A Multimodal Traffic Data Hierarchical Storage and Management Method and System Integrating Vehicle, Road, and Cloud

Abstract

This invention provides a hierarchical storage and management method and system for multimodal traffic data integrating vehicle, road, and cloud technologies, belonging to the field of data management technology. The method includes: collecting multimodal traffic data in a vehicle-road cooperative scenario; performing unified data processing; explicitly reasoning about the multi-level influence relationships between traffic events, traffic flow evolution, and traffic signal control commands based on a unified data representation model; generating a traffic causal influence diagram; and classifying and storing the multimodal traffic data according to the causal influence indicators of each node in the traffic causal influence diagram, data real-time requirements, and data access frequency. This solves the technical problem of existing technologies lacking a hierarchical scheduling mechanism in the unified storage and management of multimodal traffic data, leading to low data utilization efficiency. It achieves the technical effect of hierarchical storage of multimodal traffic data, realizing effective management and efficient utilization of multimodal traffic data.

Description

Technical Field

[0001] This invention relates to the field of data management technology, specifically to a multimodal traffic data hierarchical storage management method and system integrating vehicle, road, and cloud. Background Technology

[0002] With the rapid development of intelligent connected vehicles and autonomous driving technologies, the vehicle-road-cloud integrated system is gradually becoming an important infrastructure for intelligent transportation. This system achieves real-time perception, analysis, and control of traffic conditions through the collaborative operation of vehicles, roadside equipment, and cloud platforms. Specifically, the intelligent connected vehicle monitoring system, through multi-source data fusion and edge node management, enables unified supervision of intelligent connected devices and autonomous vehicles in vehicle-road cooperative scenarios; the unified traffic control management system dynamically optimizes traffic light timing based on traffic flow prediction and adaptive algorithms, thereby improving road traffic efficiency; the big data platform, as the data hub, undertakes functions such as data storage, analysis, and visualization, and mines the value of traffic data through machine learning technology; and the equipment management system manages roadside equipment and related hardware throughout their entire lifecycle, ensuring stable system operation. Through the collaboration between roadside sensing devices and cloud computing, real-time analysis of traffic conditions can be achieved, signal control strategies can be dynamically optimized, vehicle waiting time can be reduced, and traffic congestion can be alleviated. Simultaneously, vehicle-road cooperation can compensate for blind spots in single-vehicle perception, enabling safety functions such as collision warning and pedestrian detection.

[0003] However, during the integrated operation of vehicles, roads, and cloud, intelligent connected devices, autonomous vehicles, and roadside sensing facilities continuously generate a large amount of multimodal traffic data, such as video stream data, LiDAR point cloud data, millimeter-wave radar data, and various vehicle operating status and sensor data. This data is not only massive in scale but also characterized by high real-time requirements, diverse types, and significantly different access needs. Existing data management methods typically employ unified storage or simple distributed storage strategies, making it difficult to rationally schedule data based on its importance, real-time nature, and usage frequency. This results in low storage resource utilization efficiency and makes it difficult to simultaneously meet the needs of real-time traffic control decision-making and long-term data analysis.

[0004] In summary, existing technologies suffer from the technical problem of low efficiency in the unified storage and management of multimodal traffic data due to the lack of a hierarchical scheduling mechanism. Summary of the Invention

[0005] The purpose of this application is to provide a hierarchical storage and management method and system for multimodal traffic data that integrates vehicle, road and cloud, in order to solve the technical problem that the existing technology lacks a hierarchical scheduling mechanism for unified storage and management of multimodal traffic data, resulting in low efficiency in the utilization of stored data.

[0006] In view of the above problems, this application provides a method and system for hierarchical storage and management of multimodal traffic data that integrates vehicle, road and cloud.

[0007] The first aspect of this application provides a hierarchical storage management method for multimodal traffic data integrating vehicle, road, and cloud. This method includes: collecting multimodal traffic data in a vehicle-road cooperative scenario, wherein the multimodal traffic data includes operational status data of intelligent connected devices and autonomous vehicles, environmental perception data, and traffic control command data; traversing the multimodal traffic data to perform unified data processing and construct a unified data representation model; based on the unified data representation model, explicitly reasoning about the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control commands to generate a traffic causal influence diagram that integrates event chains, traffic flow chains, and control chains; and dividing the multimodal traffic data into storage layers according to the causal influence indicators, data real-time requirements, and data access frequency of each node in the traffic causal influence diagram, allocating high-real-time data supporting unified traffic control optimization to the high-speed storage layer of edge computing nodes, and allocating medium- and low-frequency data supporting big data platform analysis to the distributed storage layer of the central cloud platform.

[0008] Optionally, the operational status data of intelligent connected devices and autonomous vehicles includes real-time trajectory, vehicle speed, acceleration, steering wheel angle, battery status, vehicle diagnostic information, and vehicle identification; environmental perception data includes camera video streams, LiDAR point clouds, and millimeter-wave radar target lists; and traffic control command data includes real-time traffic light phase, signal timing scheme, and variable lane control commands.

[0009] Optionally, the multimodal traffic data is traversed to extract independent factors, obtaining spatiotemporal factors, environmental factors, event factors, and semantic factors; the spatiotemporal factors, environmental factors, event factors, and semantic factors are then traversed to perform a unified dimensional transformation of vectors, constructing the unified data representation model.

