AI-based multi-source heterogeneous traffic big data accident early warning intervention system

Abstract

The invention relates to the technical field of intelligent traffic, and discloses an AI-based multi-source heterogeneous traffic big data accident early warning intervention system, which comprises a data acquisition module used for acquiring multi-source heterogeneous traffic data of fixed road side equipment and mobile vehicle-mounted equipment; the edge calculation module is used for analyzing the original data in real time and generating structured local event information; the platform processing module is used for converging and fusing multi-source data, extracting multi-dimensional features such as traffic flow speed variance through artificial intelligence risk prediction models such as a long short-term memory network, and dynamically evaluating and predicting accident risk probability; and the application execution module is used for receiving and executing the active intervention strategy according to the risk assessment result. According to the invention, through the cloud edge collaborative architecture, the problems of high data analysis delay and untimely early warning are solved, dynamic and accurate prediction and active intervention of traffic accidents are realized, and a closed-loop management system capable of self-optimization is constructed.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to an AI-based multi-source heterogeneous traffic big data accident early warning and intervention system. Background Technology

[0002] With rapid socio-economic development and the continuous growth of motor vehicle ownership, road traffic safety issues are becoming increasingly serious, and traffic accidents have become a major challenge for urban traffic management. Intelligent Transportation Systems (ITS), as a key technological means to improve traffic safety and efficiency, are gradually shifting their development focus from traditional traffic monitoring and post-accident management to more proactive risk prediction and pre-accident intervention. Therefore, how to utilize advanced information technology to accurately predict potential traffic accident risks and take effective intervention measures has become a current research hotspot in the field of traffic safety.

[0003] In existing technological practices, traffic management departments have deployed various traffic information collection devices, including loop detectors, video surveillance, and microwave radar. They have also begun to utilize floating car GPS data and mobile signaling data to perceive the operational status of the road network. However, these systems still face some inherent technical bottlenecks in practical applications. Existing technical solutions often rely on single or limited data sources for analysis. For example, data from fixed detectors alone is insufficient to capture the micro-driving behavior of vehicles, while relying solely on floating car data cannot achieve full sample coverage of specific road sections. This fragmentation and one-sidedness at the data level makes it difficult for the system to form a comprehensive and three-dimensional understanding of the complex traffic environment, thereby limiting the accuracy of risk identification.

[0004] Furthermore, current mainstream traffic data processing architectures are mostly centralized cloud computing models, which transmit all raw data collected by front-end devices indiscriminately to a central server for unified processing and analysis. This architecture, when faced with ever-increasing volumes of data, especially high-definition video streams, places enormous pressure on network bandwidth and causes significant communication latency. For rapidly changing road traffic conditions, this time lag caused by data transmission and centralized computing often renders early warning information untimely, failing to meet the urgent need for immediate intervention in sudden, short-term accident risks.

[0005] At the methodological level of risk assessment and prediction, many existing systems still rely on traditional methods that set static thresholds based on traffic flow parameters (such as volume, speed, and occupancy) or employ relatively basic statistical models. While these methods are effective in identifying typical congestion scenarios, their predictive capabilities are insufficient for traffic accidents with complex causes. Traffic accidents are often the result of nonlinear coupling of multiple factors, and traditional methods struggle to effectively capture and learn the dynamic evolution and deep correlation characteristics of traffic flow data over time, thus limiting their insight into potential risks.

[0006] Furthermore, most existing technologies focus on risk detection and alerting, while remaining relatively passive in the response phase. Systems typically stop at issuing warnings to managers, lacking a closed-loop mechanism capable of automatically generating and executing proactive intervention strategies based on predicted risks. Simultaneously, the systems lack quantitative evaluation and feedback mechanisms for the actual effects of intervention measures, hindering knowledge accumulation and iterative model optimization, thus making it difficult to continuously improve the overall intelligence and adaptability of the system. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an AI-based multi-source heterogeneous traffic big data accident early warning and intervention system. This system solves the problem that existing traffic safety systems, due to their single data source, high analysis delays, and passive intervention methods, are unable to accurately and proactively predict and intervene in traffic accident risks in real time.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: an AI-based multi-source heterogeneous traffic big data accident early warning and intervention system, comprising:

[0009] The first aspect of this invention provides an AI-based multi-source heterogeneous traffic big data accident early warning and intervention system, which includes: a data acquisition module, an edge computing module, a platform processing module, and an application execution module.

