Expressway congestion meteorological risk prediction method
Through the combination of RF random forest method and gray correlation method, a highway congestion meteorological risk prediction model is constructed, which solves the problem of lack of meteorological factor assessment in the existing technology, and realizes accurate prediction and risk assessment of highway congestion events, improving Decision-making ability in traffic management.
Patent Information
- Application Number
- CN202411325454.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-23
- Publication Date
- 2025-01-07
AI Technical Summary
The lack of risk assessment based on meteorological factors in the prediction of highway congestion risk in the prior art has led to the inability to effectively manage and prevent traffic accidents caused by extreme weather.
The RF random forest method is used to screen meteorological factors, and the highway congestion meteorological risk prediction model is constructed by combining the gray correlation method. The probability of congestion events is predicted through meteorological observation data and traffic data, and the threshold interval of congestion duration and mileage is determined to evaluate meteorological risk. grade.
Accurate prediction of highway congestion incidents has been achieved, the decision-making basis for traffic management has been improved, and the risk of traffic accidents caused by extreme weather has been reduced.
Smart Images
Figure CN119272163A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for predicting meteorological risk levels of highway congestion, in particular to a method for predicting meteorological risk levels of highway congestion. Background Art
[0002] Against the backdrop of global warming, extreme weather events are becoming increasingly frequent. The dense network of expressways surrounding Beijing and Tianjin is prone to heavy fog in autumn and winter. Rain, snow, and fog often lead to congestion and traffic accidents on expressways. Therefore, meteorological factors are a significant contributor to the risk of expressway congestion. However, current predictions of expressway congestion risk based on meteorological factors are lacking, hindering safe highway traffic management. Summary of the Invention
[0003] The purpose of the present invention is to provide a method for predicting meteorological risks of highway congestion, so as to solve the problem that there is still a lack of risk prediction based on meteorological factors in highway congestion risk prediction.
[0004] The object of the present invention is achieved like this:
[0005] A method for predicting meteorological risks of highway congestion includes the following steps:
[0006] S1. Obtain information on congestion incidents that occurred along the measured section of the expressway over the past year or more. This information includes the location of the congested section, the start and end times of the congestion, the cause of the congestion incident, the season of occurrence, the time of occurrence, the road network density of the expressway at the time of the incident, and the single-item traffic volume 1 hour before the incident.
[0007] S2. Obtain meteorological observation data information from meteorological stations within a radius of no more than 10 km around the time period when the congestion event occurs. The data information includes minimum visibility, average visibility, maximum temperature, minimum temperature, average temperature, average relative humidity, maximum relative humidity, average wind speed, maximum wind speed, minimum wind speed, hourly rainfall / snow amount, and cumulative rainfall 3 hours and 6 hours before the start of the congestion, and cumulative snowfall 3 hours, 6 hours, and 12 hours before the start of the congestion.
[0008] S3. Filter out congestion events caused by adverse weather from the acquired congestion event information. When one of the following meteorological factors occurs, it is recorded as a congestion event caused by adverse weather: (1) Fog, high temperature, strong wind, low temperature, rain / snow occurs at the time of the congestion event or within 1 hour before the event; (2) Rainfall occurs within 6 hours before the event; (3) Snowfall occurs within 12 hours before the event.
[0009] S4. Define the season when the congestion event occurs, the time period when it occurs, the road network density of the highway when it occurs, and the single vehicle flow volume 1 hour before the occurrence as non-meteorological factors; define the meteorological factors set in step S3 as meteorological factors; use meteorological factors and non-meteorological factors together as independent variables affecting highway congestion events; among them, meteorological factors are numerical variables; road network density and single vehicle flow volume in non-meteorological factors are numerical variables, and the others are categorical variables; use the RF random forest method to evaluate the importance of the above independent variables respectively.
[0010] S5. Select characteristic variables that have a significant effect on the occurrence of highway congestion events from the above independent variables; use the RFE recursive feature elimination method to construct a highway congestion meteorological risk prediction model.
[0011] S6. Use the highway congestion meteorological risk prediction model to conduct prediction training on whether the collected congestion events caused by adverse weather will occur, and calculate the probability of occurrence of meteorological risks of highway congestion events; when the prediction result is "occurrence", use the grey correlation method to establish a piecewise conversion function between the probability of occurrence P of congestion events caused by adverse weather and the two sub-indicators of congestion duration b1 and congestion mileage b2; then, based on the probability of occurrence P of the prediction result and its functional relationship with the above two sub-indicators, determine the threshold range of congestion duration and congestion mileage, and obtain the meteorological risk level of congestion events in the corresponding time period in the future along the measured section of the highway.
