A big data highway accident pattern recognition system

Abstract

The application discloses a big data highway accident mode recognition system, comprising: a multi-source data acquisition module for collecting environmental data, facility operation data and vehicle behavior data of a highway section; a data processing module connected with the multi-source data acquisition module, used for fusing and processing various collected data to generate a spatio-temporal aligned joint data set; a mode recognition module connected with the data processing module, used for extracting and fusing multi-dimensional features from the joint data set, and identifying the cause mode and precursor mode of the accident based on the features; and an early warning output module connected with the mode recognition module, used for issuing a road section risk early warning information; the application integrates three heterogeneous data sources of environment, facility and vehicle behavior, constructs a multi-dimensional three-dimensional risk perception system, significantly improves the discovery ability and recognition accuracy of implicit or early risks such as thin ice and fog, and realizes the leap from single appearance monitoring to multi-dimensional essential perception of highway risks.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation systems technology, and in particular relates to a big data highway accident pattern recognition system. Background Technology

[0002] In the field of highway traffic safety, with the development of intelligent transportation systems, using technological means for accident risk early warning has become a key focus of research and application. Existing technologies mainly rely on the monitoring and analysis of single or limited data sources. One common approach is video-based event detection systems, which use high-definition cameras deployed along roadsides and computer vision algorithms to analyze traffic flow video in real time to identify abnormal events such as vehicle collisions, illegal parking, and pedestrian intrusion. Another widely used approach is early warning systems based on meteorological sensors. These systems collect environmental data by deploying visibility meters, road surface condition sensors (such as remote-sensing road surface temperature and humidity sensors), and atmospheric temperature and humidity sensors along roads, comparing the data with preset thresholds to issue warnings for severe weather such as heavy fog, icing, and water accumulation. In addition, there are traffic parameter (flow rate, speed, occupancy) monitoring systems based on fixed traffic detectors (such as loop detectors and microwave radar) to determine traffic congestion levels. These technological solutions provide effective information support for highway safety management in their respective dimensions.

[0003] However, the existing technological systems mentioned above still have significant shortcomings in achieving proactive, precise, and forward-looking risk prevention and control. First, the data dimensions are singular and isolated. Video, meteorological, and traffic flow systems typically operate independently, lacking effective correlation and fusion between data points. This makes it impossible to construct a comprehensive risk profile encompassing "environmental conditions, infrastructure status, and vehicle group behavior," leading to difficulties in identifying complex causal patterns such as "vehicle loss of control caused by thin ice and lighting failure." Second, the analytical methods are relatively superficial and passive. Existing early warning systems are mostly based on simple threshold comparisons or direct detection of already occurred events, lacking the ability to deeply mine implicit features and causal relationships from multi-source heterogeneous data. This makes it difficult to identify risk precursors through indirect signals such as abnormal facility energy consumption and subtle interventions in vehicle control systems before obvious physical signs of an accident appear, resulting in delayed early warnings. Third, the system lacks intelligent evolution capabilities. Models and rules are often fixed, making it difficult to adaptively optimize based on new cases and data feedback, potentially leading to a decline in long-term operational efficiency. Finally, insufficient consideration of privacy protection when utilizing vehicle data may also hinder the in-depth application of the technology. Therefore, there is an urgent need for a new technical solution that can deeply integrate multi-source data, achieve early risk intelligent identification and autonomous evolution, and at the same time ensure data security. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies, this invention provides a big data highway accident pattern recognition system, which solves the problems of data isolation, delayed early warning, and lack of adaptive optimization capabilities in existing technologies.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A big data-based highway accident pattern recognition system includes: The multi-source data acquisition module is used to collect environmental data, facility operation data, and vehicle behavior data of highway sections; The data processing module, connected to the multi-source data acquisition module, is used to fuse and process the acquired data of various types to generate a spatiotemporally aligned joint dataset. The pattern recognition module, connected to the data processing module, is used to extract and fuse multidimensional features from the joint dataset, and based on this, identify the causal patterns and precursor patterns of the accident. The early warning output module is connected to the pattern recognition module and is used to generate and publish road segment risk early warning information based on the recognized pattern. The model optimization module connects the early warning output module and the pattern recognition module, and is used to adaptively update the recognition model based on system feedback.