[0010] The data acquisition module is used to collect multi-source heterogeneous traffic data from fixed roadside equipment deployed along the road and mobile vehicle-mounted equipment installed on vehicles. The fixed roadside equipment may include video surveillance cameras, microwave radar, lidar, loop coils, geomagnetic sensors, weather monitoring stations, and ETC readers; the mobile vehicle-mounted equipment may include GPS terminals and on-board diagnostic system interface devices.

[0011] The edge computing module receives data collected by the data acquisition module via a communication connection. This module preprocesses and performs real-time analysis on the received multi-source heterogeneous traffic data, generating structured local event information. In one embodiment, the edge computing module runs one or more lightweight artificial intelligence models to analyze locally acquired real-time video streams to identify preset events such as traffic accidents, congestion, or debris spills. The identification results, including event type, time, and location, are then uploaded to the platform processing module as part of the structured local event information.

[0012] The platform processing module receives structured local event information uploaded by the edge computing module via a communication connection, and simultaneously receives other multi-source heterogeneous traffic data uploaded by the data acquisition module. This module aggregates, fuses, analyzes, and makes decisions based on all received data. Specifically, the platform processing module includes a big data processing platform, a stream computing engine, and an artificial intelligence algorithm model library. The platform processing module first performs spatiotemporal alignment processing on all data to unify their temporal and spatial references; subsequently, it uses a Kalman filter algorithm to fuse observation data from different sensors on the same target to generate the vehicle trajectory of that target. After completing the data fusion, the platform processing module extracts one or more feature vectors from the fused data for risk assessment. These feature vectors include, but are not limited to, the traffic flow velocity variance calculated for a specific road segment. The traffic flow velocity variance... The calculation formula is:

[0013] in is the traffic flow velocity variance; N is the total number of vehicles in the road segment. It is the speed of vehicle i; This refers to the average speed within the stated road segment. Subsequently, the platform processing module inputs the extracted feature vector into its built-in artificial intelligence risk prediction model for calculation. In one embodiment, the artificial intelligence risk prediction model is a Long Short-Term Memory (LSTM) network model. Based on the input temporal feature vector, this model calculates the probability of accident risk in the near future using a sigmoid activation function in its output layer. ;

[0014] in, Probability of accident risk; Use the Sigmoid activation function; The hidden state output of the Long Short-Term Memory network model at the last time step; The weight matrix of the output layer; The feature vector is the bias vector of the output layer. In another embodiment, the artificial intelligence algorithm model library may further include an ensemble learning model. The platform processing module simultaneously inputs the feature vector into the Long Short-Term Memory network model and the ensemble learning model, and performs a weighted fusion of the outputs of the two models to generate the final dynamic risk assessment result. Finally, based on the calculated risk assessment result, the platform processing module matches and generates a corresponding proactive intervention strategy from a preset strategy library.

[0015] The application execution module receives the active intervention strategy generated by the platform processing module via a communication connection and drives the execution devices it includes to complete the intervention actions. The execution devices may include variable message signs, variable speed limit signs, smart road studs, directional horns, vehicle-mounted terminals, and large screens in management and command centers.

[0016] In addition, the platform processing module is also configured to receive traffic status data after the application execution module performs intervention, evaluate the intervention effect by comparing the data changes before and after the intervention, and use the records containing pre-intervention features, intervention strategies and intervention effects as new samples for incremental training or periodic retraining of the artificial intelligence risk prediction model.

