Traffic flow prediction method and system based on brain-computer interface, and storage medium

Abstract

The invention discloses a traffic flow prediction method and system based on a brain-computer interface, and a storage medium, and belongs to the technical field of traffic flow prediction. The method comprises the following steps: collecting brain-computer interface data of traffic participants by using brain-computer interface equipment, and synchronously collecting traffic flow basic data of corresponding time periods and road sections; preprocessing the collected data, and extracting brain-computer interface feature data reflecting the reaction of the brain to traffic; aligning the preprocessed brain-computer interface feature data with the traffic flow basic data in time and space, and establishing preliminary association between the two; constructing a fusion model by adopting a deep learning algorithm, and inputting the multi-source data after time-space alignment into the fusion model to mine a deep relationship among the data; training a fusion model by using historical data, and adjusting model parameters through an optimization algorithm; and performing real-time traffic flow prediction based on the trained model, and dynamically adjusting and issuing traffic information according to a prediction result. According to the invention, the change of the traffic flow can be predicted more accurately.

Description

Technical Field

[0001] This invention relates to the field of traffic flow prediction technology, and in particular to a traffic flow prediction method, system and storage medium based on brain-computer interface. Background Technology

[0002] In the field of intelligent transportation systems, traffic flow prediction is a key technology for improving road traffic efficiency and alleviating congestion. Current technologies primarily rely on traditional sensors and machine learning algorithms for prediction, but these methods have many limitations and cannot fully capture the complex dynamics of traffic flow. Currently, traffic flow prediction widely employs data collection methods based on traditional sensors (such as geomagnetic sensors and video surveillance cameras), combined with machine learning algorithms (such as support vector machines) for modeling. These schemes obtain macroscopic statistical information such as traffic flow, speed, and occupancy by sensing changes in the magnetic field caused by vehicle passage or using image recognition technology, thereby predicting traffic flow trends. However, this method, which relies on a single data source and shallow models, struggles to cope with the influence of multi-source heterogeneous data and complex intrinsic factors within the transportation system.

[0003] The existing technology has the following disadvantages: 1. Limitations of Data Collection: Existing data collection methods primarily rely on traditional sensors to gather basic traffic flow data. Traditional sensors can only detect changes in the magnetic field caused by passing vehicles, thereby obtaining traffic flow information; video surveillance cameras use image recognition technology to count vehicle numbers and determine vehicle speeds. The types of data collected by these traditional devices are relatively limited, mainly focusing on macroscopic statistical information of traffic flow, and failing to capture the more complex and intrinsic information of traffic participants. Reasons for this limitation: Traffic flow conditions are not only affected by objective factors such as road infrastructure and vehicle numbers, but also closely related to subjective factors such as the psychological state and decision-making processes of traffic participants. For example, in the early stages of traffic congestion, some drivers may change their driving behavior due to anxiety, such as frequent lane changes, sudden acceleration, and sudden braking. These behaviors further influence the evolution of traffic flow. However, traditional sensors cannot capture these psychological and behavioral changes of drivers, resulting in incomplete data collection that fails to accurately reflect the true dynamics of traffic flow.

[0004] 2. Lack of consideration for the intrinsic reactions of traffic participants: Existing technologies predict traffic flow solely based on basic traffic flow data collected by traditional sensors, without considering the brain responses of traffic participants to traffic conditions. Traffic participants' brains react rapidly to different traffic scenarios, and these reactions are reflected in their driving behavior, thus influencing traffic flow. For example, when a driver sees a traffic light about to turn red at an intersection, the brain quickly assesses and decides whether to slow down and stop. This decision-making process affects the vehicle's speed and stopping position, thereby impacting the overall traffic flow. Reason for this drawback: Because existing technologies lack data on traffic participants' brain responses to traffic, they cannot deeply understand the underlying driving factors behind traffic flow changes. In complex traffic scenarios, such as traffic congestion caused by sudden events or traffic changes under special weather conditions, relying solely on traditional sensor data makes it difficult to accurately predict traffic flow trends, thus affecting the accuracy and reliability of the prediction results.

