A holographic traffic state dataset construction method based on multi-source information fusion

Abstract

The present application relates to the technical field of traffic data processing, and more particularly to a holographic traffic state dataset construction method based on multi-source information fusion, comprising: obtaining an original dataset, constructing a directed road network graph and an undirected road network graph according to a road network topology relationship, image processing traffic information according to a color matching equation, constructing a neighbor order matrix, combining a traffic information graph to preliminarily correct and complete the original traffic feature data through a space-time information compensation network, establishing an iterative completion generative adversarial network IGANI model, generating a completed dataset, building a weighted directed graph based on a road physical connection relationship model established by the neighbor order matrix and the directed road network graph, predicting traffic state data in a period through a GE-LSTM traffic state prediction model to generate a prediction dataset, and constructing a holographic traffic state dataset, which effectively overcomes the data missing problem caused by equipment damage, aging and failure, and effectively improves the control effect of the traffic control system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic data processing, and particularly relates to a holographic traffic state data set construction method based on multi-source information fusion. BACKGROUND

[0002] With the continuous increase of vehicle ownership, the problems of low efficiency of urban road intersection and slow traffic of motor vehicles gradually become the future development trend. Single sensor is limited by its physical characteristics and cannot meet the real-time monitoring problem of complex traffic network. The poor accuracy of road traffic state perception leads to low efficiency of intersection traffic, disorder and many safety hazards.

[0003] With the continuous development of intelligent transportation and vehicle-road cooperation technology, the development of advanced information collection and information interaction technology makes real-time data fusion possible, which promotes the development of intelligent networked vehicle technology and vehicle driving system. In the wireless communication environment, intelligent networked vehicles can more easily perceive and obtain surrounding environment and vehicle information, analyze in real time, effectively contact multiple influencing factors, complete the missing traffic data, and construct holographic traffic data set.

[0004] Chinese patent publication No. CN114664090A discloses a traffic data filling method and system based on recurrent neural network, which includes obtaining traffic flow data; judging whether the traffic flow data is missing, if there is missing, based on the current traffic flow data and the recurrent neural network of time dimension, introducing the correction coefficient to correct the hidden layer variables of the recurrent neural network and the current traffic flow data, obtaining the pre-filled traffic flow data; extracting the pre-filled traffic flow data characteristic variable, based on the pre-filled traffic flow data characteristic variable and the recurrent neural network of feature dimension, capturing the correlation between the pre-filled traffic flow data characteristic variables, using the full connection layer to convert the corrected hidden variables to the data space, obtaining the filled traffic flow data. The application excavates the internal relationship between time and space, uses two kinds of recurrent neural networks, respectively uses T-RNN based on time dimension improvement and F-RNN based on feature dimension improvement for dynamic filling, generalizes the dynamic influence between data on multivariate data time sequence, fills the missing data, and perceives the missing information of data. It can be seen that the above technical scheme has the following problems: the traffic original data set information obtained by the application is relatively single, and multi-dimensional data fusion is not performed, resulting in insufficient traffic state perception accuracy. SUMMARY

[0005] Therefore, the present application provides a holographic traffic state data set construction method based on multi-source information fusion, which overcomes the problem of insufficient traffic state perception accuracy caused by the lack of multi-dimensional data fusion analysis ability of the initial data information obtained in the prior art.

[0006] To achieve the above objectives, this invention provides a method for constructing a holographic traffic state dataset based on multi-source information fusion, comprising:

[0007] Step S1: Traffic signal controllers and roadside sensors are used to send signal timing data, LiDAR and high-definition camera data to intelligent vehicle-mounted unit (OBU) with V2X communication function and multi-source information fusion device (MEC). The processed data information is then uploaded to cloud server or computer terminal to form raw dataset.

[0008] Step S2: Construct directed and undirected road network graphs based on the road network topology.

[0009] Step S3: The road network map is processed into a visual representation of traffic information based on the color matching equation;

[0010] Step S4: Construct an adjacency matrix A from the original dataset based on the physical attributes of road network connectivity and distance relationships to represent the relationships between road segments;

[0011] Step S5: The adjacent order matrix A, combined with the traffic information map, performs preliminary correction and completion of the original traffic feature data through a space-time information compensation network;

[0012] Step S6: Establish the IGANI iterative completion generative adversarial network model and generate the completion dataset;

[0013] Step S7: Based on the road physical connection relationship model established by the neighbor matrix A and the directed road network graph, a weighted directed graph is built. The traffic state characteristics are represented by the graph signal matrix. A traffic state prediction model is constructed by a graph embedding long short-term memory neural network GE-LSTM with data completion mechanism to predict the traffic state data within the period.

[0014] Step S8: The traffic status data and the completed dataset are used to construct a holographic traffic status dataset.

[0015] Furthermore, according to claim 1, the method for constructing a holographic traffic state dataset based on multi-source information fusion is characterized in that, in step S2, the step of constructing a directed road network graph and an undirected road network graph according to the road network topology includes:

[0016] Step S21: Abstract each road segment in the actual road segment into adjacent blocks according to the road segment adjacency relationship;

[0017] Step S22: Delineate the intersections and road segments in the traffic network;

[0018] Step S23: Represent the speed data of each road segment using a two-dimensional information graph to construct an actual road network map, including an undirected road network map and a directed road network map.

[0019] Furthermore, according to the method for constructing a holographic traffic state dataset based on multi-source information fusion as described in claim 1, the step S3, which uses several colors to represent the average speed of each road segment and uses a color matching equation to color-represent each speed value to output a road network traffic information map, includes:

[0020] Step S31: Calculate the average speed S of each road segment. i The average speed of each road segment is represented by three colors: RGB. The average speed of each road segment is determined by formula (1) using the color matching equation.

[0021] S i =c(C)=r(R)+g(G)+b(B) (1)

[0022] Step S32: Based on the number of road segments, obtain the equation data of the average speed of each road segment, and output the road network traffic information map;

[0023] Among them, S i C represents the average speed of road segment i; C represents the color of road segment i; R represents red; G represents green; B represents blue.

[0024] Furthermore, according to the method for constructing a holographic traffic state dataset based on multi-source information fusion as described in claim 1, the step S5, in which the neighboring matrix A, combined with the traffic information map, performs preliminary correction and completion of the original traffic feature data through a space-time information compensation network, includes:

[0025] Step S51, calculate the optimal space-time information parameters, including;

[0026] Step S511: The spatial-temporal traffic information compensation is trained using machine learning methods, that is, the optimization training of the estimated parameters of the spatial-temporal traffic information network is performed, and the spatial-temporal traffic information transformation is expressed as Equation (2).