[0017] A second aspect of this invention provides an AI-based multi-source heterogeneous traffic big data accident early warning and intervention method, implemented through the aforementioned system, characterized by comprising the following steps: S1: Collect multi-source heterogeneous traffic data from fixed roadside equipment and mobile vehicle-mounted equipment through the data acquisition module; S2: The multi-source heterogeneous traffic data collected in S1 is preprocessed and analyzed in real time through the edge computing module, and structured local event information is generated. S3: The platform processing module aggregates and integrates the structured local event information generated in S2 and other multi-source heterogeneous traffic data collected in S1, uses an artificial intelligence risk prediction model to conduct dynamic risk assessment of traffic conditions, and generates proactive intervention strategies based on the assessment results. S4: Receive and execute the proactive intervention strategy generated in S3 through the application execution module to intervene in traffic.

[0018] This invention provides an AI-based multi-source heterogeneous traffic big data accident early warning and intervention system. It has the following beneficial effects: 1. This invention sets up an edge computing module near the data source to perform real-time localized analysis of raw data such as video, and uploads the identified structured event information to the platform processing module. This enables rapid identification and low-latency reporting of critical traffic events, significantly reducing data transmission bandwidth requirements and the computing load on the central platform, thereby improving the overall system's response speed to sudden risks and the timeliness of early warning intervention.

[0019] 2. This invention performs spatiotemporal alignment and deep fusion of multi-source heterogeneous data from fixed roadside equipment and mobile vehicle-mounted equipment through a platform processing module, and uses artificial intelligence models such as long short-term memory networks to learn and predict multi-dimensional temporal features, including traffic flow velocity variance. This enables a more accurate characterization of complex traffic dynamics and uncovers potential risk patterns that are difficult to detect using traditional statistical methods, thereby improving the accuracy of traffic accident risk assessment and prediction.

[0020] 3. This invention generates proactive intervention strategies through the platform processing module and executes them automatically by the application execution module. At the same time, the changes in traffic status after intervention are used as feedback for continuous incremental training or periodic retraining of the artificial intelligence risk prediction model. This not only realizes the transformation from passive early warning to proactive intervention, but also builds a closed-loop system that can self-optimize and continuously evolve, so that the system's risk prediction and intervention decision-making capabilities can adaptively improve with the accumulation of data. Attached Figure Description

[0021] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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. Example

[0023] Please see the appendix Figure 1 This invention provides an AI-based multi-source heterogeneous traffic big data accident early warning and intervention system, including: The data acquisition module is used to collect multi-source heterogeneous traffic data from fixed roadside equipment and mobile vehicle-mounted equipment; In this embodiment, the AI-based multi-source heterogeneous traffic big data accident early warning and intervention system provided by the present invention will be described in detail. In particular, the data acquisition module and its related data processing flow will be explained first.

[0024] As the perception foundation of the entire system, the data acquisition module aims to comprehensively and multidimensionally acquire raw information on road traffic operation status. Specifically, the data acquisition module is configured to access data from various physical devices and logical platforms to construct a panoramic data view covering macroscopic traffic flow, microscopic driving behavior, and the external environment.

[0025] On one hand, the data acquisition module connects to roadside equipment fixed along the road. Preferably, this equipment may include at least one of the following: video surveillance cameras for acquiring visual information about traffic scenes, microwave radar or lidar for detecting target speed and distance, loop coils or geomagnetic sensors for detecting vehicle presence, meteorological monitoring stations for acquiring environmental parameters, and ETC readers for identifying vehicles. By integrating data from these sources, the system can construct a statically deployed comprehensive monitoring network for specific road sections, providing basic data support for subsequent traffic flow characteristic analysis.

