Traffic flow speed prediction method based on soft set and integrated physical information neural network
Patent Information
- Application Number
- CN202611067611.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-17
- Publication Date
- 2026-10-23
AI Technical Summary
[0004]有鉴于此，本发明的目的在于解决现有深度学习方法可解释性差、泛化能力弱、模型本身具备不确定性的缺陷，提供一种基于软集合与集成物理信息神经网络的交通流速度预测方法
[0022]与现有技术相比，本发明具有如下有益效果：本发明提出了一种基于软集合与集成物理信息神经网络集成的交通流速度预测方法，通过将交通流基本图模型、物理信息神经网络以及软集合理论进行融合，实现了交通流速度的高精度预测。
Smart Images
Figure CN122967460A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent transportation technology and relates to a traffic flow speed prediction method based on soft set and integrated physical information neural network. Background Technology
[0002] Traffic flow velocity prediction is a crucial component of intelligent transportation systems and plays a vital role in alleviating urban traffic congestion. With accelerating urbanization, urban traffic is becoming increasingly complex, exhibiting significant spatiotemporal dynamics and nonlinear variation patterns. Accurate and efficient prediction of traffic flow velocity has become a research hotspot in traffic management and control. Accurate traffic flow prediction not only provides a scientific basis for route planning, traffic signal control, and travel guidance, but also improves road efficiency, reduces traffic congestion and energy consumption, and has significant practical implications for building smart cities and achieving green transportation. Current mainstream traffic flow prediction methods can be divided into two main types: model-driven physical models and data-driven machine learning models, but both have certain limitations. Model-driven physical models are typically built upon traffic flow theory and rigorous mathematical derivations, such as cellular automata models and car-following models. These methods offer good interpretability and can theoretically describe the operational patterns of traffic flow. However, due to the high complexity and randomness of real-world traffic systems, traffic flow is influenced by various factors such as weather, accidents, driving behavior, and road conditions. Therefore, traditional physical models often struggle to accurately depict the nonlinear spatiotemporal characteristics of complex traffic scenarios, resulting in limited prediction accuracy, especially in large-scale, complex road network environments.
[0003] With the development of big data and artificial intelligence technologies, machine learning and deep learning methods are widely used in traffic flow prediction, such as support vector machines and long short-term memory networks (LSTM). Data-driven machine learning models can automatically learn traffic flow change patterns from large amounts of historical traffic data, showing a significant advantage in prediction accuracy. However, data-driven models typically rely heavily on large-scale, high-quality datasets, and their performance tends to degrade when data is missing or noisy. Meanwhile, most deep learning models are black-box models, lacking clear physical meaning and interpretability, making it difficult to reveal the intrinsic mechanisms of traffic flow evolution. Coupled with the inherent uncertainties of deep learning models, their practical engineering applications remain limited. Summary of the Invention
[0004] In view of this, the purpose of this invention is to address the shortcomings of existing deep learning methods, such as poor interpretability, weak generalization ability, and inherent uncertainty in the models, and to provide a traffic flow speed prediction method based on soft set and integrated physical information neural network.
[0005] To achieve the above objectives, the present invention provides the following technical solution: A traffic flow speed prediction method based on soft ensemble and ensemble physical information neural networks includes the following steps: S1: Collect historical traffic flow data for road segments, including traffic volume, traffic flow speed, and other characteristics that can reflect the operational status of traffic flow. S2: Select the three basic graph models, Underwood, Greenshields and Greenberg, as physical prior knowledge, and train the three physical information neural network models respectively after calibration using historical traffic flow data. S3: Use the three trained physical information neural network models as base learners and use soft set theory to evaluate the weights of the three physical information neural network models. S4: Based on the evaluation results of soft sets, the traffic flow speed prediction values of three physical information neural network models are integrated, and the output is the prediction value of the neural network model that combines different physical constraints, so as to achieve the prediction of traffic flow speed.
[0006] Furthermore, the calculation formula for the Underwood model is as follows:
[0007] The calculation formula for the Greenshields model is as follows:
[0008] The Greenberg model calculation formula is as follows:
[0009] in, For speed, For free flow velocity; For density, Crowding density.
[0010] Furthermore, in step S2, the loss function used when training the three physical information neural network models includes the traditional data point fitting loss and the prediction error between the data-driven model and the basic graph model, as shown below:
[0011] in, The loss function for the physical information neural network model. For traditional data point fitting loss, As weight, This represents the prediction error between the data-driven model and the basic graphical model.