[0010] Optionally, traffic events, vehicle trajectory changes, and traffic light phase adjustments are treated as three types of nodes to construct a multi-type node influence graph model framework. Causal reasoning is performed on the unified data representation model using a large language model to identify the impact of events on traffic flow, the impact of traffic flow on traffic control strategies, and the reaction of traffic control instructions to subsequent events, forming a closed-loop causal chain set. Combining the influence graph model framework and the closed-loop causal chain set, a heuristic search algorithm with temporal constraints and spatial proximity constraints is used to generate a traffic causal influence graph containing multi-level propagation paths.

[0011] Optionally, traffic event nodes are used to represent events such as traffic accidents and road congestion. By combining the timestamps and spatial locations of traffic events, event location and correlation are performed through spatiotemporal data. Vehicle trajectory change nodes are used to describe changes in vehicle driving status. Based on the real-time location, speed, and direction of vehicles, analysis is performed in conjunction with road information. Traffic light phase adjustment nodes are used to represent changes in traffic lights. Based on the real-time timing data of traffic lights, analysis is performed in conjunction with traffic flow and signal cycle.

[0012] Optionally, a set of event causal chains based on the impact of events on traffic flow patterns is formed by using a large language model to infer changes in traffic flow speed, volume, and density. The large language model is also used to identify the impact of traffic flow on traffic control strategies, and by inferring the impact of traffic flow changes on traffic light timing and signal cycle adjustments, a set of feedback causal chains of traffic flow on traffic control strategies is formed. Furthermore, the large language model is used to identify the potential counter-effects of traffic control instructions on traffic events, forming a set of traffic control causal chains of traffic control instructions on events. Finally, the event causal chain set, the feedback causal chain set, and the traffic control causal chain set are fused in a closed loop to obtain a closed-loop causal chain set.

[0013] Optionally, temporal constraints are used to ensure that the causal effects of traffic events, traffic flows, and traffic control instructions propagate in the actual time sequence by sorting the timestamps of traffic events; spatial proximity constraints are used to calculate the spatial distance between traffic events and traffic flows based on geographic information system data to ensure the rationality of the impact links in physical space.

[0014] The second aspect of this application provides a vehicle-road-cloud integrated multimodal traffic data hierarchical storage management system. This system includes: multimodal traffic data comprising operational status data of intelligent connected devices and autonomous vehicles, environmental perception data, and traffic control command data; a data processing module for traversing the multimodal traffic data to perform unified data processing and construct a unified data representation model; a relationship reasoning module for explicitly reasoning about the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control commands based on the unified data representation model, generating a traffic causal influence diagram that integrates event chains, traffic flow chains, and control chains; and a data partitioning module for partitioning the multimodal traffic data into storage levels according to the causal influence indicators, data real-time requirements, and data access frequency of each node in the traffic causal influence diagram, allocating high-real-time data supporting unified traffic control optimization to the high-speed storage level of edge computing nodes, and allocating medium- and low-frequency data supporting big data platform analysis to the distributed storage level of the central cloud platform.

[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0016] The method provided in this application collects multimodal traffic data in a vehicle-road cooperative scenario. This multimodal traffic data includes operational status data of intelligent connected devices and autonomous vehicles, environmental perception data, and traffic control command data. It then traverses the multimodal traffic data to perform unified data processing and construct a unified data representation model. Based on this unified data representation model, it explicitly infers the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control commands, generating a traffic causal influence diagram that integrates event chains, vehicle flow chains, and control chains. According to the causal influence indicators, real-time data requirements, and data access frequency of each node in the traffic causal influence diagram, it divides the multimodal traffic data into storage layers. High-real-time data supporting unified traffic control optimization is allocated to the high-speed storage layer of edge computing nodes, while low-to-medium frequency data supporting big data platform analysis is allocated to the distributed storage layer of the central cloud platform. This achieves hierarchical storage of multimodal traffic data, enabling effective management and efficient utilization of multimodal traffic data.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud provided in this application.

[0020] Figure 2 A schematic diagram of the structure of the vehicle-road-cloud integrated multimodal traffic data hierarchical storage and management system provided in this application.

[0021] Figure labeling: Data acquisition module 11, data processing module 12, relation reasoning module 13, data partitioning module 14. Detailed Implementation

[0022] This application provides a hierarchical storage and management method and system for multimodal traffic data that integrates vehicle, road, and cloud technologies. It addresses the technical problem of low data utilization efficiency caused by the lack of a hierarchical scheduling mechanism in existing unified storage and management methods for multimodal traffic data. The method achieves hierarchical storage of multimodal traffic data, enabling effective management and efficient utilization of the data.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 As shown, this application provides a vehicle-road-cloud integrated multimodal traffic data hierarchical storage management method, which includes:

[0025] Collect multimodal traffic data in vehicle-road cooperative scenarios, wherein the multimodal traffic data includes operating status data of intelligent connected devices and autonomous vehicles, environmental perception data, and signal control command data.

[0026] Furthermore, the operational status data of intelligent connected devices and autonomous vehicles includes real-time trajectory, vehicle speed, acceleration, steering wheel angle, battery status, vehicle diagnostic information, and vehicle identification; environmental perception data includes camera video streams, LiDAR point clouds, and millimeter-wave radar target lists; and traffic control command data includes real-time traffic light phase, signal timing schemes, and variable lane control commands.

[0027] Specifically, intelligent connected devices are devices with network communication capabilities that enable vehicles to interact with the outside world, such as on-board sensors and communication modules. Autonomous vehicles refer to vehicles that can drive autonomously without human intervention by utilizing a variety of advanced systems and sensors on their own.