[0012] S7. Obtain hourly meteorological observation data and hourly single vehicle flow data along the measured section of the expressway on the forecast day, use these two data as calculation parameters, and calculate the meteorological risk level of congestion in the measured section of the expressway within the next hour and the possible impact of a congestion event based on the expressway congestion meteorological risk prediction model and the piecewise conversion function.
[0013] Furthermore, the screening conditions for screening congestion events caused by bad weather from the acquired congestion event information in step S3 are as follows:
[0014] ⑴ One of the following conditions occurs within 1 hour before the congestion incident: ① minimum visibility ≤ 500m, ② maximum temperature ≥ 37℃, ③ average wind speed ≥ level 6, ④ gust wind speed ≥ level 8, ⑤ minimum temperature ≤ –10℃; ⑥ rainfall in 1 hour ≥ 2mm; ⑦ snowfall in 1 hour ≥ 0.1mm.
[0015] ⑵ One of the following situations occurs within 3 hours before the congestion incident: ① The accumulated rainfall in 3 hours is ≥10mm; ② The accumulated snowfall in 3 hours is ≥0.5mm.
[0016] ⑶ One of the following situations occurs within 6 hours before the congestion incident: ① The accumulated rainfall in 6 hours is ≥30mm; ② The accumulated snowfall in 6 hours is ≥1mm.
[0017] ⑷ The cumulative snowfall for 12 hours is ≥ 2.5 mm within 12 hours before the congestion incident occurs.
[0018] Furthermore, the meteorological factors include: 1. minimum visibility, average visibility, maximum temperature, minimum temperature, average temperature, average relative humidity, maximum relative humidity, average wind speed, maximum wind speed, minimum wind speed, and hourly rainfall / snow within 1 hour before the start of the congestion; 2. cumulative rainfall 3 hours and 6 hours before the start of the congestion; 3. cumulative snowfall 3 hours, 6 hours, and 12 hours before the start of the congestion.
[0019] Furthermore, the highway congestion weather risk prediction model is:
[0020] Set the independent variable u of the fth sample of highway congestion events caused by bad weather f for:
[0021] u f =[u f1 ,u f2 …,u fn ],f=1,2,…,F
[0022] Correspondingly, the sample training matrix U is:
[0023] U={[u1,Y1],[u2,Y2]…,[u i ,Y f ]}
[0024] Among them, F is the number of samples contained in the training set; u fn is the nth characteristic variable involved in the fth sample; Y f is the same as the independent variable u f The corresponding dependent variable, and Y f ∈{1,0}.
[0025] Furthermore, the method for calculating the probability of occurrence of highway congestion weather risk in step S6 includes the following steps:
[0026] S6-1 constructs x decision trees and randomly extracts Z samples from the sample training matrix U with replacement using the bootstrap random sampling method to obtain a new training set for each decision tree.
[0027] S6-2 randomly extracts g independent variables at each node of each tree, and then selects the optimal feature for data fitting. The threshold of feature classification is determined by checking each classification point.
[0028] S6-3 forms a random forest with x decision trees, gives the probability of each tree choosing "occurrence", and takes the average of the predicted probabilities of all trees, which is the probability P of highway congestion caused by bad weather.
[0029] S6-4 identifies samples with an occurrence probability P≥0.5 as occurring, and identifies samples with an occurrence probability P<0.5 as not occurring.
[0030] Furthermore, the prediction method of the prediction training in step S6 and the segmented conversion function relationship between the P value of the prediction result and the two sub-item indicators (b1, b2) are constructed as follows:
[0031] ① When the prediction result is "P < 0.5, not occurring", the subsequent steps are terminated, and the prediction conclusion is: highway congestion will not be caused by bad weather.
[0032] ② When the prediction result is "P≥0.5, occurs", the prediction conclusion is: there is a high probability that highway congestion will occur due to bad weather, and the P value range of the prediction result is divided into three sections: 0.5≤P<0.7, 0.7≤P<0.9, 0.9≤P.
