Highway operation risk early warning method and device in dry and high-temperature environment

Abstract

The invention provides an expressway operation risk early warning method and device in a dry and high-temperature environment, and is applied to the field of data processing. Through multi-source data fusion and an intelligent algorithm, full-process automation from risk prediction to strategy generation is realized. Acquiring and cleaning weather, road and traffic data; constructing a risk prediction model based on vehicle types and road sections; dynamically evaluating and screening key early warning indexes; dynamic correction factors are introduced to adapt to emergencies; generating a personalized early warning model by using a machine learning algorithm; finally, in combination with traffic rules and weather forecast, accurate real-time early warning and time-phased management and control strategies are output, and the safety and management efficiency of expressway operation in a high-temperature environment are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a highway operation risk early warning method and device in dry high-temperature environment. BACKGROUND

[0002] In the field of highway operation management, dry high-temperature environment is one of the key inducements to cause vehicle performance degradation, traffic flow disorder and traffic accidents. The existing technology relies on independent monitoring equipment to collect single data and trigger early warning, for example, through the automatic weather station along the line to collect air temperature and road surface temperature data, when the road surface temperature exceeds 60℃, the "high temperature warning" is triggered, but the vehicle performance change and traffic flow state are not associated, and the difference in sensitivity of different vehicle types to high temperature cannot be distinguished. Through video monitoring or microwave radar to detect traffic flow and average speed, when the traffic density exceeds 25 vehicles per kilometer, the "congestion warning" is triggered, but the influence of high temperature on vehicle braking distance and following safety distance is not considered, and the early warning threshold lacks dynamic adjustment capability. Some commercial vehicles are equipped with on-board diagnostic system (OBD) to monitor engine oil temperature and brake shoe temperature, and when the parameters exceed the limit, the driver is warned, but it is not associated with the overall risk of the road section, and the road section level of coordinated early warning and control cannot be realized.

[0003] The existing technology is mostly based on historical accident data to construct a static risk map, for example, by statistically analyzing the accident distribution of a road section during the high temperature period in the past three years, marking "high risk road section" and setting a fixed early warning threshold. Such mode does not consider the dynamic evolution characteristics of risk, such as the influence of road surface temperature gradient change and traffic flow "deceleration wave" propagation on risk during high temperature period, resulting in that the early warning result lags behind the actual risk change, and the risk diffusion trend cannot be predicted in advance. The existing early warning model mostly adopts the "one-size-fits-all" design idea, and does not optimize the thermal stress characteristics of different vehicle types. For example, heavy trucks have large self-weight and low brake system heat dissipation efficiency, and the braking distance increases by 20%-30% under 60℃ road surface temperature, while the braking distance of small cars only increases by 8%-12%, but the existing model uniformly adopts the same "road surface temperature ≥ 60℃ triggers early warning" standard, resulting in insufficient early warning for heavy trucks and excessive early warning for small cars, and the early warning accuracy and practicality are insufficient.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore includes information that does not constitute prior art known to those skilled in the art. SUMMARY

[0005] According to an aspect of the present application, a highway operation risk early warning method in a dry high temperature environment is provided, comprising: acquiring real-time meteorological data, road state data and traffic flow data along the highway, using a wavelet transform algorithm to eliminate extreme weather burst pulse interference and traffic flow instantaneous fluctuation noise, and generating a standardized meteorological-road-traffic correlation data sequence; after converting the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution map, subdividing the high temperature gradient section based on a regional grid encryption algorithm, combining the differences in heat stress response of different vehicle types, and constructing a vehicle type-specific highway operation risk prediction model; based on the vehicle type-specific highway operation risk prediction model, extracting risk weight factors of different road sections to construct a dynamic risk assessment matrix, performing collaborative calculation on multi-source monitoring data, and obtaining a key early warning indicator combination under a target risk level; extracting the risk peak value, risk diffusion rate and meteorological-risk matching deviation value according to the highway operation period, dividing the risk exceeding level, and generating a multi-dimensional risk characteristic matrix; comparing real-time risk data with historical data of the same period and same section, identifying abnormal signals of short-term risk increase and traffic flow-risk matching imbalance, using a risk-weather parameter correlation curve to generate a risk warning dynamic correction factor; grouping road sections according to the administrative level of the highway, the number of lanes and the surrounding environment, using a random forest algorithm to screen key influence factors of the multi-dimensional risk characteristic matrix and the risk warning dynamic correction factor, and constructing a personalized risk warning control model by fusing risk warning threshold, traffic control constraints and emergency response limit information; based on the target warning parameter output of the personalized risk warning control model, combining the time correlation information of the highway traffic flow time series variation law and the weather forecast, generating real-time dynamic warning instructions and time period-specific risk control strategies.

[0006] In another aspect of the present application, a highway operation risk early warning device in a dry high temperature environment comprises: an acquisition module for acquiring real-time meteorological data, road state data and traffic flow data along the highway, using a wavelet transform algorithm to eliminate extreme weather burst pulse interference and traffic flow instantaneous fluctuation noise, and generating a standardized meteorological-road-traffic correlation data sequence; a processing module for converting the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution map, then subdividing the high temperature gradient section based on a regional grid encryption algorithm, combining the differences in heat stress response of different vehicle types, and constructing a vehicle type-specific highway operation risk prediction model; based on the vehicle type-specific highway operation risk prediction model, extracting risk weight factors of different road sections to construct a dynamic risk assessment matrix, performing collaborative calculation on multi-source monitoring data, and obtaining a key early warning indicator combination under a target risk level; extracting risk peak value, risk diffusion rate and meteorological-risk matching deviation value according to the highway operation period, dividing risk exceeding level, and generating a multi-dimensional risk feature matrix; comparing real-time risk data with historical same period data of the same section to identify abnormal signals of short-term risk sudden increase and traffic flow-risk matching imbalance, using a risk-weather parameter correlation curve curvature to generate a risk early warning dynamic correction factor; grouping road sections according to highway administrative level, number of lanes and surrounding environment, using a random forest algorithm to screen key influence factors of the multi-dimensional risk feature matrix and the risk early warning dynamic correction factor, and constructing an individualized risk early warning control model by fusing risk early warning threshold, traffic control constraint conditions and emergency response limit information; based on the target early warning parameter output of the individualized risk early warning control model, combining the time correlation information of highway traffic flow time series variation law and meteorological forecast, generating real-time dynamic early warning instructions and time period-specific risk control strategies.

[0007] According to another aspect of the present application, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a second processor to implement the above-mentioned highway operation risk early warning method in a dry high temperature environment.

[0008] The application provides a highway operation risk early warning method and device in a dry high-temperature environment. The method comprises the following steps: acquiring meteorological, road and traffic data, and generating a standardized sequence by wavelet transform denoising; converting the standardized sequence into a three-dimensional risk map, simulating and encrypting the three-dimensional risk map, and combining a vehicle type heat stress constraint library to construct a vehicle type prediction model; extracting a risk weight factor to construct a dynamic matrix, and screening key early warning indexes; extracting risk characteristics according to time periods to generate a multi-dimensional matrix, comparing historical data to generate a dynamic correction factor; grouping road sections, screening factors by using a random forest, constructing a personalized early warning model, and outputting real-time early warning and time period control strategies. Multi-source data fusion and a vehicle type model improve the accuracy of early warning, a dynamic correction factor and a personalized model adapt to emergency situations, and time period control strategies combined with meteorological and traffic rules significantly reduce the accident rate in a high-temperature environment, while balancing control and traffic efficiency.

[0009] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 A flow chart of a highway operation risk early warning method in a dry high-temperature environment is shown. Figure 2 A structural schematic diagram of a highway operation risk early warning device in a dry high-temperature environment is shown. DETAILED DESCRIPTION

[0011] The preferred embodiments of the present application are described below with reference to the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0012] The highway operation risk early warning method in a dry high-temperature environment according to the exemplary embodiments of the present application is described below. Figure 1 In one embodiment, the present application further provides a highway operation risk early warning method and device in a dry high-temperature environment. S101, acquiring real-time meteorological data, road state data and traffic flow data along the highway, removing extreme weather burst pulse interference and traffic flow instantaneous fluctuation noise by using a wavelet transform algorithm, and generating a standardized meteorological-road-traffic correlation data sequence.

[0013] In one implementation, first, three types of key data need to be collected from different channels, which together constitute the basis of risk analysis. Meteorological data mainly rely on the network of automatic weather stations deployed along the highway. These weather stations are professional environmental monitoring equipment that can continuously and automatically record various meteorological parameters. Specifically, air temperature and humidity directly affect the road surface state and the physiological state of the driver; light intensity affects the driver's vision, especially at noon or in the evening; wind speed and direction affect the stability of vehicle driving, especially for large vehicles; air pressure helps to determine weather system changes. For example, at 10:00 am, the weather station located at K120+300 on G101 National Highway records: air temperature 38.5°C, relative humidity 25%, light intensity 800W / m², southwest wind 3m / s.

[0014] Road condition data is obtained through embedded road sensors and road detection vehicles. Embedded sensors monitor the key state of the road in real time, while detection vehicles conduct a more comprehensive physical examination of the entire road surface on a regular basis. Road surface temperature directly determines the friction coefficient between the tire and the ground and the braking effect; road surface roughness affects the stability and comfort of vehicle driving, indirectly affecting driving risk; road surface wear / damage degree, such as rutting, cracking, etc., is a direct hidden danger to road safety. At the same time and place as the meteorological data, a temperature sensor embedded 5cm below the road surface returns data: road surface temperature 65°C. The road detection vehicle's inspection data from a week ago shows that the road surface roughness index (IRI) of this section is 2.0m / km, with no obvious rutting and cracking.

[0015] Traffic flow data is obtained through video monitoring cameras and microwave radar detectors, which are installed on the horizontal poles above the road and can detect passing vehicles in real time. Vehicle flow is the total number of vehicles passing through a section in a unit of time, reflecting the degree of traffic congestion; average speed is used to reflect the overall operation efficiency and smoothness of the traffic flow; vehicle type distribution is used to reflect the proportion of different vehicle types (such as small cars, large buses, heavy trucks), which have significant differences in thermal stress response and risk characteristics; traffic density is the number of vehicles per unit length of road, which is a core indicator of congestion level. Also at 10:00 am, through video analysis of the K120+300 section, the system detects: the vehicle flow in the past hour is 2000 vehicles / hour, the average speed is 80km / h, among which small cars account for 70%, heavy trucks account for 20%, and other vehicle types account for 10%.

[0016] The original data usually contains various interference signals (noise) that need to be removed to ensure the accuracy of the analysis. Specifically, abnormal data points caused by sudden impulse interference such as lightning, sensor instantaneous failure, etc. Transient fluctuation noise such as short-term jitter of traffic flow data caused by instantaneous acceleration or deceleration of vehicles. Using wavelet transform algorithm, the signal is decomposed into different frequency scales, so as to accurately locate and separate the noise part. For example, due to the instantaneous failure of the sensor, a -5°C error value suddenly appears in the air temperature data obtained in the previous step. Wavelet transform will identify this "burr" that is incompatible with the normal temperature pattern (about 38°C), and smooth or remove it, correcting it to a reasonable value.

[0017] After denoising, the three types of data are integrated into a unified and orderly format for subsequent model use. The same time, the same place, the weather, the road, the traffic data are bound together to form a data point, and are arranged in chronological order. After integration, we get a standardized data sequence, where a data point at a certain time can be represented as: {time: '10:00', temperature: '38.5°C', road temperature: '65°C', traffic volume: '2000 vehicles / hour', average speed: '80 km / hour'}.

[0018] S102, after converting the standardized weather-road-traffic correlation data sequence into a three-dimensional risk load distribution map, based on the regional grid encryption algorithm, the high temperature gradient section is subdivided, combined with the differences in heat stress response of different vehicle types, and the vehicle type highway operation risk prediction model is constructed.