[0012] Furthermore, in step S3, referring to and simplifying the framework of ensemble learning, specific loss functions are designed for the three physical information neural network models using three basic graph models, as follows:
[0013]
[0014]
[0015] in, The loss function for the physical information neural network based on the Underwood basic graph model is... The loss function for the physical information neural network based on the Greenshields basic graph model is... The loss function for the physical information neural network based on the Greenberg basic graph model is... The number of samples; Three different loss functions were used to train three different physical information neural network models, and the output results were retained.
[0016] Furthermore, in step S3, soft set theory is used to evaluate three physical information neural network models, specifically including: Using error as a guide, the model as the object, and time as the sequence, a soft set table is constructed, and then the following evaluation is based on the soft set:
[0017]
[0018]
[0019] Where, ξ( ij ) represents the first element in the soft set table. i At the 1st time point, the 1st j The absolute value of the error value corresponding to each model; It is the reciprocal of the absolute value of the error. This represents the evaluation result of the model, and also its weights.
[0020] Furthermore, in step S4, the results of the soft set evaluation are used as weights for model allocation. By combining the output values of the neural network model with different physical information, the final prediction results are as follows:
[0021] in, This is the model's final traffic flow speed prediction. This is the predicted value of the j-th model.
[0022] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a traffic flow speed prediction method based on the integration of soft set and integrated physical information neural network. By integrating the basic traffic flow graph model, physical information neural network and soft set theory, high-precision prediction of traffic flow speed is achieved.
[0023] (1) Improved the accuracy of traffic flow speed prediction.
[0024] This invention introduces three classic basic graph models—Underwood, Greenshields, and Greenberg—as physical constraints and constructs corresponding physical information neural network models for each. This allows the neural network to not only fit existing observation data but also satisfy physical equations as "soft constraints" integrated into the optimization objective. Compared to traditional neural network models that rely solely on historical data for training, the physical information neural network model can not only learn the nonlinear relationships in complex traffic flows but also use physical constraints to correct the model's learning direction, thereby reducing prediction errors and improving the accuracy of traffic flow speed prediction.
[0025] (2) It enhances the interpretability of the model.
[0026] Deep learning models are opaque and have complex internal mechanisms. This "black box" nature makes it difficult for models to interpret their outputs, limiting their reliability in engineering applications. This invention uses a physical information neural network that embeds a basic graph model into the neural network training process, enabling the model's predictions to reflect the intrinsic physical relationships between traffic flow density, volume, and speed. Therefore, this invention not only outputs prediction results but also explains the model's predictive behavior from a traffic flow theory perspective, improving the model's interpretability.
[0027] (3) Improved the model’s data extrapolation capability.
[0028] Existing data-driven models typically rely heavily on large-scale training data. When data is insufficient, the model is prone to underfitting, leading to a decline in predictive performance. This invention introduces physical constraints, enabling the model to maintain adherence to physical laws even when data is insufficient, sparse, or traffic conditions change, thereby enhancing the model's data extrapolation capabilities.
[0029] (4) Improved the model’s adaptability to complex traffic conditions.
[0030] Actual traffic flow exhibits significant spatiotemporal coupling characteristics and random fluctuations, making it difficult for a single model to comprehensively describe its operational patterns. This invention introduces three basic graphical models, each applicable to traffic flows at different densities. The physical information neural networks constructed from these models can predict traffic flow states from different perspectives. This multi-model integrated prediction effectively improves the model's ability to predict various traffic flows, including free-flowing, steady-flowing, and congested flows, thereby enhancing the overall prediction performance.
[0031] (5) The prediction stability and robustness are improved through model ensemble.
[0032] Because neural network models suffer from issues such as random initialization, parameter fluctuations, and local optima during training, the prediction results of a single model are easily affected by uncertainties. This invention, referencing the concept of ensemble learning, fuses multiple physical information neural network models. By integrating the prediction results of different models, it effectively reduces the impact of single model errors on the overall prediction results, thereby improving the stability and robustness of traffic flow speed prediction.
[0033] (6) Use soft set theory to effectively solve the model uncertainty problem.
[0034] This invention employs soft set theory to evaluate neural network models with different physical information and dynamically allocates model weights based on the evaluation results. Compared to traditional fixed-weight methods, this invention can more flexibly handle uncertainties and differences between models, allowing each model to fully leverage its strengths in different traffic scenarios, thereby further improving the reliability and accuracy of prediction results.
[0035] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0036] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the overall framework of the traffic flow prediction method of the present invention; Figure 2 This is a comparison diagram between the physical information neural network used in this invention and a regular neural network model; Figure 3 This is a 2D visualization of the model results in an embodiment of the present invention; Figure 4 This is a 3D visualization of the model results in an embodiment of the present invention. Detailed Implementation
[0037] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.