[0033] ③ Expressway congestion events are described using two sub-indicators: congestion duration and congestion mileage. The constructed piecewise conversion function includes a conversion function Q(b1) based on congestion duration and a conversion function Q(b2) based on congestion mileage:
[0034]
[0035]
[0036] The present invention focuses on analyzing the impact of meteorological conditions on highway congestion, adopts the random forest method to screen meteorological factors that have a high impact on congestion events, obtains the probability of occurrence of congestion events from the training model, and then determines whether a congestion event has occurred; then, based on the grey correlation theory, corresponds to the threshold range of the two evaluation indicators of occurrence probability and congestion events (congestion duration and congestion mileage), and finally gives the meteorological risk level of highway congestion, thereby providing an effective reference basis for the scheduling of highway management departments and the travel of the public. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is the importance score of each characteristic variable to the occurrence of congestion events. DETAILED DESCRIPTION
[0038] The present invention will be further described in detail below with reference to the embodiments.
[0039] This paper uses a section of the Daguan Expressway in a certain province in my country as the measured section and conducts a meteorological risk forecast for highway congestion in May 2024. The specific prediction method is as follows:
[0040] In the first step, we collected data from highway traffic management departments on 70,293 congestion incidents that occurred along a provincial section of the Daguan Expressway between 2021 and 2023. This data included the location of the congested section, the start and end times of the congestion, the cause of the congestion, the season and time of day, the highway network density at the time of the incident, and the individual traffic volume in the hour before the incident.
[0041] The second step is to obtain meteorological observation data at the time of the congestion event from national meteorological stations within a radius of no more than 10 km of the location of each congestion event. This meteorological observation data includes the following information: (1) minimum visibility, average visibility, maximum temperature, minimum temperature, average temperature, average relative humidity, maximum relative humidity, average wind speed, maximum wind speed, minimum wind speed, and hourly rainfall / snowfall within 1 hour before the congestion start time; (2) accumulated rainfall within 3 hours and 6 hours before the congestion start time; (3) accumulated snowfall within 3 hours, 6 hours, and 12 hours before the congestion start time.
[0042] The third step is to filter out congestion events caused by bad weather from the acquired congestion event data. If the following meteorological factors screening conditions are met, it will be recorded as a congestion event caused by bad weather. The specific screening conditions are:
[0043] (1) Any of the following conditions occurs within 1 hour before the congestion incident: ① minimum visibility ≤ 500m, ② maximum temperature ≥ 37°C, ③ average wind speed ≥ level 6, ④ gust wind speed ≥ level 8, ⑤ minimum temperature ≤ –10°C, ⑥ rainfall ≥ 2mm in 1 hour, ⑦ snowfall ≥ 0.1mm in 1 hour;
[0044] (2) One of the following conditions occurs within 3 hours before the congestion incident: ① 3-hour cumulative rainfall ≥ 10mm, ② 3-hour cumulative snowfall ≥ 0.5mm;
[0045] (3) One of the following conditions occurs within 6 hours before the congestion event: ① 6-hour cumulative rainfall ≥ 30mm, ② 6-hour cumulative snowfall ≥ 1mm;
[0046] ⑷ The cumulative snowfall for 12 hours is ≥ 2.5 mm within 12 hours before the congestion incident occurs.
[0047] Based on the above criteria, a total of 24,231 highway congestion events caused by bad weather were screened out, accounting for approximately 34.6%. These 24,231 highway congestion events were set as the sample set, with each event as a sample. 5,000 samples were randomly selected from them as test samples, and the remaining samples were used as control samples.
[0048] In the fourth step, first, the season when the congestion event occurs, the time period when it occurs, the road network density of the highway when it occurs, and the single vehicle flow volume 1 hour before the occurrence are defined as non-meteorological factors; the meteorological factors determined by the screening conditions in the third step are defined as meteorological factors. Meteorological factors and non-meteorological factors are taken together as independent variables affecting highway congestion events, and each factor included can also be called a characteristic variable. Among the two types of independent variables, meteorological factors are numerical variables; the road network density and single vehicle flow volume in the non-meteorological factors are numerical variables, and the other factors are categorical variables. The value settings of the non-meteorological factor independent variables are shown in the following table:
[0049]
[0050] Secondly, the importance of the above independent variables is evaluated using the RF random forest method. The evaluation method of the RF random forest method on the importance of feature variables is as follows:
[0051] Assume that there are c characteristic variables: X1, X2, X3...Xc, the importance score of the characteristic variable is represented by VIM, and the contribution of the characteristic variable is represented by the Gini index (GI).