[0005] 3. Limitations of Data Fusion and Model Building: In terms of data fusion, existing technologies primarily involve simple splicing or preliminary integration of data collected from different types of traditional sensors, without fully considering the deep-seated relationships between the data. Regarding model building, machine learning algorithms such as support vector machines, while capable of handling linear or nonlinear data of a certain scale, have limited ability to deeply mine and correlate multi-source heterogeneous data in complex traffic flow systems. Reasons for these limitations: Traffic flow systems are complex dynamic systems with intricate spatiotemporal correlations and interactions between different types of data. For example, changes in traffic flow on a certain road segment may be related to various factors such as historical traffic data from adjacent road segments and the overall psychological state of drivers in the current time period. Existing data fusion methods and model building approaches cannot fully explore these deep-seated relationships, making it difficult for models to accurately capture the changing patterns of traffic flow when dealing with complex traffic scenarios, thus affecting prediction accuracy.

[0006] In summary, existing traffic flow prediction methods based on traditional sensor data combined with machine learning algorithms suffer from problems such as incomplete data collection, lack of consideration for the intrinsic reactions of traffic participants, and limitations in data fusion and model building. This invention aims to address these shortcomings of existing technologies and provide a more accurate and comprehensive traffic flow prediction method. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a traffic flow prediction method, system and storage medium based on brain-computer interface.

[0008] The objective of this invention is achieved through the following technical solution: The first aspect of this invention provides a traffic flow prediction method based on a brain-computer interface, comprising the following steps: Data collection phase: Brain-computer interface devices are used to collect brain-computer interface data from traffic participants, and basic traffic flow data for the corresponding time period and road segment are collected simultaneously; Data preprocessing stage: The collected brain-computer interface data is preprocessed and brain-computer interface feature data reflecting the brain's response to traffic are extracted; Spatiotemporal alignment stage: The preprocessed brain-computer interface feature data is aligned with traffic flow basic data in spatiotemporal space, and a preliminary correlation is established between the two. Model building phase: Deep learning algorithms are used to build a fusion model, and spatiotemporally aligned multi-source data is input into the fusion model to explore the deep relationships between the data; Model training phase: The fusion model is trained using historical data, and the model parameters are adjusted through optimization algorithms; Traffic flow prediction phase: Real-time traffic flow prediction is performed based on the trained fusion model, and traffic information is dynamically adjusted and released based on the prediction results.

[0009] Preferably, the brain-computer interface data includes electroencephalogram (EEG) signal data, functional near-infrared spectroscopy data, eye movement data, and autonomic nervous system index data.

[0010] Preferably, the data preprocessing stage further includes the following steps: Use adaptive notch filters or independent component analysis to denoise brain-computer interface data; Principal component analysis or linear discriminant analysis methods are used to reduce the dimensionality of brain-computer interface data. Brain-computer interface feature data were obtained by extracting the prefrontal cortex θ / β ratio, event-related potential (P300) latency, and changes in oxyhemoglobin concentration to reflect cognitive load, risk perception, and spatial navigation information.

[0011] Preferably, the initial association is established through a multimodal feature fusion association method or a dynamic event-triggered association method; The multimodal feature fusion and association method includes the following steps: extracting time-frequency domain features and traffic flow dynamic parameters of EEG signals in parallel, fusing multi-source features using an adaptive weighting strategy, and constructing a nonlinear association model after dimensionality reduction and redundancy elimination through principal component analysis to reveal the potential coupling association between brain-computer interaction and traffic behavior. The dynamic event triggering association method includes the following steps: by real-time monitoring of abnormal fluctuations in EEG signals, combined with sudden states captured by traffic flow sensors, a sliding time window is used to perform bimodal event synchronous matching and causal inference; The spatiotemporal alignment stage also includes the following steps: Based on the sampling timestamps of the brain-computer interface device and the acquisition timestamps of the traffic flow sensor, the time axis of the brain-computer interface feature data is aligned with that of the basic traffic flow data using a time synchronization method. Based on the spatial coordinates of the brain-computer interface device and the GPS coordinates of the traffic flow sensor, the brain-computer interface feature data is mapped onto the road grid using a geographic information system for spatial coordinate mapping and alignment.

[0012] Preferably, the fusion model includes a hybrid network of a dual-branch 3D-CNN network and an LSTM-GCN network; The spatiotemporally aligned multi-source data is input into a dual-branch 3D-CNN network. The spatial branch processes the time-frequency domain features of EEG signals, and the temporal branch processes traffic flow dynamic parameters. Deep-level correlation features are then fused through a cross-modal attention mechanism. By utilizing a sliding time window to simultaneously detect abnormal fluctuations in EEG signals and sudden traffic flow events, a bimodal event sequence is input into an LSTM-GCN hybrid network. LSTM captures temporal dependencies, GCN models spatial interactions, and a decision tree algorithm is used to infer the causal relationship between brain-computer responses and traffic behavior.