[0027] I' t+k =T θ (I t+k (2)

[0028] Among them, I' t+k For I t+k After space-time transformation T θ (I t+k Traffic information map for the following time period; I t+k Traffic information map for time period t+k; T θ (·) represents a road segment correlation matrix model used to simulate changes in road traffic flow.

[0029] Step S512, the loss function of the space-time information compensation network is represented by a regularization method, and its optimal parameter estimation method is determined by equation (3);

[0030] θ * =argmin||I t -I' t+k ||+λ||L×I' t+k || (3)

[0031] Where, θ * For optimal estimation of space-time information parameters; λ is the regularization parameter; L is the Laplace operator; I t This is a traffic information map for time period t.

[0032] Step S513, adjust the right side of equation (3) relative to θ * Differentiate the value and set it to 0. Then, use the maximum gradient descent method for iterative analysis to finally obtain the optimal space-time information parameter θ. * ;

[0033] Step S52: The information generated after compensating the missing traffic information sequence is used as the input feature value;

[0034] Step S53: The input feature values ​​are used by generator G to generate and complete traffic information;

[0035] Step S54: The discriminator D judges the completed traffic information;

[0036] In step S55, the discriminator D outputs the completed traffic information.

[0037] Furthermore, according to claim 1, the method for constructing a holographic traffic state dataset based on multi-source information fusion is characterized in that the training steps of the generator G and the discriminator D and the calculation of the loss function include:

[0038] Step S56, given the generator G, the loss function of the discriminator D is represented by equation (4);

[0039]

[0040] Among them, J (D) (θ (G) ,θ (D) E is the cost function that minimizes the discriminator D; x Let represent the expected value calculated on the distribution of x; z is the input value of the generator G; and x is the input value of the discriminator D.

[0041] A zero-sum game is adopted, and the cost of the generator G is used to neutralize the cost of the discriminator D, as shown in equation (5).

[0042] J (G) =-J (D) (5)

[0043] The optimization of minimax is expressed as follows (6);

[0044]

[0045] Wherein, the network parameters of the generator G are θ (G) The network parameters of the discriminator D are θ (D) ,

[0046] Step S57: Calculate the cost function of the generator G;

[0047] The generator G is trained by maximizing the probability of incorrect classification or by minimizing the following cost function:

[0048]

[0049] The gradient penalty can be incorporated into the WGAN Lipschitz continuity condition constraint on the discriminator loss as Equation (7) to avoid gradient explosion or vanishing.

[0050]

[0051] Where y = tG(z) + (1-t)x, 0 ≤ t ≤ 1, λ = 10;

[0052] The cost function of generator G can be expressed as equation (8);

[0053] J (D) w (θ (G) ,θ (D) )=-E z D(z) (8)

[0054] Step S58: Determine the loss function equations for the generator G and the discriminator D;

[0055] Let the existing sample data be denoted as x, and the generated sample data be denoted as G(z). Then the loss function F of the generator G is... G (z) is defined as follows;

[0056] F G (z)=D(G(z))

[0057] The loss function F of discriminator D D (x, z) is defined as follows:

[0058] F D (x,z)=D(x)+max * (α-D(G(z)))

[0059] Where, max * (·) = max(0,·), where α is a positive real number.

[0060] Furthermore, according to the method for constructing a holographic traffic state dataset based on multi-source information fusion as described in claim 1, the step of training the generator G and discriminator D of the iteratively complete generative adversarial network (IGANI) model in step S6 includes:

[0061] Step S61: Design the data completion generator for the generative adversarial network IGANI model;

[0062] The generator G of the model can be expressed as G(x,m,z)=(u,v).

[0063]

[0064]

[0065] Where x∈Rd is a random variable, representing the observation vector value of traffic state, which includes complete observations and missing values ​​from the incomplete dataset, x∈{0,1}. d It is a random mask for x, where m j =0 is represented as x j Complete observations, m j =1 indicates a missing value; z∈R d Let x represent the noise vector; g(·) is the function to be learned; v is the data completion vector, and when x j When the observations are complete, v j =x j The completed data vector v shares complete observations with vector x; u is the data observation vector containing noise and missing values;

[0066] Step S62: Train the generator G by minimizing the loss function (9) of the generator G in the generative adversarial network IGANI model; train the discriminator D by minimizing the loss function (10) of the discriminator D.

[0067]

[0068]

[0069] in, Let y = tG(z) + (1-t) × v, 0 ≤ t ≤ 1, and λ = 10.

[0070] Furthermore, according to the method for constructing a holographic traffic state dataset based on multi-source information fusion as described in claim 1, the step of establishing an iterative completion generative adversarial network (IGANI) model to generate completion data information in step S6 includes:

[0071] Step S63: The missing observations are iteratively filled in by the data completion generator to generate a completed data matrix v;

[0072] Step S64: Based on v and the reshuffled mask matrix n, the generator G is applied again for a second iteration to generate the completed data matrix.

[0073] Step S65, the discriminator D, by respectively feeding matrix v and Provide a scalar score to distinguish between the first and second data completions, and output the completion information.

[0074] Furthermore, according to the method for constructing a holographic traffic state dataset based on multi-source information fusion as described in claim 1, the step of establishing a traffic state prediction model in step S7 includes:

[0075] S71, Construction of urban intersection road network map, representing the intersection scene as a weighted directed road network map M;

[0076] S72, extract V2X multi-source information fusion data, fuse the temporal and spatial features of the data, and construct the input X. t ;

[0077] In the road network map M, the traffic state X within the specified sliding window l. t It can be expressed as the following formula;

[0078] X t ={x t-l+1 ,x t-l+2 ,...,x t}

[0079] Here, Xt serves as the input to the LSTM neural network model, and its dimension is d.