[0052] The calculation formula of the Gini index is:
[0053]
[0054] Among them, K represents the category, P mk Indicates the proportion of category K in node m.
[0055] The importance of the characteristic variable Xj at node m, that is, the change in the Gini index VIM before and after the branching of node m is:
[0056]
[0057] Among them, GI l and GI r Represents the Gini index of the two new nodes before and after the branch respectively.
[0058] Assuming there are n trees in RF, the importance of feature variable Xj in the i-th tree is:
[0059]
[0060] Finally, all the importance scores obtained are normalized:
[0061]
[0062] Among them, VIM j That is, the importance score of the j-th feature variable. The larger the value, the higher the importance.
[0063] Step 5: First, select characteristic variables that have a significant impact on the occurrence of highway congestion events from the two categories of independent variables mentioned above. Through the above calculation method, a total of 18 characteristic variables with significant importance to the occurrence of highway congestion events were screened, including: minimum visibility, maximum temperature, minimum temperature, average temperature, maximum relative humidity, average wind speed, minimum wind speed, hourly rainfall, hourly snowfall, cumulative rainfall in the three hours before the congestion start time, cumulative rainfall in the six hours before the congestion start time, cumulative snowfall in the three hours before the congestion start time, cumulative snowfall in the six hours before the congestion start time, cumulative snowfall in the twelve hours before the congestion start time, season and time of occurrence of the congestion event, highway network density, and single-item traffic volume in the hour before the congestion event. Figure 1 The importance score values corresponding to these 18 characteristic variables are given.
[0064] Secondly, the RFE recursive feature elimination method is used to construct a highway congestion meteorological risk prediction model. The highway congestion meteorological risk prediction model constructed by the present invention is:
[0065] Set the independent variable u of the fth sample of highway congestion events caused by bad weather f for:
[0066] u f =[u f1 ,u f2 …,u fn ],f=1,2,…,F
[0067] Correspondingly, the sample training matrix U is:
[0068] U={[u1,Y1],[u2,Y2]…,[u f ,Y f ]}
[0069] Among them, F is the number of samples contained in the training set; u fn is the nth characteristic variable involved in the fth sample; Y f is the same as the independent variable u f The corresponding dependent variable, and Y f ∈{1,0}.
[0070] The sixth step is to use the highway congestion meteorological risk prediction model to predict whether each sample in the sample set has a congestion risk, and calculate the probability of highway congestion meteorological risk occurrence for each sample.
[0071] The present invention calculates the probability of occurrence of meteorological risk of highway congestion as follows:
[0072] ⑴ Build x decision trees, and randomly extract Z samples from the sample training matrix U with replacement through the bootstrap random sampling method to obtain a new training set for each decision tree.
[0073] ⑵ Randomly extract g independent variables at each node of each tree, and then select the optimal feature for data fitting. The threshold of feature classification is determined by checking each classification point.
[0074] ⑶ Organize x decision trees into a random forest, give the probability of each tree choosing "occurrence", and take the average of the predicted probabilities of all trees, which is the probability P of highway congestion caused by bad weather.
[0075] (4) Samples with a probability of occurrence P ≥ 0.5 are marked as occurring, and samples with a probability of occurrence P < 0.5 are marked as not occurring.
[0076] Let x = 5000, that is, construct 5000 decision trees. At each node of each tree, randomly select g = 3 independent variables. Then, select the optimal feature from these variables for data fitting. The probability of each tree choosing "occurrence" is determined. The predicted probabilities of all trees are averaged to obtain the probability P of the congestion event. Samples with a probability P ≥ 0.5 are marked as "occurrence", and samples with a probability P < 0.5 are marked as "non-occurrence".
[0077] The prediction method for prediction training is as follows:
[0078] ① When the prediction result is "P < 0.5, not occurring", the subsequent steps are terminated, and the prediction conclusion is: highway congestion will not be caused by bad weather.
[0079] ② When the prediction result is "P≥0.5, occurs", the prediction conclusion is: there is a high probability that highway congestion will occur due to bad weather, and the P value range of the prediction result is divided into three sections: 0.5≤P<0.7, 0.7≤P<0.9, 0.9≤P.
[0080] In the process of predictive training on the test samples in the sample set, 72.9% of the samples were successfully identified; in the control samples, 93.8% of the samples were successfully identified.
