Urban road traffic state intelligent estimation method based on microphysical representation

Abstract

The invention discloses an urban road traffic state intelligent estimation method based on microphysical representation, and the method comprises the steps: constructing a road section boundary condition based on the arrival and departure accumulated flow information of an intersection and a gate, and forming multi-source traffic perception input through combining with road section observation data; constructing a physical-data hybrid driving model, constructing a deep learning model in a data driving branch, learning a mapping relation between road section boundary cumulative flow and a traffic state in a road section, and converting an indistinguishable Newell traffic flow model into a distinguishable computational graph structure through function approximation and structural conversion in a distinguishable physical branch; and constructing a loss function of the hybrid drive model, so that data drive output and a physical model are kept coordinated, and traffic flow parameters are jointly estimated. Through deep embedding of the microphysical calculation graph, the dependence of a pure physical model on an ideal assumed condition is made up, and meanwhile, the problem that a pure data driving method is insufficient in generalization ability in a sensing blind area is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of intelligent transportation, and relates to an urban road traffic state intelligent estimation method based on a differentiable physical representation. BACKGROUND

[0002] Modern urban traffic systems have deployed a variety of sensing devices, including intersection detectors, loop cameras, millimeter wave radars, and floating car GPS systems, and traffic perception data is increasingly abundant. However, due to limitations such as device layout location, coverage range, and sampling frequency, there are still widespread perception blind spots in the road network. Intersection detectors provide boundary cross-section flow, but cannot cover the interior of the road segment; floating cars reflect part of the trajectory, but have limited penetration rate. Perception blind spots make it difficult for traffic management departments to fully grasp the road network operating conditions. Therefore, how to use multi-source perception data to estimate blind spot traffic states and reveal operating rules is an important challenge for intelligent transportation systems.

[0003] Existing traffic state modeling and estimation methods can be broadly divided into three categories: physical model-based estimation methods, data-driven estimation methods, and physical information neural network-based fusion estimation methods.

[0004] Physical model-based estimation methods take traffic flow theory as the core, establish mathematical models to describe the spatio-temporal evolution of vehicles, and use boundary observation data to infer the internal traffic state of the road segment. The estimation principle relies on the traffic flow conservation law and wave propagation theory, and according to the flow information of the upstream and downstream boundaries, the traffic state of the unobserved area is calculated using analytical solution or numerical iteration. Typical methods include the characteristic line solution based on the LWR partial differential equation, the discretized cell transmission model, and the cumulative curve reconstruction technology based on the Newell envelope theory. This type of method relies on the traffic flow conservation principle and shock wave theory, uses theoretical models such as cumulative vehicle number curves, and infers the internal traffic state of the road segment according to boundary observation data. However, these methods are based on idealized traffic flow assumptions such as first-in-first-out and steady-state conditions, which are often difficult to meet in complex and variable urban traffic environments, resulting in significant deviations between model prediction results and actual traffic conditions.

[0005] Data-driven estimation methods directly model the observed data collected by existing perception networks, and establish the mapping relationship between input observations and target states by learning the spatio-temporal distribution patterns of traffic states in historical observation data. The estimation principle is based on statistical inference or machine learning, and a large amount of historical data is used to train model parameters to realize the prediction and interpolation of unobserved locations or future traffic states. Mainly including ARIMA model based on time series analysis, support vector machine based on kernel method, random forest of ensemble learning, and deep neural network (such as convolutional neural network CNN, recurrent neural network RNN, graph neural network GNN, etc.). Such methods are essentially modeling the observable spatio-temporal region of the existing perception network, and are severely dependent on the coverage and distribution characteristics of the training data. For the perception blind area and the spatio-temporal region not fully covered by the training data, the generalization ability and estimation reliability of the model are insufficient, and it is difficult to realize accurate estimation of the full spatio-temporal traffic state. In addition, the internal mechanism of the model lacks physical meaning and has poor interpretability, and cannot reveal the internal operation law of traffic flow.

[0006] The fusion estimation method based on physical information neural network combines the mechanism of physical model and the flexibility of data-driven method, and embeds the physical constraint of traffic flow in the neural network training process. The estimation principle is to design the loss function or the network structure to make the model satisfy the physical law while fitting the observation data, and to realize the cooperation of data-driven and physical mechanism. The implementation includes adding partial differential equation residual term as physical regularization constraint in the loss function, designing network layer structure with physical perception, or using multi-task learning framework to optimize data fitting and physical consistency at the same time. Although this method tries to introduce physical constraints into the neural network, the existing work mainly realizes it by adding physical regularization term in the loss function, and lacks the differentiable reconstruction of traffic physical model from the bottom structure. Although this shallow fusion method can improve the estimation effect to a certain extent, it cannot fully play the role of physical model in revealing the operation law of traffic flow, and also limits the synchronous learning and dynamic optimization of physical parameters. SUMMARY

[0007] The present application aims to provide an intelligent estimation method for urban road traffic state based on differentiable physical representation, which can estimate the traffic state in the perception blind area of the road section and simultaneously infer the key traffic flow parameters, and provide technical support for urban road traffic operation monitoring and signal control optimization.

