Highway high-risk road section early warning method and system

Abstract

The invention relates to the technical field of intelligent traffic safety management and control, and discloses a highway high-risk road section early warning method and system, and the method comprises the steps: collecting real-time environment and vehicle data through a vehicle side and a roadside sensor, uploading the real-time environment and vehicle data to a cloud platform for data fusion, and obtaining fusion data; establishing a risk judgment model on the cloud platform, updating the risk judgment model in real time through incremental training, analyzing the fused data by using the risk judgment model, and generating a comprehensive risk score; and triggering graded early warning according to the comprehensive risk score, carrying out early warning in advance at a vehicle end, and carrying out visual display on an early warning result. According to the method, the adaptive capacity of the safety detection system to variable weather and vehicle behaviors is improved, finally, accidents are effectively reduced, and the driving experience and the traffic efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic safety control, and in particular to a method and system for early warning of high-risk highway sections. Background Art

[0002] With the rapid development of Intelligent Transportation System (ITS) and Vehicle-to-Everything (V2X) technologies, highway safety monitoring and early warning are gradually changing from traditional static deployment to dynamic perception and real-time feedback. Existing traffic management methods mainly rely on single data sources such as fixed cameras and roadside weather stations to collect road conditions or weather information. However, in complex environments such as highways and dangerous mountain roads, it is difficult to accurately perceive instantaneous traffic flow changes and micro-meteorological conditions by relying solely on fixed monitoring. At the same time, due to the limitations of network bandwidth and the difficulty of multi-source heterogeneous data fusion, most existing systems are often unable to achieve real-time quantitative assessment of sudden risk factors such as vehicle movement trajectories, road icing, and sharp reduction in visibility. In recent years, although big data and cloud computing platforms have been tried in urban signal control and highway toll dispatching, the relevant technical system for refined early warning of high-risk sections of highways has not yet fully matured, and there is a lack of an overall solution that combines real-time performance and prediction accuracy.

[0003] In response to the monitoring and early warning needs of high-risk sections of highways, traditional methods are mostly based on historical accident statistics and basic meteorological information, and specific thresholds or warning signs are manually set. However, such methods cannot grasp the dynamic traffic flow in a timely manner, and it is difficult to combine the vehicle's own driving data for comprehensive analysis, which can easily lead to delayed or excessive warnings. In addition, some existing technologies have attempted to integrate on-board sensors and roadside monitoring data, but they often lack a systematic incremental training mechanism and cannot quickly adjust risk judgment parameters under constantly updated road conditions and meteorological conditions. Therefore, in the case of sudden rain and fog, snow on the road or extreme terrain conditions, the prediction reliability is still insufficient. Especially when multi-source data fusion, vehicle-side and roadside-side linkage, and real-time upload to the cloud for in-depth analysis are required, the existing technology still has pain points such as low data processing efficiency, limited risk quantification capabilities, and a single warning response method, which makes it difficult to fully meet the high standards for safety control of highways and complex road conditions. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing high-risk highway section warning technology has limitations in real-time fusion of multi-source data, dynamic update model and hierarchical warning linkage.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: A method for early warning of high-risk highway sections, comprising: using vehicle-side and roadside sensors to collect real-time environment and vehicle data, uploading the real-time environment and vehicle data to a cloud platform for data fusion, and obtaining fused data; Establish a risk assessment model on the cloud platform, update the risk assessment model in real time through incremental training, use the risk assessment model to analyze the fusion data, and generate a comprehensive risk score; A graded warning is triggered based on the comprehensive risk score, and an early warning is issued on the vehicle side, and the warning results are displayed visually.

[0007] As a preferred solution of the high-risk highway section early warning method described in the present invention, wherein: the collection of real-time environment and vehicle data includes collecting the vehicle's own driving status and surrounding environment data using vehicle-mounted sensors; Sensor devices are deployed on high-risk sections of roads to collect road conditions and weather data.