[0026] On the other hand, the data acquisition module also connects to onboard devices that move with the vehicle. Preferably, these devices may include a GPS terminal for providing high-precision spatiotemporal trajectories, and an onboard diagnostic system interface device capable of reading the vehicle's underlying operating status, such as engine speed and throttle opening. Data from mobile devices provides the system with a microscopic perspective on the dynamic behavior of individual traffic participants, which is crucial for identifying high-risk driving behaviors such as sudden acceleration / deceleration and abnormal lane changes.

[0027] Furthermore, to enrich the data dimensions and incorporate historical experience information, the data acquisition module can also be configured to access data from third-party platforms, such as accident data recorded by police reporting platforms, road condition information released by road maintenance departments, or public road condition reports from internet map services. This data provides crucial labeling information and macro-level context for subsequent risk model training and effectiveness validation.

[0028] Understandably, since the aforementioned data originates from different types of sensors and systems, it exhibits significant differences in data format, sampling frequency, timestamp accuracy, and data quality, thus displaying heterogeneity. To ensure that this raw data can be effectively utilized by subsequent platform processing modules, a series of standardization and purification preprocessing operations are required before the data formally enters the fusion analysis stage.

[0029] To address the potential issue of missing values ​​in the original data, such as interruptions in GPS trajectory points of floating vehicles due to wireless signal obstruction, this embodiment employs a time proximity-based interpolation method for data filling. Specifically, linear or polynomial interpolation algorithms can be used to calculate the most probable state of the missing point based on the position and time information of valid data points before and after it. This step aims to ensure the integrity and continuity of the time-series data, laying the foundation for subsequent trajectory analysis and risk feature extraction.

[0030] To eliminate outliers caused by momentary sensor malfunctions or environmental interference, the system performs an outlier detection procedure. In one optional implementation, the system employs statistical methods, such as 3 Criteria, or box plot rules based on interquartile ranges, are used to identify data points that deviate from the normal data distribution range. For example, for vehicle speed data, any physically impossible instantaneous speed value will be identified as an anomaly and discarded or corrected, ensuring the physical authenticity and logical rationality of the data input into the model.

[0031] Furthermore, considering that the measurement noise inherent in the sensor itself can cause high-frequency jitter in the data sequence, affecting the accuracy of the analysis, this embodiment focuses on high-frequency time-series data, such as vehicle data. Velocity, acceleration, etc., are smoothed. The system can use a moving average filter to smooth the data. For a time series signal x, the smoothed value at time t is... It can be calculated using the following formula:

[0032] in, For each time point t, the data points are smoothed. Let be the original data point at time ti; W be the size of the sliding window. This smoothing operation effectively filters out random noise and extracts the main signal reflecting the true trend of change in the data, thus ensuring the stability of subsequent feature calculations.

[0033] After the above collection and preprocessing process, the data acquisition module and the collaborative edge computing module output a standardized and clean data stream, providing high-quality data input for the subsequent platform processing modules to perform spatiotemporal alignment, multi-dimensional feature extraction, and final dynamic risk assessment.

[0034] The edge computing module is used to preprocess and analyze multi-source heterogeneous traffic data in real time, and generate structured local event information. In this embodiment, the functional composition and operating mechanism of the edge computing module will be further described in detail. In the cloud-edge-device collaborative architecture of the entire system, the edge computing module serves as a key intermediate link connecting the front-end sensing devices and the cloud platform processing module. Its core technical purpose is to achieve localized data processing and real-time response.

[0035] During system operation, edge computing modules are deployed in physical locations close to the data source, such as roadside cabinets. They communicate directly with devices in one or more data acquisition modules, receiving the raw data streams they collect. This deployment method aims to solve the technical problems of enormous network bandwidth pressure, high transmission latency, and excessive consumption of central platform computing resources caused by transmitting all raw data over long distances to the central cloud platform.

[0036] Therefore, the edge computing module in this embodiment is endowed with dual core functions: real-time data preprocessing and lightweight intelligent analysis.

