Highway travel time prediction method and system under sparse data conditions

Abstract

The application provides a highway travel time prediction method and system under sparse data conditions, belongs to the technical field of highway management and operation, obtains highway vehicle trajectory data, and calculates a space-time speed field under real traffic conditions; a physical information deep learning model is constructed to estimate the space-time speed field; and an STA_LSTM model is constructed to predict the highway travel time. The application constructs a deep learning model based on physical information for the estimation and reconstruction of the space-time speed field, the model is equipped with prior knowledge of traffic flow theory for the deep learning neural network, and the space-time speed field is more accurately and more general to estimate by using sparse data. Considering the importance of space-time information in traffic data, combining the space and time attention mechanism, an STA_LSTM network model is constructed to predict the travel time, the space weight and time weight are dynamically allocated to the input elements, the space-time information is better captured, and the accuracy and robustness of the travel time prediction are improved.

Description

Technical Field

[0001] This invention relates to the field of highway management and operation technology, specifically to a method and system for predicting highway travel time under sparse data conditions. Background Technology

[0002] In obtaining traffic information for daily commutes, travelers are increasingly prioritizing the accuracy of travel time. With the rapid increase in the number of motor vehicles, numerous contradictions have emerged between the capacity of urban road networks and the number of vehicles, leading to escalating traffic congestion. Travel time prediction, as an indispensable key technology in intelligent transportation systems, provides valuable information for applications such as dynamic travel navigation guidance, signal control during traffic congestion, traffic accident detection, and optimal resource allocation. Furthermore, accurate and reliable travel time not only provides travelers with real-time, referable arrival times but also allows for effective assessment of road network conditions, thus providing better technical support for practical applications in road traffic management.

[0003] To address traditional travel time prediction, Wang Feiyue et al. proposed an LSTM-based method for road segment travel time prediction, comprising two parts: prediction model generation and future time prediction. In the prediction phase, current data undergoes the same processing and is input into the model for future time prediction. An Pengjin constructed a travel time prediction model based on the fusion of attention mechanisms and graph convolution methods, exploring spatial correlations in adjacent roads or trajectories. Subsequently, using deep learning methods combined with attention mechanisms, a sub-model for travel time prediction with paths as the basic unit was established, achieving temporal correlation prediction based on LSTM trees. Sun Pengcheng proposed a deep learning network based on spatiotemporal features to construct a path travel time prediction model, utilizing the spatiotemporal dependency of path travel time. Zhang Shuaiyu proposed a prediction model based on deep temporal convolutional networks and probability error compensation for traffic flow prediction in intelligent transportation systems. Li Xiantong et al. proposed a spatiotemporal feature deep learning network model that considers the spatiotemporal dependency of road segments and the time drift problem in coarse-grained models. Combining convolutional neural networks and long short-term memory networks, it achieves prediction of travel time in the next time slot. Ma Yuting constructed a bus stop travel time prediction model based on graph convolutional neural networks and long short-term memory networks, which comprehensively considers the spatiotemporal characteristics of roads.

[0004] For the application of deep learning models for physical information, Ling Dongshu explored a pedestrian flow model based on deep learning. He combined Voronoi diagrams to partition personal spaces, convolutional-long short-term memory combined neural networks, and a social force model to effectively predict future pedestrian trajectories. Simulation experiments also demonstrated self-organization phenomena, providing important references for pedestrian traffic system modeling and analysis. Yang Xiaoqian, in the context of highway networks, comprehensively considered various factors, including traffic conditions. By establishing a physical information fusion model and using methods based on Monte Carlo methods and deep learning models, she achieved the spatiotemporal distribution prediction of electric vehicle charging load and analyzed its impact on the power distribution network.