[0019] In one embodiment, the standardized weather-road-traffic correlation data sequence is converted into a three-dimensional risk load distribution map to generate basic risk visualization data. Each data point in the standardized data sequence (containing time, location, weather, traffic, etc. Information) is calculated to obtain the comprehensive risk value of the space-time point. Then, these risk values are mapped to the geographical space to form a visualization map with longitude, latitude, and risk value as three dimensions.

[0020] Based on the data at 10:00 am, the system calculates the risk value of the G101 National Highway K120-K121 km section. At K120.3 km, the risk value is 0.85 (assuming the risk value range is 0-1) due to high road surface temperature (65°C) and high traffic volume (2000 vehicles / hour); while at K120.8 km, the risk value is 0.6 due to lower road surface temperature (60°C) and smaller traffic volume (1500 vehicles / hour). Drawing these values continuously on the map forms a risk curve that appears "peak" at K120.3 km, which is a cross-section of the three-dimensional risk load distribution map.

[0021] The three-dimensional risk load distribution map is imported into the road environment and traffic flow coupling simulation software for simulation calculation to generate multi-dimensional risk field simulation data. To break through the limitation of static risk map which can only present the risk distribution at a certain point in time, professional road environment and traffic flow coupling simulation software is needed to simulate the dynamic evolution process of risk in real traffic flow to obtain multi-dimensional risk field simulation data containing time dimension. Specifically, the simulation software needs to load two types of core data to build a virtual road environment highly consistent with reality. Static environment data includes the following: road geometry information includes lane number, curve, slope, speed limit, etc. of K120 km road section. The three-dimensional risk map gives an initial risk value to every inch of the virtual road. K120.3 km is a red "mountain peak" (risk value 0.85), and other areas are "mountains" or "plains" of different heights. Dynamic traffic data includes the following: real-time traffic flow includes vehicle distribution at the start time (e.g. 10:00) of the simulation. For example, there is a heavy truck at K120.0 km, three small cars are driving in sequence at K120.1 km, and the traffic density is 2000 vehicles / hour, etc. These vehicles are initialized as "agents" with different physical properties (such as braking performance, acceleration) by the software.

[0022] The simulation software starts running based on the preset vehicle behavior model (such as the car-following model, lane-changing model). Each virtual vehicle will make real-time decisions to accelerate, decelerate or maintain a constant speed based on the behavior of the vehicle in front of it, the risk condition of the road surface, and its own physical limitations. The simulation clock starts counting (t=0s), all vehicles travel at the initial speed, and the risk value of the road surface at K120.3 km is the highest (0.85). At t=10s, a heavy truck arrives at the high-risk point at K120.3 km. The built-in vehicle model of the software takes into account the braking performance decay caused by high temperature (from the vehicle model constraint library) and combines the high-risk road surface conditions to calculate that its safe braking distance becomes longer. In order to maintain a safe following distance, the model decides to start decelerating earlier and more gently than usual. The chain reaction (deceleration wave formation) occurs at t=15s, the first small car behind the truck detects the deceleration of the front vehicle and immediately starts its car-following model to begin deceleration. At t=20s and t=25s, the second and third small cars also respond to the deceleration in turn. This response of the following cars to the behavior of the front car forms a "deceleration wave" that propagates backward.

[0023] The risk assessment module of the software runs continuously. Due to the emergence of the "deceleration wave", the speed of vehicles in the 500-meter range from K120.3 km to K120.8 km drops from 80 km / h to 60 km / h, the time interval between vehicles becomes smaller, and the risk of collision increases significantly. Therefore, the software updates the risk value of this area in real time. After a period of simulation (for example, 2 minutes), the software outputs the results. Instead of a single atlas, the results are multi-dimensional risk field simulation data that include the time dimension. Spatial slice (T = 2 minutes): After 2 minutes of simulation, the software outputs a new risk atlas. The results show that the initial high-risk "peak" concentrated at K120.3 km has spread into a high-risk "platform area" from K120.3 km to K120.8 km. Time series (specific point): The software can also output the risk value curve of K120.5 km, a fixed point, over time. The curve shows that at the beginning of the simulation (t = 0), the risk value here is 0.55; when the "deceleration wave" reaches this point (about t = 40 s), the risk value begins to rise rapidly; by t = 2 minutes, the risk value stabilizes at 0.70. The above is the multi-dimensional risk field simulation data generated, which contains information on the evolution of risk over time and space.

[0024] In the risk warning system for high-speed highway operation in dry and high-temperature environments, the core goal of the "regional grid densification algorithm" is to solve the contradiction between "risk field simulation accuracy and computational resource efficiency" - both to achieve fine risk capture in key road sections with high temperature gradients and high risks, and to avoid resource waste in low-risk areas due to excessive calculation. Its technical positioning is "a spatial optimization engine for multi-dimensional risk field simulation data", which dynamically adjusts the grid density to provide high-resolution spatial risk data support for subsequent vehicle-type risk prediction models, ensuring that the models can accurately identify the impact of details such as "local road temperature sudden change" and "vehicle thermal stress response differences" on risk.

[0025] The sub-regional grid encryption algorithm is based on the "multi-dimensional risk field simulation data" generated by the "road environment and traffic flow coupling simulation software" described above, which includes three core dimensions of spatial dimension (road segment longitude, latitude, and altitude), environmental dimension (road surface temperature, temperature gradient, and air humidity), and traffic flow dimension (vehicle flow density, vehicle type distribution, and average vehicle speed). To ensure the effective operation of the algorithm, two key preprocessing of input data are required. For data space-time alignment, non-synchronous data collected by different monitoring devices (such as embedded road sensors, weather stations, and video surveillance) are aligned according to the "5-minute time granularity + 10-meter spatial granularity" to avoid errors in temperature gradient calculation due to data space-time deviation. For example, it is necessary to ensure that the road surface temperature data at K120.2 km at 13:00 corresponds to the temperature data at K120.3 km at 13:00, rather than 13:01 or 12:59. For outlier rejection, the 3σ criterion (3 times the standard deviation under normal distribution) is used to remove extreme outliers in road surface temperature data (such as "-5°C" or "100°C" error values caused by sensor failure), and linear interpolation is used to fill in missing data segments (such as filling in the temperature mean of the previous and next 20-meter road segments), ensuring the continuity and accuracy of temperature gradient calculation.

[0026] For real-time calculation of temperature gradient change rate, the algorithm first performs "grid-by-grid temperature gradient scanning" on the preprocessed multi-dimensional risk field simulation data. The core is to calculate the temperature change rate of each grid and its adjacent grid. The specific steps are as follows: taking "20 meters x 20 meters" as the basic grid unit, the target highway road section (such as G101 National Highway K120-K121 km) is covered, each grid is assigned a unique spatial coordinate (such as "Grid ID: G101-K120-001, coordinates: longitude X1, latitude Y1"), and the real-time road surface temperature (such as 65°C), traffic flow density (such as 12 vehicles / km) and other data in the grid are associated. For each basic grid, calculate the temperature difference of its east, west, south, and north adjacent grids, and then divide by the grid spacing (20 meters) to get the "eastward temperature gradient", "westward temperature gradient", and other four direction gradient values. The absolute value of the largest gradient value is taken as the "dominant temperature gradient" of the grid. For example, a basic grid at K120.2 km (temperature 65°C), the east adjacent grid (K120.2+20 meters) temperature is 63°C, the temperature difference is 2°C, the eastward temperature gradient = 2°C ÷ 20 meters = 0.1°C / m; the west adjacent grid temperature is 65°C, the westward gradient is 0°C / m, so the dominant temperature gradient of the grid is 0.1°C / m.

[0027] A preset temperature gradient change rate threshold (which can be dynamically adjusted according to historical data of the road section, and the default threshold is "0.05°C / meter", that is, the temperature of the road surface changes by 5°C per 100 meters). If the dominant temperature gradient of a certain grid exceeds the threshold, the grid and the adjacent two grids are marked as "high temperature gradient candidate area"; if it does not exceed the threshold, it is determined as "low temperature gradient area". For example, the dominant gradient of the K120.2 km grid is 0.1°C / meter (exceeding the threshold of 0.05°C / meter), and the grid and the east and west grids (a total of 3 basic grids, covering a 60-meter road section) are marked as candidate areas.

[0028] For the marked "high temperature gradient candidate area", the algorithm starts the automatic encryption process, and realizes fine processing through "grid segmentation-precision improvement-data redistribution". To avoid false encryption caused by instantaneous temperature fluctuations, the algorithm sets a "double condition triggering" mechanism: first, "the dominant temperature gradient of the candidate area exceeds the threshold within 2 consecutive time granules (a total of 10 minutes)"; second, "the traffic flow density in the candidate area is ≥8 vehicles / km (that is, there is a certain traffic flow, which needs to be accurately evaluated for risk)". If only one condition is met, encryption is not performed, and the next time granule data is monitored. For example, the K120.2-K120.4 km road section (a total of 200 meters, containing 10 basic grids), within the two time granules of 13:00-13:10, the dominant gradient of all grids is ≥0.06°C / meter, and the traffic flow density is 10 vehicles / km, which meets the double conditions and triggers encryption.

[0029] For the high temperature gradient road section that meets the conditions, the original "20m x 20m" basic grid is divided by a ratio of "1:4", that is, 1 basic grid is divided into 4x4=16 "5m x 5m" encryption grids. For example, the original 20m x 20m grid at K120.2 km is divided into 16 5m x 5m small grids, each small grid is assigned a new ID (such as "G101-K120-001-001" to "G101-K120-001-016"), and its accurate spatial coordinates are recorded (such as the coordinates of the southeast corner of the original grid: X1+5m, Y1+5m).

[0030] The "spatial interpolation method" is used to accurately distribute the original basic grid temperature, traffic flow and other data to each encrypted grid. Taking temperature data as an example, if the original basic grid temperature is 65°C and the temperature of the adjacent basic grid to the east is 63°C, then through linear interpolation calculation: the temperature of the encrypted grid 5 meters away from the east boundary in the original grid = 65°C - (0.05°C / meter x 5 meters) = 64.75°C; the temperature of the encrypted grid 10 meters away from the east boundary in the original grid = 65°C - (0.05°C / meter x 10 meters) = 64.5°C, ensuring the continuity of the temperature gradient after encryption. Traffic flow data is allocated according to the "encrypted grid area ratio". There are 12 vehicles in the original basic grid, and each encrypted grid is allocated 12÷16≈1 vehicle (less than 1 vehicle is rounded to "0.75 vehicles" and other decimals, which is used for subsequent risk probability calculation).

[0031] For low-risk areas that do not exceed the temperature gradient threshold (such as K120.5-K120.8 km section), the algorithm uses the "low-density grid retention strategy" to avoid wasting computing resources. Continue to use the original "20m x 20m" basic grid without segmentation, and only update the temperature and traffic flow data in the grid at a 5-minute time granularity (such as the temperature of a certain grid rising from 60°C to 61°C, and the traffic flow density decreasing from 8 vehicles / km to 7 vehicles / km). For areas that have maintained a low temperature gradient for 3 consecutive time granularities (15 minutes), the data update frequency is reduced from "5 minutes / time" to "10 minutes / time", further reducing the computing load; if the temperature gradient exceeds the threshold in subsequent monitoring, the 5-minute update frequency is restored immediately, and the encryption evaluation process is started. In this way, the system can simulate the risk changes of high-risk areas with higher accuracy, while avoiding wasting computing resources in low-risk areas.