[0006] Preferably, the multi-source data acquisition module includes: The road surface meteorological data acquisition unit is used to collect data on road surface temperature and humidity gradient, freeze-thaw cycle status, ice film thickness, and friction coefficient. The facility energy consumption data acquisition unit is used to collect real-time energy consumption and operating status data of at least lighting, monitoring and snow melting equipment along the highway. The vehicle-mounted sensor acquisition unit is used to collect anonymized vehicle tire pressure, braking, and stability system intervention data.

[0007] Preferably, the fusion processing performed by the data processing module includes data cleaning, dimensional normalization, and matching and alignment based on a unified spatiotemporal reference.

[0008] Preferably, the pattern recognition module includes: The feature fusion unit is used to dynamically weight and fuse three types of features: environment, facilities and vehicles, to generate a unified feature vector. The causation analysis unit uses an unsupervised learning method to mine stable causal associations of accidents based on the unified feature vector; The precursor analysis unit uses time-series and causal analysis methods to identify the dynamic evolution precursors of accidents based on the unified feature vector.

[0009] Preferably, the road surface temperature and humidity gradient is a rate of change data at the millimeter level; the freeze-thaw cycle state is determined based on the fluctuation period of the road surface temperature around the freezing point.

[0010] Preferably, the energy loss rate collected by the facility energy consumption collection unit is calculated by comparing the actual energy consumption of the equipment with the rated energy consumption; the stable system intervention data collected by the vehicle-mounted sensor collection unit refers to the offset of the intervention timing relative to the standard setting.

[0011] Preferably, the anonymization process performed by the vehicle-mounted sensing acquisition unit refers to removing information that can directly identify a specific vehicle or driver.

[0012] Preferably, in the dynamic weighted fusion, the weight of each feature is dynamically adjusted according to its correlation strength with historical accident data.

[0013] Preferably, the stable cause correlation includes: the coupling mode of road surface temperature and humidity gradient and vehicle tire pressure fluctuation, or the correlation mode of road surface freeze-thaw cycle state and abnormal vehicle braking trigger interval.

[0014] Preferably, the precursors to dynamic evolution are identified by establishing a mapping relationship between abnormal facility energy consumption, meteorological data anomalies, and vehicle sensor data anomalies.

[0015] The technical effects and advantages of the big data highway accident pattern recognition system of the present invention are as follows: 1. This invention constructs a multi-dimensional risk perception system by integrating three major categories of heterogeneous data sources: environment, facilities, and vehicle behavior. This multi-source fusion mechanism breaks through the limitations of traditional methods that rely on a single data dimension (such as only video or weather). It can cross-verify risks from multiple perspectives, such as the micro-physical state of the road surface, the operational health of infrastructure, and the real-time dynamic response of vehicle groups. This significantly improves the ability to detect and identify hidden or early risks such as thin ice and fog, and realizes a leap from "single appearance monitoring" to "multi-dimensional essential perception" of highway risks.

[0016] 2. This invention employs a technical approach combining dynamic feature fusion and unsupervised pattern recognition. Through a dynamic weighting mechanism, the system can adaptively adjust the contribution of different data sources according to the real-time scenario, focusing the analysis on the most relevant risk clues. Combining unsupervised learning and temporal causal analysis, the system can not only identify stable accident causal patterns for long-term hazard management, but also keenly capture weak precursor signals such as abnormal energy consumption and sensor anomalies, as well as their causal chains, achieving early warning of accident risks. This significantly advances the warning window from the traditional post-event or in-event response to the pre-event prevention stage.