[0005] For applications of spatio-temporal attention long short-term memory network models, Liang Xiuxia et al. proposed a traffic flow prediction model based on multi-head self-attention networks. Considering temporal similarity and static characteristics, they introduced daily and weekly data tensors and designed temporal and spatial multi-head attention modules. Liu Yufei et al. proposed a traffic flow prediction model based on spatio-temporal graph attention networks. This model combines graph convolutional neural networks and long short-term memory neural networks to capture the spatial and temporal features of traffic flow. Simultaneously, an attention mechanism was introduced to better consider the impact of traffic information at different times on the prediction. Liao Huiruo and Yang Yan proposed a novel end-to-end deep learning model—Spatio-Temporal Attention ConvLSTM (ST-AttConvLSTM). ST-AttConvLSTM enables the model to extract local spatial features through a residual neural network via three branches for modeling. It also incorporates external factors, utilizing convolutional long short-term memory networks (ConvLSTM) and attention models to mine potential patterns in traffic flow and capture the spatio-temporal correlation of data.

[0006] Model-driven methods require substantial theoretical knowledge and prior information, but the spatiotemporal evolution of traffic flow is highly complex, making it difficult to accurately characterize certain behaviors. Furthermore, due to limitations in the model's mechanism, some spatiotemporal evolutions cannot be accurately reproduced; therefore, model-driven methods are often difficult to apply flexibly to different traffic environments. Data-driven models are highly dependent on data, making them prone to overfitting and undergeneralization. In addition, data model mechanisms lack interpretability and are susceptible to data noise and missing data. Considering data availability and quality, real traffic networks may exhibit data sparsity (leading to data gaps in certain areas and time periods) and noise issues (transmission equipment failures and traffic data storage problems), thus affecting travel time predictions. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for predicting highway travel time under sparse data conditions, so as to solve at least one of the technical problems existing in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a method for predicting highway travel time under sparse data conditions, comprising:

[0010] By acquiring vehicle trajectory data on highways, the spatiotemporal velocity field under real traffic conditions can be calculated.

[0011] Constructing a deep learning model of physical information to estimate the spatiotemporal velocity field;

[0012] Construct a STA_LSTM model to predict highway travel time.

[0013] Furthermore, feature extraction is performed using all available spatial and temporal features in the trajectory data. The data is then processed using a binning method to obtain the spatiotemporal velocity field under observation. The input features are then normalized to the range [0,1] using max-min normalization.

[0014] Furthermore, by combining physical models and deep learning models, physical laws are encoded as prior knowledge and embedded into the loss function of the neural network for training. The difference between the observed values ​​and the estimated values ​​of the deep learning model without physical information is used as the data loss, while the difference between the estimated values ​​of the deep learning model with physical information and the deep learning model without physical information is considered as the physical loss. The loss function for model training is obtained by weighting.

[0015] Furthermore, physical laws are encoded as prior knowledge and embedded into the loss function of the neural network for training. The L-BFGS-B optimization algorithm is used to optimize the parameters of the physical information deep learning model to ensure that the loss function is minimized.

[0016] Furthermore, the construction of the STA_LSTM model to predict highway travel time introduces a spatiotemporal attention mechanism. By dynamically allocating spatial and temporal weights, the LSTM model's ability to capture complex spatiotemporal relationships is enhanced. When predicting travel time, the dynamic changes in temporal and spatial features are considered, thereby improving the accuracy and reliability of the prediction.

[0017] Furthermore, a cosine annealing strategy is adopted to dynamically adjust the learning rate, thereby improving the training effect and generalization ability of the model.

[0018] Secondly, the present invention provides a highway travel time prediction system under sparse data conditions, comprising:

[0019] The acquisition module is used to acquire highway vehicle trajectory data;

[0020] The calculation module is used to calculate the spatiotemporal velocity field under real traffic conditions;

[0021] The first prediction module is used to construct a deep learning model of physical information to estimate the spatiotemporal velocity field.

[0022] The second prediction module is used to build the STA_LSTM model to predict highway travel time.

[0023] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the highway travel time prediction method under sparse data conditions as described in the first aspect.

[0024] Fourthly, the present invention provides a computer device including a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions executable by the processor, and the processor calling the program instructions to execute the highway travel time prediction method under sparse data conditions as described in the first aspect.

[0025] Fifthly, the present invention provides an electronic device comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the highway travel time prediction method under sparse data conditions as described in the first aspect.