[0038] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.
[0039] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.
[0040] Example 1: Please see Figures 1-2 This invention provides a traffic flow speed prediction method that uses a physical information neural network as a base learner and employs soft ensembles for ensemble processing, comprising the following steps: S1: Collect historical traffic flow data of road sections through facilities such as cameras and sensors. The traffic flow data needs to include traffic volume, traffic flow speed, and other characteristics that can reflect the traffic flow operation status.
[0041] S2: Three basic graph models, Underwood, Greenshields, and Greenberg, were selected as physical prior knowledge. After calibration using historical traffic flow data, three physical information neural network models were trained respectively. Because physical laws are integrated into the training of deep neural networks, the trained physical information neural network models have better interpretability and data extrapolation ability compared with traditional data-driven models.
[0042] Step S2 includes: three basic graphical models calibrated based on historical data, and a loss function redesigned based on the basic graphical models.
[0043] The three basic graphical models refer to the classic Underwood, Greenshields, and Greenberg models in the field of transportation. These three models are applicable to traffic flow states under different density ranges, and their specific formulas are as follows: Underwood model calculation formula:
[0044] Greenshields model calculation formula:
[0045] Greenberg model calculation formula:
[0046] in, For speed, For free flow velocity; For density, This refers to the crowding density.
[0047] The redesigned loss function based on the fundamental graphical model includes both the traditional data point fitting loss and the prediction error between the data-driven model and the fundamental graphical model. Its basic form is as follows:
[0048] in, The loss function for the physical information neural network model. For traditional data point fitting loss, As weight, This represents the prediction error between the data-driven model and the basic graphical model.
[0049] S3: Referring to and simplifying the framework of ensemble learning, the three physical information neural network models trained in step S2 are used as base learners to ensemble and evaluate the models. To address the inherent uncertainty of the neural network models, soft set theory is used to evaluate the weights of the three physical information neural network models.
[0050] Specifically, referring to and simplifying the framework of ensemble learning, we designed specific loss functions for three physical information neural network models using three basic graph models. The specific loss functions are as follows:
[0051]
[0052]
[0053] in, The loss function for the physical information neural network based on the Underwood basic graph model is... The loss function for the physical information neural network based on the Greenshields basic graph model is... The loss function for the physical information neural network based on the Greenberg basic graph model is... This represents the number of samples.
[0054] The three loss functions mentioned above are used for training, and the output results are retained. The three physical information neural network models will be evaluated using soft set theory, and then the output results will be integrated.
[0055] The soft set theory evaluation is error-oriented, model-oriented, and time-series-oriented, establishing soft set Table 1, as shown in Table 1.
[0056] Table 1
[0057] Where, t( i ) indicates the first i At a certain point in time, i =1,2,…,n,m( j ) indicates the first j One model, j =1,2,…,m,ξ( ij ) indicates the first i Line number j The absolute value of the error value corresponding to the column.
[0058] Once the soft set table is successfully established, the evaluation method based on soft sets is as follows:
[0059]
[0060]
[0061] in, It is the absolute value between the actual value and the model prediction. It is the reciprocal of the absolute value of the error. This represents the evaluation result of the model, and also its weights.
[0062] S4: Based on the soft set evaluation results, the traffic flow speed prediction values of three physical information neural network models are integrated, and the output combines the prediction values of the neural network models with different physical constraints to achieve accurate traffic flow speed prediction. Specifically, the results of the soft set evaluation are used as the weights for model allocation, and the output values of different physical information neural network models are combined. The final prediction results are as follows:
[0063] in, This is the model's final traffic flow speed prediction. This is the predicted value of the j-th model.
[0064] Example 2: Please see Figures 3-4 To verify the effectiveness of the traffic flow speed prediction method based on soft set and integrated physical information neural network proposed in this invention, Python was used for experimental verification. The specific experimental setup is shown below.
[0065] Table 2 shows the software versions.
[0066] Table 2
[0067] For the hyperparameters of the deep learning model, this study uses the random grid search method to search, and also regards the physical weight α as a hyperparameter. The relevant hyperparameter search space is shown in Table 3.
[0068] Table 3
[0069] Comparative Models: Three comparative models are presented, each using a single physical information neural network model and employing an LSTM model. Three basic graph models are used as physical constraints. This invention presents a physical information neural network model based on soft set ensembles, using mean squared error as the metric for comparative analysis.