[0028] In vehicle-road cooperative scenarios, multimodal traffic data is collected, including operational status data, environmental perception data, and traffic control command data from intelligent connected devices and autonomous vehicles. Specifically, operational status data for intelligent connected devices and autonomous vehicles can be collected through vehicle-mounted speed sensors, acceleration sensors, steering wheel angle sensors, etc., to capture real-time trajectory, vehicle speed, acceleration, steering wheel angle, and other data. Battery management systems can be used to obtain battery status information, vehicle electronic control units can collect vehicle diagnostic information, and vehicle identification can be obtained through license plate recognition systems or unique on-board identification codes.

[0029] In terms of environmental perception data, video streams are captured by cameras installed on vehicles or roadsides. LiDAR emits laser beams and receives reflected signals to form point cloud data. Millimeter-wave radar detects surrounding targets and generates a target list. The video streams captured by cameras are used to detect traffic signs, pedestrians, obstacles, etc., while LiDAR and millimeter-wave radar provide data on the distance, speed, and shape of surrounding objects. Environmental sensors, such as temperature and humidity sensors and visibility sensors, collect data on temperature, humidity, and visibility under current environmental conditions. Traffic control command data can be obtained through vehicle-to-everything (V2X) communication technologies, such as dedicated short-range communication or cellular vehicle-to-everything (V2X) networks, enabling data exchange between roadside traffic signal control equipment and vehicles. This allows for the acquisition of information such as real-time traffic light phase, signal timing schemes, and variable lane control commands.

[0030] By collecting multimodal traffic data, we can understand the real-time traffic conditions, traffic flow, and traffic light configuration, thus forming a complete traffic situational awareness.

[0031] Perform unified data processing on multimodal traffic data and construct a unified data representation model.

[0032] Furthermore, the multimodal traffic data is traversed to perform unified data processing and construct a unified data representation model, including: traversing the multimodal traffic data to extract independent factors, obtaining spatiotemporal factors, environmental factors, event factors, and semantic factors; traversing the spatiotemporal factors, environmental factors, event factors, and semantic factors to perform unified dimensional transformation of vectors, and constructing the unified data representation model.

[0033] Specifically, the system iterates through multimodal data, including intelligent connected device and vehicle operation status data, environmental perception data, and traffic control command data, extracting different factors from each data point to obtain spatiotemporal factors, environmental factors, event factors, and semantic factors. For vehicle operation status data, information such as timestamps, location coordinates, speed, acceleration, and steering wheel angle are parsed to generate spatiotemporal and semantic factors. The temporal factors include traffic event timestamps and peak or off-peak traffic periods, while the spatial factors include real-time vehicle location, lane number, traffic light location, and road type, used to describe the temporal and spatial distribution characteristics of events and vehicles. For environmental perception data, temperature and humidity, road conditions, and surrounding obstacles and traffic signs detected by cameras and radar are read to form environmental factors. These environmental factors include environmental conditions and road condition information. Environmental conditions include temperature, humidity, and visibility, while road condition information includes slippery road surfaces, construction areas, or traffic control status, used to reflect the impact of the external environment on traffic flow and driving behavior, providing external constraints. For traffic control command data, the system analyzes real-time traffic light phases, timing schemes, and variable lane instructions. It also combines this data with traffic accident or congestion information to generate event factors. These event factors include traffic accident events and traffic flow anomaly events. Traffic accident events include the location and type of the accident, while traffic flow anomaly events include road congestion and the area affected by the accident, reflecting the state and scope of emergencies in the traffic system. Simultaneously, the system provides semantic descriptions of traffic behavior and control behavior, obtaining semantic factors. These semantic factors include driving behavior (e.g., sudden braking, rapid acceleration, lane changing) and traffic control behavior (e.g., traffic light changes, lane control instructions).

[0034] The extracted spatiotemporal factors, environmental factors, event factors, and semantic factors are iterated one by one. Each type of factor is vectorized and its dimensions are unified: For time factors in the spatiotemporal factors, such as traffic event timestamps and peak or off-peak traffic periods, they are first converted into numerical vectors relative to a unified start time, and then sine and cosine encoding is performed to capture periodic patterns. The latitude and longitude information such as vehicle positions and traffic light positions in the spatial factors are converted into planar coordinates through Mercator projection, and then combined with lane numbers or intersection numbers through embedding vector encoding. All numerical coordinates are then subjected to min-max normalization and combined into a unified multidimensional spatial vector.

[0035] For environmental factors, continuous data such as temperature, humidity, and visibility are normalized to the [0,1] interval. Discrete road condition information, such as slippery road surfaces, construction, and traffic control, is first quantized and classified, and then a fixed-dimensional vector is generated through one-hot encoding or vector embedding. For event factors, the location of traffic accident events is processed using co-spatial factors, and category data such as accident type and traffic flow anomaly type are converted into vector representations through one-hot encoding or vector embedding, with numerical features such as the size or range of the affected area being normalized. For semantic factors, driving behavior and traffic control behavior are generated into fixed-dimensional vectors through one-hot encoding or time-series embedding to preserve the sequence information of the behavior. The multiple transformed factors are integrated into a high-dimensional unified vector using a splicing method to form a unified data representation model, thereby mapping multi-source and multi-modal traffic data to the same vector space, realizing a unified basis for comprehensive analysis and causal reasoning.

