Large-scale network traffic state multi-parameter collaborative prediction method for data loss

Abstract

The invention relates to a data loss-oriented large-scale network traffic state multi-parameter collaborative prediction method. The method comprises the following steps: S1, acquiring traffic state parameter historical information of a target traffic network and preprocessing the acquired data; s2, constructing a large-scale road network traffic multi-parameter node graph, providing a self-adaptive tensor graph network architecture oriented to various data loss conditions, inputting the self-adaptive tensor graph network architecture into historical traffic state multi-parameter data and a traffic node graph, and outputting the self-adaptive tensor graph network architecture into a traffic flow prediction result at a future moment; according to the invention, by constructing the adaptive tensor diagram network architecture, the physical coupling relationship among traffic parameters and the road network topological structure are deeply fused, and the internal relevance among multiple parameters can be effectively utilized to carry out collaborative learning and information complementation.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation, and in particular to a multi-parameter collaborative prediction method for large-scale network traffic conditions in the face of data loss. Background Technology

[0002] Traffic condition prediction is a core component of intelligent transportation systems, and its accuracy directly impacts road network management, resource allocation, and the quality of travel services. Traffic conditions are characterized by a coupling of multiple parameters, including flow rate, speed, and occupancy rate. Increased flow rate at the same node is often accompanied by increased occupancy rate and decreased speed, and this coupling relationship exhibits non-linear differences depending on road network topology, time of day (e.g., peak / off-peak), and holidays. Accurately characterizing the co-evolutionary patterns of these multiple parameters is crucial for improving prediction performance.

[0003] Existing technologies have three limitations: 1. Inadequate handling of data loss: Data loss caused by sensor failure, transmission interruption, etc. is common (such as random loss, node-level loss, time dimension loss and combined loss). Traditional interpolation or single-parameter imputation methods do not utilize the functional relationship between multiple parameters (such as the physical coupling relationship between flow rate, speed and occupancy), and the accuracy drops sharply under high loss rate. 2. Lack of multi-parameter collaborative modeling: Most methods model parameters such as flow and speed independently, or splice shallow features, making it difficult to capture high-order spatiotemporal coupling laws and ignoring the impact of "parameter feedback" on traffic state evolution; Therefore, there is an urgent need to construct a multi-parameter collaborative prediction method for large-scale network traffic conditions that is resistant to data loss. Summary of the Invention

[0004] In view of this, the present invention aims to propose a multi-parameter collaborative prediction method for large-scale network traffic states oriented towards data loss, in order to solve the problems in the prior art.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows: This invention proposes a multi-parameter collaborative prediction method for large-scale network traffic conditions oriented towards data loss. The method includes the following steps: S1, Obtain historical information on traffic state parameters of the target traffic network and preprocess the acquired data; S2 constructs a large-scale road network traffic multi-parameter node graph and proposes an adaptive tensor graph network architecture for various data loss scenarios. Its input is historical traffic state multi-parameter data and traffic node graph, and its output is the traffic flow prediction result for future time. S3 utilizes the Ivy optimization algorithm to optimize the key hyperparameter set of the adaptive tensor graph network architecture and obtain the globally optimal hyperparameters of the model. S4. The optimal hyperparameters output by the Ivy optimization algorithm are applied to the adaptive tensor graph network architecture. The model is trained using the Adam optimizer and early stopping mechanism. The parameters are updated through backpropagation to minimize the error, resulting in a well-trained multi-parameter traffic state prediction model. S5, based on the trained model, enables multi-parameter prediction of traffic conditions in large-scale networks.

[0006] Furthermore, in S1, the preprocessing includes the following steps: S101 cleans the collected traffic status data, jointly identifies and removes outliers in single parameters such as flow rate, speed, and occupancy. S102, construct a multi-parameter time series tensor of traffic status based on the temporal characteristics, spatial characteristics and data parameter types of the cleaned traffic status data; S103 divides the data obtained in S103 into training set, validation set and test set.

[0007] Furthermore, in S102, the multi-parameter time series tensor can be constructed using M features at a time step size T. And it marked the locations where traffic data was missing, obtaining... Tensors with equal dimensions for missing location information .