[0013] Preferably, the model training phase further includes the following steps: Historical data is divided into training, validation and test sets according to the proportions, and input into a dual-branch 3D-CNN network and an LSTM-GCN hybrid network. The model parameters are initially fitted by the backpropagation algorithm, and the initial loss value and prediction error are recorded. An adaptive moment estimation optimizer is used in conjunction with a validation set loss function to dynamically adjust the learning rate. Gradient pruning is used to prevent parameter explosion in the cross-modal attention layer of the dual-branch 3D-CNN network. Regularization terms are used to constrain the node weights of the graph neural network in the LSTM-GCN hybrid network. In the autoencoder-Transformer-GNN hierarchical framework, the hyperparameter combination of PCA dimensionality reduction dimension, Transformer head number and GNN neighborhood radius is searched using Bayesian optimization algorithm. The optimal model is selected after a preset number of iterations, using the mean absolute percentage error of the test set as the indicator.

[0014] Preferably, the traffic flow prediction stage further includes the following steps: Real-time data is collected through brain-computer interface devices and traffic flow sensors, and pre-trained models are used for instant denoising and feature extraction. The pre-processed real-time data is input into the trained fusion model. The fusion model predicts traffic flow in the future based on cross-modal attention mechanism and graph neural network. The prediction results are updated according to the preset frequency and deployed locally through edge computing devices to reduce latency. The prediction results trigger a tiered early warning mechanism, and traffic information is dynamically released through variable message signs or navigation apps. The actual traffic flow is then fed back to the fusion model to form a closed-loop optimization.

[0015] Preferably, the brain-computer interface device is independently set up or installed on a vehicle; the traffic participant is a specific experimental personnel or the driver of the vehicle equipped with the brain-computer interface device.

[0016] A second aspect of the present invention provides: a brain-computer interface-based traffic flow prediction system for implementing any of the above-described brain-computer interface-based traffic flow prediction methods, comprising: The data acquisition module is used to collect brain-computer interface data of traffic participants using brain-computer interface devices, and simultaneously collect basic traffic flow data for the corresponding time period and road segment. The data preprocessing module is used to preprocess the collected brain-computer interface data and extract brain-computer interface feature data that reflects the brain's response to traffic. The spatiotemporal alignment module is used to align the preprocessed brain-computer interface feature data with traffic flow basic data in spatiotemporal space and establish a preliminary correlation between the two. The model building module is used to build a fusion model using deep learning algorithms and input spatiotemporally aligned multi-source data into the fusion model to uncover deep relationships between the data. The model training module is used to train the fusion model using historical data and adjust the model parameters through optimization algorithms. The traffic flow prediction module is used to predict traffic flow in real time based on a trained fusion model, and to dynamically adjust and release traffic information based on the prediction results.

[0017] A third aspect of the present invention provides: a computer-readable storage medium storing computer-executable instructions, wherein when the computer-executable instructions are loaded and executed by a processor, any of the above-described brain-computer interface-based traffic flow prediction methods are implemented.

[0018] The beneficial effects of this invention are: 1) Enhanced Data Comprehensiveness and Richer Predictive Basis: Existing technologies primarily rely on traditional equipment such as sensors and video surveillance cameras to collect macro-level traffic flow data, including traffic volume, speed, and occupancy rates. This data is relatively limited and fails to fully reflect the complex dynamics of traffic flow. In contrast, this invention utilizes brain-computer interface devices to collect data such as brain electrical signals from traffic participants, simultaneously gathering basic traffic flow data for corresponding time periods and road segments. This brain electrical signal data contains real-time responses from traffic participants to traffic conditions, such as attention levels, emotional states, and decision-making intentions. Combining this information with basic traffic flow data significantly enriches the data sources, providing a more comprehensive and detailed basis for traffic flow prediction.

[0019] 2) Precise Feature Extraction and In-Depth Understanding of Internal Correlations: This invention preprocesses the collected brain-computer interface data through noise reduction and dimensionality reduction to extract key features reflecting the brain's response to traffic. This process removes noise and redundant information from the data, accurately capturing brain features closely related to traffic flow changes, such as the correlation between changes in specific brainwave frequencies and driver decision-making behavior. In contrast, existing technologies lack the extraction and analysis of the intrinsic reaction characteristics of traffic participants, making it difficult to deeply understand the underlying driving factors behind traffic flow changes.