[0080] S73, input it into the Long Short-Term Memory (LSTM) network unit to predict the traffic state at the next time step t;

[0081] At time t, the LSTM unit has three inputs: the current traffic state X. t The output value h of the previous hidden layer t-1and output state c t-1 It also contains three outputs, which are the hidden layer output values ​​h at this time. t and output state c t The resulting prediction is Y t The states of the three gates are i t f t o t , is a value between 0 and 1, where, in this network unit, c t and h t It will be passed to the next network, and its operation process is as follows: Equation (11)-Equation (15);

[0082] i t =σ(W xi X t +W hi h t-1 +W ci c t-1 +b i (11)

[0083] f t =σ(W xf X t +W h f h t -1 +W c fc t-1 +b f (12)

[0084] o t =σ(W xo +X t +W ho h t-1 +W co c t-1 +b o (13)

[0085] c t =f t c t-1 +tanh(W xc X t +W hc h t-1 +b c (14)

[0086] h t =o t tanh(c t (15)

[0087] Among them, W xi Expression (11) Input X of traffic status tweights, W xf Expression (12) Input X of traffic status t weights, W xo Expression (13) Input X of traffic status t weights, W xc Expression (14) Input X of traffic status t Weights; W hi Expression (11) Hidden layer h t weights, W hf Expression (12) Hidden layer h t weights, W ho Expression (13) Hidden layer h t weights, W hc Expression (14) Hidden layer h t Weights; W ci Expression (11) Output state c t weights, W cf Expression (12) output state c t weights, W co Expression (13) output state c t weights; b i The bias vector and b in expression (11) f The bias vector and b in expression (12) o The bias vector and b in expression (13) c The bias vector in equation (13); tanh represents the activation function, which maps the variables one-to-one in the range [0, 1].

[0088] σ(x) represents the sigmoid function defined in the equation, expressed as follows;

[0089]

[0090] tanh(x) represents the hyperbolic tangent function defined in the equation, determined by the following formula;

[0091]

[0092] S74, output the final result Y t ;

[0093] The network predicts the final output Y. t Calculated as: Y t =W y ·h t +b y

[0094] Among them, W y The hidden layer h represents the traffic prediction value. tThe weight, b y This represents the bias vector of the predicted value.

[0095] Furthermore, according to claim 8, the method for constructing a holographic traffic state dataset based on multi-source information fusion is characterized in that, in step S71, the method for constructing an urban intersection road network map includes:

[0096] Step S11: Install V2X sensors at the intersection.

[0097] Step S12: When the vehicle enters the sensing range of the V2X sensor, establish vehicle-to-vehicle / vehicle-to-infrastructure wireless communication within the area to collect three-dimensional traffic status information at the intersection.

[0098] Step S13: Represent the intersection scenario as a weighted directed graph, denoted as G = (V, E, X); at time t, the traffic state X of the graph M is... t The signal represented by the graph is X. t ∈R n×c ,

[0099] Where V = {v1, v2, ..., v} n} represents the set of nodes on the road segment network, where n is the number of nodes; E = {e ij} represents each vertex v i and v j A set of edges that are interconnected; X represents the traffic state within the V2X roadside sensing range; c represents the number of traffic state features, including traffic flow, traffic density, and traffic speed.

[0100] Furthermore, according to claim 9, the method for constructing a holographic traffic state dataset based on multi-source information fusion is characterized in that, in step S8, the calculation method for constructing the holographic traffic state dataset is as follows:

[0101] The real-time traffic status completion device based on multi-source vehicle-road information fusion collects the original data as the label dataset of the following formula (16);

[0102]

[0103] Where, the neighboring order matrix A represents the connection between road networks, Y a Road condition characteristics include signal phase, signal timing, queue length, and Y. b The characteristics of connected vehicle status include vehicle identification information, vehicle speed, vehicle heading angle, and Y-axis. c The status characteristics of V2X communication devices include channel bandwidth, transmission delay, packet loss rate, uplink rate, and X1~X. t Represents the feature values ​​in a time series.

[0104] Based on the road network characteristics, the effective data is fused to form the original traffic state dataset Y, which includes traffic flow, traffic flow density, and average vehicle speed.

[0105] Y = {X} t-n ,X t-n+1 ,X t-n+2 ,...,X t}

[0106] Where n is the length of the original traffic dataset time series.

[0107] The merged dataset is used to complete the traffic status data with a time step of T using the mapping function f;

[0108] [X t+1 ,...,X t+T ] = f(A; (X) t-n ,...,X -1 ,X t ))

[0109] Completed dataset labels * It can be expressed as the following formula;

[0110]

[0111] Among them, Y * This is a complete traffic status dataset.

[0112] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention leverages the technical advantages of V2X multi-source information fusion to propose a method for constructing a holographic traffic state dataset based on multi-source information fusion. It utilizes the Iterative Completion Generative Adversarial Network (IGANI) to complete missing data, learning the spatial and temporal correlations of traffic flow data. This provides a reliable traffic information dataset for related technical needs, thereby improving the efficiency of urban intersections and reducing traffic accidents. This invention effectively overcomes the problem of missing traffic data caused by equipment damage, aging, disrepair, and malfunctions. The holographic road network traffic data obtained from the completion and prediction data can effectively improve the collaborative management effect of the traffic management system and increase traffic operation efficiency. Furthermore, this invention utilizes data fusion and data completion technologies to effectively enhance the control effect of the traffic control system and alleviate traffic congestion.

[0113] Furthermore, this invention employs V2X communication technology to achieve comprehensive dynamic information interaction between vehicles and roads, and between vehicles and roads. It serves as a communication means for vehicles to coordinate and manage with roads and other vehicles. Through roadside and vehicle-mounted hardware devices, it collects road and vehicle data in real time and achieves bidirectional communication through LTE-V2X or 5G-V2X communication protocols. It features high transmission, low latency, and high stability, providing managers or drivers with dynamic traffic status information and driving advice, thereby enabling collaborative work between people, vehicles, and roads and greatly promoting a safe and efficient traffic environment.

[0114] Furthermore, this invention employs multi-source information fusion technology, relying on vehicle wireless communication technology and based on multi-source sensor fusion, including roadside perception technology with devices such as LiDAR, high-definition cameras, intelligent roadside units (RSUs), intelligent on-board units (OBUs) with V2X communication capabilities, and mobile edge computing (MEC) units, to perceive traffic observation data and generate raw datasets. This overcomes the problem that a single sensor cannot meet the real-time monitoring needs of complex traffic networks, achieving accurate perception of urban intersections, thereby improving the traffic efficiency and safety of urban road intersections.

[0115] Furthermore, the method for completing traffic datasets in this invention utilizes adversarial neural networks to learn and train on missing data over a certain time step, performing adversarial completion to generate a complete traffic dataset while ensuring a low loss rate. The dataset completion results, characterized by low error and high representativeness of evaluation indicators, will provide managers with a research foundation and data basis for traffic control, and also offer a reference for further improving the accuracy of traffic condition perception and prediction.