[0008] Furthermore, in S2, the adaptive tensor graph network architecture includes constructing a traffic network graph from the acquired traffic network information, constructing a multi-parameter graph convolutional network, constructing a fast tensor-gated recurrent unit, and designing a model training loss function.

[0009] Furthermore, the traffic network diagram output is as follows: in Represents a network node. Represents an edge. The adjacency tensor representing a network node; for Adjacency tensors in a given dimension.

[0010] Furthermore, the multi-parameter graph convolutional network is constructed as follows: While extracting spatial features of different traffic network nodes, a multidimensional tensor is constructed for multiple parameters of traffic flow. Tensor operations are used to implement circular convolution between tensors. Through block circular matrix expansion and matrix multiplication simulation, the interaction relationships between parameters are captured, and the extraction of parameter coupling rules is strengthened under data loss conditions. Therefore, the output of the multi-parameter graph convolutional network at time t can be denoted as... , represented as in, The propagation depth of a multi-parameter graph convolutional network, For tensor A time slice at time t.

[0011] Furthermore, the construction of the fast tensor-gated recurrent unit dynamically generates a low-rank kernel tensor by obtaining a trainable core tensor and factor matrix through Tucker decomposition. and They act on different gating mechanisms respectively: in, and In Desirable , respectively represent the update gate, reset gate, and hidden state of the fast tensor gating loop unit. This represents n modulo multiplication. For trainable core tensors, These are feature factors of different dimensions, which can be obtained through training.

[0012] The fast tensor-gated loop unit is represented as: in, yes The state tensor is hidden at all times. and These are the update door and the reset door. Candidate state layer; and It is an activation function. Represents the Hadamard product. Represents the tensor inner product. This is a bias term.

[0013] Furthermore, the training loss function of the design model is calculated as follows: in: To predict the loss, the mean squared error can be used to calculate the error of the non-missing data; To calculate the reconstruction loss, the theoretical reconstruction error of the missing data is calculated based on parameter correlation. The loss is L2 regularization to prevent the model from overfitting; , , The weighting coefficient indicates the data loss rate; the higher the data loss rate, the higher the weighting coefficient. The higher the weight.

[0014] Furthermore, in S3, the hyperparameter optimization involves inputting the tensor graph convolution kernel size, hidden layer dimension, learning rate, time window length, and parameter relationship weights of the adaptive tensor graph network architecture into the Ivy optimization algorithm. After initializing the population, the parameters are dynamically optimized through global roaming and local search. The fitness is calculated and the parameters are updated. The optimal solution is retained after boundary processing and survivor selection. When the termination condition is reached, the globally optimal hyperparameters are output.

[0015] Furthermore, the data output by the trained model in S5 includes different time steps of traffic network nodes at future moments, multi-parameter future data of traffic states under different loss conditions, and a visualization of the predicted and actual values ​​of data for each node.

[0016] Compared with the prior art, the present invention has the following advantages: In this invention, an adaptive tensor graph network architecture is constructed, which deeply integrates the physical coupling relationship between traffic parameters and the road network topology. This architecture can effectively utilize the inherent correlation between multiple parameters for collaborative learning and information complementarity. The architecture is adaptive to various missing data patterns, such as random loss, node-level loss, and time-dimensional loss, enabling more accurate data reconstruction and imputation. Even in extreme cases of high-proportion data loss, it maintains extremely high prediction stability and accuracy, overcoming the shortcomings of traditional interpolation and single-parameter models where accuracy drops drastically. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0019] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "back," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0020] Furthermore, in the description of this invention, unless otherwise explicitly defined, the terms "installation," "connection," "linking," and "connector" 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 in light of the specific circumstances.

[0021] The following will refer to the appendix. Figure 1 The present invention will be described in detail with reference to the embodiments.

[0022] Overall, this invention discloses a multi-parameter collaborative prediction method for large-scale network traffic states in the face of data loss, the method comprising the following steps: S1, Obtain historical information on traffic state parameters of the target traffic network and preprocess the acquired data; S2 constructs a large-scale road network traffic multi-parameter node graph and proposes an adaptive tensor graph network architecture for various data loss scenarios. Its input is historical traffic state multi-parameter data and traffic node graph, and its output is the traffic flow prediction result for future time. S3 utilizes the Ivy optimization algorithm to optimize the key hyperparameter set of the adaptive tensor graph network architecture and obtain the globally optimal hyperparameters of the model. S4. The optimal hyperparameters output by the Ivy optimization algorithm are applied to the adaptive tensor graph network architecture. The model is trained using the Adam optimizer and early stopping mechanism. The parameters are updated through backpropagation to minimize the error, resulting in a well-trained multi-parameter traffic state prediction model. S5, based on the trained model, enables multi-parameter prediction of traffic conditions in large-scale networks.