[0032] Collect the difference parameters of thermal stress response of different vehicle models in dry high-temperature environment, construct a vehicle-specific thermal stress characteristic constraint library, and generate vehicle constraint data. The core of constructing a vehicle-specific thermal stress characteristic constraint library is to realize the refinement and differentiation of risk assessment: quantify the impact of high temperature on vehicles (such as brake distance increases X%, friction coefficient decreases Y%), and combine the differences in physical and power between heavy trucks and sedans to set different risk trigger conditions and parameter decay curves. The raw data collected (such as "brake distance increases by 30% at 60°C") need to be converted into mathematical models or look-up tables that computers can understand and call. In one implementation, a table is directly constructed with "road surface temperature" as the index, corresponding to "brake distance increase percentage" and "friction coefficient decrease percentage". In another implementation, the relationship may not be linear. For example, when the temperature rises from 60°C to 65°C, the increase in brake distance may be much greater than when the temperature rises from 55°C to 60°C. At this time, other nonlinear functions (such as exponential functions, polynomial functions) are used to fit the experimental data to establish a continuous and calculable model. Specifically, the heavy truck brake distance model is D_truck(T)=D_0 (1+0.006 (T-25)) where T is the road surface temperature (°C) and is applicable when T>=25°C. This model indicates that for every 1°C increase in temperature, the brake distance increases by 0.6%.

[0033] All verified mathematical models or look-up tables for different vehicle models are stored in a structured database, i.e., a vehicle-specific thermal stress characteristic constraint library. An example of the library structure is as follows: vehicle ID: "HeavyTruck_ModelA"; brake performance model: [look-up table or function expression]; tire friction coefficient model: [look-up table or function expression]; heat dissipation performance parameters: [heat dissipation efficiency curve]; vehicle ID: "Sedan_ModelB"; brake performance model: [look-up table or function expression]; tire friction coefficient model: [look-up table or function expression]; heat dissipation performance parameters: [heat dissipation efficiency curve]. In subsequent vehicle-specific risk prediction models, when the risk at a specific location and time needs to be evaluated, the model will obtain the real-time road surface temperature at that location (e.g., 65°C at K120.3 km). According to the composition of vehicles on the current section (e.g., a heavy truck is detected), the corresponding "heavy truck" model is called from the constraint library. The road surface temperature of 65°C is input into the brake model and tire friction model of the "heavy truck" to calculate its current brake distance and friction coefficient. Using these high-temperature corrected vehicle performance parameters, the probability of an accident is calculated to obtain the accurate risk value of the vehicle model under the current conditions.

[0034] The differentiated grid data and vehicle model constraint data are fused to construct a vehicle model-specific highway operation risk prediction model. By deeply fusing the refined road section risk space data and the vehicle model-specific high-temperature response parameters, a prediction system for accurately evaluating the risks of different vehicle models is constructed by means of a machine learning model. From the data of the regionally encrypted grid, the core features strongly related to risk assessment are extracted. For the K120 kilometer section of the G101 national road, the features to be extracted from the encrypted 5m x 5m grid include: the real-time road surface temperature in the grid (such as 65°C at K120.3 kilometers), the temperature gradient change rate (the temperature difference between the grid and the adjacent grid is 2°C / 5m), the initial risk value corresponding to the grid (0.85 based on the three-dimensional risk load spectrum), and the instantaneous traffic flow density in the grid (such as 12 vehicles / km). The spatial coordinate information of the grid (such as longitude X1 and latitude Y1) is retained during extraction to ensure accurate matching with the actual road section location in the future.

[0035] The parameters in the vehicle-specific heat stress characteristic constraint library are converted into quantifiable features recognizable by the model. Taking heavy trucks and small cars as examples, for a road surface temperature of 65°C at K120.3 kilometers, the model calculates that: for heavy trucks, the braking distance increases by 30% (original braking distance 60 meters, revised 78 meters), and the tire friction coefficient decreases by 15% (original 0.8, revised 0.68); for small cars, the braking distance increases by 15% (original 40 meters, revised 46 meters), and the tire friction coefficient decreases by 8% (original 0.85, revised 0.78). At the same time, each vehicle model is assigned a unique identifier (such as "Truck_01" for heavy trucks and "Car_01" for small cars) to facilitate model differentiation.

[0036] The grid feature and vehicle constraint parameter in the same space-time dimension are bound by taking "grid spatial coordinates + time stamp" as the association key. For example, at 10:00, the grid data at K120.3 kilometers with coordinates (X1, Y1) is fused with the heavy truck constraint parameters to form a complete sample data: {grid coordinates (X1, Y1), time 10:00, road surface temperature 65°C, temperature gradient 2°C / 5m, initial risk value 0.85, traffic flow density 12 vehicles / km, vehicle model identifier Truck_01, braking distance correction coefficient 1.3, friction coefficient 0.68}. At the same time, the continuous features (such as road surface temperature and risk value) are uniformly mapped to the [0, 1] interval by using the range standardization method to ensure that the influence of different orders of magnitude features on model training is balanced.

[0037] In view of the nonlinear and multi-factor coupling characteristics of highway risks in dry and high-temperature environments, a gradient boosting decision tree (GBDT) is selected as the basic model architecture, and independent models are constructed for heavy trucks and small cars respectively. The model hierarchy and structure design is as follows: a GBDT model based on the LightGBM framework is used, which iteratively trains multiple decision trees. Each new tree fits the prediction residual of all previous trees, effectively capturing the complex nonlinear relationship between risk factors (such as road temperature, braking performance) and accident probability. Moreover, it has the advantages of fast training speed and strong adaptability to high-dimensional data, making it suitable for processing fused multi-feature data.

[0038] The model hierarchy is divided as follows: the input layer is used to receive preprocessed fused data, with a total of 12 feature dimensions, including grid space features (2 dimensions: longitude, latitude), environmental features (3 dimensions: road temperature, temperature gradient, air humidity), traffic flow features (2 dimensions: traffic flow density, average vehicle speed), vehicle performance features (3 dimensions: braking distance correction coefficient, tire friction coefficient, power attenuation rate), time features (1 dimension: hour-level timestamp), and initial risk features (1 dimension: grid initial risk value). The feature interaction layer uses the automatic feature interaction function of LightGBM to mine hidden feature correlation relationships. For example, the interaction feature of "road temperature x braking distance correction coefficient" is automatically calculated to reflect the impact of high temperature and vehicle braking performance coupling on risk; the interaction feature of "traffic flow density x tire friction coefficient" is calculated to reflect the risk changes under the combined action of traffic congestion and vehicle grip. The decision tree ensemble layer consists of 100 CART regression trees, with a maximum depth of 8 for each tree (to avoid overfitting) and a leaf node limit of 31. During training, a gradient descent strategy is used, with each tree optimizing the residual of the previous model. For example, the first tree primarily fits the relationship between risk and road temperature, and the second tree focuses on correcting the prediction bias caused by the "temperature gradient + friction coefficient" interaction term, gradually improving the model accuracy. The output layer outputs the risk probability value (range 0-1) of the corresponding vehicle type under the current grid conditions, such as a risk probability of 12% for a heavy truck and 5% for a small car in a certain grid at K120.3 km.

[0039] The key hyperparameters are set as follows: learning rate: 0.05 (controls the contribution of each tree to the model, avoiding excessive influence of a single tree); subsample ratio: 0.8 (randomly samples 80% of the samples for each tree during training, enhancing the model's generalization ability); colsample_bytree ratio: 0.8 (randomly selects 80% of the features for each tree during training, reducing the impact of feature redundancy); regularization parameter (lambda): 0.1 (uses L2 regularization to suppress model overfitting); random seed: 42 (ensures that the model training results are reproducible).

[0040] The model of each vehicle type can adapt to its unique risk characteristics, avoiding cross-vehicle data interference, by using the mode of "separate vehicle type independent training + cross-validation optimization". The historical data of G101 National Highway K120-K121 km section in the past 3 years is collected, and divided into training set and validation set according to the ratio of 7:3. The training set contains 100,000 samples, covering grid data and vehicle constraint parameters under different seasons, time periods and temperature conditions, and ensuring that the sample ratio of heavy trucks to small cars is 1:3 (matching the actual vehicle distribution of this section); the validation set contains 43,000 samples, which should be consistent with the spatio-temporal distribution characteristics of the training set, such as the sample proportion of high-temperature period (10:00-16:00) being 60%, which is consistent with the actual high-temperature risk period.

[0041] Heavy truck model training: Taking "heavy truck accident label" as the target variable (accident occurs as 1, and no accident as 0), the samples in the training set with vehicle type identifier "Truck_01" are input into the GBDT model. During the training process, the model learns through iteration that when the road surface temperature exceeds 60°C and the braking distance correction coefficient is greater than 1.2, the accident probability will increase significantly, such as a sample at K120.3 km. Due to the satisfaction of the above conditions, the model predicts the risk probability from 8% to 12%. Small car model training: The samples with vehicle type identifier "Car_01" are trained separately, and the model will capture that the sensitivity of small cars to temperature is lower than that of heavy trucks. When the road surface temperature is 65°C, only when the traffic flow density exceeds 15 vehicles / km, the risk probability will increase from 3% to 5%, which is consistent with the characteristics that the braking performance of small cars is less affected by high temperature.

[0042] S103, based on the risk prediction model of different road sections, the risk weight factor is extracted to construct a dynamic risk assessment matrix, and the multi-source monitoring data is calculated to obtain the key warning index combination under the target risk level.

[0043] In one embodiment, the risk weight factor of different road sections is extracted based on the vehicle type-based highway operation risk prediction model to construct a dynamic risk assessment matrix. From the output results of the vehicle type-based risk prediction model, the core factors reflecting the risk contribution degree of different road sections are extracted and quantitatively presented in matrix form. Taking "road section temperature gradient risk degree, traffic flow pressure intensity, and historical accident severity" as the three core dimensions, the weight of each dimension is calculated in combination with the prediction results of the vehicle type-based model. Taking the K120-K121 kilometer section of G101 National Road as an example: based on the grid temperature data in the vehicle type-based model, the temperature gradient in the road section is calculated (such as 2°C / 5m at K120.3 kilometer and 0.5°C / 5m at K120.8 kilometer), and then combined with the temperature influence coefficient of vehicle type (0.8 for heavy trucks and 0.5 for small cars), the temperature gradient risk degree is obtained. The temperature gradient risk degree of heavy trucks at K120.3 kilometer is 2x0.8=1.6, and that of small cars is 2x0.5=1.0.

[0044] According to the traffic flow density data in the vehicle type-based model (12 vehicles / km at K120.3 kilometer and 8 vehicles / km at K120.8 kilometer), combined with the vehicle type passing efficiency parameter (unit vehicle flow pressure coefficient 1.2 for heavy trucks and 0.9 for small cars), the traffic flow pressure intensity of heavy trucks at K120.3 kilometer is calculated as 12x1.2=14.4, and that of small cars is 8x0.9=7.2; the historical accident severity average value is calculated by statistically analyzing the accident data of the same period in the past three years on this road section, and assigning weights (1, 3, and 5) according to the accident grade (minor, general, and major). At K120.3 kilometer, 2 general accidents and 1 minor accident occurred in the same period in the past three years, and the average severity is (3x2+1x1) / 3≈2.33; at K120.8 kilometer, only 1 minor accident occurred, and the average value is 0.33.

[0045] Taking "road section-vehicle type" as the row and "core dimension" as the column, the calculated risk weight factors are filled into the matrix to form a dynamically updated assessment table. For example, the dynamic risk assessment matrix (simplified version) of the K120-K121 kilometer section of G101 National Road: at 10:00, the temperature gradient risk degree of heavy trucks at K120.3 kilometer is 1.6, the traffic flow pressure intensity is 14.4, and the historical accident severity is 2.33; the corresponding values for small cars at K120.3 kilometer are 1.0, 7.2, and 2.33; the corresponding values for heavy trucks at K120.8 kilometer are 0.4 (temperature gradient 0.5x0.8), 9.6 (8x1.2), and 0.33; the corresponding values for small cars at K120.8 kilometer are 0.25 (0.5x0.5), 7.2 (8x0.9), and 0.33. The matrix will be updated every 5 minutes as the real-time data (such as temperature and traffic flow) changes, ensuring the dynamic nature of the assessment.