[0116] Furthermore, this invention uses historical traffic datasets as input to study and analyze the spatiotemporal correlation between vehicles and road networks. Utilizing deep learning techniques such as convolutional neural networks, a traffic state prediction model is established to predict real-time traffic conditions at different times, thus constructing a holographic traffic state dataset. This traffic state prediction technology provides reliable algorithmic support for decision-makers to formulate control plans and prevent traffic accidents, effectively improving road traffic efficiency and safety. Attached Figure Description

[0117] Figure 1 This is a schematic diagram of the method for constructing a holographic traffic state dataset based on multi-source information fusion according to the present invention;

[0118] Figure 2 This is a schematic diagram of the data acquisition device structure in the holographic traffic status dataset construction method based on multi-source information fusion of the present invention;

[0119] In the diagram: 1. LiDAR; 2. HD camera; 3. Traffic light; 4. Switch; 5. Roadside Unit (RSU); 6. GPS module; 7. MEC unit; 8. Antenna; 9. Real-time monitoring screen; 10. Traffic light pole; 11. Traffic light pole base. Detailed Implementation

[0120] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0121] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0122] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0123] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0124] Please see Figure 1 As shown, it is a schematic diagram of the method for constructing a holographic traffic state dataset based on multi-source information fusion according to the present invention, including:

[0125] Step S1: Traffic signal controllers and roadside sensors are used to send signal timing data, LiDAR and high-definition camera data to intelligent vehicle-mounted unit (OBU) with V2X communication function and multi-source information fusion device (MEC). The processed data information is then uploaded to cloud server or computer terminal to form raw dataset.

[0126] Step S2: Construct directed and undirected road network graphs based on the road network topology.

[0127] Step S3: The road network map is processed into a visual representation of traffic information based on the color matching equation;

[0128] Step S4: Construct an adjacency matrix A from the original dataset based on the physical attributes of road network connectivity and distance relationships to represent the relationships between road segments;

[0129] Step S5: The adjacent order matrix A, combined with the traffic information map, performs preliminary correction and completion of the original traffic feature data through a space-time information compensation network;

[0130] Step S6: Establish the IGANI iterative completion generative adversarial network model and generate the completion dataset;

[0131] Step S7: Based on the road physical connection relationship model established by the neighbor matrix A and the directed road network graph, a weighted directed graph is built. The traffic state characteristics are represented by the graph signal matrix. A traffic state prediction model is constructed by a graph embedding long short-term memory neural network GE-LSTM with data completion mechanism to predict the traffic state data within the period.

[0132] Step S8: The traffic status data and the completed dataset are used to construct a holographic traffic status dataset.

[0133] This invention leverages the technological advantages of V2X multi-source information fusion to propose a method for constructing a holographic traffic state dataset based on multi-source information fusion. It utilizes an iterative completion generative adversarial network to complete missing data, learning the spatial and temporal correlations of traffic flow data. This provides a reliable traffic information dataset for related technical needs, aiming to improve the efficiency of urban intersections and reduce traffic accidents. This invention effectively overcomes the problem of missing traffic data caused by equipment damage, aging, disrepair, and malfunctions. The holographic road network traffic data obtained from the completion and prediction data can effectively improve the collaborative management effect of the traffic management system and increase traffic operation efficiency. Furthermore, by utilizing data fusion and data completion technologies, this invention can effectively enhance the control effect of the traffic control system and alleviate traffic congestion.

[0134] Specifically, in step S2, the steps of constructing directed and undirected road network graphs based on the road network topology include:

[0135] The steps for constructing directed and undirected road network graphs based on road network topology include:

[0136] Step S21: Abstract each road segment in the actual road segment into adjacent blocks according to the road segment adjacency relationship;

[0137] Step S22: Delineate the intersections and road segments in the traffic network;

[0138] Step S23: Represent the speed data of each road segment using a two-dimensional information graph to construct an actual road network map, including an undirected road network map and a directed road network map.

[0139] The directed road network is constructed based on the actual road traffic direction, including east-west and north-south road networks. In a single directed road network map, the flow direction of a single road is connected end to end to represent several traffic paths.

[0140] Specifically, in step S3, the steps of using several colors to represent the average speed of each road segment and using color matching equations to color-represent each speed value to output a road network traffic information map include:

[0141] Step S31: Calculate the average speed S of each road segment. i The average speed of each road segment is represented by three colors: RGB. The average speed of each road segment is determined by formula (1) using the color matching equation.

[0142] S i =c(C)=r(R)+g(G)+b(B) (1)

[0143] Step S32: Based on the number of road segments, obtain the equation data of the average speed of each road segment, and output the road network traffic information map;

[0144] Among them, S i C represents the average speed of road segment i; C represents the color of road segment i; R represents red; G represents green; B represents blue.

[0145] In practice, within the visible light range of 597–780 nm, the wavelength interval is used to define the range of traffic flow variation (the maximum being the maximum average speed that a road segment can withstand), and a set of color matching equations (17) corresponding to the traffic flow of each road segment in the road network can be obtained:

[0146]

[0147] Where m represents the number of road segments in the regional road network; when the data of a certain road segment is missing, the corresponding two-dimensional information color block will turn gray.

[0148] Specifically, in step S5, the step of using the adjacent-order matrix A combined with the traffic information map and a space-time information compensation network to initially correct and complete the original traffic feature data includes:

[0149] Step S51, calculate the optimal space-time information parameters, including;

[0150] Step S511: The spatial-temporal traffic information compensation is trained using machine learning methods, that is, the optimization training of the estimated parameters of the spatial-temporal traffic information network is performed, and the spatial-temporal traffic information transformation is expressed as Equation (2).

[0151] I' t+k =T θ (I t+k (2)

[0152] Among them, I' t+k For I t+k After space-time transformation T θ (I t+k Traffic information map for the following time period; I t+k Traffic information map for time period t+k; T θ (·) represents a road segment correlation matrix model used to simulate changes in road traffic flow.

[0153] Step S512, the loss function of the space-time information compensation network is represented by a regularization method, and its optimal parameter estimation method is determined by equation (3);

[0154] θ * =argmin||I t -I' t+k ||+λ||L×I' t+k || (3)

[0155] Where, θ * For optimal estimation of space-time information parameters; λ is the regularization parameter; L is the Laplace operator; I t This is a traffic information map for time period t.