[0026] The beneficial effects of this invention are as follows: Considering the quality issues of real-world traffic trajectory data, this invention leverages the advantages of physical models and deep learning models to construct a deep learning model based on physical information for estimating and reconstructing the spatiotemporal velocity field. This model equips the deep learning neural network with prior knowledge of traffic flow theory, enabling more accurate and generalizable estimation of the spatiotemporal velocity field using sparse data. Furthermore, utilizing the spatiotemporal velocity field and considering the importance of spatiotemporal information in traffic data, a spatiotemporal attention long short-term memory network model is constructed by combining spatial and temporal attention mechanisms to predict travel time. This model dynamically allocates spatial and temporal weights to input elements, thereby better capturing spatiotemporal information and improving the accuracy and robustness of travel time prediction.

[0027] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a flowchart of the highway travel time prediction method according to an embodiment of the present invention.

[0030] Figure 2 This is a diagram showing the structure of the spatiotemporal velocity field estimated by the deep learning model for physical information in an embodiment of the present invention.

[0031] Figure 3 This is a diagram illustrating the temporal structure of the spatiotemporal attention long short-term memory network for predicting travel routes, as described in an embodiment of the present invention. Detailed Implementation

[0032] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0033] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0034] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0035] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0036] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0037] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0038] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0039] To address the growing conflict between urban road network capacity and the number of motor vehicles, leading to traffic congestion on daily routes, this invention considers leveraging prior knowledge of physical information combined with the flexibility of data-driven models to construct a deep learning framework for physical information. Using limited data samples, it achieves a more accurate estimation of the spatiotemporal velocity field of the studied road segment. Furthermore, based on the spatiotemporal velocity field and considering a spatiotemporal attention mechanism, it utilizes a spatiotemporal attention long short-term memory network model to predict and analyze travel time. This more accurate and robust travel time prediction, even with limited data quality, can provide valuable information for applications such as dynamic travel navigation guidance, signal control during traffic congestion, traffic accident detection, and optimal resource allocation.

[0040] The purpose of this invention is to leverage prior knowledge of physical information combined with the flexibility of data-driven models to construct a deep learning framework for physical information, enabling more accurate estimation of the spatiotemporal velocity field of a research road segment using limited data samples. Furthermore, based on the spatiotemporal velocity field, a spatiotemporal attention long short-term memory network model is used to predict and analyze travel time. This invention aims to achieve more accurate and robust travel time prediction even with limited data quality.

[0041] Example 1

[0042] In this embodiment 1, a highway travel time prediction system under sparse data conditions is first provided, including: an acquisition module for acquiring highway vehicle trajectory data; a calculation module for calculating the spatiotemporal velocity field under real traffic conditions; a first prediction module for constructing a physical information deep learning model to estimate the spatiotemporal velocity field; and a second prediction module for constructing an STA_LSTM model to predict highway travel time.

[0043] In this embodiment 1, the above-described system is used to implement a method for predicting highway travel time under sparse data conditions. Specifically, the method includes the following steps: Step A, acquiring highway vehicle trajectory data and simultaneously calculating the spatiotemporal velocity field under real traffic conditions; Step B, constructing a physical information deep learning model to estimate the spatiotemporal velocity field; Step C, constructing an STA_LSTM model to predict highway travel time; Step D, hyperparameter sensitivity analysis and comparative analysis with benchmark model predictions.

[0044] In step A, "acquire highway vehicle trajectory data and simultaneously calculate the spatiotemporal velocity field under real traffic conditions." This involves extracting features using all available spatial and temporal features from the trajectory data, processing the data using a binning method to obtain the observed spatiotemporal velocity field, and standardizing the input features to the range [0,1] using max-min normalization.

[0045] The method described in step B for constructing a physical information deep learning model to estimate the spatiotemporal velocity field is characterized by combining a physical model and a deep learning model. Physical laws are encoded as prior knowledge and embedded into the loss function of the neural network for training, thereby improving the model's estimation accuracy of the spatiotemporal velocity field. This model utilizes the difference between observed values ​​and estimates from a physical information-free deep learning model as data loss, while also considering the difference between estimates from the physical information-free deep learning model and estimates from the physical information-free deep learning model as physical loss. The weighted average is then used to obtain the loss function for model training.