[0046] The dynamic risk assessment matrix is input into the multi-source data collaborative computing module, which synchronously fuses and analyzes multi-dimensional monitoring data such as weather, road, and traffic, to generate multiple candidate warning indicator sets. Through a distributed computing architecture, the multi-dimensional monitoring data is fused and processed, potential warning indicators are mined from different angles, and multiple candidate sets are formed to provide sufficient options for subsequent screening. The multi-source data collaborative computing module accesses real-time monitoring data of three types of weather, road, and traffic, and carries out parallel preprocessing according to the characteristics of different data sources. The weather data includes temperature (e.g., 38.6°C at K120.3 km at 10:01), humidity (24%), and wind speed (3.2 m / s) collected every 1 minute by automatic weather stations along the line, and abnormal values caused by transient sensor failures (e.g., 25°C at 10:02) need to be removed. Interpolation is used to supplement data missing periods (e.g., 38.7°C at 10:03, which is the average of the previous and next 1-minute values). Road data includes road surface temperature (K120.3 km at 10:02, 65.2°C) and roughness (IRI value 2.0) returned every 2 minutes by embedded sensors, and system errors in road detection vehicle inspection data (e.g., IRI value anomaly 1.2 caused by device calibration bias) need to be filtered. Traffic data includes vehicle flow (K120.3 km at 10:02, 2050 vehicles / hour), average speed (78 km / h), and vehicle type ratio (heavy truck 22%) counted every 30 seconds by video monitoring, and vehicle flow fluctuations caused by transient acceleration and deceleration of vehicles (e.g., vehicle flow suddenly rises to 2200 vehicles / hour at 10:02:15, which is corrected to 2080 vehicles / hour by moving average).

[0047] The module uses a distributed computing architecture (e.g., a parallel computing framework based on Hadoop) to fuse the preprocessed data in multiple dimensions and generate multiple candidate warning indicator sets. Candidate indicators based on weather-road fusion: e.g., "road temperature-air temperature difference" (K120.3 km, 65.2-38.6=26.6°C), "temperature gradient-wind speed product" (2°C / 5m x 3.2m / s=1.28°C•m / (s•m)); candidate indicators based on traffic-risk fusion: e.g., "traffic flow pressure intensity-average speed ratio" (K120.3 km, heavy truck 14.4 / 78≈0.18), "vehicle type ratio-historical accident severity product" (heavy truck 22% x 2.33≈0.51); candidate indicators based on multi-dimensional integration: e.g., the sum of "temperature gradient risk degree + traffic flow pressure intensity + historical accident severity" (K120.3 km, heavy truck 1.6+14.4+2.33≈18.33), "variation coefficient of each dimension weight factor" (measures the stability of the indicator, K120.3 km, heavy truck three-dimensional variation coefficient≈1.2). Finally, five candidate warning indicator sets are generated, each containing 8-10 indicators covering different risk influencing angles.

[0048] Based on the preset target risk level standard, the candidate early warning indicator set is screened and optimized to obtain the key early warning indicator combination under the target risk level. According to the preset risk level standard, the key indicators with both accuracy and low false alarm rate are selected from the candidate indicator set by constructing a double-target evaluation function and combining sensitivity analysis, to ensure the effectiveness of early warning. Referring to industry standards and actual road conditions, the risk level is divided into three levels of low (0-5), medium (5-10), and high (10+), corresponding to different early warning thresholds. For example, for the K120 km section of G101 National Highway, the high risk level standard for heavy trucks is "dynamic risk assessment matrix three-dimensional sum ≥ 15", the medium risk is "10-15", and the low risk is "<10"; the high risk for small cars is "≥10", the medium risk is "5-10", and the low risk is "<5". An evaluation function with "risk identification accuracy (TPR)" and "false alarm rate (FPR)" as targets is constructed to test each candidate indicator set: TPR = number of correctly warned high-risk samples / actual high-risk samples, FPR = number of low-risk samples misjudged as high-risk / actual low-risk samples. The ideal target is TPR ≥ 90%, FPR ≤ 5%.

[0049] Taking "road surface temperature-air temperature difference ≥ 25°C and traffic flow pressure intensity ≥ 12" (for heavy trucks) in a candidate indicator set as an example, 1000 sample data of the same period in the last month are selected for testing: among them, there are 100 actual high-risk samples, 92 of which are correctly warned, TPR = 92%; 900 actual low-risk samples, 40 of which are misjudged, FPR ≈ 4.4%, meeting the ideal target; another indicator "historical accident severity ≥ 2", TPR is only 65% (most high-risk samples are not identified due to insufficient historical data) and FPR = 2%, which does not meet the requirements and is preliminarily excluded. Through this function, 3 groups of 15 indicators are retained from 5 candidate indicator sets, which enter the next step of sensitivity analysis.

[0050] Sensitivity test was conducted on the 15 retained indicators to analyze the impact of small fluctuations in the indicators on the risk assessment results, and to eliminate indicators with poor stability. A ±5% small fluctuation was applied to the indicator values, and the change rate of the risk level determination results was observed (change rate = number of samples with changed results / total number of test samples). For the "road surface temperature-air temperature difference ≥ 25°C" indicator, 500 samples were selected, and the difference was fluctuated by ±5% (e.g. 26.6°C changed to 25.27°C or 27.93°C). The test found that only 3 samples changed in risk level determination results (from high risk to medium risk), with a change rate of 0.6%, low sensitivity and good stability; another indicator "wind speed ≥ 3 m / s" had 45 samples with changed determination results after fluctuation (change rate 9%), and the wind speed was easily affected by the instantaneous airflow and fluctuated greatly, so it was eliminated due to poor stability. After sensitivity analysis, 10 indicators with good stability were selected from the 15 indicators, forming 3 sets of optimized candidate early warning indicator sets, each containing 3-4 complementary indicators (e.g. one set contains "road surface temperature-air temperature difference ≥ 25°C" "traffic flow pressure intensity ≥ 12" "historical accident severity ≥ 2").

[0051] In combination with the preset target risk level (e.g. high, medium, low), the optimized candidate indicator set was finally screened to determine the key early warning indicators corresponding to different risk levels, forming a directly applicable early warning basis. For each candidate early warning indicator set, its adaptability under different target risk levels was tested. High risk level test: with "correctly identifying high risk and avoiding false alarms" as the core goal, the identification ability of the candidate set for high risk samples was tested. For example, for the candidate set containing "road surface temperature-air temperature difference ≥ 25°C" "traffic flow pressure intensity ≥ 12" "historical accident severity ≥ 2", 200 high risk samples in the same period in the last 3 months were selected for testing, and the test found that the identification accuracy rate was 93% and the false alarm rate was only 7%, meeting the high risk level warning requirements.

[0052] Medium risk level test: balancing identification accuracy and false alarm rate, the candidate set's ability to distinguish medium risk samples was tested. A candidate set contains "road surface temperature-air temperature difference 20-25°C" "traffic flow pressure intensity 8-12" "historical accident severity 0.5-2", and the test identified 150 medium risk samples with an accuracy rate of 88% and a false alarm rate of 6%, meeting the medium risk warning requirements. Low risk level test: with "reducing false alarm rate and unnecessary control" as the goal, the candidate set's ability to exclude low risk samples was tested. A candidate set contains "road surface temperature-air temperature difference < 20°C" "traffic flow pressure intensity < 8" "historical accident severity < 0.5", and the test on 1000 low risk samples had a false alarm rate of only 2.3%, meeting the low risk warning requirements.

[0053] Based on the test results, optimal indicator combinations were matched for different target risk levels. For high-risk levels: the key warning indicator combination was determined to be "road surface temperature - air temperature difference ≥ 25°C + traffic flow pressure intensity ≥ 12 + historical accident severity ≥ 2". When a heavy truck at kilometer 120.3 of G101 National Highway simultaneously meets these three indicators (e.g., at 10:05, the difference is 26.8°C, pressure intensity is 14.8, and severity is 2.33), it is judged as high-risk, triggering a Level 1 warning. For medium-risk levels: the combination is "road surface temperature - air temperature difference 20-25°C + traffic flow pressure intensity 8-12 + historical accident severity 0". "5-2", for example, at K120.8 km, a heavy truck at 10:05 has a time difference of 22°C, pressure intensity of 9.8, and severity of 0.33 (due to historical severity not meeting the standard, other supplementary indicators are needed). After supplementing "average vehicle speed ≤ 75 km / h", it is judged as medium risk and triggers a level 2 warning. Low risk level: combination is "road surface temperature - air temperature difference < 20°C + traffic flow pressure intensity < 8 + historical accident severity < 0.5". For example, at K120.8 km, a small car at 10:05 has a time difference of 18°C, pressure intensity of 7.0, and severity of 0.33. It is judged as low risk and does not need to trigger a warning.

[0054] S104 extracts risk peak value, risk diffusion rate, and weather-risk matching deviation value according to the highway operating period, classifies the risk exceeding level, and generates a multi-dimensional risk feature matrix.

[0055] In one implementation, by extracting dynamic risk features and classifying risk levels according to time periods, the preceding static indicators are upgraded into structured risk data containing a "time dimension," providing support for subsequent precise early warning and personalized control. The specific processing flow is as follows: Based on the traffic flow patterns and meteorological parameter changes on highways under dry and high-temperature conditions, the entire day is divided into multiple typical operating periods. For each period, three core features are extracted: "risk peak, risk diffusion rate, and meteorological-risk matching deviation value," to achieve dynamic characterization of risk. Referring to historical highway traffic flow data and the distribution of high-temperature periods, the entire day is divided into five typical periods: morning peak (7:00-9:00), pre-morning (9:00-12:00), midday high temperature (12:00-14:00), afternoon (14:00-17:00), and evening peak (17:00-19:00). Taking the K120 km section of G101 National Highway as an example, the core environment and traffic characteristics differ significantly at different times: during the morning rush hour, the traffic volume reaches 3,000 vehicles / hour, but the road surface temperature is only 58°C; during the midday high-temperature period, the traffic volume drops to 1,800 vehicles / hour, but the road surface temperature rises to 68°C; during the evening rush hour, the light intensity decreases, and the humidity rises from 25% to 35%, and the relationship between meteorological conditions and risks changes.

[0056] The risk peak refers to the maximum value of the risk value of a road section in a certain period, which is calculated based on the real-time risk data output by the vehicle type-specific risk prediction model. For example, for the G101 National Highway K120.3 km section, the heavy truck risk value gradually increases from 0.75 to 0.85 during the noon high temperature period (12:00-14:00), and reaches the maximum value of 0.85 at 13:30, which is the risk peak of heavy trucks during the noon period. During the morning rush hour (7:00-9:00), due to heavy traffic, the risk peak of small cars reaches 0.7 (occurring at 8:15), which is higher than the risk peak of small cars during the noon period (0.65). The risk diffusion rate refers to the average rate at which the risk value increases from the initial value of the period to the risk peak, reflecting the speed of risk spread, and the calculation formula is "(risk peak - initial risk value of the period) / rising time". Taking the K120.3 km section as an example, the initial risk value of heavy trucks during the morning rush hour (7:00) is 0.5, which increases to the risk peak of 0.8 at 8:00, with a rising time of 60 minutes, and the risk diffusion rate is (0.8-0.5) / 60=0.005 / min; during the noon high temperature period (12:00), the initial risk value is 0.7, which increases to the peak value of 0.85 at 13:30, with a rising time of 90 minutes, and the diffusion rate is (0.85-0.7) / 90≈0.0017 / min, which is significantly lower than that during the morning rush hour, reflecting the key influence of traffic volume on risk diffusion. The meteorological-risk matching deviation value refers to the difference between the "theoretical risk value" calculated based on real-time meteorological parameters (such as road surface temperature, humidity) and the actual risk value, reflecting the matching degree of meteorological conditions and risk, and the calculation formula is "actual risk value - theoretical risk value" (theoretical risk value is calculated by the historical meteorological-risk correlation model). For example, for the K120.8 km section during the evening rush hour (17:00-19:00), the real-time road surface temperature is 62°C and the humidity is 35% at 18:00, and the theoretical risk value calculated based on the historical model is 0.7; but the traffic volume during this period decreases to 1200 vehicles / hour, and the actual risk value is only 0.6, the meteorological-risk matching deviation value is 0.6-0.7=-0.1, indicating that "the meteorological parameters indicate high risk but the actual risk does not meet the expectation", which needs to be adjusted by subsequent dynamic correction factors.

[0057] The threshold values of the three core risk features for each period are set according to the industry safety standards and historical accident data on the road section. Combined with the "single feature exceeding" and "multiple feature synergistic exceeding" rules, the risk is divided into three exceeding levels: low, medium, and high, and the risk severity in different periods is determined. The feature threshold values are set as follows: based on the data statistics of the K120 kilometer section of G101 National Highway in the high temperature period in the past three years, the threshold values of the core features in each period are set (for example): the risk peak value is greater than or equal to 0.8, the risk diffusion rate is greater than or equal to 0.004 / minute, and the absolute value of the meteorological-risk matching deviation is greater than or equal to 0.15. The risk peak value is greater than or equal to 0.85, the risk diffusion rate is greater than or equal to 0.002 / minute, and the absolute value of the meteorological-risk matching deviation is greater than or equal to 0.2. The risk peak value is greater than or equal to 0.75, the risk diffusion rate is greater than or equal to 0.003 / minute, and the absolute value of the meteorological-risk matching deviation is greater than or equal to 0.18. The threshold values should reflect the differences between periods, such as the risk peak value threshold in the high temperature period is higher than that in the morning peak period due to the high road surface temperature, and the risk diffusion rate threshold in the morning peak period is higher than that in the afternoon due to the dense traffic flow.

[0058] The risk exceeding level division rules are as follows: low exceeding level: only one feature exceeds, or no feature exceeds. For example, the risk peak value of the K120.3 kilometer section in the morning peak period is 0.78 (not exceeding), the diffusion rate is 0.0035 / minute (not exceeding), and the deviation value is -0.1 (not exceeding), which is determined as low exceeding level. Medium exceeding level: two features exceed. For example, the risk peak value of the K120.3 kilometer section in the high temperature period is 0.86 (exceeding), the diffusion rate is 0.0025 / minute (exceeding), and the deviation value is -0.1 (not exceeding), which meets the two features exceeding, and is determined as medium exceeding level. High exceeding level: three features exceed, or the core features (risk peak value + diffusion rate) exceed and the absolute value of the deviation value is greater than or equal to 0.2. For example, the risk peak value of the K120.1 kilometer section in the morning peak period is 0.82 (exceeding), the diffusion rate is 0.0045 / minute (exceeding), and the deviation value is 0.18 (exceeding), which meets the three features exceeding, and is determined as high exceeding level.

[0059] The extracted risk features and divided exceeding levels are integrated into a structured matrix based on the core dimensions of "period-section-vehicle type", forming a multi-dimensional data set containing time, space, vehicle type, and risk features, providing standardized input for subsequent steps. The row dimension of the matrix is "period-section-vehicle type" combination (such as "morning peak-K120.3 kilometers-heavy truck" "high temperature-K120.8 kilometers-small car"), and the column dimension is "risk peak value, risk diffusion rate, meteorological-risk matching deviation, risk exceeding level" four types of information, ensuring that each cell corresponds to the risk data of a unique space-time and vehicle type. For example, the multi-dimensional risk feature matrix of the K120-K121 kilometer section of G101 National Highway on a high temperature day is as follows: “Early peak-K120.3km-heavy truck”: risk peak 0.82, risk diffusion rate 0.0045 / min, weather-risk matching bias value 0.18, risk exceeding level “high”. “Midday high temperature-K120.3km-heavy truck”: risk peak 0.88, risk diffusion rate 0.0022 / min, weather-risk matching bias value -0.15, risk exceeding level “medium”. “Evening peak-K120.8km-small car”: risk peak 0.72, risk diffusion rate 0.0028 / min, weather-risk matching bias value -0.12, risk exceeding level “low”. “Forenoon-K120.5km-small car”: risk peak 0.68, risk diffusion rate 0.003 / min, weather-risk matching bias value 0.08, risk exceeding level “low”. The matrix is dynamically updated with real-time data collection frequency (every 5 minutes), for example, after collecting data at 13:00 during the midday high temperature period, the risk peak of heavy truck on K120.3km is raised from 0.85 to 0.88, the “risk peak” column in the “midday high temperature-K120.3km-heavy truck” row of the matrix is updated synchronously to ensure data timeliness.

[0060] S105, compare real-time risk data with historical same-period same-section data to identify short-term risk surge and traffic flow-risk matching imbalance abnormal signals, and use risk-weather parameter correlation curve curvature to generate risk warning dynamic correction factor.

[0061] In one embodiment, by comparing real-time risk data with historical same-period same-section data, through the construction of a time series similarity model and a mutation detection algorithm, short-term risk increases, traffic flow-risk matching imbalance abnormal signals are identified, and an abnormal signal feature vector is generated. By comparing real-time risk data with historical same-period same-section data, combining a time series similarity model and a mutation detection algorithm, two types of abnormal signals, "short-term risk increases" and "traffic flow-risk matching imbalance", are accurately captured and quantitatively presented in the form of a feature vector, providing a data basis for subsequent evaluation. Select the historical risk data of the K120 km section of the G101 National Highway in the same period in the past three years (such as August 15-20 each year), and align the current real-time risk data according to "time granularity (5 minutes / time)-section-vehicle type". For example, on August 16, 2024, 13:00, the historical risk data (mean 0.72) of K120.3 km section of heavy trucks at 13:00-13:30 on August 16, 2023, 2022 and 2021 needs to be compared with the current real-time risk data (0.73 at 13:00, rising to 0.81 at 13:05, and reaching 0.89 at 13:10). The dynamic time warping (DTW) algorithm is used to calculate the similarity of the real-time risk sequence and the historical same-period sequence, and if the similarity is lower than the preset threshold (such as 0.7), it is determined as "potential anomaly". For example, on August 16, 2024, 13:00-13:10, the DTW similarity of the real-time risk sequence (0.73, 0.81, 0.89) of K120.3 km heavy trucks and the historical same-period sequence (0.72, 0.73, 0.74) is 0.62, which is lower than 0.7, and it is preliminarily determined that there is an anomaly.

[0062] The cumulative sum control chart (CUSUM) algorithm is used to detect the mutation point of the real-time risk sequence, and the traffic flow data is used to determine the type of anomaly, as follows. If the risk value increases by more than 0.15 within 10 minutes and there is no obvious traffic flow decrease, it is determined as this type of anomaly. For example, for the above K120.3 km heavy truck, the risk value increases by 0.16 from 13:00 to 13:10, and the same-period traffic flow increases from 1800 vehicles / hour to 1900 vehicles / hour, with no downward trend, which is determined as "short-term risk increase". If the traffic flow decreases but the risk value does not decrease as expected (deviation exceeds 0.1), it is determined as this type of anomaly. For example, for K120.8 km small cars, the traffic flow decreases from 1500 vehicles / hour to 1200 vehicles / hour at 14:00 (when the historical same-period traffic flow decreases by 300 vehicles / hour, the risk value should decrease by 0.12), but the real-time risk value only decreases from 0.65 to 0.62 (decrease by 0.03), with a deviation of 0.09 close to the threshold of 0.1, and after combining other period data, it is determined as "traffic flow-risk matching imbalance".

[0063] The identified abnormal type, mutation time, risk change range, etc. information is quantified as a feature vector. For example, the feature vector of the "short-term risk sudden increase" of the heavy truck at K120.3 km is: {abnormal type: 1 (representing short-term risk sudden increase), mutation start time: 13:05, 10-minute risk increase: 0.16, traffic flow change rate during the same period: +5.5%, road section: K120.3, vehicle type: heavy truck}; the feature vector of the "traffic flow-risk matching imbalance" of the small car at K120.8 km is: {abnormal type: 2 (representing matching imbalance), imbalance start time: 14:00, risk expected deviation: 0.09, traffic flow decrease: 20%, road section: K120.8, vehicle type: small car}.

[0064] The risk sudden increase intensity, duration, and impact range in the abnormal signal feature vector are classified and dynamically evaluated to generate a risk level evaluation result. For the core indicators (risk sudden increase intensity, duration, and impact range) in the abnormal signal feature vector, a grading standard is set, a weighted summation method is used to calculate the risk level score, three risk levels of high, medium, and low are classified, and the severity of the abnormality is determined. In the short-term risk sudden increase type, the maximum increase in risk value within 10 minutes is taken (such as 0.16 for the heavy truck at K120.3 km); in the traffic flow-risk matching imbalance type, the absolute value of the risk expected deviation is taken (such as 0.09 for the small car at K120.8 km). The duration of the abnormal signal from its occurrence to the present (such as 15 minutes for the abnormality at K120.3 km and 10 minutes for the abnormality at K120.8 km). The length of the road section affected by the abnormality (such as the abnormality at K120.3 km affecting only itself and the 500 meters behind, and the abnormality at K120.8 km affecting 1 km behind).

[0065] For each indicator, a grading threshold and corresponding weight are set (risk sudden increase intensity weight 0.5, duration weight 0.3, and impact range weight 0.2), and the score = (intensity score x 0.5) + (duration score x 0.3) + (impact range score x 0.2), with a total score of 10, 0-3 points for low level, 3-7 points for medium level, and 7-10 points for high level. The risk sudden increase intensity score is as follows: 0.1-0.15 for 3 points, 0.15-0.2 for 5 points, and >0.2 for 10 points; 0.05-0.1 for 2 points, 0.1-0.15 for 4 points, and >0.15 for 8 points. The duration score is 5-10 minutes for 2 points, 10-20 minutes for 4 points, and >20 minutes for 6 points. The impact range score is <500 meters for 1 point, 500 meters-1 km for 3 points, and >1 km for 5 points.

[0066] The risk level assessment result is generated as follows: K120.3 km heavy truck (short-term risk increases sharply): intensity score 5 points (increase 0.16), duration score 4 points (15 minutes), influence range score 1 point (500 meters), total score = 5x0.5+4x0.3+1x0.2=2.5+1.2+0.2=3.9 points, judged as "medium risk level"; K120.8 km small car (traffic flow-risk mismatch imbalance): intensity score 2 points (deviation 0.09), duration score 2 points (10 minutes), influence range score 3 points (1 km), total score = 2x0.5+2x0.3+3x0.2=1+0.6+0.6=2.2 points, judged as "low risk level".

[0067] Based on the risk level assessment result, a risk warning dynamic correction factor is calculated using the risk-weather parameter correlation curve curvature. The risk-weather parameter correlation curve is constructed as follows: taking "road surface temperature" as the horizontal axis and "risk value" as the vertical axis, the correlation curve of different vehicle types is constructed combined with historical data. For example, the correlation curve of K120.3 km heavy truck is: road surface temperature 60°C corresponds to risk value 0.65, 62°C corresponds to 0.72, 64°C corresponds to 0.80, 66°C corresponds to 0.90, and 68°C corresponds to 1.0 (theoretical maximum value).