[0156] Step S513, adjust the right side of equation (3) relative to θ * Differentiate the value and set it to 0. Then, use the maximum gradient descent method for iterative analysis to finally obtain the optimal space-time information parameter θ. * ;

[0157] Step S52: The information generated after compensating the missing traffic information sequence is used as the input feature value;

[0158] Step S53: The input feature values ​​are used by generator G to generate and complete traffic information;

[0159] Step S54: The discriminator D judges the completed traffic information;

[0160] In step S55, the discriminator D outputs the completed traffic information.

[0161] Specifically, the training steps for the generator G and the discriminator D, and the calculation steps for the loss function, include:

[0162] Step S56, given the generator G, the loss function of the discriminator D is represented by equation (4);

[0163]

[0164] Among them, J (D) (θ (G) ,θ (D) E is the cost function that minimizes the discriminator D; x Let represent the expected value calculated on the distribution of x; z is the input value of the generator G; and x is the input value of the discriminator D.

[0165] A zero-sum game is adopted, and the cost of the generator G is used to neutralize the cost of the discriminator D, as shown in equation (5).

[0166] J (G) =-J (D) (5)

[0167] The optimization of minimax is expressed as follows (6);

[0168]

[0169] Wherein, the network parameters of the generator G are θ (G) The network parameters of the discriminator D are θ (D) ,

[0170] Step S57: Calculate the cost function of the generator G;

[0171] The generator G is trained by maximizing the probability of incorrect classification or by minimizing the following cost function:

[0172]

[0173] The gradient penalty can be incorporated into the WGAN Lipschitz continuity condition constraint on the discriminator loss as Equation (7) to avoid gradient explosion or vanishing.

[0174]

[0175] Where y = tG(z) + (1-t)x, 0 ≤ t ≤ 1, λ = 10; WGAN Lipschitz continuity condition.

[0176] The cost function of generator G can be expressed as equation (8);

[0177] J (D) w (θ (G) ,θ (D) )=- E zD(z) (8)

[0178] Step S58: Determine the loss function equations for the generator G and the discriminator D;

[0179] Let the existing sample data be denoted as x, and the generated sample data be denoted as G(z). Then the loss function F of the generator G is... G (z) is defined as follows;

[0180] F G (z)=D(G(z))

[0181] The loss function F of discriminator D D (x, z) is defined as follows:

[0182] F D (z,z)=D(x)+max * (α-D(G(z)))

[0183] Where, max * (·) = max(0,·), where α is a positive real number.

[0184] Specifically, in step S6, the steps of training the generator G and discriminator D of the IGANI iteratively complete generative adversarial network model include:

[0185] Step S61: Design the data completion generator for the generative adversarial network IGANI model;

[0186] The generator G of the model can be expressed as G(x,m,z)=(u,v).

[0187]

[0188]

[0189] Where, x∈R d It is a random variable, representing the observed vector values ​​of traffic states. The observed vector values ​​include complete observations and missing values ​​from the incomplete dataset, x∈{0,1}. d It is a random mask for x, where m j =0 is represented as x j Complete observations, m j =1 indicates a missing value; z∈R d Let x represent the noise vector; g(·) is the function to be learned; v is the data completion vector, and when x j When the observations are complete, v j =x j The completed data vector v shares complete observations with vector x; u is the data observation vector containing noise and missing values;

[0190] Step S62: Train the generator G by minimizing the loss function (9) of the generator G in the generative adversarial network IGANI model; train the discriminator D by minimizing the loss function (10) of the discriminator D.

[0191]

[0192]

[0193] in, Let y = tG(z) + (1-t) × v, 0 ≤ t ≤ 1, and λ = 10.

[0194] Specifically, in step S6, the step of establishing the iterative completion generative adversarial network IGANI model to generate completion data information includes:

[0195] Step S63: The missing observations are iteratively filled in by the data completion generator to generate a completed data matrix v;

[0196] Step S64: Based on v and the reshuffled mask matrix n, the generator G is applied again for a second iteration to generate the completed data matrix.

[0197] Step S65, the discriminator D, by respectively feeding matrix v and Provide a scalar score to distinguish between the first and second data completions, and output the completion information.

[0198] Specifically, in step S7, the step of establishing a traffic state prediction model includes:

[0199] S71, Construction of urban intersection road network map, representing the intersection scene as a weighted directed road network map M;

[0200] S72, extract V2X multi-source information fusion data, fuse the temporal and spatial features of the data, and construct the input X. t ;

[0201] In the road network map M, the traffic state X within the specified sliding window l. t It can be expressed as the following formula;

[0202] X t ={x t-l+1 ,x t-l+2 ,...,x t}

[0203] Among them, X t As input to the LSTM neural network model, its dimension is d.

[0204] S73, input it into the Long Short-Term Memory (LSTM) network unit to predict the traffic state at the next time step t;

[0205] At time t, the LSTM unit has three inputs: the current traffic state X. t The output value h of the previous hidden layer t-1 and output state c t-1 It also contains three outputs, which are the hidden layer output values ​​h at this time. t and output state c t The resulting prediction is Y t The states of the three gates are i t f t o t , is a value between 0 and 1, where, in this network unit, c t and h t It will be passed to the next network, and its operation process is as follows: Equation (11)-Equation (15);

[0206] i t =σ(W xi X t +W hi h t-1 +W ci c t-1 +b i (11)

[0207] f t =σ(W xf X t +W h fh t-1 +W cf c t-1 +b f (12)

[0208] o t =σ(W xo X t +W ho h t-1 +W co c t-1 +b o (13)

[0209] c t =f t c t-1 +tanh(W xc X t +W hc h t-1 +b c (14)

[0210] ht =o t tanh(c t (15)

[0211] Among them, W xi Expression (11) Input X of traffic status t weights, W xf Expression (12) Input X of traffic status t weights, W xo Expression (13) Input X of traffic status t weights, W xc Expression (14) Input X of traffic status t Weights; W hi Expression (11) Hidden layer h t weights, W hf Expression (12) Hidden layer h t weights, W ho Expression (13) Hidden layer h t weights, W hc Expression (14) Hidden layer h t Weights; W ci Expression (11) Output state c t weights, W cf Expression (12) output state c t weights, W co Expression (13) output state c t weights; b i The bias vector and b in expression (11) f The bias vector and b in expression (12) o The bias vector and b in expression (13) c The bias vector in equation (13); tanh represents the activation function, which maps the variables one-to-one in the range [0, 1].

