Continuous flow traffic state estimation method and device based on physical information deep learning

Abstract

The invention discloses a continuous flow traffic state estimation method and device based on physical information deep learning, and the method comprises the steps: constructing a traffic flow conservation law through a Ross model and a plurality of fundamental diagram models; fitting a traffic state according to the basic diagram model to obtain traffic flow parameters; constructing a target deep learning model based on physical information according to the traffic flow conservation law and the traffic flow parameters; and estimating the traffic state through detector data and the target deep learning model to obtain an estimation result. The method can improve the accuracy of continuous flow traffic state estimation, and can be widely applied to the technical field of traffic state estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic state estimation, and in particular to a continuous flow traffic state estimation method and device based on deep learning with physical information. Background Art

[0002] Traffic State Estimation (TSE) is a core task in traffic management and Intelligent Transportation System (ITS). Its purpose is to understand the traffic conditions of the road network in real time and provide a basis for decision-making such as traffic flow control, route planning, and accident prediction. With the increasing complexity and scale of urban traffic, traditional traffic flow estimation methods have gradually shown deficiencies, especially in large-scale traffic data processing and high-precision traffic state prediction.

[0003] In recent years, macroscopic traffic flow models, especially those based on traffic fluid theory, such as the LWR (Lighthill-Whitham-Richards) model, have been widely applied to the analysis and estimation of traffic flow. However, in speed estimation, the LWR model needs to convert between parameters, which may introduce certain errors and thus affect the accuracy of traffic state estimation. Summary of the Invention

[0004] In view of this, the main purpose of the embodiments of the present invention is to provide a continuous flow traffic state estimation method and device based on deep learning with physical information, in order to solve at least one of the problems in the prior art. The present invention can improve the accuracy of continuous flow traffic state estimation.

[0005] To achieve the above purpose, on the one hand, an embodiment of the present invention provides a continuous flow traffic state estimation method based on deep learning with physical information, and the method includes:

[0006] Construct a traffic flow conservation law through the Ross model and several fundamental diagram models;

[0007] Fit the traffic state according to the fundamental diagram model to obtain traffic flow parameters;

[0008] Construct a target deep learning model based on physical information according to the traffic flow conservation law and the traffic flow parameters;

[0009] Estimate the traffic state through detector data and the target deep learning model to obtain an estimation result.

[0010] In some embodiments, the step of constructing a traffic flow conservation law through the Ross model and several fundamental diagram models includes the following steps:

[0011] According to the Greenshields model, a linear relationship between vehicle speed and vehicle density is obtained;

[0012] According to the Greenberg model, a linear relationship between vehicle speed and vehicle density is obtained;

[0013] According to the Van Aerde model, a non-linear relationship of vehicle traffic flow is obtained;

[0014] According to the said Ross model, the linear relationship between vehicle speed and vehicle density, the linear relationship between vehicle speed and vehicle density, and the non-linear relationship of vehicle traffic flow, the traffic flow conservation law is constructed;

[0015] Among them, the basic diagram model includes the Greenshields model, the Greenberg model, and the Van Aerde model.

[0016] In some embodiments, the expression of the Ross model is:

[0017]

[0018] In the formula, u(l,t) represents the average vehicle speed of section l at time t; u f represents the free flow speed; T represents the length of a single time interval.

[0019] In some embodiments, the expression of the Greenshields model is:

[0020] q = u×ρ

[0021]

[0022] In the formula, q represents the traffic flow; u represents the vehicle speed; ρ represents the vehicle density; u(ρ) represents the speed function of the vehicle; u f represents the free flow speed; ρ j represents the jam density; q(ρ) represents the traffic flow function of the road.

[0023] In some embodiments, the expression of the Greenberg model is:

[0024]

[0025] In the formula, u(ρ) represents the speed function of the vehicle; u m represents the vehicle speed at the maximum traffic volume; ρ j represents the jam density; ρ represents the vehicle density; q(ρ) represents the traffic flow function of the road.

[0026] In some embodiments, the expression of the Van Aerde model is as follows:

[0027]

[0028] where ρ(u) represents the vehicle density function with respect to the vehicle speed; c 1 , c 2 , c 3 represent fitting parameters; u f represents the free flow speed; u represents the vehicle speed; u c represents the critical speed; u m represents the vehicle speed at the maximum traffic volume; ρ j represents the jam density; q m represents the maximum traffic volume.

