution p of the joint representation θ (x; z).

[0082] Variational inference problems involve calculating the conditional distribution p of latent variables given observations. θ (z|x)=p θ (z,x) / p θ (x), where p θ (x)=∫p θ (z|x)p θ (z)dz.

[0083] However, directly calculating p θ The cost of (z|x) is relatively high. To ensure computational feasibility, variational inference assumes a distribution q. φ (z|x) is used to approximate the reasoning of p. θ If (z|x), then the distribution approximation can be achieved by minimizing the KL divergence between the two distributions.

[0084]

[0085] Furthermore, in order to infer the accurate distribution of latent variables, the optimization of the neural network model first requires minimizing p. θ (z|x) and q φ KL divergence between (z|x),

[0086]

[0087] However, this sampling process is non-differentiable, making the network untrainable. Therefore, referencing the reparameterization technique of variational autoencoders, we assume that the latent variable z follows a standard normal distribution. The actual traffic flow x follows a normal distribution, and the latent variable z is distributed as follows: The KL divergence between the latent variable distribution and the approximate latent variable distribution is then recalculated as follows:

[0088]

[0089] Wherein, mean μ and variance σ 2 By training a GRU network with real traffic flow data x, minimizing the above formula yields a distribution q of the latent variable z under a normal distribution condition, where the real traffic flow x follows a normal distribution. φ (z|x), and thus the true distribution p of the latent variables is obtained. θ (x|z), thus completing the encoding of the real traffic flow x into the latent variable z.

[0090] Therefore, to facilitate network training, the loss function of the variational inference module in the proposed model can be redefined as follows:

[0091]

[0092] Here, k represents the dimension of the latent variable z.

[0093] Finally, accurate latent variable features are obtained through unsupervised training, and then the latent variable features that generate the traffic distribution are passed to the output layer to complete the final traffic prediction output.

[0094] S35: The output layer contains a fully connected layer that maps the latent variable features z to the final predicted result y. The number of hidden layer units is n1. The specific mapping calculation is as follows.

[0095] y = W3*z + b3 (15)

[0096] in, These are the weights and bias parameters of the fully connected layer.

[0097] Finally, the final traffic prediction output y is obtained through output layer mapping.

[0098] Furthermore, in order to ensure relatively stable optimization parameters during the overall training of the proposed model, an adaptive moment estimation optimization algorithm is employed in one embodiment of this disclosure to optimize the proposed model. Specifically, the Adam optimization algorithm is based on gradient descent with momentum and incorporates an algorithm called RMSprop. It dynamically adjusts the learning rate of each parameter using the first-order and second-order moment estimates of the gradient, and its specific expression is as follows:

[0099]

[0100] Where, m t This represents the update of the partial first-order moment estimate; v t This is represented as an update of the partial second-moment estimate; This is expressed as a correction for the deviation of the first moment. This represents the deviation of the corrected second moment, with parameter α = 0.001, parameter β1 = 0.9, parameter β2 = 0.999, and parameter ∈ = 10. -8 The main advantage of Adam is that, after bias correction, the learning rate has a defined range for each iteration, making the parameters relatively stable.

[0101] Furthermore, the traffic prediction model prediction module S4 is executed, which inputs the data into the trained robust prediction model to output the traffic prediction results for future times.

[0102] As can be seen from the above process, the traffic flow regeneration process can use a GRU network to approximate the real training data. Therefore, the loss function of the variational autoencoder network can be optimized through the backpropagation method in neural network training, making the reconstructed traffic flow more similar to the real traffic flow. The intermediate hidden variable z generated through this optimization process will no longer be affected by data loss due to equipment or human operation problems, and can reflect the continuous potential feature distribution of the generated traffic flow, thus enabling a more robust traffic flow prediction model.

[0103] The technical effectiveness of this embodiment was demonstrated through simulation experiments using a real traffic dataset. Specifically, two months of vehicle passage data were collected from five checkpoints on the Qingdao section of the Shenhai Expressway. After processing, traffic data with a 5-minute time interval was obtained. The first month's data was used for model training, and the second month's data was used for model testing. Finally, the performance was compared with two typical deep learning models (LSTM and GRU). The experimental results used MSE, MAPE, and RMSE as evaluation metrics.