[0008] As a preferred solution of the high-risk highway section early warning method described in the present invention, the fusion data includes extracting key feature vectors according to the characteristics of the high-risk highway section, and adjusting the weights of the key feature vectors according to the real-time environment and vehicle data; According to the road condition and weather data, key environmental dimensions are set, wherein the key environmental dimensions include weather conditions, road conditions and road section characteristics, and an environmental label library is established according to the key environmental dimensions, wherein each environmental label corresponds to a type of road condition and weather data; Obtain historical data of environment and vehicle data, and set the initial weight of key feature vectors based on historical data. ; Map the collected real-time environment and vehicle data to the environment tag library, adjust the weights of key feature vectors according to the environment tags, and use the adjusted weights Multiply it with the key feature vector to obtain the fused data.

[0009] As a preferred solution of the high-risk highway section early warning method described in the present invention, the risk determination model includes an input layer, a hidden layer and an output layer, and the input layer receives the fusion data as the input of the risk determination model; The hidden layer extracts deeper feature associations based on the fused data, and divides the hidden layer into different modules according to functions, each module analyzing a type of real-time environment and vehicle data; The output layer outputs the comprehensive risk score; According to the error calculated between the output of the risk determination model and the actual risk score, the risk determination model is trained through the environment-weighted mean square error loss function, which is expressed as:

[0010] in, represents the environment-weighted mean square error loss function; Indicates the number of samples when training the risk determination model; Indicates that for Environmental weighting factor for each sample; Indicates The true risk score of samples; Represents the model for The comprehensive risk score output by each sample; Indicates samples.

[0011] As a preferred solution of the high-risk highway section early warning method described in the present invention, the environmental weighting factor includes a category penalty weight and an environmental weight, wherein the category penalty weight represents the category penalty weight for the first The penalty coefficient for each sample is adjusted based on the difference between the comprehensive risk score predicted by the model and the actual risk score:

[0012] in, represents the category penalty weight; Represents the category penalty weight magnification factor; The environmental weight is adjusted according to the real-time environment and vehicle data mapped to the environmental label obtained from the environmental label library. With initial weight The ratio of determines the environmental weight, which is expressed as:

[0013] in, represents the environmental weight, Indicates the environmental weight magnification.

[0014] As a preferred solution of the high-risk highway section early warning method described in the present invention, the real-time updating of the risk determination model through incremental training includes setting a fixed time window according to the real-time environment and vehicle data, and each time the time window passes Trigger an incremental training; The time window The real-time environment and vehicle data in the system are mapped to the environment label library, and the environment labels of the real-time environment and vehicle data are obtained. Based on the environment labels, the correlation between the current data and the hidden layer modules of the risk judgment model is determined. Only the modules with high correlation are updated, and the learning rate of the modules with low correlation is set to zero. If a continuous For modules that have not been updated in incremental training, no relevance judgment is performed using the time window The data in the module is updated accordingly; After each update, the prediction accuracy of the risk determination model is recorded, and the learning rate of the risk determination model module is adjusted according to the prediction accuracy; Set the global update period , after each global update cycle, all time windows in the cycle are summarized The data is used to train and update the risk judgment model as a whole, and to unify the weight associations between the calibration modules.

[0015] As a preferred solution of the high-risk highway section early warning method described in the present invention, wherein: triggering a graded early warning according to a comprehensive risk score includes comparing the comprehensive risk score with a preset risk threshold, dividing the risk level into four types: no risk, low risk, high risk and extremely high risk, determining risk data according to real-time environment and vehicle data and risk labels, and visually displaying the risk data together with the risk labels; When the risk level is no risk, no warning will be issued for high-risk highway sections; When the risk level is low, an early warning is given through the on-board display and voice, and a speed limit warning is issued through the roadside information board; When the risk level is high, based on the low-risk warning measures, it will be linked with the vehicle navigation system to give early warnings to vehicles whose navigation routes include high-risk sections, and prompt them to re-plan their routes; When the risk level is extremely high, road closure measures will be implemented based on high-risk prediction measures, and staff will be notified to prepare corresponding risk management measures.

[0016] A highway high-risk section warning system using any of the methods described in the present invention, wherein: a collection module collects real-time environment and vehicle data using vehicle-side and roadside sensors, uploads the real-time environment and vehicle data to a cloud platform for data fusion, and obtains fused data; The processing module builds a risk determination model on the cloud platform, updates the risk determination model in real time through incremental training, and uses the risk determination model to analyze the fused data to generate a comprehensive risk score; The early warning module triggers graded warnings based on the comprehensive risk score, issues early warnings on the vehicle side, and displays the warning results visually.

[0017] A computer device comprises: a memory and a processor; the memory stores a computer program, comprising: the steps of implementing any one of the methods of the present invention when the processor executes the computer program.

[0018] A computer-readable storage medium stores a computer program, comprising: when the computer program is executed by a processor, the steps of implementing any one of the methods of the present invention are implemented.

[0019] Beneficial effects of the present invention: The method of the present invention collects multi-dimensional real-time data through vehicle-side and roadside sensors, and integrates and processes them in the cloud to timely identify and quantify road risk factors. Based on the incremental training model, the weights can be dynamically updated with new data, which significantly improves the accuracy of risk judgment and reduces reliance on fixed rules. By calculating the loss of environmental weighting and category penalties on the fused data, the importance of extreme environments and serious errors to model training is highlighted, so as to quickly locate potential high-risk sections. The hierarchical warning strategy relies on the comprehensive risk score threshold, which can not only provide prompts through the on-board display or roadside information board, but also can be linked with the navigation system and road closures to enhance vehicle traffic safety. The combination of modular incremental training and global periodic correction can centrally update high-correlation modules while maintaining the stability of low-correlation features, further improving the adaptability to changing weather and vehicle behavior, and ultimately effectively reducing accidents and improving driving experience and traffic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 An overall flow chart of a high-risk highway section early warning method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0023] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a method for early warning of high-risk sections of a highway, comprising: S1: Use vehicle-side and roadside sensors to collect real-time environment and vehicle data, upload the real-time environment and vehicle data to the cloud platform for data fusion, and obtain fused data.

[0024] Furthermore, real-time environment and vehicle driving information are obtained from the vehicle side and the roadside side to fully reflect the dynamic conditions of high-risk sections of highways. Specifically, the vehicle side mainly uses a variety of monitoring devices such as on-board radar, camera, GPS, accelerometer and steering angle sensor to record the vehicle's driving speed, acceleration, steering angle, and environmental images or obstacle information around the vehicle in real time. These data can reflect the current operating status of the vehicle and the surrounding traffic situation, providing an important basis for subsequent risk determination. At the same time, an on-board communication unit can also be deployed inside the vehicle, so that the aforementioned data can be quickly transmitted to the back-end platform through V2I or V2X after collection.

[0025] For high-risk sections of highways, various types of sensor equipment are deployed on the roadside to achieve three-dimensional perception of the external environment. The sensor equipment includes but is not limited to: intelligent monitoring cameras, laser radars, geomagnetic sensors, meteorological sensors (such as temperature, humidity, visibility detection, etc.), and special sensor devices that can detect road conditions (such as snow, ice, water accumulation detectors, etc.). In addition, radar devices can be installed on the roadside to monitor information such as the speed, distance between vehicles and traffic flow of vehicles.

[0026] The above multi-source data is ultimately uploaded to the data processing platform or cloud in real time through wired or wireless networks for preliminary aggregation and formatting, forming fused data that can be used by subsequent risk assessment models. Through this multi-source perception and dynamic interaction, the accuracy of identifying potential dangers and the timeliness of early warning for high-risk sections of highways can be significantly improved.

[0027] Furthermore, based on the collected multi-source environment and vehicle data, key feature vectors are extracted according to the characteristics of high-risk sections of the highway, and the weights of the key feature vectors are adjusted according to the real-time environment and vehicle data.

[0028] Specifically, key environmental dimensions are set according to road conditions and weather data, and the key environmental dimensions include weather conditions, road conditions and road section characteristics. An environmental label library is established according to the key environmental dimensions, and each environmental label corresponds to a type of road condition and weather data.

[0029] Among them, meteorological conditions are divided into sunny, rainy, snowy, foggy, typhoon, etc., road conditions are divided into dry, slippery, icy, snowy, etc., and road section features are divided into flat, continuous curves, long downhill, mountainous area, near water and cliffs, etc. Each label in the environmental label library consists of meteorological conditions-road conditions-road section features. Each label feature is set to a different severity through a threshold or judgment rule. According to the pre-defined threshold or rule, the real-time environmental perception vector is mapped to one or more environmental labels, such as "moderate rainy day-slippery road section-continuous curves". The threshold of the environmental label library is set based on the specific local conditions of the high-risk section, and the collected real-time data is compared with the threshold to quickly determine the environmental label of the current section.

[0030] Obtain historical data of environment and vehicle data, and set the initial weight of key feature vectors based on historical data. , the weight is used to identify the relative importance of each feature in risk determination under general conditions (such as sunny days and straight roads).

[0031] The collected real-time environment and vehicle data are mapped to the environment label library for matching, the environment type of the current road section is determined, the corresponding weight bias mapping table is called according to the environment label, the key feature vector is weighted, and the adjusted weight is used. Multiply it with the key feature vector to obtain the fused data.

[0032] The adjustment value of the initial vector in the weight bias mapping table can be pre-set based on expert experience or statistical analysis (such as "on rainy days, the importance of road humidity and braking frequency increases by 20%"), or the system can continuously learn during long-term operation and store it in the weight bias mapping table to form an adaptive mechanism.

[0033] It should be noted that in order to prevent the weights of certain features from being excessively amplified or reduced in a single correction, the adjustment amount of the weights is constrained by setting upper and lower thresholds. When the system detects that the environmental labels change frequently (such as from "light rain" to "heavy rain" and then to "no rain" in a short period of time), a smoothing strategy will be introduced to make progressive corrections to the weights to prevent the model from excessive oscillation due to too frequent environmental fluctuations.

[0034] The method of the present invention not only makes static corrections to single weather or road conditions, but also supports multi-label superposition, and allocates weighted superposition or the highest priority strategy according to the actual degree of danger, thereby achieving finer-grained feature tuning. In the risk prediction of high-risk sections of highways, it can better adjust the feature weights in real time for complex and changeable weather and road conditions, amplify or highlight the most representative features, and thus improve the overall accuracy and robustness of the risk judgment model.

[0035] S2: Establish a risk assessment model on the cloud platform, update the risk assessment model in real time through incremental training, use the risk assessment model to analyze the fused data, and generate a comprehensive risk score.

[0036] Furthermore, the risk determination model includes an input layer, a hidden layer and an output layer. The input layer accepts the fused data as the input of the risk determination model; the hidden layer extracts deeper feature associations based on the fused data, and divides the hidden layer into different modules according to function, each module analyzing a type of real-time environment and vehicle data; the output layer outputs a comprehensive risk score.

[0037] The error is calculated based on the output of the risk determination model and the real risk score. The real risk score is based on the real risk events that have occurred in the historical data (such as vehicle accidents, failures, etc.), and is obtained through expert evaluation to reflect the actual risk level. The risk determination model is trained through the environment-weighted mean square error loss function, which can effectively highlight the weight of harsh environments or critical errors. The loss function is expressed as:

[0038] in, represents the environment-weighted mean square error loss function; Indicates the number of samples when training the risk determination model; Indicates that for Environmental weighting factor for each sample; Indicates The true risk score of samples; Represents the model for The comprehensive risk score output by each sample; Indicates samples.

[0039] Furthermore, the environmental weighting factor includes a category penalty weight and an environmental weight, and the value of the environmental weighting factor is the product of the category penalty weight and the environmental weight.

[0040] The category penalty weight is used to distinguish the impact of the model's error type on the loss during prediction, and the environment weight is used to amplify (or reduce) the error ratio in different environment scenarios. The environment weight and category penalty focus on "in which environment the error occurred" and "what type of error" respectively. These two dimensions are not contradictory, and often need to be considered simultaneously to obtain a weighting strategy that better meets actual needs.

[0041] The category penalty weights represent The penalty coefficient for each sample is adjusted based on the difference between the comprehensive risk score predicted by the model and the actual risk score:

[0042] in, represents the category penalty weight; Represents the category penalty weight magnification factor.

[0043] The environmental weight is adjusted according to the real-time environment and vehicle data mapped to the environmental label obtained from the environmental label library. With initial weight The ratio of determines the environmental weight, which is expressed as:

[0044] in, represents the environmental weight, Indicates the magnification of the environment weight. If both the environment weight and the category penalty are too large, the cumulative multiplication will cause gradient explosion or training instability. In practice, it is necessary to limit the maximum penalty coefficient or use measures such as gradient clipping.

[0045] It should be noted that the environmental weight and the adjusted weight All are judged based on the environmental tags in the environmental tag library, and the environmental weight magnification is Obtaining the adjusted value of the environmental weight can effectively reduce the amount of calculation and further deepen the impact and correlation of environmental factors on the output of the risk judgment model.

[0046] Feature weights are often adjusted for each feature individually, while loss weighting factors are used to amplify or reduce the prediction error of the entire sample. The two weights have different "granularity" and cover different contents, so the two cannot reuse the same value. For real-time environment and vehicle data, there are multiple different The problem is usually chosen relative to The largest adjustment To calculate the environmental weight, The larger the change in , the more severe the corresponding environmental conditions are. When used to calculate the loss function, the penalty for extreme weather forecast errors can be increased.

[0047] It is also worth noting that by optimizing the loss function through the environmental weighting factor, the risk judgment model has a larger penalty coefficient when there is a deviation in the extreme weather forecast, and a smaller penalty coefficient when there is a deviation in good weather, which strengthens the prediction ability of extreme weather. Take 2) and the true risk score = high risk but predicted as low risk ( Take 1.5), then:

[0048] In the loss function, the error of this sample is magnified 3 times. The model will significantly adjust the relevant weights during back propagation to quickly correct such serious underreporting. When the environment is relatively safe and there is a false alarm (the model gives a higher risk score even though the actual risk is not high), if only a small penalty coefficient (such as 1.2) is set, the overall magnification will not be excessive.

[0049] Furthermore, the real-time updating of the risk determination model through incremental training includes setting a fixed time window according to the real-time environment and vehicle data, and each time the time window passes Trigger an incremental training. The time window For a short time window, such as the last five or ten minutes, the most recent road conditions and weather conditions are obtained and the model is incrementally trained to increase the prediction accuracy of the risk determination model for changes in short-term risk factors.

[0050] The time window The real-time environment and vehicle data in the system are mapped to the environment label library, and the environment labels of the real-time environment and vehicle data are obtained. Based on the environment labels, the correlation between the current data and the hidden layer modules of the risk judgment model is determined. Only the modules with high correlation are updated, and the learning rate of the modules with low correlation is set to zero.

[0051] Specifically, if it is found that this batch of data is mainly related to weather changes, such as sudden heavy rain, only the modules that are highly related to meteorological judgment will be given a higher update priority, and modules that are not related to or weakly related to traffic flow will not be updated. If certain new feature combinations appear in this batch of data sets (such as night + rain and snow + curves + high traffic), a higher learning rate will be set on the connection weights of the corresponding modules so that the model can adapt to new situations more quickly.

[0052] If a continuous For modules that are not updated during the incremental training, no relevance judgment is performed and the corresponding modules are updated using the data within the time window.

[0053] After each update, the prediction accuracy of the risk determination model is recorded, and the learning rate of the risk determination model module is adjusted according to the prediction accuracy. After completing a small batch incremental training, the system deploys the new parameters to the online inference module to observe its prediction performance on actual data in the next time period. If the prediction effect of the model (such as accuracy, error rate and other indicators) keeps improving steadily or remains at a high level after several consecutive incremental updates, it means that the local update is effective. When it is found that the error of some modules of the model increases due to too frequent or improper updates, the system will actively reduce the learning rate of the module or temporarily freeze the update of the module to allow other modules to compensate; it can also unfreeze the module when a new round of changes in the environment occurs to prevent excessive interference with the knowledge learned in normal situations.

[0054] Set the global update period , after each global update cycle, all time windows in the cycle are summarized The risk assessment model is trained and updated as a whole based on the data, and the weight associations between the calibration modules are unified. The purpose of global fine-tuning is to further correct the long-term local accumulated errors while ensuring adaptation to the latest scenarios, and to unify the weight associations between the calibration modules. When new large-scale historical traffic data or special scenario data (such as congestion data on bad weather and special holidays) are available, offline supplementary training is performed again to enhance the overall generalization performance of the model, and then it is redeployed to the online system.

[0055] It should be noted that traditional online learning methods often perform a gradient update on the entire network when new data arrives, which can easily cause knowledge forgetting or update conflicts; the present invention adopts the "environment matching + priority identification + local training" method to effectively reduce disturbances to irrelevant modules and improve the learning rate or update frequency on key modules, significantly enhancing the pertinence and efficiency of real-time warnings for high-risk sections.