[0008] Technical scheme: In order to achieve the above invention purpose, the intelligent estimation method for urban road traffic state based on differentiable physical representation provided by the present application comprises the following steps:

[0009] The road segment boundary conditions are constructed based on the arrival and departure cumulative flow information provided by the intersection detector and the toll gate equipment, and a multi-source traffic perception input is formed in combination with the observation data available in the road segment;

[0010] A physical-data hybrid driving model is constructed, in which a data-driven branch and a differentiable physical branch are combined in parallel; a deep learning model is constructed in the data-driven branch to learn the mapping relationship between the road segment boundary cumulative flow and the traffic state in the road segment, and in the differentiable physical branch, a non-differentiable Newell traffic flow model is converted into a differentiable calculation graph structure through function approximation and structure conversion; the data-driven branch and the differentiable physical branch share the same input data, and independently output traffic state estimation results, respectively;

[0011] A loss function of the hybrid driving model is constructed to keep the data-driven output and the physical model in coordination, and to jointly estimate the traffic flow parameters; the hybrid model is optimized and trained to realize traffic state estimation and obtain traffic flow characteristic parameters.

[0012] Preferably, the input data of the physical-data hybrid driving model includes boundary conditions and internal observation data, wherein the boundary conditions include time series of arrival cumulative curves and departure cumulative curves, which are extracted from intersection detector data; the internal observation data are extracted from multi-source perception information including floating car trajectory data and millimeter wave radar detection data.

[0013] Preferably, the deep learning model of the data-driven branch adopts an encoder-decoder framework to process time series input, and introduces an attention mechanism to enhance the modeling of the dependency relationship between the input historical state and the output spatial distribution; the encoder compresses the input cumulative flow time series into a fixed-dimensional vector representation, and the decoder generates a full-space traffic state profile at the target time based on the vector representation; the input tensor shape of the deep learning model is , wherein is the time window length, 2 represents the cumulative arrival-departure sequence, the target output is a full-space discrete profile at the prediction time, and the dimension is equal to the number of spatial grids.

[0014] Preferably, the encoder adopts a single-layer LSTM encoder to update the hidden state step by step for the input sequence, obtain the encoding representation of each time step, and calculate the attention weight to obtain the weighted representation of each time step; the context vector is obtained by aggregating in the time dimension, which is used to initialize the decoder; in the decoding stage, a step-by-step LSTM structure is adopted, which only processes one time step at a time, and the hidden state of the previous time step and the current input are jointly transmitted into the LSTM to obtain the hidden state of the current step; then, it is mapped to a dimension corresponding to the number of spatial grids through a fully connected layer to obtain the spatial traffic state profile at the target time.

[0015] As preferred, in the differentiable physical branch, the Newell traffic flow model is transformed into a differentiable function expression as follows:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021] wherein, denotes time, denotes spatial position, is the cumulative vehicle number at the space-time point, denote the upper and lower bounds of the differentiable cumulative, respectively, denote the continuous upstream arrival sequence and downstream departure sequence, respectively, and denote the differentiable weight functions of the upstream and downstream at time , position , respectively, is the time window length, is the free flow speed, is the jam wave speed, is the congestion density, is the road segment length, denotes the Sigmoid function, is the smoothing parameter.

[0022] As preferred, in the physical-data hybrid driving model, the input and supervision target of the data-driven model are as follows: based on the cumulative arrival curve and the cumulative departure curve of the upstream and downstream, a time series input is constructed, and a relative processing is performed through the maximum cumulative value in the window to form an encoder input sequence of the data-driven model; the relative amount of the full-space cumulative vehicle number at the end time of the window is taken as the supervision target, wherein the observation position is provided by the multi-source perception data inside the road segment to form a limited observation point set; the input of the differentiable physical branch includes the sliding window data of the arrival and departure, the spatial discrete grid of the road segment and the corresponding relative reference value.