[0212] σ(x) represents the sigmoid function defined in the equation, expressed as follows;

[0213]

[0214] tanh(x) represents the hyperbolic tangent function defined in the equation, determined by the following formula;

[0215]

[0216] S74, output the final result Y t ;

[0217] The network predicts the final output Y. t Calculated as: Yt =W y ·h t +b y

[0218] Among them, W y The hidden layer h represents the traffic prediction value. t The weight, b y This represents the bias vector of the predicted value.

[0219] This invention uses historical traffic datasets as input to study and analyze the spatiotemporal correlation between vehicles and road networks. Utilizing deep learning techniques such as convolutional neural networks, it establishes a traffic state prediction model to predict real-time traffic conditions at different times, thus constructing a holographic traffic state dataset. This traffic state prediction technology provides reliable algorithmic support for decision-makers to formulate control plans and prevent traffic accidents, effectively improving road traffic efficiency and safety.

[0220] Specifically, in step S71, the method for constructing the urban intersection road network map includes:

[0221] Step S11: Install V2X sensors at the intersection.

[0222] Step S12: When the vehicle enters the sensing range of the V2X sensor, establish vehicle-to-vehicle / vehicle-to-infrastructure wireless communication within the area to collect three-dimensional traffic status information at the intersection.

[0223] Step S13: Represent the intersection scenario as a weighted directed graph, denoted as G = (V, E, X); at time t, the traffic state X of the graph M is... t The signal represented by the graph is X. t ∈R n×c ,

[0224] Where V = {v1, v2, ..., v} n} represents the set of nodes on the road segment network, where n is the number of nodes; E = {e ij} represents a set of edges connecting vertices vi and vj; X represents the traffic state within the V2X roadside sensing range; c represents the number of traffic state features, including traffic flow, traffic density, and traffic speed.

[0225] The present invention provides a method for completing traffic datasets. This method utilizes adversarial neural networks to learn and train on missing data over a certain time step, performing adversarial completion to generate a complete traffic dataset while ensuring a low loss rate. The dataset completion results, characterized by low error and high representativeness of evaluation indicators, will provide managers with a research foundation and data basis for traffic control, and also offer a reference for further improving the accuracy of traffic condition perception and prediction.

[0226] Specifically, in step S8, the calculation method for constructing the holographic traffic state dataset is as follows:

[0227] The real-time traffic status completion device based on multi-source vehicle-road information fusion collects the original data as the label dataset of the following formula (16);

[0228]

[0229] Where, the neighboring order matrix A represents the connection between road networks, Y a Road condition characteristics include signal phase, signal timing, queue length, and Y. b The characteristics of connected vehicle status include vehicle identification information, vehicle speed, vehicle heading angle, and Y-axis. c The status characteristics of V2X communication devices include channel bandwidth, transmission delay, packet loss rate, uplink rate, and X1~X. t Represents the feature values ​​in a time series.

[0230] Based on the road network characteristics, the effective data is fused to form the original traffic state dataset Y, which includes traffic flow, traffic flow density, and average vehicle speed.

[0231] Y = {X} t-n ,X t-n+1 ,X t-n+2 ,...,X t}

[0232] Where n is the length of the original traffic dataset time series.

[0233] The merged dataset is used to complete the traffic status data with a time step of T using the mapping function f;

[0234] [X t+1 ,...,X t+T ] = f(A; (X) t-n ,...,X t-1 ,X t ))

[0235] Completed dataset labels * It can be expressed as the following formula;

[0236]

[0237] Among them, Y * This is a complete traffic status dataset.

[0238] This invention can effectively overcome the problem of missing traffic data caused by equipment damage, aging, disrepair, and equipment failure. Based on the holographic road network traffic data obtained by supplementing and predicting data, this invention can effectively improve the collaborative management effect of the traffic management system and improve traffic operation efficiency. This invention utilizes data fusion technology and data supplementation technology to effectively improve the control effect of the traffic control system and alleviate traffic congestion.

[0239] Example 1:

[0240] Please see Figure 2 This is a schematic diagram of the data acquisition device structure in the holographic traffic state dataset construction method based on multi-source information fusion of the present invention. The specific steps for installing the data acquisition device include:

[0241] Step 1: The device is equipped with a 32-line mechanical lidar 1, which is installed at the top of the traffic light 3. To ensure data synchronization, the sampling frequency is set to 10Hz. It is used to identify and acquire basic information such as the position of vehicles on the road. It is connected to the MEC unit through a gigabit switch.

[0242] Step 2: Install a high-definition camera 2 on the top of the traffic light 3 to realize the function of machine vision recognition to obtain real-time vehicle location information, and connect it to the MEC unit through a gigabit switch;

[0243] Step 3: Install the gigabit switch 4 on the upper part of the traffic signal pole 10 for data transmission between the components of the device;

[0244] Step 4: Install the roadside unit RSU5 in the middle of the traffic signal pole 10. It is used to establish a connection with the intelligent vehicle unit OBU through V2I communication and receive intelligent vehicle information data.

[0245] Step 5: Install GPS module 6 on the lower part of traffic signal pole 10 to obtain the geographical location of the smart traffic signal pole;

[0246] Step 6: Install antenna 8 on the traffic signal pole base 11 for connection between the device and the Internet, and exchange device data with cloud data.

[0247] Step 7: Install MEC unit 7 on the device, in a stable, well-ventilated and heat-dissipating location on the traffic signal pole base 11, for edge computing tasks such as multi-source data fusion;

[0248] Step 8: Install the real-time monitoring screen 9 inside the traffic signal pole base 11 to monitor the OBU task unloading process, V2X communication status, real-time road traffic status, and MEC task processing status, etc.

[0249] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0250] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a holographic traffic state dataset based on multi-source information fusion, characterized in that, include: Step S1: Traffic signal controllers and roadside sensors are used to send signal timing data, LiDAR and high-definition camera data to intelligent vehicle-mounted unit (OBU) with V2X communication function and multi-source information fusion device (MEC). The processed data information is then uploaded to cloud server or computer terminal to form raw dataset. Step S2: Construct directed and undirected road network graphs based on the road network topology. Step S3: The road network map is processed into a visual representation of traffic information based on a color matching equation; Step S4: The original dataset is used to construct an adjacency matrix A based on the physical attributes of road network connectivity and distance relationships to represent the relationships between road segments; Step S5: The adjacent order matrix A, combined with the traffic information map, performs preliminary correction and completion of the original traffic feature data through a space-time information compensation network; Step S6: Establish the IGANI iterative completion generative adversarial network model and generate the completion dataset; Step S7: Based on the road physical connection relationship model established by the neighbor matrix A and the directed road network graph, a weighted directed graph is constructed. The traffic state characteristics are represented by the graph signal matrix. A traffic state prediction model is constructed using a graph embedding long short-term memory neural network (GE-LSTM) with data completion mechanism to predict the traffic state data within the period. In step S7, the step of establishing the traffic state prediction model includes: S71, Construction of urban intersection road network map, representing the intersection scenario as a weighted directed road network map M; methods for constructing urban intersection road network maps include: Step S11: Install V2X sensors at the intersection. Step S12: When the vehicle enters the sensing range of the V2X sensor, establish vehicle-to-vehicle / vehicle-to-infrastructure wireless communication within the area to collect three-dimensional traffic status information at the intersection. Step S13: Represent the intersection scenario as a weighted directed graph, denoted as G = (V, E, X); at time t, the traffic state X of the graph M is... t The signal represented by the graph is X. t ∈R n×c , Where V = {v1, v2, ..., v} n } represents the set of nodes on the road segment network, where n is the number of nodes; E = {e ij } represents a set of edges connecting vertices vi and vj; X represents the traffic state within the V2X roadside sensing range; c represents the number of traffic state features, including traffic flow, traffic density, and traffic speed; S72, extract V2X multi-source information fusion data, fuse the temporal and spatial features of the data, and construct the input X. t ; In the road network map M, the traffic state X within the specified sliding window l. t It can be expressed as the following formula; X t ={x t-l+1 ,x t-l+2 ,...,x t } Among them, X t As input to the LSTM neural network model, its dimension is d. S73, input it into the Long Short-Term Memory (LSTM) network unit to predict the traffic state at the next time step t; At time t, the LSTM unit has three inputs: the current traffic state X. t The output value h of the previous hidden layer t-1 and output state c t-1 It also contains three outputs, which are the hidden layer output values ​​h at this time. t and output state c t The resulting prediction is Y t The states of the three gates are i t f t o t , is a value between 0 and 1, where, in this network unit, c t and h t It will be passed to the next network, and its operation process is as follows: Equation (11)-Equation (15); i t =σ(W xi X t +W hi h t-1 +W ci c t-1 +b i ) (11) f t =σ(W xf X t +W hf h t-1 +W cf c t-1 +b f ) (12) o t =σ(W xo X t +W ho h t-1 +W co c t-1 +b o ) (13) c t =f t c t-1 +tanh(W xc X t +W hc h t-1 +b c ) (14) h t = no t fishy(c) t ) (15) Among them, W xi Expression (11) Input X of traffic status t weights, W xf Expression (12) Input X of traffic status t weights, W xo Expression (13) Input X of traffic status t weights, W xc Expression (14) Input X of traffic status t Weights; W hi Expression (11) Hidden layer h t weights, W hf Expression (12) Hidden layer h t weights, W ho Expression (13) Hidden layer h t weights, W hc Expression (14) Hidden layer h t Weights; W ci Expression (11) Output state c t weights, W cf Expression (12) output state c t weights, W co Expression (13) output state c t weights; b i The bias vector and b in expression (11) f The bias vector and b in expression (12) o The bias vector and b in expression (13) c The bias vector in equation (13); tanh represents the activation function, which maps the variables one-to-one in the range [0, 1]. σ(x) represents the sigmoid function defined in the equation, expressed as follows; tanh(x) represents the hyperbolic tangent function defined in the equation, determined by the following formula; S74, output the final result Y t ; The network predicts the final output Y. t Calculated as: Y t =W y ·h t +b y Among them, W y The hidden layer h represents the traffic prediction value. t The weight, b y The bias vector representing the predicted value; Step S8: The traffic state data and the completed dataset are used to construct a holographic traffic state dataset. The calculation method for constructing the holographic traffic state dataset in step S8 is as follows: The data acquisition device for real-time traffic status based on multi-source vehicle-road information fusion collects raw data as the label dataset of the following formula (16); Where, the neighboring order matrix A represents the connection between road networks, Y a Road condition characteristics include signal phase, signal timing, queue length, and Y. b The characteristics of connected vehicle status include vehicle identification information, vehicle speed, vehicle heading angle, and Y-axis. c The status characteristics of V2X communication devices include channel bandwidth, transmission delay, packet loss rate, uplink rate, and X1~X. t Represents the feature values ​​in a time series. Based on the road network characteristics, the effective data is fused to form the original traffic state dataset Y, which includes traffic flow, traffic flow density, and average vehicle speed. Y={X t-n ,X t-n+1 ,X t-n+2 ,...,X t } Where n is the length of the original traffic dataset time series. The merged dataset is used to complete the traffic status data with a time step of T using the mapping function f; [X t+1 ,...,X t+T ]=f(A;(X t-n ,...,X t-1 ,X t )) Completed dataset labels * It can be expressed as the following formula; Among them, Y * For a complete traffic status dataset, The specific steps for installing the data acquisition device include: Step 1: The device is equipped with a 32-line mechanical lidar, which is installed at the top of the traffic light. The sampling frequency is set to 10Hz to identify and acquire basic information about the position of vehicles on the road. It is connected to the MEC unit through a gigabit switch. Step 2: The device is equipped with a high-definition camera, which is installed on the top of the traffic light to realize the function of machine vision recognition to obtain real-time vehicle location information. It is connected to the MEC unit through a gigabit switch. Step 3: Install a gigabit switch on the upper part of the traffic signal pole for data transmission between the various components of the device; Step 4: Install the roadside unit (RSU5) in the middle of the traffic signal pole. It is used to establish a connection with the intelligent vehicle unit (OBU) via V2I communication and receive intelligent vehicle information data. Step 5: Install a GPS module on the device at the bottom of the traffic signal pole to obtain the geographical location of the smart traffic signal pole; Step 6: Install the antenna on the device at a stable point on the base of the traffic signal pole. This antenna is used to connect the device to the Internet and to exchange data between the device and the cloud. Step 7: Install the MEC unit on the traffic signal pole base for edge computing tasks such as multi-source data fusion; Step 8: Install a real-time monitoring screen inside the traffic signal pole base to monitor the OBU task offloading process, V2X communication status, real-time road traffic status, and MEC task processing status.