[0070] Using mean squared error as the indicator, the experimental results are shown in Table 4.
[0071] Table 4
[0072] The experimental results lead to the following conclusions: The traffic flow speed prediction method based on soft set theory and ensemble physical information neural networks proposed in this invention can effectively improve traffic flow prediction accuracy. This invention enhances the model's generalization ability and interpretability through physical constraints, reduces the error of a single model through multi-model ensemble, and optimizes the model fusion weights through soft set theory, achieving synergistic gains among the three. It effectively overcomes the shortcomings of existing physical models in terms of prediction accuracy, as well as the poor interpretability and weak generalization ability of traditional data-driven models. It demonstrates high prediction accuracy, stability, reliability, and engineering application value in the field of traffic flow speed prediction.
[0073] Example 3: An electronic device, comprising a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method described in Embodiment 1 when executing the computer program.
[0074] Example 4: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 1.
[0075] Example 5: A computer program product includes a computer program that, when executed by a processor, implements the method described in Example 1.
[0076] In the above embodiments, the reference to "this embodiment" in the specification indicates that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple appearances of "this embodiment" do not necessarily refer to the same embodiment.
[0077] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.
[0078] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0079] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.
[0080] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.
[0081] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0082] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.
[0083] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A traffic flow speed prediction method based on soft set and ensemble physical information neural network, characterized in that: Includes the following steps: S1: Collect historical traffic flow data for road segments, including traffic volume, traffic flow speed, and other characteristics that can reflect the operational status of traffic flow. S2: Select the three basic graph models, Underwood, Greenshields and Greenberg, as physical prior knowledge, and train the three physical information neural network models respectively after calibration using historical traffic flow data. S3: Use the three trained physical information neural network models as base learners and use soft set theory to evaluate the weights of the three physical information neural network models. S4: Based on the evaluation results of soft sets, the traffic flow speed prediction values of three physical information neural network models are integrated, and the output is the prediction value of the neural network model that combines different physical constraints, so as to achieve the prediction of traffic flow speed.
2. The traffic flow speed prediction method based on soft set and ensemble physical information neural network according to claim 1, characterized in that: The calculation formula for the Underwood model is as follows: The calculation formula for the Greenshields model is as follows: The Greenberg model calculation formula is as follows: in, For speed, For free flow velocity; For density, This refers to the crowding density.
3. The traffic flow speed prediction method based on soft set and ensemble physical information neural network according to claim 1, characterized in that: In step S2, the loss function used when training the three physical information neural network models includes the traditional data point fitting loss and the prediction error between the data-driven model and the basic graph model, as shown below: in, The loss function for the physical information neural network model. For traditional data point fitting loss, As weight, This represents the prediction error between the data-driven model and the basic graphical model.
4. The traffic flow speed prediction method based on soft set and ensemble physical information neural network according to claim 1, characterized in that: In step S3, referring to and simplifying the framework of ensemble learning, specific loss functions are designed for the three physical information neural network models using three basic graph models. The specific loss functions are as follows: in, The loss function for the physical information neural network based on the Underwood basic graph model is... The loss function for the physical information neural network based on the Greenshields basic graph model is... The loss function for the physical information neural network based on the Greenberg basic graph model is... The number of samples; Three different loss functions were used to train three different physical information neural network models, and the output results were retained.
5. The traffic flow speed prediction method based on soft set and ensemble physical information neural network according to claim 1, characterized in that: In step S3, soft set theory is used to evaluate three physical information neural network models, specifically including: Using error as a guide, the model as the object, and time as the sequence, a soft set table is constructed, and then the following evaluation is based on the soft set: Where, ξ( ij ) represents the first element in the soft set table. i At the 1st time point, the 1st j The absolute value of the error value corresponding to each model; It is the reciprocal of the absolute value of the error. This represents the evaluation result of the model, and also its weights.
6. The traffic flow speed prediction method based on soft set and ensemble physical information neural network according to claim 1, characterized in that: In step S4, the results of the soft set evaluation are used as the weights assigned to the model. By combining the output values of the neural network model with different physical information, the final prediction results are as follows: in, This is the model's final traffic flow speed prediction. This is the predicted value of the j-th model.
7. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement the traffic flow speed prediction method based on soft set and integrated physical information neural network as described in any one of claims 1-6.
8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the traffic flow speed prediction method based on soft set and integrated physical information neural network as described in any one of claims 1-6.
9. A computer program product, characterized in that: The method includes a computer program that, when executed by a processor, implements the traffic flow speed prediction method based on soft set and integrated physical information neural network as described in any one of claims 1-6.