[0068] The curve curvature calculation and sensitivity judgment use numerical differentiation method to calculate the curvature of the curve point corresponding to the real-time road surface temperature. The curvature value >0.05 is judged as "high sensitivity", 0.03-0.05 is "medium sensitivity", and <0.03 is "low sensitivity". For example, on August 16, 2024, at 13:10, the road surface temperature of K120.3 km is 65°C, the curvature of the corresponding correlation curve point is 0.062, which exceeds 0.05, and is judged as "high sensitivity", that is, a small change in road surface temperature (such as an increase of 1°C) will cause a significant increase in risk value (from 0.85 to 0.90).

[0069] The dynamic correction factor is calculated as follows: correction factor = base coefficient x risk level coefficient x curvature sensitivity coefficient, wherein the base coefficient is fixed at 1.0, the risk level coefficient (low level 1.0, medium level 1.2, high level 1.5), and the curvature sensitivity coefficient (low sensitivity 1.0, medium sensitivity 1.1, high sensitivity 1.3). For K120.3 km heavy truck: risk level coefficient 1.2 (medium level), curvature sensitivity coefficient 1.3 (high sensitivity), correction factor = 1.0 x 1.2 x 1.3 = 1.56, which means that the current risk prediction value (0.89) needs to be multiplied by 1.56, and the correction is 1.40 (because the risk value is maximum 1.0, finally according to 1.0), to match the sudden risk. For K120.8 km small car: risk level coefficient 1.0 (low level), curvature sensitivity coefficient 1.0 (assuming that the road surface temperature is 63°C, the curvature is 0.028, and the sensitivity is low), correction factor = 1.0 x 1.0 x 1.0 = 1.0, the current risk prediction value (0.62) does not need to be corrected, and the original value is kept.

[0070] The curvature of the curve corresponding to the road surface temperature and the risk level are recalculated every 5 minutes, and the correction factor is updated. For example, at 13:15, the road surface temperature of K120.3 km rises to 66°C, the curvature becomes 0.075 (still high sensitivity), the risk level is still medium level, and the correction factor remains 1.56; if the road surface temperature rises to 68°C at 13:20, the curvature is 0.09, and the risk level is upgraded to high level (score 7.5), the correction factor is updated to 1.0 x 1.5 x 1.3 = 1.95, and the risk warning is further strengthened.

[0071] S106, combining the highway administrative level, the number of lanes and the surrounding environment, the road section is grouped, the random forest algorithm is used to screen the key influence factors of the multi-dimensional risk feature matrix and the risk warning dynamic correction factor, and the personalized risk warning control model is constructed by fusing the risk warning threshold, the traffic control constraint condition and the emergency response limit information.

[0072] In one embodiment, the importance of the risk peak value, risk diffusion rate, and meteorological-risk matching deviation value in the multi-dimensional risk characteristic matrix and the risk early warning dynamic correction factor is evaluated, and a redundancy analysis is performed to generate a key impact factor ranking feature, a risk factor contribution weight, and a dynamic correction factor adjustment coefficient, thereby forming a risk factor screening basic information. The random forest algorithm is used to perform importance ranking and redundancy elimination on the multi-dimensional risk characteristic matrix (including the risk peak value, risk diffusion rate, and meteorological-risk matching deviation value) and the risk early warning dynamic correction factor, thereby determining the contribution of each factor to the risk early warning and providing core input for subsequent model construction. The multi-dimensional risk characteristic data (such as the early morning peak K120.3 km heavy truck risk peak value 0.82, diffusion rate 0.0045 / min, and deviation value 0.18) and the corresponding dynamic correction factor (such as 1.56) of the K120-K121 km section of the G101 national road in the last three months are analyzed using the random forest algorithm (based on the Scikit-learn framework). The algorithm can output the factor importance and redundancy index through the voting mechanism of multiple decision trees.

[0073] The contribution of each factor to the "risk early warning accuracy" is calculated to generate a key impact factor ranking feature. For example, the analysis result shows that the importance score of the risk peak value is the highest (0.35), because it directly reflects the upper limit of the risk in the period and is the core basis for triggering high-level early warning; the risk diffusion rate is the second (0.28), which can predict the risk spreading trend in advance; the dynamic correction factor (0.22) is more important than the meteorological-risk matching deviation value (0.15) because it can adapt to sudden situations. The final ranking is: risk peak value > risk diffusion rate > dynamic correction factor > meteorological-risk matching deviation value. Redundant factors (correlation coefficient > 0.8) are removed by calculating the Pearson correlation coefficient between the factors. For example, the correlation coefficient between the "risk peak value" and the "road surface temperature" is 0.85 (high road surface temperature in high temperature period leads to high risk peak value), but the "risk peak value" has integrated the comprehensive influence of road surface temperature and traffic flow, so the "road surface temperature" is removed; the correlation coefficient between the "dynamic correction factor" and the "risk diffusion rate" is only 0.42, which is not redundant and is retained.

[0074] Based on the above analysis, the key impact factor ranking feature (risk peak value > risk diffusion rate > dynamic correction factor > meteorological-risk matching deviation value), the risk factor contribution weight (risk peak value 0.35, diffusion rate 0.28, and deviation value 0.15), and the dynamic correction factor adjustment coefficient (set to 1.2 according to the importance of the factor, which is higher than the 1.0 of the deviation value) are output, thereby forming a risk factor screening basic information.

[0075] The grouping constraints information is formed by converting the grouping basis of the highway administrative level, the number of lanes, and the surrounding environment, dividing the type boundary, generating the road segment grouping feature weight, the differentiated early warning threshold parameter, and the hierarchical control strategy constraint parameter. The "national highway (such as G101)" and "provincial highway (such as S201)" are converted into the grouping feature weight (the national highway weight is 1.2 because of large traffic flow and high priority of control; the provincial highway weight is 1.0). According to the "double six-lane" and "double four-lane" division, the differentiated early warning threshold parameter is generated (the risk peak threshold of the six-lane road segment is 0.85 because of more lanes and strong traffic diversion capacity, which is higher than the threshold of 0.8 of the four-lane road). If the surrounding of the road segment is "urban area (such as K120.1 km near residential area)", the "emergency response time" in the hierarchical control strategy constraint parameter is shortened to 10 minutes; if it is "suburban area (such as K120.8 km)", the emergency response time can be relaxed to 15 minutes.

[0076] To avoid grouping ambiguity, clear boundary standards are set. For example, the administrative level is taken as the boundary of "national highway number (G at the beginning of national highway)"; the number of lanes is taken as the boundary of "actual number of lanes ≥ 6" for double six-lane; and the surrounding environment is taken as the boundary of "residential households within 500 meters ≥ 50 households" for urban area. Based on the boundary division, the road segment grouping feature weight (such as K120.3 km of G101 national highway is 1.2), the differentiated early warning threshold parameter (the risk peak threshold of six-lane K120.3 km is 0.85), and the hierarchical control strategy constraint parameter (the emergency response of urban area K120.1 km is 10 minutes) are generated, forming the grouping constraint information.

[0077] The objective function of the individualized risk early warning control model is constructed, the parameter adaptation and the target achievement degree evaluation are performed in combination with the risk identification accuracy, the false alarm rate and the response timeliness, the model parameter optimization feature and the strategy matching degree index are generated, and the objective function optimization information is formed. Taking "risk identification accuracy, false alarm rate and response timeliness" as the core target, a multi-objective optimization function is constructed, the optimal parameters of the model are determined through parameter adaptation and target achievement degree evaluation, and the balanced performance of the model is ensured. The objective function is set as F = α × accuracy rate - β × false alarm rate - γ × response time, wherein α, β and γ are weight coefficients (calibrated based on expert experience and historical data, and set as 0.4, 0.3 and 0.3 respectively). For example, when the model accuracy is 92%, the false alarm rate is 4.4%, and the response time is 8 minutes, F = 0.4 × 0.92 - 0.3 × 0.044 - 0.3 × 0.08 = 0.368 - 0.0132 - 0.024 = 0.3308, the higher the function value, the better the comprehensive performance of the model.

[0078] Gradient Boosting Decision Tree (GBDT) is selected as the base model (based on the LightGBM framework) because it can handle nonlinear feature interactions and has better generalization ability than neural networks on small sample data. Specifically, the input layer (key factors: risk peak value, diffusion rate, dynamic correction factor, bias value) → feature interaction layer (automatically calculate interaction features such as "risk peak value x dynamic correction factor") → decision tree integration layer (100 CART trees, maximum depth 8) → output layer (risk warning level: low / medium / high).

[0079] The learning rate of the model is 0.05 (control parameter update step), the sub-sample ratio is 0.8 (to prevent overfitting), the column sampling ratio is 0.8 (randomly select 80% features to train each tree), and the regularization parameter λ = 0.1 (to suppress model complexity). 5-fold cross-validation is used to iteratively optimize the parameters. For example, when the initial parameters are "decision tree maximum depth 6", the model accuracy is 88%, the false alarm rate is 6%, the response time is 10 minutes, and the F value is 0.298. After adjusting to "maximum depth 8", the accuracy is improved to 92%, the false alarm rate is reduced to 4.4%, the response time is shortened to 8 minutes, the F value is increased to 0.3308, the target achievement degree is increased by 11%, and this is determined as the optimal parameters. The final generated model parameter optimization features (maximum depth 8, learning rate 0.05), strategy matching degree indicators (F value 0.3308, higher than the preset threshold 0.3), and target function optimization information are formed.

[0080] Integrate the risk factor screening basic information, grouping constraint information, and target function optimization information to generate personalized risk warning thresholds, hierarchical warning rules, and linkage control strategy suggestions for corresponding road segment groups, forming a personalized risk warning control model. Take "road segment grouping" as a unit, bind the screened key factors (such as risk peak value, diffusion rate), grouping constraint parameters (such as threshold 0.85), and optimized model parameters (such as maximum depth 8) to ensure that the model adapts to the characteristics of this group of road segments. For example, for the "national highway - two-way 6-lane - urban area" grouping (such as G101 national highway K120.1-K120.3 kilometers), combined with the screening basic information (risk peak value weight 0.35) and the grouping constraint (threshold 0.85), set the warning threshold: high risk (risk peak value ≥ 0.85 and diffusion rate ≥ 0.004 / min), medium risk (0.8 ≤ risk peak value < 0.85 or 0.003 ≤ diffusion rate < 0.004 / min), and low risk (risk peak value < 0.8 and diffusion rate < 0.003 / min).

[0081] The hierarchical early warning rule and control strategy are as follows: when the risk peak value is greater than or equal to 0.85 and the dynamic correction factor is greater than or equal to 1.5, a first-level early warning is triggered. The speed of heavy trucks is limited to 80 km / h, and temporary cooling reminder signs are set every 2 km, and the emergency team arrives at the scene within 10 minutes. When the risk peak value is 0.8-0.85 and the diffusion rate is 0.003-0.004 / min, a second-level early warning is triggered. The "high temperature, drive carefully" prompt is displayed on the information board, and the patrol frequency is increased to once every 30 minutes.

[0082] The finally generated personalized risk early warning control model outputs differentiated results for different groups. For example, for G101 National Road K120.3 km (national highway - 6 lanes - urban area), when the risk peak value is 0.88, the diffusion rate is 0.0045 / min, and the correction factor is 1.56, the model determines that it is a high-risk, and outputs a first-level early warning instruction and the corresponding control strategy; while for K120.8 km (provincial highway - 4 lanes - suburban area), when the risk peak value is 0.72 and the diffusion rate is 0.0028 / min, the model determines that it is a low-risk, and no early warning is needed. The model risk identification accuracy rate reaches 93% and the false alarm rate is controlled within 5%, which fully adapts to the actual characteristics of the road section and meets the personalized early warning needs.