[0036] For example, the following data was collected at an intersection: A vehicle was in the north lane at 8:15 AM, with a speed of 30 km / h, an acceleration of 0.5 m / s², and a slight left turn of the steering wheel; the environmental conditions were: temperature 22℃, humidity 60%, good visibility, and dry road surface; a minor collision occurred on the south side of the intersection, affecting traffic in the north lane; the green light had 20 seconds remaining. After processing, the time factor was mapped to a two-dimensional vector with relative seconds plus sine and cosine encoding. After spatial coordinate normalization, it was concatenated with the lane embedding vector. The environmental factor was normalized to a vector in the [0,1] interval. The event type was generated as a vector using one-hot encoding. Driving behavior and signal control behavior were generated as vectors through one-hot or time series embedding. Finally, all vectors were uniformly mapped and concatenated to form the representation of this data in a unified data representation model, which can be directly used for traffic causal reasoning and multimodal analysis.

[0037] By traversing and extracting independent factors from the data, such as spatiotemporal factors, environmental factors, event factors, and semantic factors, and performing unified dimensional transformation, a unified data representation model is constructed. This solves the problem of large differences in the formats and structures of different types of data, making it difficult to process them uniformly. It enables the effective integration of data from different sources and improves data processing efficiency.

[0038] Based on the unified data representation model, explicit reasoning is performed on the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control instructions to generate a traffic causal influence diagram that integrates event chains, vehicle flow chains, and control chains.

[0039] Furthermore, based on the unified data representation model, explicit reasoning is performed on the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control instructions to generate a traffic causal influence graph that integrates event chains, traffic flow chains, and control chains. This includes: constructing an influence graph model framework with traffic events, vehicle trajectory changes, and traffic light phase adjustments as three types of nodes; performing causal reasoning on the unified data representation model using a large language model to identify the impact of events on traffic flow, the impact of traffic flow on traffic control strategies, and the counter-effects of traffic control instructions on subsequent events, forming a closed-loop causal chain set; and combining the influence graph model framework and the closed-loop causal chain set, using a heuristic search algorithm with temporal constraints and spatial proximity constraints to generate a traffic causal influence graph containing multi-level propagation paths.

[0040] Furthermore, traffic event nodes are used to represent events such as traffic accidents and road congestion. By combining the timestamps and spatial locations of traffic events, event location and correlation are performed through spatiotemporal data. Vehicle trajectory change nodes are used to describe changes in vehicle driving status. Based on the real-time location, speed, and direction of vehicles, and combined with road information, analysis is performed. Traffic light phase adjustment nodes are used to represent changes in traffic lights. Based on the real-time timing data of traffic lights, and combined with traffic flow and signal cycle, analysis is performed.

[0041] Furthermore, temporal constraints are used to sort traffic events by timestamps to ensure that the causal effects of traffic events, traffic flow, and traffic control instructions propagate in the actual time sequence of occurrence; spatial proximity constraints are used to calculate the spatial distance between traffic events and traffic flow based on geographic information system data to ensure the rationality of the impact links in physical space.

[0042] Specifically, traffic events, vehicle trajectory changes, and traffic light phase adjustments are categorized into three types of nodes, resulting in traffic event nodes, vehicle trajectory change nodes, and traffic light phase adjustment nodes. Traffic event nodes represent events such as traffic accidents and road congestion. By parsing the event's timestamp and spatial location, they are mapped to specific roads or intersections on the road network, and linked with adjacent events to form a complete event chain. Vehicle trajectory change nodes describe changes in vehicle driving status. Based on the vehicle's real-time location, speed, and direction, combined with road topology, lane information, and traffic flow analysis, traffic flow dynamics are captured. Traffic light phase adjustment nodes represent changes in traffic lights. Real-time acquisition of traffic light phase, timing schemes, and signal cycles is combined with analysis of their impact based on the current traffic flow status.

[0043] Then, based on spatiotemporal data and a unified data representation model, the impact of traffic events on the trajectories of surrounding vehicles is analyzed. Directed edges are established between traffic event nodes and affected vehicle trajectory change nodes. Based on the changes in traffic flow and speed caused by vehicle trajectory changes, the potential triggering effect on traffic light phase adjustments is derived, and vehicle trajectory change nodes are connected to traffic light phase adjustment nodes. Furthermore, based on the reaction of traffic light phase adjustments to subsequent traffic flow and potential events, traffic light nodes are connected to subsequently affected traffic event nodes, thereby generating a logically sound impact graph model framework with multi-type node interactions.

[0044] By utilizing large language models, such as the GPT, BERT, and T5 series, causal reasoning is performed on a unified data representation model to identify the impact of traffic events on traffic flow. For example, after a traffic accident, how does it lead to a decrease in traffic speed and an increase in queue length on surrounding roads? The impact of traffic flow on traffic control strategies is analyzed, such as how traffic lights adjust their timing schemes to optimize traffic flow when traffic volume on a certain road segment continues to increase. Simultaneously, the adverse effects of traffic control commands on subsequent events are analyzed; for example, whether extending the green light time will cause new traffic congestion or reduce the probability of accidents. By using large language models for comprehensive causal reasoning, a closed-loop causal chain set is formed, ensuring that the influence relationships of each link in the traffic system are accurately represented.