[0037] On the one hand, the edge computing module performs some or all of the operations in the aforementioned data preprocessing process. Specifically, for time-series data received from various sensors, the edge computing module can perform operations such as data cleaning, missing value imputation, outlier removal, and data smoothing. By completing these basic processes near the source of the data, the data uploaded to the platform processing module can be guaranteed to have higher consistency and quality, while reducing the computational burden on the central platform.

[0038] More importantly, on the other hand, the edge computing module is the primary platform for performing real-time, low-latency analysis tasks. Preferably, one or more specifically optimized lightweight artificial intelligence models, such as lightweight convolutional neural networks or object detection models, are deployed and run within the edge computing module. These models are specifically trained to analyze locally accessed real-time video streams.

[0039] Specifically, after the video surveillance camera transmits the video stream to the edge computing module, the built-in lightweight artificial intelligence model analyzes the images frame by frame or at specific time intervals to automatically identify preset key traffic events. These events may include: traffic accidents caused by collisions or malfunctions, congestion caused by a surge in traffic flow, or road debris caused by vehicle spills or foreign object intrusion.

[0040] Upon identifying any of the above events, the edge computing module does not directly upload the original video clip that led to the judgment. Instead, it immediately generates a structured local event information. This information is a highly condensed metadata record, which may include: event type code, precise timestamp of the event, geographical coordinates or road segment markers of the event, number of vehicles involved, and a confidence score.

[0041] This approach uploads only high-value structured event information to the platform processing module, eliminating the need to transmit raw, large-volume video streams. This significantly reduces network bandwidth consumption and minimizes unnecessary computational load on the platform processing module. Furthermore, since initial event identification is performed locally, the delay in information reporting is shortened, providing valuable time for subsequent global risk assessment and decision-making by the platform processing module.

[0042] In summary, the edge computing module plays the role of a distributed intelligent outpost in this invention. By performing preliminary data purification and real-time event detection at the data source, it achieves effective filtering of massive amounts of data and rapid extraction of key information, forming an important technical support for the entire system to achieve efficient and low-latency early warning capabilities.

[0043] The platform processing module is used to aggregate and integrate structured local event information and multi-source heterogeneous traffic data. It uses a built-in artificial intelligence risk prediction model to dynamically assess traffic conditions and generate proactive intervention strategies based on the assessment results. This embodiment will describe in detail the core platform processing module of the system of the present invention. As the computational and decision-making center of the system, the platform processing module undertakes core functions such as the fusion of massive amounts of data, the analysis of complex models, and the generation of global strategies. Preferably, this module can logically be composed of components such as a big data processing platform, a stream computing engine, and an artificial intelligence algorithm model library.

[0044] During system operation, the platform processing module receives pre-processed structured local event information uploaded from various edge computing modules via the communication network. Simultaneously, it aggregates more raw, multi-source, heterogeneous traffic data from the data acquisition module. Its core technological objective is to deeply integrate and mine this data from different sources, in varying formats, and with different spatiotemporal references, in order to achieve accurate and dynamic assessment and prediction of road traffic risks.

[0045] To achieve this, the platform processing module first performs a spatiotemporal data alignment and fusion process. This is a crucial preliminary step, designed to unify all input data into the same spatiotemporal coordinate system, providing a foundation for subsequent correlation analysis. Specifically, in the time dimension, the timestamps of all nodes within the system are synchronized with a standard time source via network time protocols, and the platform processing module discretizes all event and data records to a unified time step. Above. In terms of spatial dimension, all geospatial data is uniformly transformed to a standard geographic coordinate system and matched with a precise digital road network model, assigning a unique spatial identifier to each smallest unit of analysis. .