[0017] 3. This invention continuously optimizes the identification model using historical early warning feedback and new data, enabling the early warning accuracy to improve over time and reducing reliance on manual rule adjustments during long-term operation and maintenance. Simultaneously, a rigorous anonymization process ensures personal privacy and data security, allowing the system to be widely applied while complying with regulatory requirements. Ultimately, the system's tiered early warning commands can be efficiently integrated with multiple aspects such as maintenance management, emergency response, and public information services, directly translating the results of intelligent analysis into concrete actions to improve proactive highway safety management. Attached Figure Description

[0018] Figure 1 This is a system block diagram of a big data highway accident pattern recognition system proposed in this invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "includes..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0021] refer to Figure 1This invention discloses a highway accident risk identification system, belonging to the field of intelligent transportation and active safety technology. The system aims to solve the problems of single data dimension, delayed early warning, and lack of adaptive capability in existing technologies. The system includes a multi-source data acquisition module, a data processing module, a pattern recognition module, an early warning output module, and a model optimization module connected in sequence, forming a closed loop. The multi-source data acquisition module collects road surface meteorological data, facility energy consumption data, and anonymized vehicle sensor data; the data processing module cleans, normalizes, and spatiotemporally aligns the multi-source data to generate a joint dataset; the pattern recognition module identifies accident causal correlation patterns and dynamic evolution precursor patterns respectively through dynamic weighted feature fusion and unsupervised learning and temporal causal analysis; the early warning output module performs risk assessment and graded early warning on the identified patterns; and the model optimization module continuously optimizes the identification model based on system feedback. This invention achieves early, accurate, and adaptive early warning of highway accident risks through deep fusion of multi-source data and closed-loop intelligent analysis.

[0022] Example 1: Standard accident risk identification process: Purpose of implementation: This embodiment aims to demonstrate the complete closed-loop workflow of the system of the present invention in multi-source data acquisition, fusion processing, pattern recognition and early warning release, and to verify the collaborative operation mechanism and basic performance of the system in a typical mountainous highway tunnel scenario.

[0023] Implementation System: The implementation system comprises five core components that are sequentially connected and interconnected: The multi-source data acquisition section consists of millimeter-wave radar, infrared sensors, smart meters, and roadside communication units (RSUs) deployed inside the tunnel, used to collect environmental, facility, and vehicle data.

[0024] Data processing section: Composed of a server cluster in the road section monitoring center, responsible for merging and processing the collected raw data.

[0025] Pattern recognition section: A software module running on the same server cluster, used to identify risk patterns from processed data.

[0026] Early warning output section: The early warning release platform integrated in the monitoring center is responsible for generating and releasing early warning information.

[0027] Model optimization section: Runs as a background service, responsible for updating the recognition model based on feedback.

[0028] Implementation steps: (1) Data acquisition: The meteorological sensor collects the temperature and humidity at different depths of the road surface every 5 minutes and calculates the gradient value and friction coefficient; the smart meter collects the power and energy consumption of lighting, fan and other equipment in real time; the roadside RSU receives the anonymized tire pressure, ABS trigger and other vehicle sensor data of passing vehicles.

[0029] (2) Data processing: The central server cleans and normalizes the received raw data, and performs spatiotemporal alignment based on the unified road segment station number and timestamp to form a spatiotemporally aligned joint dataset.

[0030] (3) Pattern Recognition: The system first extracts three types of features from the dataset: meteorological, energy consumption, and vehicle features. A pre-trained weighted model dynamically assigns feature weights based on the real-time scene, and then performs feature fusion. Subsequently, an unsupervised clustering algorithm is used to mine stable causal patterns, and a time-series causal analysis algorithm is used to identify dynamic precursor signals. In this example, the system identifies the causal pattern of "high humidity environment at the tunnel entrance accompanied by insufficient lighting" and the precursor pattern of "specific energy consumption fluctuations of the fan indicating the intervention of the vehicle stability system."

[0031] (4) Early warning output: The system will assess the cause mode as a medium-level long-term risk and push it to the maintenance department; and assess the precursor mode as a high-level short-term warning and release it to the driver through the information board and vehicle terminal.