[0104] Comparison of Highway Traffic Flow Prediction Results from Three Different Models

[0105]

[0106] The comparative analysis of the results in the table above shows that the traffic prediction model based on variational autoencoder network has the ability to resist noise in abnormal traffic data on highways and exhibits more accurate traffic prediction results.

[0107] Compared to LSTM and GRU neural network prediction models, the embodiments of this disclosure exhibit lower traffic flow prediction errors and achieve more accurate prediction precision in terms of MSE, RMSE, and MAPE metrics. This is due to two main reasons: firstly, the use of a variational autoencoder to infer the potential distribution of highway traffic flow, accurately modeling the characteristics of this potential distribution through variational inference, and thus generating accurate traffic flow data; secondly, the use of a GRU recurrent neural network to characterize the temporal characteristics of traffic flow sequence evolution, enabling the model to accurately model the nonlinear temporal dependence of highway traffic flow and accurately characterize the long-term and short-term sequence features of the traffic flow sequence. Furthermore, since variational inference is used to model the probability distribution of highway traffic flow data, and the probability distribution of variables learned in the latent space is continuous, it exhibits robustness to noise caused by factors such as equipment malfunctions and human error, thus demonstrating more reliable and accurate traffic flow prediction performance.

[0108] The highway traffic flow prediction method based on variational autoencoder networks (GRUs) provided in this disclosure fully utilizes the feature extraction and noise resistance capabilities of GRUs, combined with the advantages of long and short period time series modeling in gated recurrent unit (GRU) neural networks. Taking into account phenomena such as missing traffic data samples due to malfunctions in highway traffic monitoring equipment or manual operation, a robust and high-precision traffic flow prediction model is achieved. Specifically, the GRU network is used for the posterior probability q. φ Modeling the parameters of (z|x) effectively extracts the time dependence in the flow sequence data. On the other hand, the variational encoder encodes the latent variables through random sampling, which can reduce the interference of noise in the original sensor data to a certain extent. The prediction model built on the reliable dataset shows more robust and accurate prediction results, thus providing reliable technical support for high-speed safety and flow diversion.

[0109] Having introduced a traffic flow prediction method according to an exemplary embodiment of the present disclosure, we will now introduce an electronic device according to another exemplary embodiment of the present disclosure.

[0110] Those skilled in the art will understand that various aspects of this disclosure can be implemented as systems, methods, or program products. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: entirely in hardware, entirely in software (including firmware, microcode, etc.), or in a combination of hardware and software, collectively referred to herein as “circuit,” “module,” or “system.”

[0111] In some possible implementations, the electronic device according to this disclosure may include at least one processor and at least one computer storage medium. The computer storage medium stores program code that, when executed by the processor, causes the processor to perform the steps in the traffic flow prediction method according to the various exemplary embodiments of this disclosure described above.

[0112] The following reference Figure 4 To describe an electronic device 600 according to such an embodiment of the present disclosure. Figure 4 The electronic device 600 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0113] like Figure 4 As shown, the electronic device 600 is presented in the form of a general-purpose electronic device. The components of the electronic device 600 may include, but are not limited to: at least one processor 601, at least one computer storage medium 602, and a bus 603 connecting different system components (including the computer storage medium 602 and the processor 601).

[0114] Bus 603 represents one or more of several bus structures, including a computer storage media bus or computer storage media controller, peripheral bus, processor, or local bus using any of the various bus structures.

[0115] Computer storage medium 602 may include readable media in the form of volatile computer storage media, such as random access computer storage medium (RAM) 621 and / or cache storage medium 622, and may further include read-only computer storage medium (ROM) 623.