[0023] In this embodiment, an adaptive tensor graph network architecture is constructed, which deeply integrates the physical coupling relationship between traffic parameters and the road network topology. This architecture can effectively utilize the inherent correlation between multiple parameters (flow rate, speed, occupancy rate) for collaborative learning and information complementarity. This architecture has adaptive capabilities for various missing data patterns, such as random loss, node-level loss, and time-dimensional loss, enabling more accurate data reconstruction and imputation. Therefore, even in extreme cases of high-proportion data loss, it maintains extremely high prediction stability and accuracy, overcoming the shortcomings of traditional interpolation and single-parameter models where accuracy drops sharply.

[0024] The Ivy League (IVY) optimization algorithm is introduced to automatically and globally optimize the key hyperparameters of the model, effectively avoiding the blindness, time-consuming nature, and tendency to get trapped in local optima of traditional manual hyperparameter tuning. This process ensures that the model can adapt to traffic network datasets of different sizes and characteristics, always running at the optimal configuration, significantly improving the universality, automation level, and upper limit of the final performance of the method.

[0025] The model training adopts a strategy that combines the Adam optimizer with an early stopping mechanism. On the one hand, the efficient convergence characteristics of the Adam optimizer are used to accelerate the training process. On the other hand, the early stopping mechanism monitors the performance of the validation set in real time and terminates the training in time before the model overfits. This ensures the excellent generalization ability of the final model, making it perform well on unknown data as well, and enhancing the reliability of actual deployment.

[0026] In summary, this method exhibits high scalability for large-scale networks. The accurate and robust multi-parameter prediction results obtained can be directly applied to advanced applications in intelligent transportation systems, such as real-time traffic guidance, intelligent traffic light timing, congestion warning, and travel assistance (ETA), providing core decision support for improving the operational efficiency, management level, and public travel experience of urban road networks, and possessing significant socio-economic value.

[0027] Based on the above settings, in S1, the preprocessing includes the following steps: S101 cleans the collected traffic state data, jointly identifying and removing outliers in single parameters such as flow rate, speed, and occupancy. This effectively identifies and removes implicit anomalies where a single parameter may appear within a reasonable range, but when considered in conjunction with other parameters, they violate the principles of traffic flow dynamics. This ensures the rationality and consistency of the input data from the source, significantly reducing the interference of noisy data on model learning and laying a solid foundation for training a stable and reliable prediction model.

[0028] S102 constructs a multi-parameter time series tensor for traffic state based on the temporal and spatial characteristics and data parameter types of the cleaned traffic state data. This tensor organically integrates the three dimensions of time, space, and features, transforming the complex road network state into an efficient and structured format suitable for processing by modern deep learning models (especially tensor decomposition and graph neural networks). This representation method preserves the inherent spatiotemporal correlations and parameter correlations in traffic data, providing an information source for models to deeply mine these complex dependencies.

[0029] S103 divides the data obtained in S103 into training set, validation set and test set.

[0030] In step S102, the multi-parameter time series tensor can be constructed by selecting M features under a time step size T. And it marked the locations where traffic data was missing, obtaining... Tensors with equal dimensions for missing location information .

[0031] It should be noted that in S2, the adaptive tensor graph network architecture includes constructing a traffic network graph from the acquired traffic network information, constructing a multi-parameter graph convolutional network, constructing a fast tensor-gated recurrent unit, and designing a model training loss function.

[0032] By constructing a traffic network graph, the road network topology (such as node connections, road grades, and distances) is transformed into graph-structured data, enabling the model to understand complex spatial relationships. A multi-parameter graph convolutional network (GCN) is constructed to simultaneously perform joint convolution operations on multiple parameters such as traffic flow, speed, and occupancy on the graph structure. A Fast Tensor Gated Recurrent Unit (Fast Tensor GRU) extends the traditional GRU to a time-series model capable of directly processing multi-dimensional tensor inputs. A specially designed model training loss function typically incorporates multi-task learning ideas (such as jointly optimizing prediction and reconstruction losses).