[0032] (5) Model optimization: The system collects actual traffic and accident data for the next 24 hours, calculates the accuracy and false alarm rate of this warning, and automatically adjusts the confidence threshold of the precursor recognition algorithm to complete the adaptive update of the model. Implementation effect

[0033] The system enables 24 / 7 continuous monitoring and early warning of tunnel environmental risks. Statistics show that on this section of road, the system's accuracy rate in predicting accidents caused by slippery conditions leading to loss of control exceeds 75%, with an average lead time of 10-15 minutes for short-term warnings, effectively assisting road management in proactive prevention and control decisions.

[0034] Example 2: Specific identification of causal patterns on icy and snowy road surfaces: Purpose of implementation: This embodiment focuses on demonstrating the system's ability to accurately identify data collection, feature analysis, and causal patterns in a specific high-risk scenario of "freeze-thaw cycle - thin ice" in frigid regions.

[0035] Implementation System: The implementation system also consists of five core components, including: Multi-source data acquisition section: It is equipped with high-precision road surface temperature sensors, energy consumption monitoring modules for electric snow melting systems, and enhanced roadside communication units.

[0036] The data processing, pattern recognition, early warning output, and model optimization components are deployed in the road section monitoring center.

[0037] Implementation steps: (1) The system continuously monitors the temperature of the bridge pavement, records the number of fluctuations and duration around the freezing point (0°C), and accurately defines the "freeze-thaw cycle".

[0038] (2) Synchronously monitor the energy consumption status of the snow melting system. When it is detected that the road surface temperature has completed 3 freeze-thaw cycles in the early morning of a certain day, and the energy consumption of the snow melting system is zero (indicating that it has not been started), the system will take this as a critical abnormal signal.

[0039] (3) At the same time, the on-board data feedback showed that the ABS triggering frequency of vehicles passing through the bridge surface was abnormally high.

[0040] (4) The pattern recognition part first performs feature fusion, assigning high weights to three types of features: “freeze-thaw state”, “failure of snow melting facilities”, and “high frequency triggering of vehicle ABS”. Then, through an unsupervised learning algorithm, a clear causal association pattern is successfully clustered from the data.

[0041] (5) The early warning output section determines this mode as high risk, immediately issues a bridge surface icing warning and initiates an emergency response.

[0042] (6) The model optimization section records the entire process data of this event to enhance the identification features of this type of pattern.

[0043] Implementation results: This system enables automated and precise identification of "black ice" (thin, transparent ice), a significant hazard that is difficult to detect with the naked eye. Compared to traditional manual inspections or single weather warnings, this system improves the reliability and timeliness of warnings several times over by analyzing the correlation between collective feedback from vehicle behavior and facility status, providing crucial decision-making basis for timely activation of snow melting and de-icing emergency responses.

[0044] Example 3: Example of precursor pattern recognition for abnormal facility energy consumption: Purpose of implementation: This embodiment aims to verify the technical feasibility of identifying and warning of potential accident risks in the early stages by using abnormal changes in facility energy consumption, an indirect safety indicator.

[0045] Implementation System: The system is deployed on foggy highway sections. Its implementation system comprises five parts, with the multi-source data acquisition component enhancing the real-time monitoring capability of energy consumption for fog-prone guidance lights.

[0046] Implementation steps: (1) The data acquisition section continuously monitors the current and voltage of the fog zone induction lamps and calculates their real-time energy consumption loss rate.

[0047] (2) One night, the data processing department discovered that the energy consumption rate of a section of the guide light had increased abnormally in a short period of time.

[0048] (3) The pattern recognition part starts the analysis and integrates the three types of data, namely “abnormal energy consumption”, “slight decrease in visibility” and “increased vehicle speed fluctuation”, and makes causal inferences. The algorithm eliminates the interference of traffic flow changes and confirms that “guide light failure” and “initial fogging” jointly caused “unstable vehicle driving”.