[0046] The method described above encodes physical laws as prior knowledge, uses a physics-based LWR model for parameter estimation, and ensures model consistency at both the physical and data levels by solving an optimization problem. Encoding physical laws as prior knowledge and embedding them into the loss function of the neural network for training is characterized by using the L-BFGS-B optimization algorithm to perform parameter optimization of the physical information deep learning model, ensuring the minimization of the loss function. The method described above obtains the model training loss function through weighted summation, and is characterized by the design of the physical information deep learning model loss function, particularly including both data loss and physical loss components. The method described above for constructing a physical information deep learning model to estimate the spatiotemporal velocity field under sparse data conditions can maintain high prediction accuracy even with incomplete data.

[0047] In step C, the STA_LSTM model is constructed to predict highway travel time. A spatiotemporal attention mechanism is introduced, which enhances the LSTM model's ability to capture complex spatiotemporal relationships by dynamically allocating spatial and temporal weights. This spatiotemporal attention mechanism fully considers the dynamic changes in temporal and spatial features when predicting travel time, improving the accuracy and reliability of the prediction. The STA_LSTM model for predicting highway travel time also employs a cosine annealing strategy to dynamically adjust the learning rate, further improving the model's training effect and generalization ability.

[0048] Example 2

[0049] like Figures 1 to 3 As shown in Embodiment 2, a method for predicting highway travel time under sparse data conditions is provided. The specific steps are as follows:

[0050] Step A: Obtain highway vehicle trajectory data, including vehicle number, vehicle location, speed, and timestamp. Simultaneously, calculate the spatiotemporal velocity field under real traffic conditions.

[0051] Step B: Construct a deep learning model of physical information to estimate the spatiotemporal velocity field.

[0052] Step C: Based on the spatiotemporal velocity field, construct the STA_LSTM model to predict highway travel time.

[0053] Step D involves performing parameter sensitivity analysis to examine the impact of selecting key parameters in the STA_LSTM model on the optimization results.

[0054] Specifically, the "construction of urban rail transit network" mentioned in step A is carried out as follows: feature extraction is performed using all available spatial and temporal features in the trajectory data, the data is processed using the binning method to obtain the spatiotemporal velocity field under observation, and the input features are standardized to the range [0,1] using max-min normalization.

[0055] Specifically, in step B, "constructing a deep learning model for physical information," the advantages of physical models and data-driven models are leveraged. Physical laws are encoded as prior knowledge, and by embedding physical principles into a neural network, the model can better understand and capture the physical laws of traffic flow. Combined with the neural network model, it can better adapt to estimation problems under sparse data conditions, generating a data-efficient approximator to process input information and provide estimation results. Based on the physical computation graph, v is calculated. (i) The neural network is denoted as f. λ(s|λ). This network structure combines a physical model and deep learning methods. The prior knowledge of the physical model constrains and guides the estimation problem of the spatiotemporal velocity field, and the parameter set is used to adjust the network structure to optimize the estimation performance.

[0056] Physically based deep learning models encode known physics-based models to acquire prior physical knowledge. Physically unbased deep learning networks can be described as follows: This network structure does not directly contain physical information equations, but rather estimates them by learning the relationships between data. It is a pre-calibrated set of physical model parameters, where θ represents the parameters of the deep learning model without physical information, used to estimate the spatiotemporal velocity field v. (i) This makes the actual observation speed With the estimated velocity v (i) To minimize the error, it can be expressed as:

[0057]

[0058] In the problem of estimating the spatiotemporal velocity field using a deep learning model based on physical information, this invention utilizes observational data to estimate the parameters of a physics-based LWR model. The calibration process of the physical model is defined as an optimization problem, where the objective function Obj can be expressed as:

[0059]

[0060] In this embodiment, the loss function for estimating the spatiotemporal velocity field using deep learning based on physical information consists of two parts: one is minimizing the difference between the observed values ​​and the estimated values ​​from the deep learning model without physical information, i.e., data loss; the other is minimizing the difference between the estimated values ​​based on physical information and the calculated values ​​from the deep learning model without physical information, i.e., physical loss. The mean squared error (MSE) can be expressed as:

[0061]

[0062] In the formula: α — the weights of the balancing data and the physical loss function;

[0063] —Pre-calibrated physical parameters;

[0064] θ — Parameters of a deep learning network without physical information;

[0065] f θ (·)——A deep learning network without physical information, parameterized by θ;

[0066] —The i-th observed feature;

[0067] s (j) —The j-th supplementary feature.