2. The method for constructing a holographic traffic state dataset based on multi-source information fusion according to claim 1, characterized in that, In step S2, the steps of constructing directed and undirected road network graphs based on the road network topology include: Step S21: Abstract each road segment in the actual road segment into adjacent blocks according to the road segment adjacency relationship; Step S22: Delineate the intersections and road segments in the traffic network; Step S23: Represent the speed data of each road segment using a two-dimensional information graph to construct an actual road network map, including an undirected road network map and a directed road network map.

3. The method for constructing a holographic traffic state dataset based on multi-source information fusion according to claim 1, characterized in that, In step S3, the steps of using several colors to represent the average speed of each road segment and using color matching equations to color-code each speed value to output a road network traffic information map include: Step S31: Calculate the average speed S of each road segment. i The average speed of each road segment is represented by three colors: RGB. The average speed of each road segment is determined by formula (1) using the color matching equation. S i =c(C)=r(R)+g(G)+b(B) (1) Among them, S i C represents the average speed of road segment i; R represents red; G represents green; B represents blue. Step S32: Based on the number of road segments, obtain the equation data of the average speed of each road segment, and output the road network traffic information map.

4. The method for constructing a holographic traffic state dataset based on multi-source information fusion according to claim 1, characterized in that, In step S5, the step of using the adjacent-order matrix A combined with the traffic information map and a space-time information compensation network to initially correct and complete the original traffic feature data includes: Step S51, calculate the optimal space-time information parameters, including; Step S511: The spatial-temporal traffic information compensation is trained using machine learning methods, that is, the optimization training of the estimated parameters of the spatial-temporal traffic information network is performed, and the spatial-temporal traffic information transformation is expressed as Equation (2). I' t+k =T θ (I t+k ) (2) Among them, I' t+k For I t+k After space-time transformation T θ (I t+k Traffic information map for the following time period; I t+k Traffic information map for time period t+k; T θ (·) represents a road segment correlation matrix model used to simulate changes in road traffic flow. Step S512, the loss function of the space-time information compensation network is represented by a regularization method, and its optimal parameter estimation method is determined by equation (3); i * =argmin||I t -I' t+k ||+λ||L×I' t+k || (3) Where, θ * For optimal estimation of space-time information parameters; λ is the regularization parameter; L is the Laplace operator; I t This is a traffic information map for time period t. Step S513, adjust the right side of equation (3) relative to θ * Differentiate the value and set it to 0. Then, use the maximum gradient descent method for iterative analysis to finally obtain the optimal space-time information parameter θ. * ; Step S52: The information generated after compensating the missing traffic information sequence is used as the input feature value; Step S53: The input feature values ​​are used by generator G to generate and complete traffic information; Step S54: The discriminator D judges the completed traffic information; In step S55, the discriminator D outputs the completed traffic information.

5. The method for constructing a holographic traffic state dataset based on multi-source information fusion according to claim 4, characterized in that, The training steps for the generator G and the discriminator D, and the calculation steps for the loss function, include: Step S56, given the generator G, the loss function of the discriminator D is represented by equation (4); Among them, J (D) (θ (G) ,θ (D) E is the cost function that minimizes the discriminator D; x Let represent the expected value calculated on the distribution of x; z is the input value of the generator G; and x is the input value of the discriminator D. A zero-sum game is adopted, and the cost of the generator G is used to neutralize the cost of the discriminator D, as shown in equation (5). J (G) -J (D) (5) The optimization of minimax is expressed as follows (6); Wherein, the network parameters of the generator G are θ (G) The network parameters of the discriminator D are θ (D) , Step S57: Calculate the cost function of the generator G; The generator G is trained by maximizing the probability of incorrect classification or by minimizing the following cost function: The gradient penalty can be incorporated into the WGAN Lipschitz continuity condition constraint on the discriminator loss as Equation (7) to avoid gradient explosion or vanishing. Where y = tG(z) + (1-t)x, 0 ≤ t ≤ 1, λ = 10; The cost function of generator G can be expressed as equation (8); J (D) w (i (G) ,i (D) )=-E z D(z) (8) Step S58: Determine the loss function equations for the generator G and the discriminator D; Let the existing sample data be denoted as x, and the generated sample data be denoted as G(z). Then the loss function F of the generator G is... G (z) is defined as follows; F G (z)=D(G(z)) The loss function F of discriminator D D (x, z) is defined as follows: F D (x,z)=D(x)+max * (α-D(G(z))) Where, max * (·) = max(0,·), where α is a positive real number.

6. The method for constructing a holographic traffic state dataset based on multi-source information fusion according to claim 5, characterized in that, In step S6, the steps of training the generator G and discriminator D of the IGANI iteratively complete generative adversarial network model include: Step S61: Design the data completion generator for the generative adversarial network IGANI model; The generator G of the model can be expressed as G(x,m,z)=(u,v). Where, x∈R d It is a random variable, representing the observed vector values ​​of traffic states. The observed vector values ​​include complete observations and missing values ​​from the incomplete dataset, x∈{0,1}. d It is a random mask for x, where m j =0 is represented as x j Complete observations, m j =1 indicates a missing value; z∈R d Let x represent the noise vector; g(·) is the function to be learned; v is the data completion vector, and when x j When the observations are complete, v j =x j The completed data vector v shares complete observations with vector x; u is the data observation vector containing noise and missing values; Step S62: Train the generator G by minimizing the loss function (9) of the generator G in the generative adversarial network IGANI model; train the discriminator D by minimizing the loss function (10) of the discriminator D. in, Let y = tG(z) + (1-t) × v, 0 ≤ t ≤ 1, and λ = 10.

7. The method for constructing a holographic traffic state dataset based on multi-source information fusion according to claim 6, characterized in that, In step S6, the steps of establishing the iterative completion generative adversarial network IGANI model to generate completion data information include: Step S63: The missing observations are iteratively filled in by the data completion generator to generate a completed data matrix v; Step S64: Based on v and the reshuffled mask matrix n, the generator G is applied again for a second iteration to generate the completed data matrix. Step S65, the discriminator D, by respectively feeding matrix v and Provide a scalar score to distinguish between the first and second data completions, and output the completion information.

Images