[0049] (4) The warning output section determines that this is a precursor to a fog accident, and then increases the flashing frequency of the guide lights on the road section and issues warning information through the variable message sign.

[0050] (5) The model optimization section incorporates this successful early warning case into the knowledge base and optimizes the correlation weight between energy consumption anomalies and meteorological and vehicle data. Implementation effect

[0051] This embodiment demonstrates that the system can uncover potential safety risks from seemingly unrelated equipment malfunction information. Through abnormal energy consumption warnings, the system can identify risk trends approximately 20-30 minutes before visibility reaches the traditional meteorological warning threshold, achieving earlier and more proactive early warning intervention.

[0052] Example 4: Example of dynamic model optimization and data anonymization: Purpose of implementation: This embodiment focuses on demonstrating the system's specific measures for data privacy protection, as well as its ability to continuously learn and optimize its model.

[0053] Implementation System: Based on the standard five-part architecture, the system has set up a dedicated anonymization processing server at the vehicle data access terminal in the data acquisition part, and designed a closed-loop feedback learning process in the model optimization part.

[0054] Implementation steps: (1) Anonymization: Before being connected to the system, vehicle data first flows through an anonymization server. The server performs the following operations: permanently deletes the license plate number and vehicle identification number (VIN) from the data packet; performs irreversible hashing and salting on the vehicle ID used for temporary sessions; and retains only the vehicle type, timestamp, geographical location, and de-identified sensor data.

[0055] (2) Model optimization process: In the early stages of system operation, the accuracy of identifying the risk of "overheating of truck brakes on long downhill slopes" was not high. The model optimization part continuously collected data for up to six months, including warning records, real accident reports, and data from all sensors during the same period. Through comparative analysis and incremental learning algorithms, this part automatically adjusted the fusion strategy of relevant features in the pattern recognition part and introduced new discriminative features. After several iterations, the system's warning accuracy for this type of risk was significantly improved. Implementation effect

[0056] The system achieves high-quality data utilization while strictly adhering to data privacy regulations. Model optimization capabilities enable the system's recognition accuracy to continuously improve over time, with an average quarterly increase of 5%-15% in the accuracy of identifying major risk patterns, demonstrating the system's intelligent evolution characteristics.

[0057] Example 5: Examples of system stress resistance and coordinated response under extreme weather conditions: Purpose of implementation: This embodiment tests and demonstrates the system's robustness, resilience, and cross-departmental collaborative response level under extreme weather conditions such as typhoons, and in the face of complex situations such as data loss and noise interference.

[0058] Implementation System: The system adopts a distributed architecture, with redundancy in its five core components. The early warning output section has established standard interfaces with external navigation platforms and emergency management department systems.

[0059] Implementation steps: (1) During typhoon weather, some road surface meteorological sensors were damaged. The data processing part activated the fault tolerance mechanism and used spatial interpolation to estimate the missing data using data from nearby sensors and regional meteorological radar.

[0060] (2) The pattern recognition part dynamically adjusts the feature fusion strategy in the face of multiple complex features such as strong winds, low visibility, and abnormal equipment status, and gives higher weight to features that directly reflect the stability of handling, such as the vehicle yaw rate.

[0061] (3) The system successfully identified the combined cause pattern of “strong crosswinds” and “obstructed video surveillance view”, and discovered a strong causal precursor relationship between “surge in collective power consumption of information boards” and “precipitous drop in average vehicle speed on road sections”.

[0062] (4) The early warning output section not only issues regular road section warnings, but also pushes regional detour suggestions to the public navigation map through the interface, and automatically generates event reports to be distributed to maintenance and emergency departments.

[0063] (5) The model optimization part saves the data and disposal effects in this extreme scenario as a special case library to enhance the system resilience in future similar scenarios.

[0064] Implementation effect: In the extreme weather stress test, the core functions of the system remained normal, and the warning accuracy rate remained above 70%. Its cross-system linkage ability extended the reach of warning information from single road users to the regional traffic management network, significantly improving the overall disposal efficiency and safety guarantee level for major adverse weather events.