[0045] This paper combines a multi-type node influence graph model framework with a closed-loop causal chain set to map the causal relationships of traffic events, vehicle trajectory changes, and traffic light phase adjustments onto the nodes and edges of the influence graph model framework. A heuristic search algorithm is then used for path expansion. During the search process, temporal constraints are employed. By comparing the timestamps of each node in the traffic event, it is ensured that the causal propagation of events, traffic flow, and traffic control instructions follows the actual chronological order of occurrence, avoiding logical reversal. Simultaneously, spatial proximity constraints are used. Utilizing road network coordinates and node latitude and longitude provided by a geographic information system, the spatial distance between traffic events and traffic flow is calculated. Edge connections are only allowed between geographically adjacent nodes or nodes within the influence range, ensuring that the causal paths are physically reasonable. During the heuristic search process, each time a path is expanded, the heuristic search algorithm prioritizes nodes and edges with strong influence based on causal influence indicators, such as the degree of influence of an event on the speed change of traffic flow or the regulatory effect of traffic control on subsequent events. This generates multi-level propagation paths along causal relationships until the entire closed-loop causal chain set is covered. Finally, a multi-level traffic causal influence graph containing event chains, traffic flow chains, and control chains is generated, enabling comprehensive visualization and analysis of the interaction between events, traffic flow, and traffic control in the traffic system.

[0046] For example, a minor collision occurs in the northbound lane of an intersection at 8:15 AM, serving as a traffic event node. Vehicle trajectories show the speed of vehicles ahead decreasing to 10 km / h, serving as a trajectory change node. The remaining 15 seconds of the green light on the traffic light serve as a traffic light phase adjustment node. The large language model inference shows that the accident causes a decrease in traffic speed in the northbound lane. This decrease in traffic speed triggers a 30-second extension of the green light to alleviate congestion. The extended green light time further affects traffic flow in the southbound lane, forming a closed-loop causal chain. Through temporal and spatial proximity constraints, the event, traffic flow, and traffic control nodes are connected into a multi-level propagation path, generating a final traffic causal impact diagram that visually demonstrates the multi-level impact of the accident on traffic flow and traffic control strategies.

[0047] By constructing a traffic causal impact diagram, the multi-level causal relationships between traffic events, traffic flow evolution, and traffic control instructions can be reflected intuitively and comprehensively. This not only improves the accuracy of traffic state changes and the ability to optimize traffic control strategies, but also enables scientific hierarchical storage and management of multimodal traffic data based on the causal influence indicators of each node, combined with data real-time requirements and access frequency. This achieves efficient management and control of intelligent transportation while ensuring the collaborative capabilities of real-time decision-making and historical data analysis.

[0048] Furthermore, causal reasoning is performed on the unified data representation model using a large language model to identify the impact of events on traffic flow, the impact of traffic flow on traffic control strategies, and the feedback effects of traffic control instructions on subsequent events, forming a closed-loop causal chain set. This includes: using the large language model to infer changes in traffic speed, flow rate, and density of traffic events, forming an event causal chain set based on the impact of events on traffic flow patterns; using the large language model to identify the impact of traffic flow on traffic control strategies, and inferring the impact of traffic flow changes on traffic light timing and signal cycle adjustments, forming a feedback causal chain set of traffic flow on traffic control strategies; using the large language model to identify the potential feedback effects of traffic control instructions on traffic events, forming a traffic control causal chain set of traffic control instructions on events; and performing closed-loop fusion of the event causal chain set, the feedback causal chain set, and the traffic control causal chain set to obtain a closed-loop causal chain set.

[0049] Specifically, multimodal data vectors from the unified data representation model are input into the large language model. The large language model possesses powerful semantic understanding and logical reasoning capabilities, enabling in-depth analysis of relevant information about traffic events within the unified data representation model, such as event type, occurrence time, and spatial location. Combined with knowledge from the transportation domain, the large language model can infer how traffic events cause changes in traffic flow speed, volume, and density. For example, by analyzing a large amount of historical traffic data and event cases, the large language model understands that after a traffic accident, the traffic flow speed at the accident site will significantly decrease, the flow rate will drop sharply in a short period, and the traffic density will gradually increase as vehicles queue. Based on this reasoning, a set of causal chains based on the impact of events on traffic flow patterns is formed, clarifying the causal relationships between different types of events and changes in traffic flow.

[0050] The large language model continuously monitors changes in traffic flow, including dynamic fluctuations in traffic speed, volume, and density. Through analysis of traffic signal control principles and actual operational data, the large language model can infer the impact of traffic flow changes on traffic light timing and cycle adjustments. For example, when traffic flow on a certain road segment continuously increases and traffic density exceeds a certain threshold, the large language model will infer that the green light time for that direction needs to be extended to increase vehicle throughput and alleviate congestion. Simultaneously, it may adjust the signal cycle accordingly to make the traffic signal timing at the entire intersection more reasonable. This forms a set of feedback causal chains between traffic flow and traffic control strategies, reflecting the interaction between them.

[0051] Furthermore, the large language model is used to identify the potential counter-effects of traffic control commands on traffic events. The large language model analyzes the impact of traffic control commands, such as traffic light timing adjustments and signal cycle changes, on the traffic environment. For example, if a traffic light extends the green light time in a certain direction, it may lead to increased traffic flow in that direction, potentially causing new traffic accident risks, such as vehicles cutting in at intersections. This forms a set of causal chains between traffic control commands and events, reflecting the reverse causal relationship between traffic control commands and traffic events. The large language model uses a graph fusion algorithm to comprehensively analyze the event causal chain set, the feedback causal chain set, and the traffic control causal chain set. These sets are then fused in a closed loop to obtain a complete closed-loop causal chain set. This closed-loop causal chain set comprehensively reflects the mutual influence between traffic events, traffic flow changes, and traffic control strategies, with each chain retaining the node order, timestamp, and causal relationship.

[0052] For example, a traffic accident occurs at an urban intersection, leading to the closure of part of the main road on the east side of the intersection. The Big Language Model first analyzes the traffic event, using semantic understanding technology to determine the event type as a traffic accident, the time of occurrence as 9:00 AM, and the spatial location as the main road on the east side of the intersection. Combining historical data and traffic domain knowledge, the Big Language Model infers that after the accident, the traffic speed on this road segment decreased from an average of 40 km / h to 10 km / h, the flow rate decreased from 1000 vehicles per hour to 200 vehicles per hour, and the traffic density increased from 20 vehicles per kilometer to 80 vehicles per kilometer, forming a causal chain of events. As traffic flow changes, the Big Language Model detects that the traffic flow on the east side of the intersection continues to increase and the traffic density exceeds a threshold. Through data mining and optimization algorithms, the Big Language Model infers that the traffic light needs to extend the green light time for the east direction from 30 seconds to 50 seconds, while simultaneously adjusting the signal cycle from 90 seconds to 110 seconds, forming a causal chain of feedback from traffic flow to traffic control strategies. However, after extending the green light time, the large language model, through risk assessment analysis, found that increased traffic flow on the east side might lead to vehicles rushing through intersections, increasing the risk of secondary accidents. For example, as the green light is about to end, some vehicles may still attempt to accelerate through the intersection, potentially colliding with vehicles traveling normally in adjacent directions, forming a set of causal chains between traffic control instructions and events. Finally, the large language model uses a graph fusion algorithm to perform closed-loop fusion of the event causal chain set, the feedback causal chain set, and the traffic control causal chain set, forming a closed-loop causal chain set. This closed-loop causal chain set clearly demonstrates how traffic accidents affect traffic flow, how changes in traffic flow adjust traffic control strategies, and how adjustments to traffic control instructions might trigger new traffic incidents, providing comprehensive and accurate data for traffic management departments to formulate response measures.

[0053] By using a closed-loop causal chain set, the multi-level causal relationship between traffic events, traffic flow, and traffic control commands can be accurately reflected, thereby identifying the nodes with the greatest impact on traffic and their corresponding data types, realizing the rational allocation of storage resources and the optimization of data access efficiency, while ensuring the synergistic capability of real-time traffic control optimization and big data analysis.

[0054] Based on the causal influence index of each node in the traffic causal influence diagram, the data real-time requirements, and the data access frequency, the multimodal traffic data is divided into storage layers. High real-time data supporting unified traffic control optimization is allocated to the high-speed storage layer of edge computing nodes, while medium- and low-frequency data supporting big data platform analysis is allocated to the distributed storage layer of the central cloud platform.

[0055] Specifically, based on the generated traffic causal impact map, the causal influence indicators, data real-time requirements, and data access frequency of each node are evaluated. First, the importance of each node in the traffic causal impact map is analyzed, such as the causal influence indicators of traffic accident nodes, critical traffic flow nodes, or traffic light adjustment nodes. For traffic event nodes, the causal influence indicators are quantified by statistically analyzing changes in traffic speed, flow rate, and congestion index on affected roads within a certain time window before and after the event. The degree of impact on traffic is quantified by calculating the rate of speed reduction, the percentage reduction in flow rate, or the expansion range of congestion areas. For vehicle trajectory change nodes, the disturbance amplitude is calculated as a causal influence indicator by comparing the overall traffic flow density, average speed, and flow rate fluctuations of the corresponding road segments before and after the trajectory change. For traffic light adjustment nodes, the downstream lane efficiency before and after the traffic light adjustment is statistically analyzed, such as changes in the average number of vehicles passing through or queue length, and the magnitude of the change is mapped to an influence score. Based on this, historical data statistics, such as average influence values ​​or node contribution scores output by large language model inference, can be combined for comprehensive normalization processing, mapping the causal influence indicators of all nodes to a unified quantitative scale.

[0056] Then, the real-time performance of each node's data is determined based on the node type and application scenario. For example, real-time traffic signal timing data and real-time vehicle location data are crucial for achieving real-time and precise traffic signal control and dynamic vehicle navigation and scheduling, requiring extremely fast updates and processing, typically demanding millisecond-level response times. Historical traffic flow statistics and long-term traffic accident trend analysis data, mainly used for traffic planning, have lower real-time requirements. Simultaneously, the data access frequency is determined by analyzing the query frequency and usage scenarios of different data types. For example, real-time trajectory and signal control data are accessed frequently, while historical event or environmental data is accessed less frequently.

[0057] Based on the causal influence indicators, real-time data requirements, and data access frequency of each node, multimodal traffic data is divided into storage layers: high-real-time, high-frequency data supporting unified traffic signal control optimization, such as real-time vehicle location, speed, acceleration, and current traffic light phase, are allocated to the high-speed storage layer of edge computing nodes. Edge computing nodes, characterized by their proximity to the data source and low latency, can quickly process and store high-real-time data. For example, real-time vehicle location data can be transmitted in real-time to the high-speed cache of the edge computing nodes. The traffic signal control system can immediately obtain this data and, combined with the analysis results of the traffic causal influence diagram, quickly adjust traffic light timings to achieve real-time optimized traffic control.

[0058] For low- to medium-frequency data supporting big data platform analysis, such as historical traffic events, changes in environmental conditions, and long-term traffic flow statistics, these are allocated to the distributed storage layer of the central cloud platform. The central cloud platform possesses powerful storage and computing capabilities, capable of storing massive amounts of historical data. For example, historical traffic flow data and traffic accident statistics can be stored in the central cloud platform's distributed file system, such as Hadoop HDFS. The traffic big data platform can retrieve this data from the central cloud platform to perform traffic trend analysis and pattern recognition, providing a scientific basis for traffic planning and policy formulation. Distributed storage technology improves data reliability and scalability by distributing data across multiple nodes.