[0068] This embodiment employs the L-BFGS-B (Limited-memory Broyden-Fletcher-Goldfarb-Shanno with Bounds) optimization algorithm. The optimization process iterates continuously on the training data, constantly reducing the value of the loss function until the maximum number of iterations is reached or the loss function converges.

[0069] The phrase "constructing a STA_LSTM model to predict highway travel time" in step C means employing a Spatiotemporal Attention-Based Long Short-Term Memory (STA_LSTM) model to predict travel time. This model combines temporal and spatial attention mechanisms to dynamically allocate spatial and temporal weights to input elements. First, within a single time step, the STA_LSTM model dynamically assigns spatial weights to the input elements, and then adjusts the temporal attention weights in the LSTM hidden layer states at each time step. This structure allows the model to dynamically adjust attention weights based on the spatiotemporal characteristics of the input, thereby more effectively capturing the complex relationships in spatiotemporal data and improving the accuracy and reliability of travel time prediction.

[0070] This embodiment uses velocity fields and historical travel times under different spatiotemporal conditions as input features, and the output is a prediction of the travel time at the next time step. To better capture spatiotemporal relationships and historical trends, this embodiment embeds temporal and spatial attention weights into the LSTM unit. These weights affect the input and output of the LSTM unit. Through spatial and temporal attention modules, the STA_LSTM model can adaptively adjust its attention weights according to specific spatiotemporal conditions, thereby improving the performance and prediction accuracy of the LSTM unit and enabling the model to more effectively cope with different spatiotemporal conditions.

[0071] In the spatial attention mechanism, the input feature matrix x can be divided into k m-dimensional vectors, which can be represented as:

[0072] After the model computes a single layer of neurons, the input feature vector will pass through Sigmoid(x) = 1 / (1+e) -x Function activation, and further through Standardization is then used to generate spatial attention weights. This can be expressed as:

[0073]

[0074] The model performing the Hadamard product operation, i.e., term-by-term product, can be represented as:

[0075]

[0076] Subsequently, the temporal attention weight β, after undergoing ReLU activation ReLU(x) = max(0,x) and Softmax normalization operations, can be expressed as:

[0077] β=TA(H)=[β1,β2,...,β k ] 1×k

[0078] The product of matrices, where the output layer generates the final prediction without activation, can be represented as: h att ∈R 1×s

[0079] This embodiment uses the root mean square error (RMSE) to evaluate the model's performance on the test data. The root mean square error is calculated from the MSE and can be expressed as:

[0080]

[0081]

[0082] Where: N test — Test dataset size.

[0083] This embodiment employs a strategy of dynamically adjusting the learning rate using cosine annealing.

[0084] Specifically, step D, "performing parameter sensitivity analysis to analyze the impact of key parameter selection in the STA_LSTM model on the optimization results," involves adjusting hyperparameters, including learning rate, batch size, period, and key parameters such as the initial cosine period length and period length multiple under the cosine annealing algorithm. This allows for the analysis of the model's predictive performance.

[0085] Example 3

[0086] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When executed by a processor, the computer instructions implement the highway travel time prediction method under sparse data conditions as described above. The method includes:

[0087] Step A: Obtain highway vehicle trajectory data, including vehicle number, vehicle location, speed, and timestamp. Simultaneously, calculate the spatiotemporal velocity field under real traffic conditions.

[0088] Step B: Construct a deep learning model of physical information to estimate the spatiotemporal velocity field.

[0089] Step C: Based on the spatiotemporal velocity field, construct the STA_LSTM model to predict highway travel time.