[0081] When the prediction result is "occurrence", the grey correlation method is used to establish a piecewise conversion function between the probability P of congestion events caused by bad weather and the two sub-item indicators of congestion duration b1 and congestion mileage b2.
[0082] The relationship between the predicted probability of occurrence P and the two sub-items of congestion duration b1 and congestion mileage b2 is as follows: Expressway congestion events are described by the two sub-items of congestion duration b1 and congestion mileage b2. The constructed piecewise conversion function includes the conversion function Q(b1) based on congestion duration and the conversion function Q(b2) based on congestion mileage:
[0083]
[0084] Based on the three value intervals divided by the P value of the prediction result in the prediction training, and based on the piecewise conversion function relationship between the above-mentioned probability of occurrence P and the congestion duration b1 and congestion mileage b2, the threshold intervals of congestion duration b1 and congestion mileage b2 are determined, and the meteorological risk level of highway congestion and the corresponding risk consequence description are obtained as shown in the table below.
[0085]
[0086] The seventh step is to obtain hourly meteorological observation data and hourly single vehicle flow data along the measured section of the highway on the forecast day. These two data are used as calculation parameters. Based on the highway congestion meteorological risk prediction model and the piecewise conversion function, the meteorological risk level of congestion in the measured section of the highway within the next hour and the possible impact of a congestion event are finally calculated.
[0087] The proposed prediction method was tested using congestion events along the tested highway sections during the winter and spring of 2023-2024. The accuracy rate for identifying congestion events caused by adverse weather was 71.8%, with a false alarm rate of 15.6% and an underreporting rate of 14.3%. Furthermore, 57.3% of the predicted results were within one level of the corresponding congestion risk level. This demonstrates the reliability and feasibility of the proposed prediction method.
Claims
1. A method for predicting meteorological risk of highway congestion, characterized in that: The following steps are involved: S1. Obtain information on congestion incidents along the measured section of the expressway over the past year or more, including the location of the congested section, the start and end time of the congestion, the cause of the congestion incident, the season of occurrence, the time of occurrence, the road network density of the expressway at the time of occurrence, and the single-item traffic volume 1 hour before the occurrence; S2. Obtain meteorological observation data information from meteorological stations within a radius of no more than 10 km around the time when the congestion event occurs, including minimum visibility, average visibility, maximum temperature, minimum temperature, average temperature, average relative humidity, maximum relative humidity, average wind speed, maximum wind speed, minimum wind speed, hourly rainfall / snow amount within 1 hour before the start of the congestion, and the accumulated rainfall 3 hours and 6 hours before the start of the congestion, and the accumulated snowfall 3 hours, 6 hours and 12 hours before the start of the congestion; S3. Filter out congestion events caused by bad weather from the acquired congestion event information. When one of the following meteorological factors occurs, it is recorded as a congestion event caused by bad weather: (1) fog, high temperature, strong wind, low temperature, rain / snow occurs when the congestion event occurs or within 1 hour before the congestion event occurs; (2) rain occurs within 6 hours before the congestion event occurs; (3) snow occurs within 12 hours before the congestion event occurs; S4, defining the season when the congestion event occurs, the time period when it occurs, the road network density of the expressway when it occurs, and the single vehicle flow volume 1 hour before the occurrence as non-meteorological factors; defining the meteorological factors set in step S3 as meteorological factors; using meteorological factors and non-meteorological factors together as independent variables affecting expressway congestion events; wherein the meteorological factors are numerical variables; the road network density and single vehicle flow volume in the non-meteorological factors are numerical variables, and the others are categorical variables; using the RF random forest method to evaluate the importance of the above independent variables respectively; S5. Select characteristic variables that have a significant effect on the occurrence of highway congestion events from the above independent variables; use the RFE recursive feature elimination method to build a highway congestion meteorological risk prediction model; S6. Use the highway congestion meteorological risk prediction model to predict whether the collected congestion events caused by bad weather will occur, and calculate the probability of the meteorological risk of highway congestion events; when the prediction result is "occurrence", use the grey correlation method to establish a piecewise conversion function between the probability of occurrence P of congestion events caused by bad weather and the two sub-item indicators of congestion duration b1 and congestion mileage b2; then determine the threshold range of congestion duration and congestion mileage based on the prediction result of the probability of occurrence P and its functional relationship with the above two sub-item indicators, and obtain the meteorological risk level of congestion events in the corresponding period of time in the future along the measured section of the highway; S7. Obtain hourly meteorological observation data and hourly single vehicle flow data along the tested section of the expressway on the forecast day, use these two data as calculation parameters, and calculate the meteorological risk level of congestion in the tested section of the expressway within the next hour and the possible impact of congestion events based on the expressway congestion meteorological risk prediction model and piecewise conversion function.