[0059] By rationally dividing storage levels, it is possible to ensure that high real-time data is processed and stored quickly, meeting the needs of real-time traffic signal control, improving the operational efficiency and safety of the traffic system. At the same time, storing medium and low frequency data on the central cloud platform facilitates big data analysis and mining, providing strong support for traffic planning and decision-making, realizing the efficient utilization and intelligent management of traffic data, and thus improving the overall operational efficiency and intelligence level of the vehicle-road-cloud integration.

[0060] Example 2, based on the same inventive concept as the vehicle-road-cloud integrated multimodal traffic data hierarchical storage and management method in the foregoing examples, such as... Figure 2 As shown, this application provides a vehicle-road-cloud integrated multimodal traffic data hierarchical storage and management system, wherein the vehicle-road-cloud integrated multimodal traffic data hierarchical storage and management system includes:

[0061] The data acquisition module 11 is used to collect multimodal traffic data in the vehicle-road cooperative scenario, wherein the multimodal traffic data includes the operating status data of intelligent connected devices and autonomous vehicles, environmental perception data, and traffic control command data; the data processing module 12 is used to traverse the multimodal traffic data to perform unified data processing and construct a unified data representation model; the relationship reasoning module 13 is used to perform explicit reasoning on the multi-level influence relationship between traffic events, traffic flow evolution, and traffic control commands based on the unified data representation model, and generate a traffic causal influence diagram that integrates event chains, vehicle flow chains, and control chains; the data partitioning module 14 is used to partition the multimodal traffic data into storage layers according to the causal influence index of each node in the traffic causal influence diagram, the data real-time requirements, and the data access frequency, allocating high real-time data supporting unified traffic control optimization to the high-speed storage layer of edge computing nodes, and allocating medium and low frequency data supporting big data platform analysis to the distributed storage layer of the central cloud platform.

[0062] Furthermore, the data acquisition module 11 is also used for: the operation status data of intelligent connected devices and autonomous vehicles, including real-time trajectory, vehicle speed, acceleration, steering wheel angle, battery status, vehicle diagnostic information and vehicle identification; environmental perception data, including camera video stream, lidar point cloud, and millimeter-wave radar target list; and signal control command data, including real-time phase of traffic lights, signal timing scheme, and variable lane control commands.

[0063] Furthermore, the data processing module 12 is also used to: traverse the multimodal traffic data to extract independent factors, obtain spatiotemporal factors, environmental factors, event factors and semantic factors; traverse the spatiotemporal factors, environmental factors, event factors and semantic factors to perform vector unified dimension transformation, and construct the unified data representation model.

[0064] Furthermore, the relational reasoning module 13 is also used to: construct an influence graph model framework for multiple types of nodes, taking traffic events, vehicle trajectory changes, and traffic light phase adjustments as three types of nodes; perform causal reasoning on the unified data representation model through a large language model to identify the impact of events on traffic flow, the impact of traffic flow on traffic control strategies, and the counter-effects of traffic control instructions on subsequent events, forming a closed-loop causal chain set; and combine the influence graph model framework and the closed-loop causal chain set, using a heuristic search algorithm with temporal constraints and spatial proximity constraints to generate a traffic causal influence graph containing multi-level propagation paths.

[0065] Furthermore, the relational reasoning module 13 also includes: traffic event nodes for representing events such as traffic accidents and road congestion, which combine the timestamps and spatial locations of traffic events to locate and associate events using spatiotemporal data; vehicle trajectory change nodes for describing changes in vehicle driving status, which analyze the changes based on the real-time location, speed, and direction of the vehicle, combined with road information; and traffic light phase adjustment nodes for representing changes in traffic lights, which analyze the changes based on the real-time timing data of the traffic lights, combined with traffic flow and signal cycle.

[0066] Furthermore, the relational reasoning module 13 is also used to: utilize a large language model to reason about the changes in traffic flow speed, flow rate, and density caused by traffic events, forming a set of event causal chains based on the impact of events on traffic flow patterns; utilize a large language model to identify the impact of traffic flow on traffic control strategies, and by reasoning about the impact of traffic flow changes on traffic light timing and signal cycle adjustments, forming a set of feedback causal chains of traffic flow on traffic control strategies; utilize a large language model to identify the potential counter-effects of traffic control instructions on traffic events, forming a set of traffic control causal chains of traffic control instructions on events; and perform closed-loop fusion of the event causal chain set, the feedback causal chain set, and the traffic control causal chain set to obtain a closed-loop causal chain set.

[0067] Furthermore, the relational reasoning module 13 also includes: a temporal constraint for sorting traffic events by timestamps to ensure that the causal effects of traffic events, traffic flow, and traffic control instructions propagate in the actual time sequence of occurrence; and a spatial proximity constraint for calculating the spatial distance between traffic events and traffic flow based on geographic information system data to ensure the rationality of the impact link in physical space.