[0046] After completing spatiotemporal alignment, to obtain more accurate and robust state estimates for individual traffic participants on the road, the platform processing module employs a Kalman filter algorithm to fuse observation data from different sensors. This aims to overcome the limitations of a single sensor; for example, radar speed measurement is accurate but positioning may be inaccurate, while video positioning is accurate but speed measurement may be affected by frame rate. The system state vector of the vehicle at discrete time k is defined as follows: This vector can contain its position and velocity information. The Kalman filter implementation follows two iterative steps: prediction and update. The first step is the prediction step, which estimates the posterior state at time k-1. Predict the prior state at time k

[0047]

[0048] Where A is the state transition matrix, which describes how the state evolves over time based on the kinematic model; B is the control input matrix. This is the control vector, which can be ignored in vehicle tracking scenarios without external control input. The next step is the update step, which utilizes the actual measurements obtained from the sensors at time t. The predicted state is corrected to obtain the posterior state estimate. :

[0049] in, H is the measurement vector; H is the observation matrix, which maps the real state space to the observation space. This is the Kalman gain, whose value is dynamically calculated based on the process noise covariance Q and the measurement noise covariance R, used to balance the weights between the predicted and measured values. Through this iterative process, the platform processing module can generate a smooth and accurate motion trajectory for each traffic participant on the road.

[0050] After obtaining high-quality fused data, the platform processing module enters the dynamic risk assessment and prediction phase. The core of this phase is to utilize artificial intelligence models to mine deep patterns related to traffic accident risks from the data. To this end, the module first performs feature engineering on traffic risks. The system extracts multi-dimensional features from the fused data to construct feature vectors for model input. These characteristics can be categorized into several types: for example, micro-driving behavior characteristics that characterize individual driving risk, such as instantaneous speed v, acceleration a, and headway THW; and macro-traffic flow characteristics that characterize the overall operating state of a road segment, such as traffic flow q, density k, and average speed. and the velocity variance that can measure traffic flow stability ; and environmental and historical characteristics that characterize external conditions, such as weather, visibility, and frequency of historical accidents. Among these, velocity variance... It is an important indicator reflecting the dispersion of traffic flow, and its calculation formula is: Where N is the total number of vehicles in this road segment. It is the speed of vehicle i, and It is the average speed of all vehicles within the road segment.

[0051] Subsequently, the platform processing module inputs time-series data containing the aforementioned characteristics into its built-in artificial intelligence risk prediction model. Considering the significant time dependence of traffic risk evolution, this embodiment preferably uses a Long Short-Term Memory (LSTM) network as the core prediction model because it can effectively capture long-term temporal dependencies in the data. The core of the LSTM network is its memory unit, which regulates the information flow through three gating structures: the forget gate, the input gate, and the output gate. At time t, for the input feature vector... and the hidden state of the previous moment The computation process of an LSTM unit can be summarized as: forget gate The decision is based on the cell state at the previous moment. Forgotten information:

[0052] Input gate With candidate cell state The decision to store new information:

[0053] Subsequently, the cell state is updated by combining forgetting and input:

[0054] Finally, output gate It determines which information to output from the updated cell state to form the hidden state at the current moment. :

[0055] In the above formula, W and b represent the weight matrix and bias vector of each gate, respectively. It is the Sigmoid activation function, and tanh is the hyperbolic tangent activation function. This indicates element-wise multiplication.

[0056] After processing the entire input time series, the model outputs the hidden state of the last time step. The data is fed into a fully connected layer and activated by a sigmoid function to calculate the probability of an accident on that road segment in the near future. :

[0057] in, and These are the weight matrix and bias vector of the output layer, respectively. This is a continuous value between 0 and 1. It intuitively quantifies the level of future risk.

[0058] When the platform processing module calculates the risk probability When one or more preset thresholds are exceeded, the proactive intervention strategy generation process is triggered. The module has a built-in strategy knowledge base that stores the correspondence between different risk levels, risk types, and available intervention methods in the form of rules. Based on the current risk assessment results, the module automatically matches the optimal combination of intervention strategies, encapsulates it into standardized instructions, and sends them to the corresponding application execution modules.