[0020] 3) Precise spatiotemporal alignment of data, leading to more effective association establishment: This invention aligns preprocessed brain-computer interface feature data with basic traffic flow data spatiotemporally, establishing a preliminary association between the two. This precise spatiotemporal alignment ensures accurate matching of data from different sources in both time and space, enabling the model to more rationally analyze the relationships between the data. Existing technologies in data fusion often simply splice or preliminarily integrate data collected by different types of sensors without fully considering the spatiotemporal relationships between the data, making it difficult for the model to accurately capture the dynamic changes in traffic flow.

[0021] 4) Deep Mining of Fusion Models Leads to Significantly Improved Prediction Accuracy: This invention employs deep learning algorithms to construct a fusion model, inputting aligned multi-source data to mine deep-seated relationships between the data. Deep learning algorithms possess powerful nonlinear mapping and feature learning capabilities, automatically learning complex patterns and rules from multi-source data, thereby more accurately predicting changes in traffic flow. In contrast, existing machine learning algorithms, such as support vector machines, have limited capabilities for deep mining and correlation analysis of multi-source heterogeneous data in complex traffic flow systems, resulting in lower prediction accuracy.

[0022] 5) Real-time prediction and dynamic adjustment to improve traffic management efficiency: This invention uses a trained model to predict traffic flow in real time, and dynamically adjusts and releases traffic information based on the prediction results. This real-time and dynamic nature enables traffic management departments to promptly grasp changes in traffic flow and take corresponding traffic management measures in advance, such as adjusting traffic light timings and releasing traffic guidance information, thereby effectively alleviating traffic congestion and improving road traffic efficiency. Existing technologies often lack the ability to predict and dynamically adjust in real time, and cannot respond promptly to sudden changes in traffic flow.

[0023] 6) By introducing brain-computer interface technology, this invention has significant advantages in terms of data comprehensiveness, feature extraction, data association, model prediction and traffic management. It can more accurately predict changes in traffic flow and provide more reliable decision support for traffic management departments. It has broad application prospects and important practical value. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] See Figure 1 The first aspect of this invention provides: a traffic flow prediction method based on a brain-computer interface, comprising the following steps: Data collection phase: Brain-computer interface devices are used to collect brain-computer interface data from traffic participants, and basic traffic flow data for the corresponding time period and road segment are collected simultaneously; Data preprocessing stage: The collected brain-computer interface data is preprocessed and brain-computer interface feature data reflecting the brain's response to traffic are extracted; Spatiotemporal alignment stage: The preprocessed brain-computer interface feature data is aligned with traffic flow basic data in spatiotemporal space, and a preliminary correlation is established between the two. Model building phase: Deep learning algorithms are used to build a fusion model, and spatiotemporally aligned multi-source data is input into the fusion model to explore the deep relationships between the data; Model training phase: The fusion model is trained using historical data, and the model parameters are adjusted through optimization algorithms; Traffic flow prediction phase: Real-time traffic flow prediction is performed based on the trained fusion model, and traffic information is dynamically adjusted and released based on the prediction results.

[0027] In some embodiments, the brain-computer interface data includes electroencephalogram (EEG) signal data, functional near-infrared spectroscopy (FIR) data, eye movement data, and autonomic nervous system index data.

[0028] In this embodiment, brain-computer interface devices are used to collect brain-computer interface data from traffic participants, mainly including: electroencephalogram (EEG) data, functional near-infrared spectroscopy (fNIRS) data, eye movement data, and autonomic nervous system index data. At the same time, basic traffic flow data for the corresponding time period and road segment are collected simultaneously, including traffic flow data for road segments, bridges, and other areas within the corresponding time period.

[0029] In some embodiments, the data preprocessing stage further includes the following steps: Use adaptive notch filters or independent component analysis to denoise brain-computer interface data; Principal component analysis or linear discriminant analysis methods are used to reduce the dimensionality of brain-computer interface data. Brain-computer interface feature data were obtained by extracting the prefrontal cortex θ / β ratio, event-related potential (P300) latency, and changes in oxyhemoglobin concentration to reflect cognitive load, risk perception, and spatial navigation information.

[0030] In this embodiment, the collected brain-computer interface data undergoes preprocessing such as denoising and dimensionality reduction to extract key features reflecting the brain's response to traffic. The main methods are as follows: Data denoising methods utilize adaptive notch filters or independent component analysis (ICA) to denoise the brain signals of traffic participants. Data dimensionality reduction methods utilize principal component analysis (PCA) or linear discriminant analysis (LDA) to reduce the dimensionality of the participants' brain data. The method for extracting key features of the brain's response to traffic involves collecting brain data in traffic scenarios through a multimodal brain-computer interface. After denoising and dimensionality reduction, key features reflecting cognitive load, risk perception, and spatial navigation, such as the prefrontal cortex θ / β ratio, event-related potential (P300) latency, and changes in oxyhemoglobin concentration, are further extracted.

[0031] In some embodiments, the initial association is established through a multimodal feature fusion association method or a dynamic event-triggered association method; The multimodal feature fusion and association method includes the following steps: extracting time-frequency domain features and traffic flow dynamic parameters of EEG signals in parallel, fusing multi-source features using an adaptive weighting strategy, and constructing a nonlinear association model after dimensionality reduction and redundancy elimination through principal component analysis to reveal the potential coupling association between brain-computer interaction and traffic behavior. The dynamic event triggering association method includes the following steps: by real-time monitoring of abnormal fluctuations in EEG signals, combined with sudden states captured by traffic flow sensors, a sliding time window is used to perform bimodal event synchronous matching and causal inference; The spatiotemporal alignment stage also includes the following steps: Based on the sampling timestamps of the brain-computer interface device and the acquisition timestamps of the traffic flow sensor, the time axis of the brain-computer interface feature data is aligned with that of the basic traffic flow data using a time synchronization method. Based on the spatial coordinates of the brain-computer interface device and the GPS coordinates of the traffic flow sensor, the brain-computer interface feature data is mapped onto the road grid using a geographic information system for spatial coordinate mapping and alignment.

[0032] In this embodiment, the preprocessed brain-computer interface feature data is spatiotemporally aligned with traffic flow basic data to establish a preliminary association between the two. The data alignment method is as follows: timestamp-based synchronization alignment: the sampling timestamp recorded by the brain-computer interface device is matched with the acquisition timestamp of traffic flow sensors (such as GPS, cameras), and high-precision clock synchronization technology (such as NTP protocol) is used to align the EEG signal with the time axis of traffic flow data.

[0033] Spatial coordinate mapping alignment: By combining the spatial coordinates of the driver's EEG acquisition device (such as the position of the helmet electrode) with the GPS coordinates of traffic flow sensors (such as vehicle-mounted cameras and roadside units), the EEG data is mapped to the road grid using a Geographic Information System (GIS) for spatial coordinate mapping alignment.

[0034] The preliminary association establishment method is as follows: Multimodal feature fusion association method: By extracting the time-frequency domain features of EEG signals (such as power spectral density and entropy value) and traffic flow dynamic parameters (spatiotemporal distribution of flow, density and speed) in parallel, an adaptive weighting strategy is adopted to fuse multi-source features. After dimensionality reduction and redundancy elimination by principal component analysis, a nonlinear association model is constructed to reveal the potential coupling association between brain-computer interaction and traffic behavior.

[0035] Dynamic event triggering association method: By real-time monitoring of abnormal fluctuations in EEG signals (such as sudden changes in blink frequency, alpha wave inhibition representing distraction), combined with sudden states captured by traffic flow sensors (such as speed drops caused by sudden braking of vehicles, congestion caused by multiple vehicles gathering), a sliding time window is used to perform bimodal event synchronous matching and causal inference.

[0036] In some embodiments, the fusion model includes a hybrid network of a dual-branch 3D-CNN network and an LSTM-GCN network; The spatiotemporally aligned multi-source data is input into a dual-branch 3D-CNN network. The spatial branch processes the time-frequency domain features of EEG signals, and the temporal branch processes traffic flow dynamic parameters. Deep-level correlation features are then fused through a cross-modal attention mechanism. By utilizing a sliding time window to simultaneously detect abnormal fluctuations in EEG signals and sudden traffic flow events, a bimodal event sequence is input into an LSTM-GCN hybrid network. LSTM captures temporal dependencies, GCN models spatial interactions, and a decision tree algorithm is used to infer the causal relationship between brain-computer responses and traffic behavior.

[0037] In this embodiment, a deep learning algorithm is used to construct a fusion model, inputting aligned multi-source data to mine deep relationships between data. The specific method is as follows: Based on spatiotemporal alignment, multimodal deep feature fusion is performed. Brain-computer interface features (prefrontal cortex θ / β ratio, P300 latency) and traffic flow data (flow rate, speed) are aligned using timestamps (NTP protocol) and GIS spatial coordinate mapping, and then input into a dual-branch 3D-CNN network. The spatial branch processes EEG time-frequency features, and the temporal branch processes traffic flow dynamic parameters. Deep-level correlation features are then fused through a cross-modal attention mechanism.

[0038] Construction of a dynamic event-driven deep causal reasoning model: By using a sliding time window to simultaneously monitor abnormal EEG fluctuations (alpha wave suppression, blink frequency) and sudden traffic flow states (sudden braking, congestion), the dual-modal event sequence is input into an LSTM-GCN hybrid network. LSTM captures temporal dependencies, GCN models spatial interactions, and the causal relationship between brain-computer responses and traffic behavior is inferred through a decision tree algorithm.

[0039] In some embodiments, the model training phase further includes the following steps: Historical data is divided into training, validation and test sets according to the proportions, and input into a dual-branch 3D-CNN network and an LSTM-GCN hybrid network. The model parameters are initially fitted by the backpropagation algorithm, and the initial loss value and prediction error are recorded. An adaptive moment estimation optimizer is used in conjunction with a validation set loss function to dynamically adjust the learning rate. Gradient pruning is used to prevent parameter explosion in the cross-modal attention layer of the dual-branch 3D-CNN network. Regularization terms are used to constrain the node weights of the graph neural network in the LSTM-GCN hybrid network. In the autoencoder-Transformer-GNN hierarchical framework, the hyperparameter combination of PCA dimensionality reduction dimension, Transformer head number and GNN neighborhood radius is searched using Bayesian optimization algorithm. The optimal model is selected after a preset number of iterations, using the mean absolute percentage error of the test set as the indicator.

[0040] In this embodiment, a fusion model is trained using historical data, and parameters are adjusted through algorithm optimization to improve the model's prediction accuracy for traffic flow. The specific steps are as follows: Initial training of the fusion model based on historical multi-source data: Aligned brain-computer interface features (such as prefrontal cortex θ / β ratio, P300 latency) and traffic flow data (flow rate, speed) within historical time periods are collected and divided into training, validation, and test sets in a 7:2:1 ratio. The data is input into a dual-branch 3D-CNN or LSTM-GCN hybrid network, and the model parameters are initially fitted using the backpropagation algorithm. The initial loss value and prediction error are recorded.

[0041] Optimization algorithm-driven dynamic parameter adjustment: An adaptive moment estimation (Adam) optimizer is employed, combined with a validation set loss function (mean squared error, MSE) to dynamically adjust the learning rate. For 3D-CNN networks, gradient pruning is used to prevent parameter explosion in cross-modal attention layers; for LSTM-GCN models, regularization terms (L2 norm) are used to constrain the node weights of the graphical neural network.

[0042] Accuracy Iterative Optimization of the Multi-Scale Feature Fusion Framework: In the autoencoder-Transformer-GNN hierarchical framework, a Bayesian optimization algorithm is used to search for hyperparameter combinations of PCA dimensionality reduction dimension, Transformer head number, and GNN neighborhood radius. Using the test set MAPE (Mean Absolute Percentage Error) as an indicator, the optimal model is selected after a specified number of iterations, ultimately ensuring that the traffic flow prediction accuracy meets a preset threshold.

[0043] In some embodiments, the traffic flow prediction phase further includes the following steps: Real-time data is collected through brain-computer interface devices and traffic flow sensors, and pre-trained models are used for instant denoising and feature extraction. The pre-processed real-time data is input into the trained fusion model. The fusion model predicts traffic flow in the future based on cross-modal attention mechanism and graph neural network. The prediction results are updated according to the preset frequency and deployed locally through edge computing devices to reduce latency. The prediction results trigger a tiered early warning mechanism, and traffic information is dynamically released through variable message signs or navigation apps. The actual traffic flow is then fed back to the fusion model to form a closed-loop optimization.

[0044] In this embodiment, real-time traffic flow prediction is performed based on a trained model, and traffic information is dynamically adjusted and released according to the prediction results. The specific steps are as follows: Real-time data acquisition and preprocessing are accessed simultaneously: Multi-source data is collected in real time through brain-computer interface devices and traffic flow sensors (such as vehicle-mounted GPS and roadside cameras). Pre-trained Notch Filter and ICA model are used to denoise the EEG signals in real time. PCA dimensionality reduction is used simultaneously to extract key features (such as the θ / β ratio). After spatiotemporal alignment with the NTP protocol and GIS system, the data is input into the trained fusion model.

[0045] Dynamic traffic flow prediction and real-time result generation: The preprocessed real-time data is input into the optimized 3D-CNN / LSTM-GCN hybrid model. The model is based on cross-modal attention mechanism and graph neural network to quickly calculate traffic flow parameters (flow rate, speed) for the next 5-10 minutes. The prediction results are updated every 30 seconds, and the latency is reduced by local deployment of edge computing devices.

[0046] Multi-channel dynamic release and feedback closed loop of traffic information: Automatically trigger a graded early warning mechanism based on prediction results: Speed ​​suggestions are released through variable message signs when there is mild congestion; When there is severe congestion, navigation apps are linked to plan detour routes, and the actual traffic flow is fed back to the model to form a closed loop optimization.

[0047] In some embodiments, the brain-computer interface device is independently installed or mounted on a vehicle; the traffic participant is a specific experimental subject or the driver of the vehicle equipped with the brain-computer interface device.

[0048] In this embodiment, the solution is currently in the experimental stage. The brain-computer interface device is installed on the experimenter for testing. When it is promoted to the market in the future, it can cooperate with various car companies to install in-vehicle brain-computer interface devices in vehicles in advance, thereby completing the collection of driver's brain data.

[0049] A second aspect of the present invention provides: a brain-computer interface-based traffic flow prediction system for implementing any of the above-described brain-computer interface-based traffic flow prediction methods, comprising: The data acquisition module is used to collect brain-computer interface data of traffic participants using brain-computer interface devices, and simultaneously collect basic traffic flow data for the corresponding time period and road segment. The data preprocessing module is used to preprocess the collected brain-computer interface data and extract brain-computer interface feature data that reflects the brain's response to traffic. The spatiotemporal alignment module is used to align the preprocessed brain-computer interface feature data with traffic flow basic data in spatiotemporal space and establish a preliminary correlation between the two. The model building module is used to build a fusion model using deep learning algorithms and input spatiotemporally aligned multi-source data into the fusion model to uncover deep relationships between the data. The model training module is used to train the fusion model using historical data and adjust the model parameters through optimization algorithms. The traffic flow prediction module is used to predict traffic flow in real time based on a trained fusion model, and to dynamically adjust and release traffic information based on the prediction results.

[0050] A third aspect of the present invention provides: a computer-readable storage medium storing computer-executable instructions, wherein when the computer-executable instructions are loaded and executed by a processor, any of the above-described brain-computer interface-based traffic flow prediction methods are implemented.

[0051] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A traffic flow prediction method based on brain-computer interface, characterized in that: Includes the following steps: Data collection phase: Brain-computer interface devices are used to collect brain-computer interface data from traffic participants, and basic traffic flow data for the corresponding time period and road segment are collected simultaneously; Data preprocessing stage: The collected brain-computer interface data is preprocessed and brain-computer interface feature data reflecting the brain's response to traffic are extracted; Spatiotemporal alignment stage: The preprocessed brain-computer interface feature data is aligned with traffic flow basic data in spatiotemporal space, and a preliminary correlation is established between the two. Model building phase: Deep learning algorithms are used to build a fusion model, and spatiotemporally aligned multi-source data is input into the fusion model to explore the deep relationships between the data; Model training phase: The fusion model is trained using historical data, and the model parameters are adjusted through optimization algorithms; Traffic flow prediction phase: Real-time traffic flow prediction is performed based on the trained fusion model, and traffic information is dynamically adjusted and released based on the prediction results.

2. The traffic flow prediction method based on brain-computer interface according to claim 1, characterized in that: The brain-computer interface data includes electroencephalogram (EEG) signal data, functional near-infrared spectroscopy (FIR) data, eye movement data, and autonomic nervous system index data.

3. The traffic flow prediction method based on brain-computer interface according to claim 1, characterized in that: The data preprocessing stage also includes the following steps: Use adaptive notch filters or independent component analysis to denoise brain-computer interface data; Principal component analysis or linear discriminant analysis methods are used to reduce the dimensionality of brain-computer interface data. Brain-computer interface feature data were obtained by extracting the prefrontal cortex θ / β ratio, event-related potential (P300) latency, and changes in oxyhemoglobin concentration to reflect cognitive load, risk perception, and spatial navigation information.

4. The traffic flow prediction method based on brain-computer interface according to claim 1, characterized in that: The initial association is established through a multimodal feature fusion association method or a dynamic event-triggered association method; The multimodal feature fusion and association method includes the following steps: extracting time-frequency domain features and traffic flow dynamic parameters of EEG signals in parallel, fusing multi-source features using an adaptive weighting strategy, and constructing a nonlinear association model after dimensionality reduction and redundancy elimination through principal component analysis to reveal the potential coupling association between brain-computer interaction and traffic behavior. The dynamic event triggering association method includes the following steps: by real-time monitoring of abnormal fluctuations in EEG signals, combined with sudden states captured by traffic flow sensors, a sliding time window is used to perform bimodal event synchronous matching and causal inference; The spatiotemporal alignment stage also includes the following steps: Based on the sampling timestamps of the brain-computer interface device and the acquisition timestamps of the traffic flow sensor, the time axis of the brain-computer interface feature data is aligned with that of the basic traffic flow data using a time synchronization method. Based on the spatial coordinates of the brain-computer interface device and the GPS coordinates of the traffic flow sensor, the brain-computer interface feature data is mapped onto the road grid using a geographic information system for spatial coordinate mapping and alignment.

5. The traffic flow prediction method based on brain-computer interface according to claim 1, characterized in that: The fusion model includes a hybrid network of a dual-branch 3D-CNN network and an LSTM-GCN network; The spatiotemporally aligned multi-source data is input into a dual-branch 3D-CNN network. The spatial branch processes the time-frequency domain features of EEG signals, and the temporal branch processes traffic flow dynamic parameters. Deep-level correlation features are then fused through a cross-modal attention mechanism. By utilizing a sliding time window to simultaneously detect abnormal fluctuations in EEG signals and sudden traffic flow events, a bimodal event sequence is input into an LSTM-GCN hybrid network. LSTM captures temporal dependencies, GCN models spatial interactions, and a decision tree algorithm is used to infer the causal relationship between brain-computer responses and traffic behavior.

6. The traffic flow prediction method based on brain-computer interface according to claim 5, characterized in that: The model training phase also includes the following steps: Historical data is divided into training, validation and test sets according to the proportions, and input into a dual-branch 3D-CNN network and an LSTM-GCN hybrid network. The model parameters are initially fitted by the backpropagation algorithm, and the initial loss value and prediction error are recorded. An adaptive moment estimation optimizer is used in conjunction with a validation set loss function to dynamically adjust the learning rate. Gradient pruning is used to prevent parameter explosion in the cross-modal attention layer of the dual-branch 3D-CNN network. Regularization terms are used to constrain the node weights of the graph neural network in the LSTM-GCN hybrid network. In the autoencoder-Transformer-GNN hierarchical framework, the hyperparameter combination of PCA dimensionality reduction dimension, Transformer head number and GNN neighborhood radius is searched using Bayesian optimization algorithm. The optimal model is selected after a preset number of iterations, using the mean absolute percentage error of the test set as the indicator.

7. The traffic flow prediction method based on brain-computer interface according to claim 1, characterized in that: The traffic flow prediction phase also includes the following steps: Real-time data is collected through brain-computer interface devices and traffic flow sensors, and pre-trained models are used for instant denoising and feature extraction. The pre-processed real-time data is input into the trained fusion model. The fusion model predicts traffic flow in the future based on cross-modal attention mechanism and graph neural network. The prediction results are updated according to the preset frequency and deployed locally through edge computing devices to reduce latency. The prediction results trigger a tiered early warning mechanism, and traffic information is dynamically released through variable message signs or navigation apps. The actual traffic flow is then fed back to the fusion model to form a closed-loop optimization.

8. The traffic flow prediction method based on brain-computer interface according to any one of claims 1-7, characterized in that: The brain-computer interface device is independently set up or installed on a vehicle; the traffic participant is a specific experimental personnel or the driver of a vehicle equipped with a brain-computer interface device.

9. A traffic flow prediction system based on a brain-computer interface, characterized in that: A method for implementing the brain-computer interface-based traffic flow prediction method as described in any one of claims 1-8 includes: The data acquisition module is used to collect brain-computer interface data of traffic participants using brain-computer interface devices, and simultaneously collect basic traffic flow data for the corresponding time period and road segment. The data preprocessing module is used to preprocess the collected brain-computer interface data and extract brain-computer interface feature data that reflects the brain's response to traffic. The spatiotemporal alignment module is used to align the preprocessed brain-computer interface feature data with traffic flow basic data in spatiotemporal space and establish a preliminary correlation between the two. The model building module is used to build a fusion model using deep learning algorithms and input spatiotemporally aligned multi-source data into the fusion model to uncover deep relationships between the data. The model training module is used to train the fusion model using historical data and adjust the model parameters through optimization algorithms. The traffic flow prediction module is used to predict traffic flow in real time based on a trained fusion model, and to dynamically adjust and release traffic information based on the prediction results.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the brain-computer interface-based traffic flow prediction method as described in any one of claims 1-8.

Images