[0065] Comparative example 1: Single video event detection system: Implementation purpose: By comparing with the traditionally widely used single data source (video) analysis system, the significant advantages of the multi-source data fusion analysis technology adopted in the present invention in terms of risk identification prospectiveness, accuracy, and environmental adaptability are highlighted.

[0066] Implementation system: The comparative example system is the high-definition video event detection system commonly deployed in the current industry. It installs intelligent cameras along the road section and directly detects traffic accidents (such as parking, spillage) or traffic anomalies (such as congestion) by real-time analyzing the video stream through computer vision algorithms.

[0067] Comparison scenario and results: Select the same foggy road section scenario as in Example 3 for comparison.

[0068] In the initial stage of fog (the system of the present invention has issued a warning through abnormal energy consumption): The video system did not detect any events and was in a "no warning" state because the visibility was尚可 but there was no direct accident picture.

[0069] In the stage of thickening fog: Due to the blurred picture and decreased contrast, the algorithm function of the video system was severely degraded or失效, and the false alarm rate increased, still unable to provide reliable warnings.

[0070] Comparative conclusion: Throughout the development process of the fog situation, the single video system was always unable to provide effective warnings, while the system of the present invention achieved early warnings.

[0071] Compared with Comparative example 1 for Examples 1 - 5, the multi-source data fusion accident risk identification system shown in Examples 1 to 5 of the present invention forms a sharp contrast with the single video event detection system represented by the traditional Comparative example 1 in terms of the core technical path, warning efficiency, and applicable boundaries. Its先进性 is mainly reflected in the breadth of data dimensions, the depth of risk perception, and the temporal prospectiveness of warnings.

[0072] From the fundamental differences in data foundation and recognition mechanism, the implementation system constructs a three-dimensional dynamic data system covering the environment (road surface weather), facilities (equipment energy consumption), and vehicles (anonymized sensors), and indirectly derives risks through the fusion processing and pattern recognition of these heterogeneous data. For example, Implementation 2 determines the risk of transparent thin ice by analyzing the coupling relationship between "freeze-thaw cycle," "zero energy consumption of snow melting facilities," and "high-frequency triggering of vehicle ABS"; Implementation 3 warns of fog by causal inference between "abnormal increase in energy consumption of guide lights," "decline in initial visibility," and "vehicle speed fluctuations." This multi-source correlation analysis enables it to perceive hidden risks that cannot be directly perceived by the naked eye or a single sensor. In contrast, the comparative system relies solely on a single data source—two-dimensional video images—and its recognition depends entirely on the algorithm's direct detection of visible events (such as parking spaces and scattered objects) in the image. This causes it to fail in the initial fog stage of Implementation 3 due to the lack of visible events, and to miss or falsely report in the thin ice scenario of Implementation 2 because it cannot identify transparent or reflective objects. Essentially, this invention infers "possibility" from multi-dimensional "state" associations, while the comparative example identifies "occurrence" from a single-dimensional "image." The two differ in their logical levels of risk perception.

[0073] This fundamental difference leads to comprehensive advantages in early warning performance, environmental adaptability, and system intelligence. Regarding early warning performance, the example system achieves a leap from "post-event response" to "pre-event early warning," as shown in Examples 1, 3, and 5. It can provide early warnings of risks 10-30 minutes in advance, reserving a critical time window for proactive intervention. In contrast, the comparative system only alarms when an event occurs or is about to occur, representing a passive response. In terms of environmental adaptability, the example system exhibits strong robustness, maintaining reliable operation under various harsh conditions such as nighttime (Example 2), foggy weather (Example 3), and extreme weather causing partial sensor failure (Example 5), through data complementarity and algorithmic fault tolerance. The comparative system, however, experiences a sharp decline in performance or even failure under insufficient light, low visibility, and severe weather. Regarding system intelligence, the model optimization module described in Examples 4 and 5 enables the system to continuously learn and evolve, with early warning accuracy improving over time, forming a dynamically enhanced intelligent closed loop. The comparative system, on the other hand, only uses static rules or model judgments, lacking a self-optimization mechanism. Comprehensive testing shows that the system of this invention is significantly superior to traditional single video detection systems in terms of early warning accuracy, false alarm rate reduction, and effective lead time, proving its innovative technical value in achieving early, accurate, and reliable risk perception through multi-source data fusion.