[0033] The traffic network diagram output is as follows: in Represents a network node. Represents an edge. Represents the adjacency tensor between network nodes; for Adjacency tensors in a given dimension.

[0034] The multi-parameter graph convolutional network is constructed as follows: While extracting spatial features of different traffic network nodes, a multidimensional tensor is constructed for multiple parameters of traffic flow. Tensor operations are used to implement circular convolution between tensors. Through block circular matrix expansion and matrix multiplication simulation, the interaction relationships between parameters are captured, and the extraction of parameter coupling rules is strengthened under data loss conditions. Therefore, the output of the multi-parameter graph convolutional network at time t can be denoted as... , represented as in, The propagation depth of a multi-parameter graph convolutional network, For tensor A time slice at time t.

[0035] The construction of the fast tensor-gated recurrent unit dynamically generates a low-rank kernel tensor by obtaining a trainable core tensor and factor matrix through Tucker decomposition. and They act on different gating mechanisms respectively: in, and In Desirable , respectively represent the update gate, reset gate, and hidden state of the fast tensor gating loop unit. This represents n modulo multiplication. For trainable core tensors, These are feature factors of different dimensions, which can be obtained through training.

[0036] The fast tensor-gated loop unit is represented as: in, yes The state tensor is hidden at all times. and These are the update door and the reset door. Candidate state layer; and It is an activation function. Represents the Hadamard product. Represents the tensor inner product. This is a bias term.

[0037] The training loss function of the design model is calculated as follows: in: To predict the loss, the mean squared error can be used to calculate the error of the non-missing data; To calculate the reconstruction loss, the theoretical reconstruction error of the missing data is calculated based on parameter correlation. The loss is L2 regularization to prevent the model from overfitting; , , The weighting coefficient indicates the data loss rate; the higher the data loss rate, the higher the weighting coefficient. The higher the weight.

[0038] It should be further explained that in S3, the hyperparameter optimization is to input the tensor graph convolution kernel size, hidden layer dimension, learning rate, time window length, and parameter relationship weights of the adaptive tensor graph network architecture into the Ivy optimization algorithm. After initializing the population, the parameters are dynamically optimized through global roaming and local search, the fitness is calculated and the parameters are updated, and the optimal solution is retained after boundary processing and survivor selection. When the termination condition is reached, the globally optimal hyperparameters are output.

[0039] By employing a dynamic strategy that combines population initialization, global roaming, and local search, key hyperparameters such as tensor graph convolution kernel size, hidden layer dimension, learning rate, time window length, and parameter relationship weights are optimized collaboratively.

[0040] In addition, the data output by the trained model in S5 includes different time steps of traffic network nodes at future times, multi-parameter future data of traffic state under different loss conditions, and a visualization of the predicted and actual values ​​of each node's data.

[0041] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-parameter collaborative prediction method for traffic conditions in large-scale networks facing data loss, characterized in that: The method includes the following steps: S1, Obtain historical information on traffic state parameters of the target traffic network and preprocess the acquired data; S2 constructs a large-scale road network traffic multi-parameter node graph and proposes an adaptive tensor graph network architecture for various data loss scenarios. Its input is historical traffic state multi-parameter data and traffic node graph, and its output is the traffic flow prediction result for future time. S3 utilizes the Ivy optimization algorithm to optimize the key hyperparameter set of the adaptive tensor graph network architecture and obtain the globally optimal hyperparameters of the model. S4. The optimal hyperparameters output by the Ivy optimization algorithm are applied to the adaptive tensor graph network architecture. The model is trained using the Adam optimizer and early stopping mechanism. The parameters are updated through backpropagation to minimize the error, resulting in a well-trained multi-parameter traffic state prediction model. S5, based on the trained model, enables multi-parameter prediction of traffic conditions in large-scale networks.