[0116] The computer storage medium 602 may also include a program / utility 625 having a set (at least one) of program modules 624, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0117] Electronic device 600 can also communicate with one or more external devices 604 (e.g., keyboard, pointing device, etc.), and with one or more devices that enable a user to interact with electronic device 600, and / or with any device that enables electronic device 600 to communicate with one or more other electronic devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 605. Furthermore, electronic device 600 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 606. As shown, network adapter 606 communicates with other modules used in electronic device 600 via bus 603. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0118] In some possible implementations, various aspects of the traffic flow prediction method provided in this disclosure may also be implemented in the form of a program product, which includes program code that, when the program product is run on a computer device, causes the computer device to perform the steps in the image search method according to the various exemplary embodiments of this disclosure described above.

[0119] The program product may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access computer storage media (RAM), read-only computer storage media (ROM), erasable programmable read-only computer storage media (EPROM or flash memory), optical fibers, portable compact disk read-only computer storage media (CD-ROM), optical computer storage media, magnetic computer storage media, or any suitable combination thereof.

[0120] The image search program product of the embodiments of this disclosure can be a portable compact disc read-only computer storage medium (CD-ROM) and include program code, and can run on an electronic device. However, the program product of this disclosure is not limited thereto. In this document, the readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0121] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0122] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0123] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's electronic device, partially on the user's device, as a standalone software package, partially on the user's electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In cases involving remote electronic devices, the remote electronic device can be connected to the user's electronic device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external electronic device (e.g., via the Internet using an Internet service provider).

[0124] It should be noted that although several modules of the apparatus have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided and embodied by multiple modules.

[0125] Furthermore, although the operations of the methods disclosed herein are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all of the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0126] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk computer storage media, CD-ROMs, optical computer storage media, etc.) containing computer-usable program code.

[0127] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce functions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] These computer program instructions may also be stored in a computer-readable computer storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable computer storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0130] Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from its spirit and scope. Therefore, if such modifications and variations fall within the scope of the claims of this disclosure and their equivalents, this disclosure is also intended to include such modifications and variations.

Claims

1. A method of predicting traffic flow on a highway, characterized by, The method includes: It receives highway vehicle passage data collected from highway monitoring equipment and forms highway traffic flow sequence data at fixed intervals. The highway traffic flow sequence data is Z-score standardized to obtain normalized traffic flow sequence data; The traffic flow sequence data and the correlation between multiple dimensions are mapped to the latent space through a variational gated recursive unit autoencoder network to extract the long and short-term time series features of the traffic flow sequence data. The variational gated recursive unit autoencoder network models the probability distribution of the highway traffic flow sequence data through variational inference, and the probability distribution of the variables in the latent space is continuous. The currently collected real-time data is input into the variational gated recursive unit autoencoder network to obtain traffic prediction information for future time moments; The variational gated recurrent unit autoencoder network includes an input layer, which is used to accept the normalized flow sequence data. The joint distribution of the latent variable z and the true variable is defined as follows: The latent variable z is derived from the prior distribution The data points x were obtained by sampling from the z-based distribution and reconstructed from the z-based distribution. Produced in, among which This represents the distribution of the normalized flow sequence data x given the latent variable z; The variational gated recurrent unit autoencoder network includes an encoding layer, which establishes a multivariate Gaussian model with a diagonal covariance structure. This is used to encode the input traffic sequence data x to obtain the feature distribution of the latent variable z, wherein the traffic sequence data x is input into the GRU recurrent neural network, and the output of the GRU recurrent neural network is an encoding vector. And the Gaussian distribution parameters of the latent variable z, including the mean, are obtained through linear transformation. and variance This yields the approximate posterior distribution of the latent variable z. ; The variational gated recurrent unit autoencoder network includes a sampling layer, which is used to sample the encoded latent variable z to obtain the feature representation vector of the flow sequence, from a normal distribution. The sampled feature representation z is generated, where the feature is derived from the standard normal distribution. Upsampling And use reparameterization to compute z. ； ； ； Where k represents the dimension of the latent variable z, W1 and W2 represent the weight matrix parameters, and b2 and b3 represent the bias vectors. , , The variational gated recurrent unit autoencoder network includes a decoding layer, which encodes the sampled hidden vector. The decoding layer uses a GRU network with the same length as the corresponding encoding layer, and each GRU has m hidden units. The output of the decoding layer is defined as: ； ； in, Let be the hidden state output of the GRU network at time t. This represents the initial cell state value in the decoding layer, where , The output of the decoding layer is represented as .