[0074] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0075] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A big data-based highway accident pattern recognition system, characterized in that, include: The multi-source data acquisition module is used to collect environmental data, facility operation data, and vehicle behavior data of highway sections; The data processing module, connected to the multi-source data acquisition module, is used to fuse and process the acquired data of various types to generate a spatiotemporally aligned joint dataset. The pattern recognition module, connected to the data processing module, is used to extract and fuse multidimensional features from the joint dataset, and based on this, identify the causal patterns and precursor patterns of the accident. The early warning output module is connected to the pattern recognition module and is used to generate and publish road segment risk early warning information based on the recognized pattern. The model optimization module connects the early warning output module and the pattern recognition module, and is used to adaptively update the recognition model based on system feedback.

2. The big data highway accident pattern recognition system as described in claim 1, characterized in that, The multi-source data acquisition module includes: The road surface meteorological data acquisition unit is used to collect data on road surface temperature and humidity gradient, freeze-thaw cycle status, ice film thickness, and friction coefficient. The facility energy consumption data acquisition unit is used to collect real-time energy consumption and operating status data of at least lighting, monitoring and snow melting equipment along the highway. The vehicle-mounted sensor acquisition unit is used to collect anonymized vehicle tire pressure, braking, and stability system intervention data.

3. The big data highway accident pattern recognition system as described in claim 1, characterized in that, The data processing module performs fusion processing, including data cleaning, dimensional normalization, and matching and alignment based on a unified spatiotemporal reference.

4. The big data highway accident pattern recognition system as described in claim 1, characterized in that, The pattern recognition module includes: The feature fusion unit is used to dynamically weight and fuse three types of features: environment, facilities and vehicles, to generate a unified feature vector. The causation analysis unit uses an unsupervised learning method to mine stable causal associations of accidents based on the unified feature vector; The precursor analysis unit uses time-series and causal analysis methods to identify the dynamic evolution precursors of accidents based on the unified feature vector.

5. The big data highway accident pattern recognition system as described in claim 2, characterized in that, The road surface temperature and humidity gradient is measured as the rate of change at the millimeter depth level; the freeze-thaw cycle state is determined based on the fluctuation period of the road surface temperature around the freezing point.

6. The big data highway accident pattern recognition system as described in claim 2, characterized in that, The energy loss rate collected by the facility energy consumption acquisition unit is calculated by comparing the actual energy consumption of the equipment with the rated energy consumption; the stable system intervention data collected by the vehicle-mounted sensor acquisition unit refers to the deviation of the intervention timing relative to the standard setting.

7. The big data highway accident pattern recognition system as described in claim 2, characterized in that, The anonymization process performed by the vehicle-mounted sensor acquisition unit refers to removing information that can directly identify a specific vehicle or driver.

8. The big data highway accident pattern recognition system as described in claim 4, characterized in that, In the dynamic weighted fusion, the weights of each feature are dynamically adjusted based on their correlation strength with historical accident data.

9. The big data highway accident pattern recognition system as described in claim 4, characterized in that, The stable causal correlations include: the coupling mode between road surface temperature and humidity gradient and vehicle tire pressure fluctuation, or the correlation mode between road surface freeze-thaw cycle state and abnormal vehicle braking trigger interval.

10. A big data highway accident pattern recognition system as described in claim 4, characterized in that, The precursors to dynamic evolution are identified by establishing a mapping relationship between abnormal facility energy consumption, meteorological data anomalies, and vehicle sensor data anomalies.

Images