2. The method for multi-parameter collaborative prediction of large-scale network traffic states oriented towards data loss, as described in claim 1, is characterized in that: In S1, the preprocessing includes the following steps: S101 cleans the collected traffic status data, jointly identifies and removes outliers in single parameters such as flow rate, speed, and occupancy. S102, construct a multi-parameter time series tensor of traffic status based on the temporal characteristics, spatial characteristics and data parameter types of the cleaned traffic status data; S103 divides the data obtained in S103 into training set, validation set and test set.

3. The method for multi-parameter collaborative prediction of large-scale network traffic states oriented towards data loss, as described in claim 2, is characterized in that: In step S102, the multi-parameter time series tensor can be constructed by selecting M features under a time step size T. And it marked the locations where traffic data was missing, obtaining... Tensors with equal dimensions for missing location information .

4. The method for multi-parameter collaborative prediction of large-scale network traffic states oriented towards data loss, as described in claim 3, is characterized in that: In S2, the adaptive tensor graph network architecture includes constructing a traffic network graph from the acquired traffic network information, constructing a multi-parameter graph convolutional network, constructing a fast tensor-gated recurrent unit, and designing a model training loss function.

5. The method for multi-parameter collaborative prediction of large-scale network traffic states oriented towards data loss, as described in claim 4, is characterized in that: The traffic network diagram output is as follows: in Represents a network node. Represents an edge. Represents the adjacency tensor between network nodes; Let m be the adjacency tensor in m dimensions.

6. The method for multi-parameter collaborative prediction of large-scale network traffic states oriented towards data loss, as described in claim 5, is characterized in that: The multi-parameter graph convolutional network is constructed as follows: While extracting spatial features of different traffic network nodes, a multidimensional tensor is constructed for multiple parameters of traffic flow. Tensor operations are used to implement circular convolution between tensors. Through block circular matrix expansion and matrix multiplication simulation, the interaction relationships between parameters are captured, and the extraction of parameter coupling rules is strengthened under data loss conditions. Therefore, the output of the multi-parameter graph convolutional network at time t can be denoted as... , represented as in, The propagation depth of a multi-parameter graph convolutional network, For tensor exist A time slice of a moment.

7. A multi-parameter collaborative prediction method for large-scale network traffic states oriented towards data loss, as described in claim 6, is characterized in that: Low-rank kernel tensors are dynamically generated by obtaining trainable core tensors and factor matrices through Tucker decomposition. and They act on different gating mechanisms respectively: in, and In Desirable , respectively represent the update gate, reset gate, and hidden state of the fast tensor gating loop unit. This represents n modulo multiplication. For trainable core tensors, These are feature factors of different dimensions, which can be obtained through training; The fast tensor-gated loop unit is represented as: in, yes The state tensor is hidden at all times. and These are the update door and the reset door. Candidate state layer; and It is an activation function. Represents the Hadamard product. Represents the tensor inner product. This is a bias term.

8. The method for multi-parameter collaborative prediction of large-scale network traffic states oriented towards data loss, as described in claim 7, is characterized in that: The training loss function of the design model is calculated as follows: in: To predict the loss, the mean squared error can be used to calculate the error of the non-missing data; To calculate the reconstruction loss, the theoretical reconstruction error of the missing data is calculated based on parameter correlation. The loss is L2 regularization to prevent the model from overfitting; , , The weighting coefficient indicates the data loss rate; the higher the data loss rate, the higher the weighting coefficient. The higher the weight.

9. A multi-parameter collaborative prediction method for large-scale network traffic states oriented towards data loss, as described in claim 1, is characterized in that: In S3, hyperparameter optimization involves inputting the tensor graph convolution kernel size, hidden layer dimension, learning rate, time window length, and parameter relationship weights of the adaptive tensor graph network architecture into the Ivy optimization algorithm. After initializing the population, the parameters are dynamically optimized through global roaming and local search. The fitness is calculated and the parameters are updated. The optimal solution is retained after boundary processing and survivor selection. When the termination condition is reached, the globally optimal hyperparameters are output.

10. A multi-parameter collaborative prediction method for large-scale network traffic states oriented towards data loss, as described in claim 1, characterized in that: The data output by the trained model in S5 includes different time steps of traffic network nodes at future moments, multi-parameter future data of traffic states under different loss conditions, and a visualization of the predicted and actual values ​​of data for each node.

Images