[0083] In one embodiment, the target early warning parameters output by the personalized risk early warning control model are combined with the time sequence change law of the highway traffic flow and the time correlation information of the weather forecast to generate real-time dynamic early warning instructions and time-periodic risk control strategies.

[0084] In one embodiment, the target early warning parameters output by the personalized risk early warning control model are combined with the time sequence change law of the highway traffic flow and the time correlation information of the weather forecast to generate real-time dynamic early warning instructions and time-periodic risk control strategies. In one embodiment, the target early warning parameters output by the personalized risk early warning control model are combined with the time sequence change law of the highway traffic flow and the time correlation information of the weather forecast to generate real-time dynamic early warning instructions and time-periodic risk control strategies. For example, the personalized model output for G101 National Road K120.3 km (national highway - 6 lanes - urban area) is "risk level: high, dynamic correction factor: 1.56"; the real-time traffic flow data collected simultaneously is "vehicle average speed 70 km / h, vehicle type composition (25% heavy trucks, 70% small cars, 5% others), traffic flow density 22 vehicles / km". The correlation logic is: the early warning parameters and traffic flow indicators are combined through weighted calculation to highlight the influence weight of key indicators of traffic flow under high risk (such as the proportion of heavy trucks is higher than that of small cars).

[0085] The formula of risk-speed correlation factor is "(model risk level coefficient x dynamic correction factor) ÷ (actual average speed ÷ standard safety speed)", wherein the high risk level coefficient is set as 1.2, and the standard safety speed (in high temperature environment) is set as 80 km / h. Substituting the data into the formula, (1.2 x 1.56) ÷ (70 ÷ 80) = 1.872 ÷ 0.875 ≈ 2.14, the factor > 2.0, indicating that the current speed is lower than the safety standard, and the risk superposition effect is significant. The formula of risk-vehicle type correlation factor is "model risk level coefficient x (heavy truck proportion x 1.5 + small car proportion x 0.8 + other vehicle type proportion x 1.0)", wherein the weight of heavy truck is 1.5 (high risk of heat stress), and the weight of small car is 0.8 (lower risk). Substituting the data into the formula, 1.2 x (25% x 1.5 + 70% x 0.8 + 5% x 1.0) = 1.2 x (0.375 + 0.56 + 0.05) = 1.2 x 0.985 ≈ 1.18, the factor > 1.0, indicating that the proportion of high-risk vehicle types in the current vehicle composition is high, and the targeted control needs to be strengthened.

[0086] The formula of risk-density correlation factor is "dynamic correction factor x (actual traffic flow density ÷ standard flow density threshold value)", and the standard flow density threshold value of 6-lane high temperature period is set as 25 vehicles / km. Substituting the data into the formula, 1.56 x (22 ÷ 25) = 1.56 x 0.88 ≈ 1.37, the factor < 1.5, indicating that the current flow density does not reach the congestion threshold, and the risk is mainly caused by high temperature rather than traffic congestion. Integrating the above calculation results, the risk warning basic quantitative factor set of K120.3 km section is formed: {risk-speed correlation factor: 2.14, risk-vehicle type correlation factor: 1.18, risk-density correlation factor: 1.37, section: K120.3, period: noon high temperature 13:00-14:00}.

[0087] The temperature, humidity, and light intensity parameters in the meteorological forecast information and real-time monitoring data along the expressway are matched in time and space and trend prediction analysis is performed to generate meteorological trend quantization factors. The future 6-hour meteorological forecast information along the expressway is matched in time and space with the real-time meteorological monitoring data, and the potential impact of meteorological changes on risks is quantized to generate meteorological trend quantization factors to provide meteorological basis for time-period control. Taking the three meteorological monitoring points (K119.5, K120.3, and K120.8) along the K120 kilometer section as an example, the real-time monitoring data is “13:00 road surface temperature 65°C, humidity 23%, and light intensity 850W / m²”; the corresponding meteorological forecast information is “13:00-15:00 road surface temperature maintains 64-66°C (trend stable), 15:00-17:00 humidity rises to 30% and light intensity drops to 600W / m² (risk alleviation trend)”. In the time-space matching, the forecast period of each monitoring point is completely corresponding to the real-time data collection period (such as 13:00-14:00 forecast corresponding to 13:00 real-time data).

[0088] The formula of the temperature trend factor is “(future 1-hour forecast temperature mean ÷ real-time temperature) × risk temperature sensitivity coefficient”, and the temperature sensitivity coefficient under high risk is set to 1.3. The 13:00-14:00 forecast temperature mean is 65.5°C, and the substitution gives: (65.5 ÷ 65) × 1.3 ≈ 1.31, the factor > 1.3, indicating that the temperature is still in the high-risk interval and needs to be continuously monitored. The formula of the humidity-light comprehensive factor is “(real-time humidity ÷ forecast humidity after 1 hour) × (forecast light intensity after 1 hour ÷ real-time light intensity) × 0.9” (0.9 is the humidity-light joint weight). The 13:00-14:00 forecast humidity is 24% and the light intensity is 820W / m², and the substitution gives: (23 ÷ 24) × (820 ÷ 850) × 0.9 ≈ 0.958 × 0.965 × 0.9 ≈ 0.83, the factor < 1.0, indicating that the humidity will slightly rise and the light will slightly decrease in the future 1 hour, and the risk has a weak alleviation trend.

[0089] The formula of the meteorological risk duration factor is “(duration of high-risk meteorological conditions (temperature ≥ 60°C) in the future 2 hours ÷ 2) × 1.2”, the forecast 13:00-15:00 temperature is ≥ 60°C, and the duration is 2 hours, the substitution gives: (2 ÷ 2) × 1.2 = 1.2, the factor = 1.2, indicating that the high-risk meteorological conditions will continue, and the control period needs to be extended. The integrated results give the meteorological trend quantization factor set of the K120.3 kilometer section at 13:00-14:00: {temperature trend factor: 1.31, humidity-light comprehensive factor: 0.83, meteorological risk duration factor: 1.2, forecast period: 13:00-14:00}.

[0090] Based on the risk early warning basic quantitative factor, meteorological trend quantitative factor, combined with the hierarchical architecture of the whole link prediction and control model of highway operation risk, the risk level, risk diffusion rate and the implementation effect of control measures in different time periods are fused to generate early warning-trend-control related quantitative characteristics. The input layer of the whole link model receives two types of quantitative factors; the feature fusion layer adopts the weighted summation method, giving the basic quantitative factor (traffic flow correlation) a weight of 0.6, and the meteorological trend factor a weight of 0.4 (the influence of meteorology on risk is slightly lower than traffic flow in high temperature environment); the control evaluation layer introduces historical control measure effect data (such as "accident rate reduced by 30% when speed limit is 80km / h"), which corrects the fusion result; the output layer generates related quantitative characteristics.

[0091] Correlation characteristic calculation (take the high temperature from 13:00 to 14:00 as an example): The formula of period risk comprehensive characteristic value is "(basic quantitative factor mean value x 0.6) + (meteorological trend factor mean value x 0.4) x control effect correction coefficient", basic factor mean value = (2.14 + 1.18 + 1.37) ÷ 3 ≈ 1.56, meteorological factor mean value = (1.31 + 0.83 + 1.2) ÷ 3 ≈ 1.11, control effect correction coefficient (assuming "speed limit 75km / h" is adopted) is set to 0.95 (historical data shows that this measure can reduce 5% risk). Substituting: (1.56 x 0.6 + 1.11 x 0.4) x 0.95 = (0.936 + 0.444) x 0.95 = 1.38 x 0.95 ≈ 1.31, characteristic value > 1.2, it is determined that the control needs to be strengthened in this period. The formula of risk diffusion-control response characteristic value is "(risk diffusion rate x dynamic correction factor) ÷ (emergency response time ÷ standard response time)", K120.3km risk diffusion rate 0.0045 / min, emergency response time (urban area) 10min, standard response time 8min. Substituting: (0.0045 x 1.56) ÷ (10 ÷ 8) = 0.00702 ÷ 1.25 ≈ 0.0056, characteristic value < 0.01, indicating that the current emergency response time can cover the risk diffusion speed, and the control response is timely.

[0092] The formula for controlling the characteristic value of the time period is "time period risk comprehensive characteristic value x (time period traffic flow peak value coefficient x 1.2 + time period meteorological risk coefficient x 0.8)", the traffic flow peak value coefficient of the noon high temperature period is 0.9 (the traffic flow is lower than the morning peak), and the meteorological risk coefficient is 1.3 (the temperature is the highest). Substituting, we get: 1.31 x (0.9 x 1.2 + 1.3 x 0.8) = 1.31 x (1.08 + 1.04) = 1.31 x 2.12 ≈ 2.78, the characteristic value > 2.5, indicating that high-intensity control measures (such as speed limit + temporary prompt) need to be taken in this period. The integrated K120.3 km road section warning-trend-control correlation quantitative characteristic set of the noon high temperature 13:00-14:00 is: {time period risk comprehensive characteristic value: 1.31, risk diffusion-control response characteristic value: 0.0056, time period control adaptation characteristic value: 2.78, time period: 13:00-14:00}.

[0093] Based on the highway operation risk full-link prediction and control model, the warning-trend-control correlation quantitative characteristics are analyzed and processed to generate real-time dynamic warning instructions and sub-period risk control strategy comprehensive evaluation information for highway operation risk in dry and hot environments, including risk identification accuracy, early warning response timeliness, and sub-period control measure effectiveness. The full-link model compares the correlation characteristic value with the preset threshold (such as time period risk comprehensive characteristic value > 1.2 triggers high-level control, and sub-period control adaptation characteristic value > 2.5 triggers high-intensity measures), and generates targeted strategies combined with the historical control effect database; at the same time, risk identification accuracy, early warning response timeliness and other evaluation indexes are calculated to verify the effectiveness of the strategy.

[0094] For the correlation characteristics of K120.3 km section 13:00-14:00, the output instruction is: "

Real-time warning instruction

[0095] Through model backtracking verification and real-time data comparison, evaluation indicators are generated. Comparing the model prediction of high-risk period with the actual accident hidden danger record (such as vehicle abnormal deceleration, tire overheating alarm) within nearly 1 hour, the accuracy rate = (correctly identified hidden danger number ÷ total hidden danger number) × 100% = 18 ÷ 20 × 100% = 90%. Calculate the average time from model output high risk to emergency measures landing (12 minutes), timeliness = (standard response time - actual response time difference) ÷ standard response time × 100% = (15-3) ÷ 15 × 100% = 80%. After implementing the control measures during the high temperature in the afternoon, the average speed of traffic flow is increased to 75 km / h, and the illegal road occupation rate of heavy trucks is decreased by 40%, the effectiveness = (risk relief degree after implementing the measures ÷ expected relief degree) × 100% = (0.35 ÷ 0.4) × 100% = 87.5%.

[0096] The comprehensive evaluation information finally formed is "G101 National Highway K120.3 km section dry high temperature risk control comprehensive evaluation: risk identification accuracy rate 90%, early warning response timeliness 80%, midday control measure effectiveness 87.5%, overall in line with the preset control target (accuracy rate ≥ 85%, timeliness ≥ 75%, effectiveness ≥ 80%), can continue to implement the current time period strategy, and it is suggested that after 15:00, the running frequency of the road cooling device should be appropriately reduced according to the rising of the meteorological humidity".