[0059] Finally, to enable continuous learning and self-optimization of the system, the platform processing module is also responsible for building a feedback and model optimization closed loop. After the intervention command is issued, the module continuously monitors changes in traffic flow indicators in the target area, and quantitatively evaluates the effectiveness of the intervention measures by comparing data before and after the intervention. These complete event records, including risk characteristics, intervention strategies, and intervention effects, will be stored in the database as valuable training samples for periodic retraining or incremental learning of the artificial intelligence risk prediction model, thereby enabling its predictive capabilities to continuously evolve and adapt.

[0060] The application execution module is used to receive and execute proactive intervention strategies to intervene in traffic.

[0061] In this embodiment, the functional implementation and workflow of the application execution module will be described. The application execution module constitutes the final execution link of the closed loop of the system of the present invention. Its core technical purpose is to transform the digital proactive intervention strategy output by the platform processing module into actual intervention actions on the physical world traffic environment.

[0062] In the system's workflow, the application execution module continuously listens for and receives instructions from the platform processing module through one or more communication interfaces. The proactive intervention strategy is encapsulated into a standardized instruction format, preferably using a data exchange format with good scalability and readability such as JSON. Each instruction clearly defines the key information required for this intervention, such as: the unique identifier of the target execution device, the specific action code to be performed, and the detailed parameters related to that action.

[0063] To achieve effective intervention in the traffic environment, the application execution module can be physically integrated or logically controlled by various types of terminal execution devices. Preferably, the application execution module may specifically include: variable message signs for issuing text or graphic warning information to drivers; variable speed limit signs for dynamically adjusting the legally mandated maximum speed on road sections; smart road studs that can guide traffic flow or warn of danger by changing their luminous color or flashing frequency; directional horns for broadcasting voice messages in specific areas; vehicle-mounted terminals that can directly push warning information to in-vehicle displays; and a large management and command center screen for traffic management personnel to monitor and manually intervene.

[0064] Upon receiving the proactive intervention strategy, the parsing and scheduling unit within the application execution module begins operation. This unit first parses the instruction content and identifies the identifier of the target execution device. Then, based on the action code and related parameters defined in the instruction, it generates a low-level control command conforming to the communication protocol of that specific device. Finally, through the appropriate communication link, this low-level control command is precisely sent to the target execution device.

[0065] It is understandable that this module's function is to alter the current or anticipated state of road traffic through immediate physical or informational intervention, thereby suppressing or eliminating the risks predicted by the platform processing module. For example, when the platform processing module predicts a sharp increase in collision risk due to foggy weather, its generated intervention strategy may be decomposed by the application execution module into multiple parallel actions: driving variable speed limit signs to lower the speed limit, driving variable message signs to display fog warnings, and driving smart road studs along the route to operate in a high-frequency flashing mode. Through multi-dimensional and multi-channel intervention, these actions collectively guide drivers to reduce speed and increase vigilance.

[0066] It is worth noting that the work of the application execution module is not the end of the entire system process. After an intervention is completed, the resulting changes in road traffic conditions are immediately sensed and recorded by data acquisition modules deployed in the same area. This new data, containing the effects of the intervention, is then fed back into the platform processing module as feedback information, thus forming a complete technical loop of "risk prediction - proactive intervention - effect evaluation - model optimization," providing a data foundation for the continuous learning and adaptive evolution of the entire system.

[0067] Please see the appendix Figure 2 An AI-based multi-source heterogeneous traffic big data accident early warning and intervention method includes the following steps: S1: Collect multi-source heterogeneous traffic data from fixed roadside equipment and mobile vehicle-mounted equipment through the data acquisition module; S2: Preprocesses and analyzes multi-source heterogeneous traffic data in real time through the edge computing module, and generates structured local event information; S3: The platform processing module aggregates and integrates structured local event information and multi-source heterogeneous traffic data, uses artificial intelligence risk prediction models to conduct dynamic risk assessment of traffic conditions, and generates proactive intervention strategies based on the assessment results. S4: Receive and execute proactive intervention strategies through the application execution module to intervene in traffic.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based multi-source heterogeneous traffic big data accident early warning and intervention system, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous traffic data from fixed roadside equipment and mobile vehicle-mounted equipment; The edge computing module is used to preprocess and analyze the multi-source heterogeneous traffic data in real time, and generate structured local event information. The platform processing module is used to aggregate and fuse the structured local event information and the multi-source heterogeneous traffic data, perform dynamic risk assessment of traffic conditions through the built-in artificial intelligence risk prediction model, and generate proactive intervention strategies based on the assessment results. The application execution module is used to receive and execute the proactive intervention strategy to intervene in traffic.