2. The method for predicting highway traffic flow according to claim 1, characterized in that, The variational gated recurrent unit autoencoder network includes an output layer, which maps the latent variable features to the target output through a fully connected layer. The latent variable features z are mapped to the final prediction result y, wherein the number of hidden layer units is [number missing]. The specific mapping is as follows: ； in, , These are the weights and bias parameters of the fully connected layer.

3. A method for predicting highway traffic flow according to any one of claims 1-2, characterized in that, The loss function of the variational gated recurrent unit autoencoder network consists of a first loss function and a second loss function. The first loss function is the negative log-likelihood function, which is used to measure the reconstruction error of the input data. The second loss function is the KL divergence between the true distribution and the approximate posterior distribution, which is used to measure the similarity between the two distributions.

4. A method for predicting highway traffic flow according to any one of claims 1-2, characterized in that, The variational gated recurrent unit autoencoder network employs adaptive moment estimation, which includes RMSprop of the momentum term. The learning rate of each parameter is dynamically adjusted using the first-order moment estimation and second-order moment estimation of the gradient. After bias correction, the learning rate has a defined range for each iteration.

5. An electronic device, characterized in that, Includes storage units and a processor; wherein: The processor is configured as follows: It receives highway vehicle passage data collected from highway monitoring equipment and forms highway traffic flow sequence data at fixed intervals. The highway traffic flow sequence data is Z-score standardized to obtain normalized traffic flow sequence data; The traffic flow sequence data and the correlation between multiple dimensions are mapped to the latent space through a variational gated recursive unit autoencoder network to extract the long and short-term time series features of the traffic flow sequence data. The variational gated recursive unit autoencoder network models the probability distribution of the highway traffic flow sequence data through variational inference, and the probability distribution of the variables in the latent space is continuous. The currently collected data is input into the variational gated recursive unit autoencoder network to obtain traffic prediction information for future moments; The variational gated recurrent unit autoencoder network includes an input layer, which is used to accept the normalized flow sequence data. The joint distribution of the latent variable z and the true variable is defined as follows: The latent variable z is derived from the prior distribution The data points x were obtained by sampling from the z-based distribution and reconstructed from the z-based distribution. Produced in, among which This represents the distribution of the normalized flow sequence data x given the latent variable z; The variational gated recursive unit autoencoder network includes an encoding layer, which establishes a multivariate Gaussian model with a diagonal covariance structure. This is used to encode the input traffic sequence data x to obtain the feature distribution of the latent variable z, wherein the traffic sequence data x is input into the GRU recurrent neural network, and the output of the GRU recurrent neural network is an encoding vector. And the Gaussian distribution parameters of the latent variable z, including the mean, are obtained through linear transformation. and variance This yields the approximate posterior distribution of the latent variable z. ; The variational gated recurrent unit autoencoder network includes a sampling layer, which is used to sample the encoded latent variable z to obtain the feature representation vector of the flow sequence, from a normal distribution. The sampled feature representation z is generated, where the feature is derived from the standard normal distribution. Upsampling And use reparameterization to compute z. ； ； ； Where k represents the dimension of the latent variable z, W1 and W2 represent the weight matrix parameters, and b2 and b3 represent the bias vectors. , ; The variational gated recurrent unit autoencoder network includes a decoding layer, which encodes the sampled hidden vector. The decoding layer employs a GRU network with the same length as the corresponding encoding layer, and each GRU hidden unit has the following number of units: The output of the decoding layer is defined as: ； ； in, Let be the hidden state output of the GRU network at time t. This represents the initial cell state value in the decoding layer, where , The output of the decoding layer is represented as .

Images