[0097] In one embodiment, as shown in Figure 2 The application also provides a highway operation risk early warning device in a dry high temperature environment, comprising: An acquisition module 201 is configured to acquire real-time meteorological data, road state data and traffic flow data along the highway, remove extreme weather burst pulse interference and traffic flow transient fluctuation noise by using a wavelet transform algorithm, and generate a standardized meteorological-road-traffic correlation data sequence; The processing module 202 is used to convert the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution atlas, subdivide the high temperature gradient section based on a regional grid encryption algorithm, combine the thermal stress response differences of different vehicle types, and construct a vehicle type-specific expressway operation risk prediction model; based on the vehicle type-specific expressway operation risk prediction model, extract the risk weight factor of different sections to construct a dynamic risk assessment matrix, perform collaborative calculation on the multi-source monitoring data, and obtain the key early warning index combination under the target risk level; according to the expressway operation period, extract the risk peak value, risk diffusion rate, and meteorological-risk matching deviation value, divide the risk exceeding level, and generate a multi-dimensional risk feature matrix; compare the real-time risk data with the historical same period data of the same section, identify the abnormal signals of short-term risk surge and traffic flow-risk matching imbalance, use the risk-weather parameter correlation curve curvature to generate a risk warning dynamic correction factor; group the sections according to the expressway administrative level, the number of lanes, and the surrounding environment, use the random forest algorithm to screen the key influence factors of the multi-dimensional risk feature matrix and the risk warning dynamic correction factor, fuse the risk warning threshold, the traffic control constraint condition, and the emergency response limit information to construct a personalized risk warning control model; based on the target warning parameter output of the personalized risk warning control model, combine the time correlation information of the expressway traffic flow time sequence change rule and the meteorological forecast to generate real-time dynamic warning instructions and time period-specific risk control strategies.

[0098] Each of the embodiments in the present application is described in a related manner, and the same and similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the evaluation of the dry high temperature environment expressway operation risk warning method, electronic device, electronic equipment, and readable storage medium embodiments, since they are basically similar to the dry high temperature environment expressway operation risk warning method embodiments described above, the description is relatively simple, and the relevant parts can be referred to the above-mentioned dry high temperature environment expressway operation risk warning method embodiments.

Claims

1. A method for early warning of highway operation risk in a dry high-temperature environment, characterized in that, The application comprises the following steps: Obtain real-time meteorological data, road state data, and traffic flow data along the expressway, and use wavelet transform algorithm to eliminate extreme weather burst pulse interference and traffic flow instantaneous fluctuation noise, and generate standardized meteorological-road-traffic correlation data sequence; After converting the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution map, the high temperature gradient section is subdivided based on a regional grid encryption algorithm, and a vehicle-specific expressway operation risk prediction model is constructed in combination with the differences in thermal stress response of different vehicle types; Based on the vehicle-specific expressway operation risk prediction model, the risk weight factor of different road sections is extracted to construct a dynamic risk assessment matrix, and the target risk level is obtained by co-computing the multi-source monitoring data and obtaining the key warning index combination; According to the risk peak value, risk diffusion rate, and meteorological-risk matching deviation value of the expressway operation period, the risk exceeding level is divided, and a multi-dimensional risk feature matrix is generated; By comparing the real-time risk data with the historical data of the same period and the same section, the abnormal signals of short-term risk increase and traffic flow-risk matching imbalance are identified, and the risk warning dynamic correction factor is generated by using the risk-weather parameter correlation curve curvature; The road sections are grouped according to the administrative level of the expressway, the number of lanes, and the surrounding environment, the random forest algorithm is used to screen the key influence factors of the multi-dimensional risk feature matrix and the risk warning dynamic correction factor, and the individualized risk warning control model is constructed by fusing the risk warning threshold, the traffic control constraint condition, and the emergency response limit information; Based on the target warning parameter output of the individualized risk warning control model, the real-time dynamic warning instruction and the time-period risk control strategy are generated by combining the time correlation information of the expressway traffic flow time series variation law and the meteorological forecast.

2. The method of claim 1, wherein, After converting the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution map, the high temperature gradient section is subdivided based on a regional grid encryption algorithm, and a vehicle-specific expressway operation risk prediction model is constructed in combination with the differences in thermal stress response of different vehicle types, including: Convert the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution map to generate basic risk visualization data; Import the three-dimensional risk load distribution map into a road environment and traffic flow coupling simulation software for simulation calculation to generate multi-dimensional risk field simulation data; Use a regional grid encryption algorithm to grid-subdivide the high temperature gradient section in the multi-dimensional risk field simulation data to generate differential grid data, wherein the regional grid encryption algorithm identifies the temperature gradient change rate in real time, and when the change rate exceeds a preset threshold, it automatically triggers grid encryption, and high-density grids are used for high-risk road sections and low-density grids are used for low-risk road sections; Collect thermal stress response difference parameters of different vehicle types in dry and high temperature environment to construct a vehicle-specific thermal stress characteristic constraint library to generate vehicle constraint data, wherein the vehicle-specific thermal stress characteristic constraint library customizes parameters for thermal sensitivity, brake performance attenuation, and tire friction coefficient change of heavy trucks and small cars; Fuse the differential grid data and vehicle constraint data to construct a vehicle-specific expressway operation risk prediction model.

3. The method of claim 1, wherein, Based on the risk prediction model of highway operation by vehicle type, the risk weight factors of different road sections are extracted to construct a dynamic risk assessment matrix. The key warning indicator combination under the target risk level is obtained by collaborative calculation of multi-source monitoring data, including: Based on the risk prediction model of highway operation by vehicle type, the risk weight factors of different road sections are extracted to construct a dynamic risk assessment matrix. The risk weight factors are extracted based on the road temperature gradient risk degree, traffic flow pressure intensity, and historical accident severity. The dynamic risk assessment matrix is input into the multi-source data collaborative calculation module for synchronous fusion analysis of meteorological, road, and traffic multi-dimensional monitoring data. A plurality of candidate warning indicator sets are generated. The multi-source data collaborative calculation module adopts a distributed computing architecture to perform parallel data cleaning, feature extraction, and risk quantification according to the differences in time and space resolution and noise characteristics of different data sources. Based on the preset target risk level standard, the candidate warning indicator set is screened and optimized to obtain the key warning indicator combination under the target risk level. The screening process uses a dual-objective evaluation function of risk identification accuracy and false alarm rate, and introduces index sensitivity analysis to predict the risk impact of small fluctuations in key warning indicators.

4. The method of claim 1, wherein, By comparing real-time risk data with historical data of the same period and the same road section, abnormal signals of short-term risk increase and traffic flow-risk mismatch are identified. A risk warning dynamic correction factor is generated using the curvature of the risk-weather parameter correlation curve, including: By comparing real-time risk data with historical data of the same period and the same road section, abnormal signals of short-term risk increase and traffic flow-risk mismatch are identified by constructing a time series similarity model and a mutation detection algorithm. An abnormal signal feature vector is generated. The risk increase intensity, duration, and impact range in the abnormal signal feature vector are classified and dynamically evaluated based on risk level to generate risk level evaluation results. Based on the risk level evaluation results, a risk warning dynamic correction factor is calculated using the curvature of the risk-weather parameter correlation curve. The curvature value reflects the sensitivity of risk to changes in weather parameters. When the curvature value exceeds a preset threshold, the dynamic correction factor weight is automatically increased to respond quickly to sudden weather changes.

5. The method of claim 1, wherein, The road sections are grouped based on highway administrative level, number of lanes, and surrounding environment. Random forest algorithm is used to screen key influencing factors from multi-dimensional risk feature matrix and risk warning dynamic correction factor. A personalized risk warning control model is constructed by integrating risk warning threshold, traffic control constraints, and emergency response limit information, including: The risk peak value, risk diffusion rate, and meteorological-risk matching deviation value in the multi-dimensional risk feature matrix and risk warning dynamic correction factor are evaluated for importance and analyzed for redundancy. Key influencing factor sorting features, risk factor contribution weights, and dynamic correction factor adjustment coefficients are generated to form risk factor screening basic information. The grouping constraints are converted and the type boundaries are divided based on highway administrative level, number of lanes, and surrounding environment. Road section grouping feature weights, differentiated warning threshold parameters, and hierarchical control strategy constraint parameters are generated to form grouping constraint information. The target function of the personalized risk early warning control model is constructed, parameter adaptation and target achievement evaluation are performed in combination with risk identification accuracy, false alarm rate and response timeliness, model parameter optimization characteristics and strategy matching degree indexes are generated, and target function optimization information is formed; The personalized risk early warning threshold, hierarchical early warning rule and linkage control strategy suggestion corresponding to the grouping of the road section are generated by integrating the risk factor screening basic information, grouping constraint information and target function optimization information, and the personalized risk early warning control model is formed.

6. The method of claim 5, wherein, Based on the target early warning parameter output of the personalized risk early warning control model, in combination with the time correlation information of the highway traffic flow time sequence change rule and the weather forecast, real-time dynamic early warning instructions and time period risk control strategies are generated, including: The target early warning parameters output by the personalized risk early warning control model are quantitatively analyzed and processed in combination with the vehicle average speed, vehicle type composition ratio and traffic flow density in the real-time traffic flow data, and risk early warning basic quantitative factors are generated; The temperature, humidity, light intensity and other parameters in the weather forecast information and real-time monitoring data along the highway are spatio-temporally matched and trend prediction analyzed, and meteorological trend quantitative factors are generated; Based on the risk early warning basic quantitative factors and the meteorological trend quantitative factors, in combination with the hierarchical architecture of the highway operation risk whole-link prediction and control model, the risk level, risk diffusion rate and implementation effect of the control measures in different time periods are fused and processed, and early warning-trend-control correlation quantitative characteristics are generated; Based on the highway operation risk whole-link prediction and control model, the early warning-trend-control correlation quantitative characteristics are analyzed and processed, and the real-time dynamic early warning instruction and time period risk control strategy comprehensive evaluation information for the highway operation risk in the dry and hot environment are generated, including risk identification accuracy, early warning response timeliness and time period control measure efficiency.

7. A device for early warning of the risk of highway operation in a dry high-temperature environment, characterized in that, The device comprises: An acquisition module is configured to acquire real-time weather data, road state data and traffic flow data along the highway, remove extreme weather burst pulse interference and traffic flow instantaneous fluctuation noise by using a wavelet transform algorithm, and generate standardized weather-road-traffic correlation data sequences; The processing module is used for converting the standardized meteorological-road-traffic correlation data sequence into a three-dimensional risk load distribution atlas, subdividing the high temperature gradient section based on a regional grid encryption algorithm, combining the thermal stress response differences of different vehicle types, and constructing a vehicle type-specific expressway operation risk prediction model; based on the vehicle type-specific expressway operation risk prediction model, the risk weight factors of different road sections are extracted to construct a dynamic risk assessment matrix, the multi-source monitoring data are cooperatively calculated, and the key early warning index combination under the target risk level is obtained; according to the expressway operation period, the risk peak value, the risk diffusion rate and the meteorological-risk matching deviation value are extracted, the risk exceeding level is divided, and a multi-dimensional risk feature matrix is generated; by comparing the real-time risk data with the historical same period data of the same section, the abnormal signals of short-term risk sudden increase and traffic flow-risk matching imbalance are identified, the risk warning dynamic correction factor is generated by using the risk-weather parameter correlation curve curvature; the road section is grouped in combination with the expressway administrative level, the number of lanes and the surrounding environment, the random forest algorithm is used to screen the key influence factors of the multi-dimensional risk feature matrix and the risk warning dynamic correction factor, the personalized risk warning control model is constructed by fusing the risk warning threshold, the traffic control constraint condition and the emergency response limit information; based on the target warning parameter output of the personalized risk warning control model, in combination with the time correlation information of the expressway traffic flow time sequence change law and the meteorological forecast, real-time dynamic warning instructions and time period-specific risk control strategies are generated.

8. An electronic device, comprising: It comprises: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the drying high-temperature environment expressway operation risk warning method by executing the executable instructions.

Images