[0029] In some embodiments, according to the basic diagram model, fitting the traffic state to obtain traffic flow parameters includes the following steps:

[0030] Obtain traffic flow data;

[0031] According to the traffic flow data, respectively fit the traffic state through several of the basic diagram models to obtain the traffic flow parameters;

[0032] wherein the traffic flow parameters include the relationship parameters of flow, the relationship parameters of vehicle speed, and the relationship parameters of vehicle density.

[0033] In some embodiments, constructing a target deep learning model based on physical information according to the traffic flow conservation law and the traffic flow parameters includes the following steps:

[0034] Construct an initial deep learning model according to the neural network model; the neural network model includes an LSTM layer and multiple fully connected layers;

[0035] Take the traffic flow conservation law as a constraint condition;

[0036] Construct a first loss function based on the error between the estimation result and the detector data;

[0037] Construct a second loss function based on the error between the estimation result and the traffic flow conservation law;

[0038] Obtain the target loss function according to the first loss function and the second loss function;

[0039] Initialize the initial deep learning model through the traffic flow parameters, and combine the constraint condition and the target loss function to obtain the target deep learning model.

[0040] In some embodiments, estimating the traffic state by using the detector data and the target deep learning model to obtain an estimation result includes the following steps:

[0041] Obtain the detector data through traffic detectors at fixed positions;

[0042] Input the detector data into the target deep learning model to estimate the traffic state and obtain the estimation result.

[0043] To achieve the above object, another aspect of the embodiments of the present invention provides a continuous flow traffic state estimation device based on physics-informed deep learning, where the device includes:

[0044] A first module for constructing a traffic flow conservation law by using the Lighthill-Whitham-Richards (LWR) model and several fundamental diagram models;

[0045] A second module for fitting the traffic state according to the fundamental diagram model to obtain traffic flow parameters;

[0046] A third module for constructing a target deep learning model based on physics information according to the traffic flow conservation law and the traffic flow parameters;

[0047] A fourth module for estimating the traffic state by using the detector data and the target deep learning model to obtain an estimation result.

[0048] To achieve the above object, another aspect of the embodiments of the present invention provides an electronic device, where the electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the above-mentioned continuous flow traffic state estimation method based on physics-informed deep learning is implemented.

[0049] To achieve the above object, another aspect of the embodiments of the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned continuous flow traffic state estimation method based on physics-informed deep learning is implemented.

[0050] To achieve the above object, another aspect of the embodiments of the present invention provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, the computer device executes the above-mentioned continuous flow traffic state estimation method based on physics-informed deep learning.

[0051] The embodiments of the present invention at least include the following beneficial effects: The present invention provides a method and device for continuous traffic state estimation based on deep learning of physical information. Through the Ross model and several basic diagram models, the traffic flow conservation law is constructed. Several different basic diagrams are adopted, and the deep learning model based on physical information can adapt to the changes of more complex road conditions, further enhancing its applicability; according to the basic diagram model, the traffic state is fitted to obtain traffic flow parameters; according to the traffic flow conservation law and the traffic flow parameters, a target deep learning model based on physical information is constructed. The deep learning model based on physical information can effectively reduce the error in the process of traffic parameter conversion and improve the accuracy of traffic flow estimation; through detector data and the target deep learning model, the traffic state is estimated to obtain an estimation result. By combining the traffic flow conservation law and deep learning, the present invention significantly improves the estimation accuracy, and the present invention can improve the accuracy of continuous traffic state estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0053] Figure 1 is a flowchart of a method for continuous traffic state estimation based on deep learning of physical information provided by an embodiment of the present invention;

[0054] Figure 2 is a schematic structural diagram of a device for continuous traffic state estimation based on deep learning of physical information provided by an embodiment of the present invention;

[0055] Figure 3 is a schematic hardware structure diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following further describes the present invention in detail with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the embodiments of the present invention. They are only examples of devices and methods consistent with some aspects of the embodiments of the present invention described in detail in the appended claims.