[0090] Step D involves performing parameter sensitivity analysis to examine the impact of selecting key parameters in the STA_LSTM model on the optimization results.

[0091] Example 4

[0092] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, and the memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the highway travel time prediction method under sparse data conditions as described above, the method including:

[0093] Step A: Obtain highway vehicle trajectory data, including vehicle number, vehicle location, speed, and timestamp. Simultaneously, calculate the spatiotemporal velocity field under real traffic conditions.

[0094] Step B: Construct a deep learning model of physical information to estimate the spatiotemporal velocity field.

[0095] Step C: Based on the spatiotemporal velocity field, construct the STA_LSTM model to predict highway travel time.

[0096] Step D involves performing parameter sensitivity analysis to examine the impact of selecting key parameters in the STA_LSTM model on the optimization results.

[0097] Example 5

[0098] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the highway travel time prediction method under sparse data conditions as described above, the method including:

[0099] Step A: Obtain highway vehicle trajectory data, including vehicle number, vehicle location, speed, and timestamp. Simultaneously, calculate the spatiotemporal velocity field under real traffic conditions.

[0100] Step B: Construct a deep learning model of physical information to estimate the spatiotemporal velocity field.

[0101] Step C: Based on the spatiotemporal velocity field, construct the STA_LSTM model to predict highway travel time.

[0102] Step D involves performing parameter sensitivity analysis to examine the impact of selecting key parameters in the STA_LSTM model on the optimization results.

[0103] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0105] These computer program instructions may also be stored in a computer-readable 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 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.

[0106] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed 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.

[0107] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A method for predicting highway travel time under sparse data conditions, characterized in that, include: Step A: Obtain highway vehicle trajectory data and calculate the spatiotemporal velocity field under real traffic conditions; Step B: Construct a physical information deep learning model to estimate the spatiotemporal velocity field; this includes: combining the physical model and the deep learning model, encoding physical laws as prior knowledge, embedding them into the loss function of the neural network for training, using the difference between the observed values ​​and the estimated values ​​of the deep learning model without physical information as the data loss, and simultaneously considering the difference between the estimated values ​​of the physical information deep learning model and the deep learning model without physical information as the physical loss, obtaining the loss function for model training through weighting; encoding physical laws as prior knowledge, embedding them into the loss function of the neural network for training, and using the L-BFGS-B optimization algorithm to perform parameter optimization of the physical information deep learning model to ensure the minimization of the loss function; Step C: Based on the estimated spatiotemporal velocity field, construct an STA_LSTM model to predict highway travel time; this includes: introducing a spatiotemporal attention mechanism, which enhances the LSTM model's ability to capture complex spatiotemporal relationships by dynamically allocating spatial and temporal weights; considering the dynamic changes in temporal and spatial features when predicting travel time to improve the accuracy and reliability of the prediction; and employing a cosine annealing strategy to dynamically adjust the learning rate to improve the model's training effect and generalization ability.

2. The method for predicting highway travel time under sparse data conditions according to claim 1, characterized in that, Step A includes: extracting features using all available spatial and temporal features in the trajectory data, processing the data using a binning method to obtain the spatiotemporal velocity field under observation, and standardizing the input features to the range [0,1] using max-min normalization.

3. A highway travel time prediction system under sparse data conditions based on the method as described in claim 1 or 2, characterized in that, include: The acquisition module is used to acquire highway vehicle trajectory data; The calculation module is used to calculate the spatiotemporal velocity field under real traffic conditions; The first prediction module is used to construct a deep learning model of physical information to estimate the spatiotemporal velocity field. The second prediction module is used to build the STA_LSTM model to predict highway travel time.

4. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the highway travel time prediction method under sparse data conditions as described in claim 1 or 2.

5. A computer device, characterized in that, The system includes a memory and a processor, which communicate with each other. The memory stores program instructions that can be executed by the processor, and the processor invokes the program instructions to execute the highway travel time prediction method under sparse data conditions as described in claim 1 or 2.

6. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions to implement the highway travel time prediction method under sparse data conditions as described in claim 1 or 2.

Images