[0068] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The vehicle-road-cloud integrated multimodal traffic data hierarchical storage management method and specific examples in the aforementioned embodiment one are also applicable to the vehicle-road-cloud integrated multimodal traffic data hierarchical storage management system of this embodiment. Through the foregoing detailed description of the vehicle-road-cloud integrated multimodal traffic data hierarchical storage management method, those skilled in the art can clearly understand the vehicle-road-cloud integrated multimodal traffic data hierarchical storage management system of this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0069] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0070] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud, characterized in that: The method includes: Collect multimodal traffic data in vehicle-road cooperative scenarios, wherein the multimodal traffic data includes operating status data of intelligent connected devices and autonomous vehicles, environmental perception data, and traffic control command data; Perform unified data processing on multimodal traffic data and construct a unified data representation model; Based on the unified data representation model, explicit reasoning is performed on the multi-level influence relationship between traffic events, traffic flow evolution, and traffic control instructions to generate a traffic causal influence diagram that integrates event chains, vehicle flow chains, and control chains. Based on the causal influence index of each node in the traffic causal influence diagram, the data real-time requirements, and the data access frequency, the multimodal traffic data is divided into storage layers. High real-time data supporting unified traffic control optimization is allocated to the high-speed storage layer of edge computing nodes, while medium- and low-frequency data supporting big data platform analysis is allocated to the distributed storage layer of the central cloud platform.

2. The multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud as described in claim 1, characterized in that, The operational status data of intelligent connected devices and autonomous vehicles includes real-time trajectory, vehicle speed, acceleration, steering wheel angle, battery status, vehicle diagnostic information, and vehicle identification. Environmental perception data includes camera video streams, LiDAR point clouds, and millimeter-wave radar target lists; Traffic control command data includes real-time signal phase, signal timing scheme, and variable lane control commands.

3. The multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud as described in claim 1, characterized in that, Perform unified data processing on multimodal traffic data to construct a unified data representation model, including: Independent factor extraction is performed by traversing the multimodal traffic data to obtain spatiotemporal factors, environmental factors, event factors, and semantic factors; By traversing the spatiotemporal factors, environmental factors, event factors, and semantic factors, a unified dimensional transformation of the vectors is performed to construct the unified data representation model.

4. The multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud as described in claim 1, characterized in that, Based on the unified data representation model, explicit reasoning is performed on the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control instructions to generate a traffic causal influence diagram that integrates event chains, traffic flow chains, and control chains, including: By classifying traffic incidents, changes in vehicle trajectories, and adjustments to traffic light phases as three types of nodes, a multi-type node impact graph model framework is constructed. By using a large language model to perform causal reasoning on a unified data representation model, the impact of events on traffic flow, the impact of traffic flow on traffic control strategies, and the reaction of traffic control instructions to subsequent events are identified, forming a closed-loop causal chain set. Combining the aforementioned influence graph model framework and closed-loop causal chain set, a heuristic search algorithm with temporal constraints and spatial proximity constraints is used to generate a traffic causal influence graph containing multi-level propagation paths.

5. The multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud as described in claim 4, characterized in that, Traffic event nodes are used to represent events such as traffic accidents and road congestion. By combining the timestamps and spatial locations of traffic events, event location and correlation are achieved through spatiotemporal data. Vehicle trajectory change nodes are used to describe changes in vehicle driving status, based on the vehicle's real-time position, speed, and direction, combined with road information for analysis; Traffic light phase adjustment nodes are used to represent changes in traffic lights, and are analyzed based on real-time timing data of traffic lights, combined with traffic flow and signal cycle.

6. The multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud as described in claim 4, characterized in that, By using a large language model to perform causal reasoning on a unified data representation model, the impact of events on traffic flow, the impact of traffic flow on traffic control strategies, and the reaction of traffic control instructions to subsequent events are identified, forming a closed-loop causal chain set, including: By using a large language model to infer the changes in traffic flow speed, volume, and density caused by traffic events, a set of event causal chains based on the impact of events on traffic flow patterns is formed. By using a large language model to identify the impact of traffic flow on traffic control strategies, and by reasoning about the impact of traffic flow changes on traffic light timing and signal cycle adjustments, a set of feedback causal chains of traffic flow on traffic control strategies is formed. Large language models are used to identify the potential counter-effects of traffic control instructions on traffic events, forming a set of causal chains of traffic control instructions on events; The event causal chain set, feedback causal chain set, and signal control causal chain set are fused in a closed loop to obtain a closed-loop causal chain set.

7. The multimodal traffic data hierarchical storage and management method integrating vehicle, road, and cloud as described in claim 4, characterized in that, Temporal constraints are used to ensure that the causal effects of traffic events, traffic flow, and traffic control instructions propagate in the actual chronological order of their occurrence by sorting traffic events by timestamps. Spatial proximity constraints are used to calculate the spatial distance between traffic events and traffic flows based on geographic information system data, ensuring the rationality of impact links in physical space.

8. A multimodal traffic data hierarchical storage and management system integrating vehicle, road, and cloud, characterized in that: The steps for implementing the vehicle-road-cloud integrated multimodal traffic data hierarchical storage management method according to any one of claims 1 to 7, wherein the vehicle-road-cloud integrated multimodal traffic data hierarchical storage management system comprises: The data acquisition module is used to collect multimodal traffic data in vehicle-road cooperative scenarios, wherein the multimodal traffic data includes operating status data of intelligent connected devices and autonomous vehicles, environmental perception data, and signal control command data; The data processing module is used to traverse multimodal traffic data, perform unified data processing, and build a unified data representation model. The relational reasoning module is used to explicitly reason about the multi-level influence relationships between traffic events, traffic flow evolution, and traffic control instructions based on the unified data representation model, and generate a traffic causal influence diagram that integrates event chains, traffic flow chains, and control chains. The data partitioning module is used to partition the multimodal traffic data into storage layers based on the causal influence index of each node in the traffic causal influence diagram, the data real-time requirements, and the data access frequency. The high real-time data supporting unified traffic control optimization is allocated to the high-speed storage layer of the edge computing nodes, while the medium- and low-frequency data supporting big data platform analysis is allocated to the distributed storage layer of the central cloud platform.

Images