[0057] It should be noted that although the functional modules are divided in the system schematic diagram and the logical sequence is shown in the flowchart, in some cases, the steps shown or described can be executed in a different module division from that in the system or in a different order from that in the flowchart. The terms "first / S100" and "second / S200" in the specification, claims and the above-mentioned drawings may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "when" as used herein may be interpreted as "when...", "while...", or "in response to determining".

[0058] The terms "at least one", "a plurality of", "each", "any one" and the like used in the present invention, "at least one" includes one, two or more, "a plurality of" includes two or more, "each" refers to each one of the corresponding plurality, and "any one" refers to any one of the plurality.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0060] Traffic State Estimation (TSE) is a core task in traffic management and Intelligent Transportation System (ITS). Its purpose is to understand the traffic conditions of the road network in real time and provide a basis for decision-making such as traffic flow control, route planning, and accident prediction. With the increasing complexity and scale of urban traffic, traditional traffic flow estimation methods have gradually revealed deficiencies, especially in large-scale traffic data processing and high-precision traffic state prediction.

[0061] In recent years, macroscopic traffic flow models, especially those based on traffic fluid theory, such as the LWR (Lighthill-Whitham-Richards) model, have been widely applied to the analysis and estimation of traffic flow. The LWR model is a continuous flow model that describes the evolution of traffic flow on the road network, assuming that the traffic flow is continuous and smooth on the road, and there is a certain mathematical relationship between traffic density and flow. However, in speed estimation, the LWR model needs to convert between parameters, which may introduce certain errors and thus affect the accuracy of traffic state estimation.

[0062] In view of this, as Figure 1As shown, an embodiment of the present invention provides a continuous traffic state estimation method based on physics-informed deep learning, which may include but is not limited to steps S100 to S400:

[0063] Step S100, construct the traffic flow conservation law through the Ross model and several fundamental diagram models;

[0064] Step S200, fit the traffic state according to the fundamental diagram model to obtain traffic flow parameters;

[0065] Step S300, construct a physics-informed target deep learning model according to the traffic flow conservation law and the traffic flow parameters;

[0066] Step S400, estimate the traffic state through detector data and the target deep learning model to obtain an estimation result.

[0067] In step S100 of some embodiments, the Ross model is combined with three different fundamental diagram models to construct the conservation law describing traffic flow, which involves the relationships between key parameters such as vehicle speed, density, and flow rate on the road section.

[0068] In some embodiments, step S100 may include but is not limited to steps S110 to S140:

[0069] Step S110, obtain the linear relationship between vehicle speed and vehicle density according to the Greenshields model;

[0070] Step S120, obtain the linear relationship between vehicle speed and vehicle density according to the Greenberg model;

[0071] Step S130, obtain the non-linear relationship of vehicle traffic flow according to the Van Aerde model;

[0072] Step S140, construct the traffic flow conservation law according to the Ross model, the linear relationships between vehicle speed and vehicle density, the linear relationship between vehicle speed and vehicle density, and the non-linear relationship of vehicle traffic flow;

[0073] Among them, the fundamental diagram models include the Greenshields model, the Greenberg model, and the Van Aerde model.

[0074] In steps S110 to S140 of some embodiments, the Ross Model is used in combination with three basic traffic flow models (Greenshields Model, Greenberg Model, and Van Aerde Model) to establish the conservation law of traffic flow. These models provide a theoretical basis for the dynamic characteristics of traffic flow by describing the relationships among vehicle density, vehicle speed, and traffic flow. For example, the Greenshields Model assumes a linear relationship between vehicle speed and vehicle density; the Greenberg Model assumes a logarithmic relationship between vehicle speed and vehicle density; the Van Aerde Model describes vehicle traffic flow through complex non-linear relationships. By combining these models, the conservation law of traffic flow can be derived.

[0075] In some embodiments, step S200 may include, but is not limited to, steps S210 to S220:

[0076] Step S210, obtaining traffic flow data;

[0077] Step S220, according to the traffic flow data, respectively fitting the traffic state through several of the basic traffic flow models to obtain the traffic flow parameters;

[0078] Wherein, the traffic flow parameters include relationship parameters of traffic flow, relationship parameters of vehicle speed, and relationship parameters of vehicle density.

[0079] In steps S210 to S220 of some embodiments, according to different traffic flow conditions and corresponding different speed-density conditions on the road, different basic traffic flow models are used to obtain a parameter set. Exemplarily, using known traffic flow data, the traffic state under different traffic flow conditions is fitted by combining three basic traffic flow models. Through the fitting process, key traffic flow parameters can be obtained, such as free flow speed, jam density, maximum traffic flow, etc. These parameters reflect the characteristics of traffic flow under different states (such as free flow, transitional flow, and congested flow). Optionally, three different types of basic traffic flow models are selected, including the Greenshields Model, the Greenberg Model, and the Van Aerde Model. Under these three basic traffic flow models, different fixed parameters (including relationship parameters of traffic flow, speed, and density) are fitted. By fitting these models, the position speed u(l,t) at each moment (i.e., the vehicle speed at road section position l at moment t) can be obtained. In subsequent steps, the calculation results of the position speed at each moment are substituted into the Ross Model, and the loss function can be further calculated to optimize the traffic state estimation.

[0080] In some embodiments, the expression of the Ross Model is:

[0081]

[0082] Wherein, u(l, t) represents the average vehicle speed of road section l at time t; u f represents the free flow speed; T represents the length of a single time interval.

[0083] The Ross model used to calculate the loss function is:

[0084]

[0085] Wherein, represents the model estimated speed; represents the average vehicle speed.

[0086] In some embodiments, the expression of the Greenshields model is:

[0087] q = u × ρ

[0088]

[0089] Wherein, q represents the traffic flow; u represents the vehicle speed; ρ represents the vehicle density; u(ρ) represents the vehicle speed function; u f represents the free flow speed; ρ j represents the jam density; q(ρ) represents the traffic flow function of the road.

[0090] In some embodiments, the expression of the Greenshields model is:

[0091] q = u × ρ

[0092]

[0093] Wherein, q represents the traffic flow; u represents the vehicle speed; ρ represents the vehicle density; u(ρ) represents the vehicle speed function; u f represents the free flow speed; ρ j represents the jam density; q(ρ) represents the traffic flow function of the road.

[0094] In some embodiments, the expression of the Greenberg model is:

[0095]

[0096] Wherein, u(ρ) represents the vehicle speed function; u m represents the vehicle speed at the maximum traffic volume; ρ j represents the jam density; ρ represents the vehicle density; q(ρ) represents the traffic flow function of the road.

[0097] In some embodiments, the expression of the Van Aerde model is:

[0098]

[0099] where ρ(u) represents the vehicle density function with respect to the vehicle speed; c 1 , c 2 , c 3 represent fitting parameters; u f represents the free flow speed; u represents the vehicle speed; u c represents the critical speed; u m represents the vehicle speed at the maximum traffic volume; ρ j represents the jam density; q m represents the maximum traffic volume.

[0100] In some embodiments, step S300 may include, but is not limited to, steps S310 to S360:

[0101] Step S310, constructing an initial deep learning model according to the neural network model; the neural network model includes an LSTM layer and multiple fully connected layers;

[0102] Step S320, taking the traffic flow conservation law as a constraint condition;

[0103] Step S330, constructing a first loss function based on the error between the estimation result and the detector data;

[0104] Step S340, constructing a second loss function based on the error between the estimation result and the traffic flow conservation law;

[0105] Step S350, obtaining a target loss function according to the first loss function and the second loss function;

[0106] Step S360, initializing the initial deep learning model through the traffic flow parameters, and combining the constraint condition and the target loss function to obtain the target deep learning model.

[0107] In step S310 of some embodiments, an initial deep learning model is constructed according to the neural network model. The neural network model includes multiple fully connected layers and an LSTM layer, and is used to extract dynamic features in time series data. Its operation steps are:

[0108] f t = σ(W f ·[h t-1 , x t +b f )

[0109] i t = σ(W i ·[ht-1 , x t + b i )

[0110]

[0111] o t = σ(W o · [h t-1 , x t + b o )

[0112] h t = o t · tanh(C t )

[0113] In the formula, f t represents the activation value of the forgetting gate at time t; σ represents the sigmoid activation function; W f represents the weight matrix of the forgetting gate; h t-1 represents the hidden state at the previous time; x t represents the input at the current time; b f represents the bias term of the forgetting gate; i t represents the activation value of the input gate at time t; W i represents the weight matrix of the input gate; b i represents the bias term of the input gate; represents the value of the candidate memory cell at time t; tanh represents the hyperbolic tangent activation function; W C represents the weight of the candidate memory cell; b C represents the bias term of the candidate memory cell; C t represents the state of the memory cell at the current time t; C t-1 represents the state of the memory cell at the previous time t - 1; o t represents the activation value of the output gate at the current time t; W o represents the weight matrix of the output gate; b o represents the bias term of the output gate; h t represents the hidden state at the current time t.

[0114] In some embodiments, assume that a certain road section is U = L × T, define u(l, t) as the average vehicle speed at road section position l at time t, and ρ(l, t) as the average vehicle density at road section position l at time t. Assume that a certain number of estimation detectors are arranged on the road. Then, for the traffic state at the road section (l o , t o ) with a detector, use s(l o , t o) It can be represented. Let M be a deep learning model. For all (l,t) ∈ U, the velocity u(l,t) can be estimated from the detector data s(l o ,t o ) by the deep learning model M. Then there is the following expression:

[0115]

[0116] In step S320 of some embodiments, the traffic flow conservation law can be embedded into the model as a physical constraint condition.

[0117] In steps S330 to S350 of some embodiments, the loss function calculation consists of two parts: the error of the model estimation result relative to the data collected by the fixed detector, and the error of the estimation result relative to the traffic flow conservation law. Then the error calculation expression is:

[0118]

[0119] J = α×J dl +β×J phy

[0120] In the formula, J dl represents the fitting error of the data collected from the fixed detector, and the first loss function; N o represents the number of positions of the fixed detections; u(i,t) represents the actual value of the vehicle speed; represents the estimated value of the vehicle speed; J phy represents the fitting error of traffic flow conservation, that is, the second loss function; N represents the number of data on the road; J represents the target loss function; α, β represent weights. In order to iteratively optimize the model, the embodiments of the present invention specify the target loss function J as the loss function of the target deep learning model.

[0121] In step S360 of some embodiments, the deep learning model can be initialized and optimized using traffic flow parameters to ensure that the model can accurately reflect the physical characteristics of traffic flow. Embedding the traffic flow conservation law into the model as a physical constraint can significantly improve the performance of the model, enabling it to play a greater role in traffic flow prediction and management, and specifying the target loss function J as the loss function of the target deep learning model to obtain the target deep learning model.

[0122] In some embodiments, step S400 may include but is not limited to steps S410 to S420:

[0123] Step S410, obtaining the detector data through traffic detectors at fixed positions;

[0124] Step S420: Input the detector data into the target deep learning model to estimate the traffic state and obtain the estimation result.

[0125] In some embodiments, based on the traffic flow conservation law and combined with the deep learning method, a deep learning model based on physical information is proposed. The constructed traffic flow conservation law is used as a constraint condition, and a deep neural network is utilized to estimate the traffic state. Traffic detector data is obtained using traffic detectors at fixed positions. Based on the relationship between vehicle speed and density, a deep learning model is employed to estimate the traffic state of the entire road section and output the speed prediction result at each moment. Exemplarily, data such as speed and density are obtained from the detector. The detector data is input into the deep learning model, and combined with physical constraints and traffic flow parameters, the traffic state of the entire road section is estimated. The accuracy of the estimation result is verified, and the model is adjusted if necessary. The final traffic state estimation result is output for traffic management. Through this process, the effective utilization of detector data is ensured, and the accurate estimation of the traffic state of the entire road section is achieved through the deep learning model.

[0126] In the embodiments of the present invention, the target deep learning model is respectively applied to the NGSIM (Next Generation Simulation) dataset of US highways and the open-source vehicle dataset UTE developed by Southeast University. NGSIM is a dataset of highways in California, USA, and is traffic data of US highways collected by the Federal Highway Administration of the United States. This dataset is collected through various sensor devices (such as video cameras and geomagnetic sensors) deployed in the highway road network, providing detailed traffic flow data for traffic flow modeling and simulation research. The statistical interval of the dataset is 0.1 second, covering various characteristics of traffic flow, including vehicle speed, vehicle distance, lane occupancy rate, traffic volume, etc. In addition, the NGSIM dataset also provides lane information and specific position data of each vehicle, which is suitable for studying highway traffic flow dynamics, lane change behavior, the relationship between traffic flow and road network spatial structure, etc. The open-source data UTE (Urban Traffic Environment Data) of Southeast University is an urban traffic environment dataset developed and made public by the School of Transportation of Southeast University. This dataset contains vehicle trajectory data of multiple stations, and the data is extracted from aerial videos and verified manually to ensure accuracy. Each file represents a set of samples, and the data comes from different stations and times and is extracted under dynamic traffic conditions. The UTE dataset includes the following: video files with object recognition frames, original trajectory data, and data format descriptions. The original vehicle trajectory data includes parameters such as vehicle number, position coordinates, lane number, vehicle length, vehicle width, driving speed, headway time, headway distance, acceleration, etc. The time accuracy of the data is 0.1 second, and the position accuracy is 0.01 meter.

[0127] For the target datasets NGSIM and UTE, obtain their road network and vehicle basic information (including road segment length, road segment width, number of vehicles, vehicle speed, time, etc.); and sample according to the sampling interval.

[0128] In some embodiments, the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used as evaluation metrics throughout the process.

[0129] In the embodiments of the present invention, 20% of the experimental data is used to construct the training set and validation set for model training and optimization, and the remaining 80% is used as the test set for model performance evaluation and horizontal comparison with the baseline model. The experimental results are shown in the model performance in Table 1. Among them, the Non-physics neural network is the pure neural network method, which includes 10 fully connected layers and 2 LSTM (long short-term memory) layers; the Huang et al. method is the model proposed by Huang et al., which integrates the LWR model and the Greenshields model and consists of a deep learning network with 10 fully connected layers, and each fully connected layer has 40 neurons.

[0130] Table 1

[0131]

[0132] When comparing the PIDL model constructed using the LWR model with the PIDL model constructed using the Ross model, the estimation accuracy of the latter is significantly improved. Specifically, on the US101 dataset, taking the Greenberg model with the best fitting performance as an example, the MAE accuracy of the Greenberg model combined with the Ross equation is 1.18, which is about 40% higher than the MAE accuracy of the combination of the Greenberg model and the LWR model. Other metrics also show advantages. The addition of the Ross model demonstrates its advantage in improving the traffic state estimation accuracy. It should be noted that regardless of which basic graph model is used, the model containing the Ross model has a significant estimation advantage compared with the model using the LWR.

[0133] In summary, the embodiments of the present invention introduce the Ross model to develop the traffic flow conservation law, combine the traffic flow conservation law with deep learning, further construct a deep learning loss function, and propose a continuous traffic state estimation framework based on physics-informed deep learning. At the same time, three different fundamental diagrams are used to construct the traffic flow conservation law for traffic parameter estimation to guide the fitting process of the deep neural network, which can effectively reduce the errors caused by the conversion of three traffic parameters, adapt to the influence of more road conditions, and enhance the robustness of the PIDL model to different fundamental diagrams. To verify the effectiveness and robustness of the model, the present invention conducts traffic state estimation experiments on three different roads. The results show that compared with the existing models and the PIDL model based on the LWR model, the PIDL model based on the Ross equation has higher accuracy and better robustness.

[0134] As Figure 2 shown, the embodiments of the present invention also provide a continuous traffic state estimation device 500 based on physics-informed deep learning, which can implement the above-mentioned continuous traffic state estimation method based on physics-informed deep learning. The device includes:

[0135] The first module 501 is used to construct the traffic flow conservation law through the Ross model and several fundamental diagram models;

[0136] The second module 502 is used to fit the traffic state according to the fundamental diagram model to obtain traffic flow parameters;

[0137] The third module 503 is used to construct a target deep learning model based on physical information according to the traffic flow conservation law and the traffic flow parameters;

[0138] The fourth module 504 is used to estimate the traffic state through detector data and the target deep learning model to obtain an estimation result.

[0139] It can be understood that the content in the above method embodiments is applicable to the device embodiments of the present invention. The functions specifically implemented by the device embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0140] The embodiments of the present invention also provide an electronic device, which includes a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the above-mentioned continuous traffic state estimation method based on physics-informed deep learning. The electronic device can be any intelligent terminal including a tablet computer, an in-vehicle computer, etc.

[0141] It can be understood that the content in the above method embodiments is applicable to the device embodiments of the present invention. The functions specifically implemented by the device embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0142] Reference Figure 3 , Figure 3 schematically shows the hardware structure of an electronic device according to another embodiment. The electronic device includes:

[0143] A processor 601, which can be implemented in ways such as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided by the embodiments of the present invention;

[0144] A memory 602, which can be implemented in forms such as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 602 can store an operating system and other application programs. When implementing the technical solutions provided by the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 602 and are called by the processor 601 to execute the continuous traffic state estimation method based on physical information deep learning according to the embodiments of the present invention;

[0145] An input / output interface 603, which is used to implement information input and output;

[0146] A communication interface 604, which is used to implement communication and interaction between the device and other devices, and can achieve communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0147] A bus 605, which transmits information between various components of the device (such as the processor 601, the memory 602, the input / output interface 603, and the communication interface 604);

[0148] Among them, the processor 601, the memory 602, the input / output interface 603, and the communication interface 604 are communicatively connected to each other inside the device through the bus 605.

[0149] An embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which when executed by a processor implements the above-mentioned continuous traffic state estimation method based on physics-informed deep learning.

[0150] It can be understood that the content in the above method embodiments is applicable to this storage medium embodiment. The functions specifically implemented by this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0151] An embodiment of the present invention also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to execute the aforementioned continuous traffic state estimation method based on physics-informed deep learning.

[0152] In summary, the continuous traffic state estimation method and device according to the embodiments of the present invention have the following advantages:

[0153] 1. The embodiment of the present invention belongs to the cross-field of deep learning and intelligent transportation systems, involving related methods of deep neural networks, data science and analysis, and traffic flow theory. Using the basic framework of deep learning and its powerful fitting ability, it explores the relationship between observed data and traffic parameters, and combines the Ross model in the traffic flow model to describe continuous traffic flow, better capturing the dynamic changes of traffic flow, improving the interpretability and accuracy of the model, and finally forming a continuous traffic state estimation method based on physics-informed deep learning (PIDL).

[0154] 2. By adopting three different fundamental diagrams, the physics-informed deep learning model according to the embodiment of the present invention can adapt to the changes of more complex road conditions, further enhancing its applicability. The physics-informed deep learning model can also effectively reduce the errors in the conversion process of three traffic parameters, improving the accuracy of traffic flow estimation. By combining the traffic flow conservation law and deep learning, the estimation accuracy is significantly improved, and strong robustness is demonstrated under different road conditions.

[0155] In some alternative embodiments, the functions / operations recited in the block diagrams may not occur in the order noted in the operational illustrations. For example, depending on the functionality / operation involved, two blocks shown in succession may actually be executed substantially concurrently or the blocks may sometimes be executed in the reverse order. Further, the embodiments presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more thorough understanding of the technology. The disclosed methods are not limited to the operations and logical flows presented herein. Alternative embodiments are envisioned in which the order of various operations is altered and in which sub-operations described as part of a larger operation are performed independently.

[0156] Moreover, although the present invention has been described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the described functions and / or features may be integrated in a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It should also be understood that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, given the attributes, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the modules will be understood within the ordinary skill of an engineer. Thus, those of ordinary skill in the art will be able to implement the present invention as set forth in the claims without undue experimentation. It should also be understood that the particular concepts disclosed are illustrative only and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0157] If the described functions are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or part of the technical solution, may be embodied in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0158] The logic and / or steps represented in the flowchart or otherwise described herein can, for example, be considered as a definitional sequence of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device. As used in this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0159] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection (electronic device) having one or more wirings, a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, as the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0160] It should be understood that the various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the multiple steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0161] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0162] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents.

[0163] The above has specifically described the preferred embodiments of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art can also make various equivalent deformations or substitutions without violating the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of the present invention.

Claims

1. A continuous flow traffic state estimation method based on deep learning of physical information, characterized in that: The following steps are involved: The conservation law of traffic flow is constructed through Ross model and several basic graph models; According to the basic graph model, the traffic state is fitted to obtain traffic flow parameters; According to the traffic flow conservation law and the traffic flow parameters, construct a target deep learning model based on physical information; The traffic status is estimated through the detector data and the target deep learning model to obtain an estimation result.