2. The system according to claim 1, characterized in that, The fixed roadside equipment of the data acquisition module includes at least one of the following: video surveillance camera, microwave radar, lidar, loop coil, geomagnetic sensor, meteorological monitoring station, and ETC reader / writer; The mobile vehicle-mounted device includes at least one of a GPS terminal and an on-board diagnostic system interface device.

3. The system according to claim 1, characterized in that, The edge computing module is configured to run a lightweight artificial intelligence model to analyze local real-time video streams to identify traffic accidents, congestion, or littering events, and upload the identified events as part of the structured local event information to the platform processing module.

4. The system according to claim 1, characterized in that, The platform processing module includes a big data processing platform, a stream computing engine, and an artificial intelligence algorithm model library; the platform processing module is also configured to perform spatiotemporal alignment on the received data and use a Kalman filter algorithm to fuse observation data of the same target from different sensors to generate an accurate vehicle motion trajectory.

5. The system according to claim 4, characterized in that, The AI ​​risk prediction model is a Long Short-Term Memory (LSTM) network model; the platform processing module is configured to extract feature vectors from the fused data and input the feature vectors into the LTM network model; the output layer of the LTM network model calculates the probability of accident risk in the near future using a Sigmoid activation function. : in, Probability of accident risk; Use the Sigmoid activation function; The hidden state output of the Long Short-Term Memory network model at the last time step; The weight matrix of the output layer; is the bias vector of the output layer.

6. The system according to claim 4, characterized in that, The platform processing module extracts feature vectors from the fused data, including traffic flow velocity variance calculated for specific road segments. ; The velocity variance The calculation formula is: in, is the traffic flow velocity variance; N is the total number of vehicles in the road segment. It is the speed of vehicle i; It is the average speed within the stated road segment.

7. The system according to claim 1, characterized in that, The application execution module includes at least one of the following: variable message signs, variable speed limit signs, smart road studs, directional horns, vehicle-mounted terminals, and large screens in management and command centers.

8. The system according to claim 1, characterized in that, The system is also configured to form a feedback loop; the platform processing module evaluates the intervention effect based on the traffic state changes after the application execution module performs the intervention, and uses the evaluation results to perform incremental training or periodic retraining of the artificial intelligence risk prediction model.

9. The system according to claim 4, characterized in that, The artificial intelligence algorithm model library also includes an ensemble learning model; the platform processing module is configured to simultaneously input the feature vector into the long short-term memory network model and the ensemble learning model, and to perform weighted fusion of the output results of the two models to generate the final dynamic risk assessment result.

10. An AI-based multi-source heterogeneous traffic big data accident early warning and intervention method, as described in any one of claims 1-9, characterized in that, Includes the following steps: S1: Collect multi-source heterogeneous traffic data from fixed roadside equipment and mobile vehicle-mounted equipment through the data acquisition module; S2: The multi-source heterogeneous traffic data is preprocessed and analyzed in real time through the edge computing module, and structured local event information is generated; S3: The platform processing module aggregates and integrates the structured local event information and the multi-source heterogeneous traffic data, uses an artificial intelligence risk prediction model to conduct dynamic risk assessment of traffic conditions, and generates proactive intervention strategies based on the assessment results. S4: Receive and execute the active intervention strategy through the application execution module to intervene in traffic.

Images