[0056] Secondly, some existing incremental learning technologies do not have a dedicated global fine-tuning link after updating small batches of data. However, the present invention can integrate more real traffic scene data within a large time range by regularly reviewing and retraining the incremental update results, and comprehensively correct the network parameters, taking into account both short-term response and long-term optimization. Specifically, the incremental update of the short-term window responds to sudden changes in weather, road conditions, etc., while the global update responds to long-term changes such as seasonal and periodic traffic flow. The combination of the two ensures the accuracy of the risk assessment model prediction.

[0057] S3: Trigger graded warnings based on comprehensive risk scores, issue early warnings on the vehicle side, and visualize the warning results.

[0058] According to the comprehensive risk score, a graded warning is triggered, and an early warning is issued through the vehicle-mounted navigation system, and the warning results are displayed visually. In the present invention, based on the comprehensive risk score generated by the aforementioned risk determination model, the score is compared with the pre-set risk threshold, so as to preliminarily divide the potential danger level of the current road section or vehicle driving scene into four levels of no risk, low risk, high risk and extremely high risk. Combined with real-time environment and vehicle data and the aforementioned environmental labels, not only can the comprehensive risk score be further corrected or verified, but also specific extreme environments (such as nighttime heavy rain, strong wind areas, etc.) can be additionally identified to form more accurate warning results. Subsequently, the warning results are displayed visually together with the corresponding environmental labels, which facilitates the driver, roadside equipment or management platform to quickly identify dangerous conditions.

[0059] Specifically, when the risk level is no risk, no warning will be issued for high-risk highway sections, but the system will still record relevant data for subsequent incremental training and model optimization.

[0060] If the risk level is low, the driver will be given advance warning through the on-board display and voice prompts, and a speed limit warning can be issued on the roadside information board at the same time to remind other vehicles entering the section to maintain a safe distance and appropriate speed.

[0061] When the risk level is upgraded to high risk, the system will further link with the vehicle navigation based on the low-risk warning measures: the system will issue a "high-risk road ahead" warning to the driver in advance through a comprehensive analysis of the vehicle's destination and road section information, and provide suggestions for re-planning the route or slowing down.

[0062] If the comprehensive risk score and environmental label jointly indicate that the risk level is extremely high, a more stringent response strategy will be initiated based on the high-risk warning plan, including the implementation of temporary traffic control such as road closures, and simultaneously notifying relevant emergency or maintenance departments to prepare corresponding risk management measures to minimize the possibility of accidents occurring and expanding.

[0063] During this process, the vehicle navigation system can combine the model output with the records in the environmental tag library to dynamically update the navigation path or speed limit plan, so that the vehicle can avoid dangerous areas as much as possible or take corresponding mitigation measures in harsh environments. When the management center or roadside equipment receives extremely high-risk information pushed by the system, it can also implement warning broadcasts, instructions or personnel evacuation in accordance with the actual situation on site. Overall, through multi-level warning and visualization methods, drivers and traffic control personnel can clearly grasp the specific risk status of high-risk sections of the highway, and take corresponding protective or diversion actions in advance, thereby achieving efficient management and optimization of highway traffic safety.

[0064] Embodiment 2: In an exemplary embodiment, a highway high-risk section warning system is also provided, including a collection module, a processing module and a warning module.

[0065] The acquisition module is connected to the processing module. The acquisition module uploads the real-time environment and vehicle data obtained by the vehicle-end and roadside sensors to the cloud platform through a wired or wireless network. After receiving the above data, the processing module integrates, cleans and formats it, and conducts a comprehensive analysis based on the risk judgment model.

[0066] The processing module is connected to the acquisition module and the early warning module. The processing module uses incremental training to update the risk determination model in real time and outputs a comprehensive risk score. The comprehensive risk score and corresponding risk information are passed to the early warning module so that the latter can trigger graded warnings and perform visual displays accordingly.

[0067] The early warning module is connected to the vehicle / roadside terminal. The early warning module decides whether to issue an early warning or a higher level of control based on the level of the comprehensive risk score, and sends the early warning instruction or visual information to the vehicle navigation system, vehicle display terminal or roadside information board. In extremely high-risk situations, higher-level response measures such as road closure instructions can also be sent to the management center or emergency department.

[0068] Through the transmission of the above data and instructions, the three modules form a closed-loop system process from bottom to top (data collection → data processing → risk assessment → early warning execution), and continuously improve the efficiency of road safety monitoring based on real-time iteration and modular updates.

[0069] Furthermore, the collection module is deployed on the vehicle side and the roadside side of high-risk sections to obtain real-time data from multiple sources, and sends it to the cloud platform after preliminary integration.

[0070] The specific implementation method includes data collection and data transmission on the vehicle side and the roadside side. The vehicle side can obtain the vehicle's own driving status and surrounding environment information based on on-board radar, camera, GPS, accelerometer, steering angle sensor, etc.; the roadside side deploys intelligent monitoring cameras, lidar, meteorological sensors, roadside radar and other equipment to monitor road conditions in high-risk sections to detect road water accumulation, ice, terrain changes, etc.; the above multi-source data is transmitted to the cloud platform via wired (such as optical fiber) or wireless (such as V2X) methods to provide comprehensive and accurate real-time data support for the processing module.

[0071] The processing module runs on the cloud platform and is mainly responsible for data fusion, model training and updating, and risk score calculation. It is the core analysis center of the entire system.

[0072] Specifically, the multi-source real-time data uploaded by the acquisition module is formatted, noise filtered, and timestamp aligned, and the processed data is used as the input of the risk determination model; Establish a network model with input layer, hidden layer and output layer, and update the model parameters in real time through incremental training; Set fixed time windows based on the real-time environment and vehicle data collected, trigger small batch incremental training when each time window arrives, and use mechanisms such as environment weighted mean square error to focus on extreme environments or serious errors; The potential risk level of the current road section is digitally quantified and the results are sent to the early warning module.

[0073] The warning module triggers warning measures of corresponding levels based on the comprehensive risk score and real-time environmental information output by the processing module, and visualizes the risk data to the driver or traffic management department.

[0074] Specifically, the comprehensive risk score is compared with the pre-set risk threshold to automatically determine the four levels of no risk, low risk, high risk, and extremely high risk; For high-risk or extremely high-risk sections of road, the vehicle navigation system will provide early warning and prompts, and guide the vehicle to re-plan its route when necessary; Present risk data and environmental labels on the vehicle display, roadside information board or backend management platform interface, so that drivers and managers can understand the risk situation intuitively; If the risk level reaches an extremely high level, the early warning module will also send an alarm message to the highway management center or emergency department to coordinate the implementation of higher-level response plans such as road closures and personnel evacuation.

[0075] Through the close linkage between the above three functional modules, the present invention can carry out real-time monitoring, accurate judgment and dynamic warning of high-risk sections of highways; based on the collection of rich multi-source data, the prediction accuracy is continuously improved through continuous incremental training of cloud-based models, and potential dangers are promptly communicated to relevant parties with the help of a graded warning mechanism, thereby achieving efficient and safe management of high-risk sections.

[0076] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0077] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0078] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0079] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0080] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for early warning of high-risk highway sections, characterized in that: include: Use vehicle-side and roadside sensors to collect real-time environment and vehicle data, upload the real-time environment and vehicle data to the cloud platform for data fusion, and obtain fused data; Establish a risk assessment model on the cloud platform, update the risk assessment model in real time through incremental training, use the risk assessment model to analyze the fusion data, and generate a comprehensive risk score; Trigger graded warnings based on comprehensive risk scores, issue early warnings on the vehicle side, and visualize the warning results; The collecting of real-time environment and vehicle data includes collecting the vehicle's own driving status and surrounding environment data using vehicle-mounted sensors; Install sensors on high-risk roads to collect road condition and weather data; The fused data includes extracting key feature vectors based on the characteristics of high-risk sections of highways, and adjusting the weights of the key feature vectors based on real-time environment and vehicle data; According to the road condition and weather data, key environmental dimensions are set, wherein the key environmental dimensions include weather conditions, road conditions and road section characteristics, and an environmental label library is established according to the key environmental dimensions, wherein each environmental label corresponds to a type of road condition and weather data; Obtain historical data of environment and vehicle data, and set the initial weight of key feature vectors based on historical data. ; Map the collected real-time environment and vehicle data to the environment tag library, adjust the weights of key feature vectors according to the environment tags, and use the adjusted weights Multiply it with the key feature vector to obtain the fused data.

2. A highway high-risk section early warning method as claimed in claim 1, characterized in that: The risk determination model comprises an input layer, a hidden layer and an output layer, and the input layer receives the fusion data as input of the risk determination model; The hidden layer extracts deeper feature associations based on the fused data, and divides the hidden layer into different modules according to functions, each module analyzing a type of real-time environment and vehicle data; The output layer outputs the comprehensive risk score; According to the error calculated between the output of the risk determination model and the actual risk score, the risk determination model is trained through the environment-weighted mean square error loss function, which is expressed as: ;in, represents the environment-weighted mean square error loss function; Indicates the number of samples when training the risk determination model; Indicates that for Environmental weighting factor for each sample; Indicates The true risk score of samples; Represents the model for The comprehensive risk score output by each sample; Indicates samples.

3. A highway high-risk section early warning method as claimed in claim 2, characterized in that: The environmental weighting factor includes a category penalty weight and an environmental weight. The category penalty weight represents the The penalty coefficient for each sample is adjusted based on the difference between the comprehensive risk score predicted by the model and the actual risk score: ;in, represents the category penalty weight; Represents the category penalty weight magnification factor; The environmental weight is adjusted according to the real-time environment and vehicle data mapped to the environmental label obtained from the environmental label library. With initial weight The ratio of determines the environmental weight, which is expressed as: ;in, represents the environmental weight, Indicates the environmental weight magnification.

4. A highway high-risk section early warning method as claimed in claim 3, characterized in that: The real-time updating of the risk determination model through incremental training includes setting a fixed time window according to the real-time environment and vehicle data, and each time the time window passes Trigger an incremental training; The time window The real-time environment and vehicle data in the system are mapped to the environment label library, and the environment labels of the real-time environment and vehicle data are obtained. Based on the environment labels, the correlation between the current data and the hidden layer modules of the risk judgment model is determined. Only the modules with high correlation are updated, and the learning rate of the modules with low correlation is set to zero. If a continuous For modules that have not been updated in incremental training, no relevance judgment is performed using the time window The data in the module is updated accordingly; After each update, the prediction accuracy of the risk determination model is recorded, and the learning rate of the risk determination model module is adjusted according to the prediction accuracy; Set the global update period , after each global update cycle, all time windows in the cycle are summarized The data is used to train and update the risk judgment model as a whole, and to unify the weight associations between the calibration modules.

5. A highway high-risk section early warning method as claimed in claim 4, characterized in that: The triggering of graded warning according to the comprehensive risk score includes comparing the comprehensive risk score with a preset risk threshold, classifying the risk level into four types: no risk, low risk, high risk and extremely high risk, determining risk data according to real-time environment and vehicle data and risk labels, and visually displaying the risk data together with the risk labels; When the risk level is no risk, no warning will be issued for high-risk highway sections; When the risk level is low, an early warning is given through the on-board display and voice, and a speed limit warning is issued through the roadside information board; When the risk level is high, based on the low-risk warning measures, it will be linked with the vehicle navigation system to give early warnings to vehicles whose navigation routes include high-risk sections, and prompt them to re-plan their routes; When the risk level is extremely high, road closure measures will be implemented based on high-risk prediction measures, and staff will be notified to prepare corresponding risk management measures.

6. A highway high-risk section early warning system, applied to a highway high-risk section early warning method according to any one of claims 1 to 5, characterized in that: include, The acquisition module uses vehicle-side and roadside sensors to collect real-time environment and vehicle data, uploads the real-time environment and vehicle data to the cloud platform for data fusion, and obtains fused data; The processing module builds a risk determination model on the cloud platform, updates the risk determination model in real time through incremental training, and uses the risk determination model to analyze the fused data to generate a comprehensive risk score; The early warning module triggers graded warnings based on the comprehensive risk score, issues early warnings on the vehicle side, and displays the warning results visually.

7. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the high-risk highway section warning method as described in any one of claims 1-5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a high-risk highway section warning method as described in any one of claims 1-5 are implemented.

Images