2. The continuous flow traffic state estimation method based on physical information deep learning according to claim 1 is characterized in that: The method of constructing the traffic flow conservation law by using the Ross model and several basic graph models includes the following steps: According to the Greenshields model, a linear relationship between vehicle speed and vehicle density is obtained; According to the Greenberg model, a linear relationship between vehicle speed and vehicle density is obtained; According to the Van Aerde model, the nonlinear relationship of vehicle traffic flow is obtained; Constructing the traffic flow conservation law according to the Ross model, the linear relationship between vehicle speed and vehicle density, the linear relationship between vehicle speed and vehicle density, and the nonlinear relationship between vehicle traffic flow; The basic graph models include the Greenshields model, the Greenberg model and the Van Aerde model.

3. The continuous flow traffic state estimation method based on physical information deep learning according to claim 1 is characterized in that: The expression of the Ross model is: Where u(l, t) represents the average vehicle speed of road section l at time t; u f represents free flow speed; T represents the length of a single time interval.

4. The continuous flow traffic state estimation method based on physical information deep learning according to claim 2 is characterized in that: The expression of Greenshields model is: q=u×ρ In the formula, q represents the flow rate; u represents the vehicle speed; ρ represents the vehicle density; u(ρ) represents the speed function of the vehicle; u f represents the free flow velocity; ρ j represents the blocking density; q(ρ) represents the flow function of the road.

5. The continuous flow traffic state estimation method based on physical information deep learning according to claim 2 is characterized in that: The expression of the Greenberg model is: Where u(ρ) represents the speed function of the vehicle; u m represents the vehicle speed at maximum traffic volume; ρ j represents the blocking density; ρ represents the vehicle density; q(ρ) represents the flow function of the road.

6. The continuous flow traffic state estimation method based on physical information deep learning according to claim 2 is characterized in that: The expression of the Van Aerde model is: Where p(u) represents the vehicle density function with respect to vehicle speed; c1, c2, c3 represent fitting parameters; u f represents the free flow speed; u represents the vehicle speed; u c represents the critical speed; u m represents the vehicle speed at maximum traffic volume; ρ j represents the blocking density; q m Represents the maximum traffic volume.

7. The continuous flow traffic state estimation method based on physical information deep learning according to claim 1 is characterized in that: The method of fitting the traffic state according to the basic graph model to obtain traffic flow parameters includes the following steps: Obtain traffic flow data; According to the traffic flow data, the traffic state is fitted by using a plurality of the basic graph models to obtain the traffic flow parameters; The traffic flow parameters include flow rate-related parameters, vehicle speed-related parameters and vehicle density-related parameters.

8. The continuous flow traffic state estimation method based on physical information deep learning according to claim 1 is characterized in that: The method of constructing a target deep learning model based on physical information according to the traffic flow conservation law and the traffic flow parameters includes the following steps: According to the neural network model, construct an initial deep learning model; the neural network model includes an LSTM layer and multiple fully connected layers; Taking the traffic flow conservation law as a constraint condition; constructing a first loss function based on an error between the estimation result and the detector data; constructing a second loss function based on an error between the estimation result and the traffic flow conservation law; Obtaining a target loss function according to the first loss function and the second loss function; The initial deep learning model is initialized by using the traffic flow parameters, and the target deep learning model is obtained by combining the constraint conditions and the target loss function.

9. The continuous flow traffic state estimation method based on physical information deep learning according to claim 1 is characterized in that: The method of estimating the traffic state through the detector data and the target deep learning model to obtain an estimation result includes the following steps: Obtaining the detector data through a traffic detector at a fixed position; The detector data is input into the target deep learning model, the traffic state is estimated, and the estimation result is obtained.

10. A continuous flow traffic state estimation device based on deep learning of physical information, characterized in that: include: The first module is used to construct the conservation law of traffic flow through the Ross model and several basic graph models; The second module is used to fit the traffic state according to the basic graph model to obtain traffic flow parameters; The third module is used to construct a target deep learning model based on physical information according to the traffic flow conservation law and the traffic flow parameters; The fourth module is used to estimate the traffic status through the detector data and the target deep learning model to obtain an estimation result.

Images