[0023] As preferred, the loss function of the physical-data hybrid driving model considers consistency loss and data-driven supervision loss; the consistency loss is expressed as:

[0024]

[0025] wherein, denotes a set of full space-time grid points, denotes a data-driven branch output, denotes a physical branch output; the data-driven supervised loss is represented as:

[0026]

[0027] wherein, denotes a true value; a total objective function is:

[0028]

[0029] wherein, for balancing between physical mechanism fitting and data fitting.

[0030] As preferred, the training process of the physical-data hybrid driven model comprises: firstly, data-driven branch pre-training, wherein only the data-driven branch parameters are updated to achieve a stable fitting on the full space supervision target, early stopping is adopted and the optimal weight is saved; secondly, joint training, wherein the pre-training weight is loaded, the joint forward of the two branches is enabled, and the joint optimization is enabled; the optimizer acts on the parameters of the two branches respectively, and an adaptive learning rate annealing strategy based on the total loss is used to improve the late convergence accuracy.

[0031] The application further provides a computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the intelligent estimation method of urban road traffic state based on a differentiable physical representation.

[0032] The application further provides a computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the steps of the intelligent estimation method of urban road traffic state based on a differentiable physical representation.

[0033] Beneficial effects: the intelligent estimation method of urban road traffic state based on a differentiable physical representation provided by the application converts a traditional non-differentiable traffic flow model into a differentiable computational graph form through function approximation and structure reconstruction, realizes deep embedding of the physical model in a deep learning framework, and on this basis, constructs a physical-data hybrid driven estimation framework, comprehensively utilizes boundary flow information of an intersection detector and multi-source perception data in a road section, realizes full space-time estimation of traffic states in a perception blind area, and simultaneously learns key traffic flow characteristic parameters such as free flow speed, blocked wave speed and crowded density in a training process. Compared with the prior art, the application has the following obvious advantages:

[0034] 1. Physical model differentiability and parameter synchronous estimation integration. The present application converts the traditional non-differentiable Newell traffic flow model into a computable graph that can be back-propagated, enabling it to be trained end-to-end with data-driven networks in a unified framework, and simultaneously learning key traffic flow parameters while reconstructing traffic states, without the need for prior independent calibration, significantly improving modeling accuracy and adaptability.

[0035] 2. Hybrid model fusion and traffic flow rule revelation. The physical-data hybrid driving model constructed by the present application can be compatible with multi-source heterogeneous traffic perception data, fully utilizing the complementary advantages of various data sources. Through the deep embedding of the differentiable physical computational graph, the dependence of pure physical models on ideal assumptions is compensated, and the problem of insufficient generalization ability in the perception blind area of pure data-driven methods is overcome. At the same time, the hybrid model can accurately estimate the traffic state while fully revealing the internal operation rules of traffic flow and synchronously identifying key characteristic parameters, providing physically interpretable decision-making basis for traffic management and control. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The method flowchart of the embodiment of the present application.

[0037] Figure 2 The technical flowchart of the embodiment of the present application.

[0038] Figure 3 The approximate differentiable table illustration of the Newell traffic flow model in the embodiment of the present application.

[0039] Figure 4 The physical-data hybrid driving model framework diagram in the embodiment of the present application.

[0040] Figure 5 The loss reduction curve diagram of the first training stage (pure LSTM model) in the embodiment of the present application.

[0041] Figure 6 The loss reduction curve diagram of the second training stage (hybrid model) in the embodiment of the present application.

[0042] Figure 7 The Newell model traffic parameter estimation result screenshot of the computational graph in the second stage of the embodiment of the present application.

[0043] Figure 8 The performance comparison diagram of various models under different permeability in the embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions of the present application will be further described below in combination with the drawings and specific embodiments.

[0045] For better understanding of the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0046] Multi-source traffic perception data: refers to the traffic operation data collected by different sensing devices such as intersection detectors, loop cameras, millimeter wave radars and floating car GPS in urban traffic system.

[0047] Computational graph: a computational structure representing mathematical models as directed graphs, with nodes representing operation operations and edges representing data flow, supporting automatic differentiation and backpropagation.

[0048] Traffic state estimation: using limited observation data, the traffic operation state of the area not directly observed is inferred by modeling and calculation method.

[0049] Perception blind area: due to the limitations of sensing device layout position, coverage range, sampling frequency and the like, the road space-time area that cannot be directly observed or the observation data is sparse.

[0050] As shown in Figure 1 The embodiments of the present application disclose a kind of intelligent estimation method of urban road traffic state based on differentiable physical representation, first, based on the cumulative traffic information of arrival and departure provided by intersection detector and loop device, the boundary condition of road section is constructed, combined with part of observation data (millimeter wave radar, floating car trajectory) available in road section to form multi-source traffic perception input;Then, a physical-data hybrid driven model is constructed, and the data driven branch and the differentiable physical branch are combined in parallel;In the data driven branch, a deep learning model is constructed, and the mapping relationship between the cumulative traffic of road section boundary and the traffic state in road section is learned, in the differentiable physical branch, function approximation and structure conversion are used to convert the non-differentiable Newell traffic flow model into a differentiable computational graph structure;Data driven branch and differentiable physical branch share the same input data, and respectively independently output traffic state estimation result;Loss function of hybrid driven model is constructed, so that data driven output and physical model keep coordination, and traffic flow parameters are estimated jointly;The hybrid model is optimized and trained, traffic state estimation is realized, and traffic flow characteristic parameters are obtained.

[0051] The intelligent estimation method of urban road traffic state based on differentiable physical representation provided by the embodiments of the present application can fully compatible with multi-source traffic perception data, realize the estimation of road section perception blind area traffic state under the AI calculation and training framework, and simultaneously infer key traffic flow parameters, to provide technical support for urban road traffic operation monitoring and signal control optimization.

[0052] Exemplary, Figure 2 The detailed implementation process of the intelligent estimation method of urban road traffic state based on differentiable physical representation is shown in the figure, which specifically includes the following steps:

[0053] S1, input data and preliminary state construction. Obtain intersection detector license plate data, construct boundary conditions (arrival and departure cumulative flow curves) upstream and downstream of the road segment, combine multi-source traffic perception data such as floating car trajectories and millimeter wave radar to form limited observations within the road segment.

[0054] S2, deep learning model construction. Construct a neural network structure based on sequence modeling, use an encoder-decoder framework to process time series input, and introduce an attention mechanism to enhance the modeling of the dependence between input historical state and output spatial distribution.

[0055] S3, construction of a differentiable physical traffic flow calculation graph model. Based on the Newell model, establish the spatio-temporal evolution relationship of road traffic flow, and perform function approximation and structure conversion to convert the original discrete physical model form into a differentiable calculation graph structure.

[0056] S4, mixed model construction and joint training. In a unified framework, construct a physical-data hybrid driving model: combine the data-driven branch constructed in step S2 with the differentiable physical calculation graph branch constructed in step S3 in parallel, both branches share the same input data (boundary conditions and internal observations), and independently output traffic state estimation results, and realize deep fusion through consistency constraints. Define a joint objective function that includes consistency loss and data-driven supervision loss, and optimize and train the hybrid model. In the training process, use the limited observation data within the road segment as supervision to learn the estimation ability of the full spatio-temporal traffic state, and simultaneously learn the traffic flow characteristic parameters (free flow speed, jam wave speed, congestion density, etc.) through the physical branch.

[0057] The specific implementation process of step S1 can include:

[0058] S11, establish a full spatio-temporal grid system, discretize the research road segment by 1 meter in space and 1 second in time, form a refined spatio-temporal grid structure, and use it to describe the spatio-temporal distribution of traffic state.

[0059] S12, boundary condition data collection. Collect intersection vehicle license plate matching data to form two types of records: entering and leaving; remove duplicates for the same vehicle and sort them to obtain entering and leaving sequences that can be used for accumulation.

[0060] S13, internal observation data collection. Collect multi-source perception information such as floating car trajectory data and millimeter wave radar detection data within the road segment, including vehicle identification, timestamp, and position along the road segment. Perform unified origin translation and length clipping on the position to obtain a consistent set of time-position trajectory points. Establish a mapping from license plate to cumulative number, and use it to filter trajectories, retaining only vehicles that appear in the license plate data to ensure consistency between the entering / leaving boundary and the internal trajectory.

[0061] S14. Boundary Condition Construction and Preliminary State Reconstruction. Based on license plate matching data, cumulative arrival curves are constructed respectively. Cumulative curve of departure The time series was analyzed; after adding points at the beginning and end of the interval, linear interpolation was used to obtain two time-continuous boundary curves. These two boundary curves were used as deterministic boundary conditions for the upstream and downstream of the road segment, providing anchor points for the subsequent reconstruction of the cumulative vehicle state within the road segment at each time slice.

[0062] S15. At each discrete time point, a segmented polygonal line is formed by combining three types of anchor points: upstream boundary points (from the arrival cumulative curve), the actual location points of floating cars within the road segment (from the floating car trajectory set), and downstream boundary points (from the departure cumulative curve). These anchor points are connected in an orderly manner along the road segment direction, and monotonically linear interpolation is used to reconstruct the spatial profile at that moment, obtaining the cumulative traffic state at that moment. To avoid physical conflicts, discrete points that do not satisfy boundary consistency are removed or corrected. If individual time slices lack internal trajectory points, and the profile obtained solely from the extrapolation of upstream and downstream boundaries has uncertainties, a simplified propagation strategy based on traffic waves or temporal smoothing of adjacent times is used to fill in the temporal continuity and ensure robust reconstruction from a small number of trajectories to the state of the entire road segment. The spatial profiles of each time slice are pieced together one by one to form a complete spatiotemporal traffic state matrix. .

[0063] The specific implementation process of step S2 may include:

[0064] S21. Encoder-Decoder LSTM Model Construction. The encoder-decoder architecture is a sequence-to-sequence neural network structure. The encoder compresses the input cumulative traffic time series into a fixed-dimensional vector representation, and the decoder generates a full-space traffic state profile at the target time based on this vector representation. LSTM (Long Short-Term Memory) networks control the flow of information through three gating mechanisms: the forget gate, the input gate, and the output gate. The forget gate determines what information is discarded from the cell state, the input gate determines what new information is stored in the cell state, and the output gate controls what part is output based on the cell state, thus effectively solving the gradient vanishing problem of traditional recurrent neural networks.

[0065] For each training batch, the model performs the following core computational steps:

[0066] (1) Encoder forward propagation:

[0067]

[0068] in, Represented as the encoder LSTM layer, and is the hidden state and cell state of the encoder at time is the normalized cumulative flow feature (including both arrival and departure channels) at time

[0069] (2) Attention mechanism computation:

[0070]

[0071] where is the attention weight, is the length of the time window, are the attention weight vector, attention weight matrix and bias vector (learnable parameters) respectively, is the attention energy function.

[0072] (3) Context vector generation:

[0073]

[0074] where is the weighted context vector, which is used to initialize the decoder.

[0075] (4) Decoder state update:

[0076]

[0077] where denotes the decoder LSTM layer, is the output at the previous time, denotes the vector concatenation operation.

[0078] (5) Output layer mapping:

[0079]

[0080] where is the predicted spatial traffic state profile, and are the output layer parameters.

[0081] S22, Input organization and feature representation. Collect the cumulative flow sequence of the upstream and downstream boundaries in a fixed-length time window, and make zero-base translation relative to the maximum value of the upstream cumulative flow in the window , , , forming a two-dimensional feature pair . This relative processing eliminates the dimensional difference of absolute cumulative values in different time periods, normalizes the input data, and improves the numerical stability of training.​

[0082] Model input tensor shape is , where is the time window length, and 2 is the cumulative arrival-departure sequence. The target output is the predicted time step full-space discrete profile, with dimensions equal to the number of spatial grids .

[0083] S23, the encoder uses a single-layer LSTM encoder to update the hidden state step by step for the input sequence, obtain the encoding representation of each time step, and output a time feature sequence , where is the hidden dimension size, and the final is obtained as the initial state of the decoder, where is the hidden state, is the memory state.

[0084] S24, in the decoding process, the additive attention mechanism is used to calculate the attention degree of each decoding time to each time of the encoder. Specifically, the current hidden state of the decoder and the hidden state of each time of the encoder are linearly transformed and summed, and then mapped to an energy score through activation function and attention weight vector, and then the attention weight distribution is obtained through normalization.

[0085] S25, in the decoding stage, a step-by-step LSTM structure is used, which only processes one time step at a time, and the hidden state of the last time step and the current input are transmitted into the LSTM together to obtain the hidden state of the current step; then, through a fully connected layer, it is mapped to dimension, directly obtaining the spatial traffic state profile of the target time step . In training, the teacher forcing strategy is used, that is, the real target sequence is used as the input of the decoder during the training process of the decoder, instead of using the prediction output of the last step, which can avoid error accumulation, provide decoder input with the same dimension as output, accelerate model convergence and improve training stability.

[0086] The specific implementation process of the step S3 can include:

[0087] S31, Newell physical model basic expression. Let the link length be , the upstream cumulative arrival curve be , the downstream cumulative departure curve be , the free flow speed be , the jam density be (the physical propagation is upstream, and the formula is recorded as positive speed), and the congestion density be The standard Newell envelope model then represents the cumulative number of vehicles at spatiotemporal points. for:

[0088]

[0089] S32. Based on the cumulative arrival curve obtained in step S1 With cumulative departure curve Interpolate the sequences at equal time intervals to construct a continuous set of upstream arrival sequences and downstream departure sequences. It is stored in tensor form and serves as the boundary condition for the physical model.

[0090] S33, Define free-flow propagation delay With the propagation delay of the blocking wave For each spatial location Two types of time delay indexes are established to describe the time offset of the vehicle cumulative curve.

[0091] S34. Due to time shift in the original model Since it is not differentiable, a continuous approximation is made using a logic function, which has the following form:

[0092]

[0093]

[0094] in, For time, The central position parameter of the Logistic function determines the midpoint of the transition from 0 to 1. The smoothing parameter controls the width of the 0-1 step transition in the Logistic function and can be used as a fixed variable or a parameter to be learned.

[0095] For both upstream free-flow propagation and downstream blocking wave propagation, the center position parameter of their time delay is defined respectively:

[0096]

[0097] in, For vehicles to flow freely from the upstream boundary propagation to location Time delay, To block the wave from the downstream boundary at the wave speed propagation to location The time delay.

[0098] For each Construct differentiable shiftable kernels (time-indexed) count):

[0099]

[0100]

[0101] where, and denote the differentiable weight functions upstream and downstream at time , location , is the time window length, is the function.

[0102] The upper and lower bounds of the differentiable accumulations are thus obtained:

[0103]

[0104]

[0105] The differentiable function of the number of vehicles in the Newell model is thus obtained and , as shown in Figure 3 .

[0106] S35, the differentiable computation graph is packaged and output. The minimum value operation of the Newell envelope relationship remains unchanged:

[0107]

[0108] Set as a learnable parameter; , tensor dot product, and ) together constitute a differentiable computation graph, supporting automatic differentiation and backpropagation.

[0109] As shown in Figure 4 , the specific implementation process of the step S4 can include:

[0110] S41, data organization and batch input: organize the data generated in steps S1-S3 into batches that can be directly used by the network: based on the cumulative arrival curve and the cumulative departure curve of the upstream and downstream, construct a time series input, and through the maximum cumulative value in the window, perform relative processing to form the encoder input sequence of the data-driven model; the decoder end is initialized with a zero vector during training. The relative amount of the total space cumulative vehicle number at the end of the window is used as the supervision target, where the observation position is provided by the multi-source perception data available inside the road segment (such as floating car trajectory, millimeter wave radar detection, etc.), forming a limited set of observation points. Organize the inputs required for physical branches, including sliding window data of arrival and departure, road space discrete grid, and corresponding relative benchmark values. Finally, the above inputs and supervision targets are uniformly packaged into batches and input into the hybrid model framework for subsequent joint training and parameter estimation.

[0111] S42, to link the physical model and the data-driven model under a unified framework, set the following training target composed of two parts, and add the physical error as a monitoring quantity:

[0112] (1) consistency loss: constrain the data-driven branch output to be consistent with the physical branch output, using spatial mean square error MSE:

[0113]

[0114] (2) data-driven supervision loss: constrain the data-driven branch output to be consistent with the reconstructed true value , using spatial mean square error:

[0115]

[0116] (3) total objective function: weight the consistency loss and the data-driven supervision loss by the weight coefficient to form the final optimization target:

[0117]

[0118] wherein is used to balance between physical mechanism fitting and data fitting.

[0119] The above objective function is used to guide joint training: while ensuring the consistency of the two branches, the data-driven branch is supervised using the reconstructed true value, thereby promoting accurate reconstruction of the full-space-time traffic state.

[0120] S43, two-stage training process and optimization strategy: ​

[0121] (1) First, data-driven branch pre-training: only update the data-driven branch parameters, so that it can first reach a stable fitting on the full-space supervision target, and use early stopping and save the optimal weight to avoid large residual at the initial stage, which can drag the physical branch and cause convergence direction deviation or slow convergence.

[0122] (2) Subsequent joint training: load the pre-trained weights, enable joint forward and joint optimization of the two branches; the optimizer acts on the parameters of the two branches, and uses a total loss-based adaptive learning rate annealing strategy (such as reducing the learning rate according to the index stagnation) to improve the convergence accuracy at the later stage.

[0123] S44, state estimation and parameter output: input the upstream and downstream cumulative curves and internal observation data of the road section at any time period, and the relative estimation of the cumulative vehicle number in the full space can be obtained through the hybrid framework, and combined with the relative reference to restore to the absolute amount, and then estimate the full-time traffic state matrix according to time. At the same time, output the key traffic flow parameters (such as free flow speed, jam wave speed, congestion density and smoothing coefficient) learned in the physical branch as road section parameter estimation, which can be aggregated and persisted according to needs for subsequent management and control applications.

[0124] Case analysis: The invention selects a typical road section in the central urban area of a city as a method example scene, and uses the SUMO simulation platform to build a test environment consistent with the real traffic conditions. The total length of the road section is about 327m, and the virtual detectors are set according to the actual signal timing and lane arrangement. Taking the part of the floating car trajectory observation available in the road section as an example, the vehicle arrival, departure time and floating car trajectory data are collected.

[0125] The simulation time is set to 2 hours each time, and a total of 7 simulation programs are run to obtain the experimental data set, the vehicle length is set to 5 meters, the vehicle acceleration is set to 2.6m / s 2 , and the deceleration is set to 4.5m / s 2 . The vehicle arrival probability of each road section in the 7 experiments is set to 0.25, 0.22, 0.205, 0.19, 0.16, 0.13. Other parameters use the default traffic flow model parameters of the simulation software. The data set with a vehicle arrival probability of 0.205 is selected as the test set, and the remaining data sets are divided into training sets.

[0126] The hybrid model is implemented based on PyTorch 2.4.1, the experimental environment is Intel(R) Core(TM) i9-13980HX 2.20GHz, the memory is 64G, the graphics card is RTX4080, and the Windows system. The optimizer uses Adam, the initial learning rate is 0.01, and the batchsize is 128.

[0127] Hidden dimensions of the data-driven branch The value is 128, the input dimension is 2, and the encoder time window length is... The value is 240. All parameters in the physics branch are trainable: free-flow velocity. Reverse wave speed Blocking density Smoothing coefficient The initial values ​​are set to 20.0, 10.0, 0.20, and -10.0, respectively.

[0128] The loss reduction curve of the reconstructed spatiotemporal traffic state obtained using the method proposed in this invention is shown below. Figure 5 and Figure 6 As shown (floating car trajectory penetration rate is set to 1.0, loss fusion coefficient α = 0.5). The traffic parameters estimated in the second stage are as follows... Figure 7 As shown.

[0129] This study compares the reconstruction accuracy and convergence performance of the Newell model based on a physical model, the purely data-driven Encoder–Decoder LSTM model, and the proposed physical-data-driven hybrid model (fusion coefficient α=0.2) under different trajectory penetration conditions. Experiments were conducted by training the three models at multiple penetration levels (20%, 40%, 60%, 80%, and 100%), and recording their loss functions (MSE) on the test set.

[0130] Table 1. Performance Comparison of Traffic State Estimation Models under Different Penetration Rates

[0131]

[0132] When floating car penetration is low, the physical model Newell outperforms the purely data-driven Encoder-Decoder LSTM model, but its accuracy is inferior to the hybrid model. This indicates that when data volume is insufficient or data quality is poor, the hybrid model can effectively utilize the prior traffic knowledge provided by the physical traffic flow model Newell, thus compensating to some extent for the performance shortcomings of the purely data-driven Encoder-Decoder LSTM model in the case of insufficient data.

[0133] When data penetration is high, the performance of the fusion model may be slightly inferior to that of the pure data-driven method. This is mainly because, with sufficient data, the pure data-driven model can fully explore and learn complex traffic states without relying on the prior constraints provided by the physical traffic model, thus achieving better reconstruction accuracy.

[0134] Figure 8The traffic state reconstruction performance of the three models under different data penetration rates is intuitively shown: when the data penetration rate is in the interval of 20%-70%, the fusion model shows obvious advantages; when the data penetration rate is higher than about 70%, the data-driven Encoder-Decoder LSTM model has higher accuracy in state reconstruction.

[0135] The embodiment of the present application further discloses a computer system, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the computer program implements the steps of the intelligent estimation method for urban road traffic state based on a differentiable physical representation when executed by the processor.

[0136] The embodiment of the present application further discloses a computer program product, comprising a computer program, and the computer program implements the steps of the intelligent estimation method for urban road traffic state based on a differentiable physical representation when executed by the processor.

[0137] The program code for implementing the method of the present application can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, causes the steps of the method of the present application to be implemented. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server. The present application is not detailed in the above, which is the known technology of those skilled in the art.

Claims

1. A method for intelligent estimation of urban road traffic state based on differentiable physical representation, characterized in that, Includes the following steps: Based on the arrival and departure cumulative traffic information provided by intersection detectors and checkpoint equipment, road segment boundary conditions are constructed, and multi-source traffic perception inputs are formed by combining the observation data available within the road segment. A physics-data hybrid driven model is constructed, combining the data-driven branch and the differentiable physics branch in parallel. In the data-driven branch, a deep learning model is built to learn the mapping relationship between the cumulative flow at the road segment boundary and the traffic state within the road segment. In the differentiable physics branch, the non-differentiable Newell traffic flow model is transformed into a differentiable computational graph structure through function approximation and structural transformation. The data-driven branch and the differentiable physics branch share the same input data and output traffic state estimation results independently. A loss function for the hybrid-driven model is constructed to ensure that the data-driven output is consistent with the physical model and to jointly estimate traffic flow parameters. The hybrid model is then optimized and trained to achieve traffic state estimation and obtain traffic flow characteristic parameters.

2. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 1, characterized in that, The input data of the physical-data hybrid driven model includes boundary conditions and internal observation data. The boundary conditions include time series of arrival and departure cumulative curves, which are extracted from intersection detector data. The internal observation data of the road segment are extracted from multi-source sensing information, including floating car trajectory data and millimeter-wave radar detection data.

3. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 1, characterized in that, The deep learning model in the data-driven branch uses an encoder-decoder framework to process time-series inputs and introduces an attention mechanism to enhance the modeling of the dependency relationship between the input historical state and the output spatial distribution. The encoder compresses the input cumulative traffic time series into a fixed-dimensional vector representation, and the decoder generates a full-space traffic state profile at the target time based on the vector representation; the deep learning model input tensor has the following shape: ,in The time window length is 2, which represents the cumulative arrival-departure sequence. The target output is a discrete profile of the entire space at the predicted time, with the dimension equal to the number of spatial grids.

4. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 3, characterized in that, The encoder uses a single-layer LSTM encoder to progressively update the hidden state of the input sequence, obtain the encoded representation at each time step, and calculate the attention weights to obtain the weighted representation at each time step. The context vector is obtained by aggregation in the time dimension and used to initialize the decoder. In the decoding stage, an LSTM structure is used to execute stepwise, processing only one time step at a time. The hidden state of the previous time step and the current input are fed into the LSTM to obtain the hidden state of the current step. Then, it is mapped to the dimension corresponding to the number of spatial grids through a fully connected layer to obtain the spatial traffic state profile at the target time.

5. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 1, characterized in that, In the differentiable physics branch, the differentiable functional expression of the Newell traffic flow model is as follows: ； ； ； ； ； in, Indicates time, Indicates spatial location, The cumulative number of vehicles at a given time and space point. Let these represent the differentiable cumulative sums of the upper and lower bounds, respectively. These represent the continuous upstream arrival sequence and downstream departure sequence, respectively. and These represent the upstream and downstream times respectively. ,Location Differentiable weight function at point, The length of the time window. For free flow velocity, For blocking wave velocity, For crowding density, For the length of the road segment, This represents the Sigmoid function. For smoothing parameters.

6. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 1, characterized in that, In the physical-data hybrid driven model, the input and supervision targets of the data-driven model are as follows: based on the cumulative arrival curves and cumulative departure curves of upstream and downstream, a time series input is constructed, and the maximum cumulative value within the window is relativized to form the encoder input sequence of the data-driven model; the relative quantity of the cumulative number of vehicles in the entire space at the end of the window is used as the supervision target, wherein the observation location is provided by multi-source sensing data within the road segment to form a finite set of observation points; The inputs to the differentiable physics branch include arriving and departing sliding window data, a discrete grid of road segment space, and the corresponding relativized reference values.

7. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 1, characterized in that, The loss function of the physical-data hybrid driven model considers consistency loss and data-driven supervision loss; the consistency loss is expressed as: ； in, Represents the set of all space points throughout the time. This indicates data-driven branch output. The physical branch output is represented; the data-driven supervision loss is expressed as: ； in, Represent the truth value; the overall objective function is: ； in, It is used to make a trade-off between physical mechanism fit and data fit.

8. The intelligent estimation method for urban road traffic state based on differentiable physical representation according to claim 1, characterized in that, The training process of the physical-data hybrid driven model includes: first, pre-training the data-driven branch, in which only the parameters of the data-driven branch are updated to achieve a stable fit on the supervised target in the whole space, and early stopping is adopted and the optimal weights are saved; then, joint training is performed, in which the pre-trained weights are loaded and the joint forward pass and joint optimization of the two branches are enabled; the optimizer is applied to the parameters of the two branches respectively, and an adaptive learning rate annealing strategy based on the total loss is used to improve the convergence accuracy in the later stage.

9. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent estimation method for urban road traffic state based on differentiable physical representation as described in any one of claims 1-8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent estimation method for urban road traffic state based on differentiable physical representation as described in any one of claims 1-8.

Images