2. The highway congestion weather risk prediction method according to claim 1 is characterized in that: The screening conditions for screening congestion events caused by bad weather from the acquired congestion event data information in step S3 are as follows: (1) One of the following conditions occurs within 1 hour before the congestion incident: ① minimum visibility ≤ 500m, ② maximum temperature ≥ 37℃, ③ average wind force ≥ level 6, ④ gust wind force ≥ level 8, ⑤ minimum temperature ≤ –10℃, ⑥ rainfall ≥ 2mm in 1 hour, ⑦ snowfall ≥ 0.1mm in 1 hour; (2) One of the following situations occurs within 3 hours before the congestion event: ① 3-hour cumulative rainfall ≥ 10mm, ② 3-hour cumulative snowfall ≥ 0.5mm; ⑶ One of the following situations occurred within 6 hours before the congestion incident: ① 6-hour cumulative rainfall ≥ 30mm, ② 6-hour cumulative snowfall ≥ 1mm; ⑷ The accumulated snowfall for 12 hours is ≥ 2.5 mm within 12 hours before the congestion incident occurs.
3. The highway congestion weather risk prediction method according to claim 1 is characterized in that: The meteorological factors include: ① minimum visibility, average visibility, maximum temperature, minimum temperature, average temperature, average relative humidity, maximum relative humidity, average wind speed, maximum wind speed, minimum wind speed, and hourly rainfall / snow within 1 hour before the start of the congestion; ② the accumulated rainfall 3 hours and 6 hours before the start of the congestion; ③ the accumulated snowfall 3 hours, 6 hours, and 12 hours before the start of the congestion.
4. The highway congestion weather risk prediction method according to claim 1 is characterized in that: The highway congestion meteorological risk prediction model is: Set the independent variable u of the fth sample of highway congestion events caused by bad weather f for: in f =[in f1 ,in f2 …,in fn ],f=1,2,...,F Correspondingly, the sample training matrix U is: U={[u1,Y1],[u2,Y2]…,[u f ,Y f ]} Among them, F is the number of samples contained in the training set; u fn is the nth characteristic variable involved in the fth sample; Y f is the same as the independent variable u f The corresponding dependent variable, and Y f ∈{1,0}.
5. The highway congestion weather risk prediction method according to claim 4 is characterized in that: The method for calculating the probability of occurrence of highway congestion meteorological risk in step S6 includes the following steps: S6-1 constructs x decision trees, and randomly extracts Z samples from the sample training matrix U with replacement using the bootstrap random sampling method to obtain a new training set for each decision tree; S6-2 randomly extracts g independent variables at each node of each tree, and then selects the best feature for data fitting. The threshold of feature classification is determined by checking each classification point; S6-3 forms a random forest with x decision trees, gives the probability of each tree choosing "occurrence", and takes the average of the predicted probabilities of all trees, which is the probability P of highway congestion caused by bad weather; S6-4 marks samples with an occurrence probability of P≥0.5 as occurring, and marks samples with an occurrence probability of P<0.5 as not occurring.
6. The highway congestion weather risk prediction method according to claim 5 is characterized in that: The prediction method of the prediction training in step S6 and the piecewise conversion function relationship between the occurrence probability P of the prediction result and the two sub-item indicators are constructed as follows: ① When the prediction result is "P<0.5, not happening", the subsequent steps are terminated, and the prediction conclusion is: highway congestion will not be caused by bad weather; ② When the prediction result is "P≥0.5, occurs", the prediction conclusion is: there is a high probability that highway congestion will occur due to bad weather, and the P value range of the prediction result is divided into three sections: 0.5≤P<0.7, 0.7≤P<0.9, 0.9≤P; ③ Expressway congestion events are described by two sub-item indicators: congestion duration b1 and congestion mileage b2. The constructed piecewise conversion function includes the conversion function Q(b1) based on congestion duration and the conversion function Q(b2) based on congestion mileage: