Short-term visibility forecast and correction method for highways

Abstract

This invention relates to a method for short-term visibility forecasting and correction on highways, comprising the following steps: acquiring 10-minute historical meteorological observation data from traffic meteorological observation stations along the measured section of the highway; calculating the Pearson correlation coefficient between visibility and various meteorological factors, and performing a significance test; acquiring hourly real-time meteorological observation data along the measured section of the highway; training an LSTM prediction model to obtain visibility forecast results for 10 minutes and station numbers within the next 2 hours; performing probabilistic correction on the visibility forecast results using visibility observation data acquired from automatic weather stations, performing trend correction on the visibility forecast results using visibility data retrieved from video images along the highway, and using the correction results to correct subsequent new visibility forecast results, thereby obtaining visibility forecast results for station numbers and 10-minute intervals along the measured section of the highway within a short-term near-term timeframe of 0-2 hours.

Description

Technical Field

[0001] This invention relates to a method for forecasting highway visibility, specifically a method for short-term forecasting and correcting visibility on expressways. Background Technology

[0002] Dense fog is a major factor contributing to highway safety accidents and traffic disruptions, and the extent of fog's impact on highways is primarily reflected in visibility levels. Therefore, accurate forecasting of fog and visibility, especially highway visibility forecasts for the next two hours, is crucial for improving highway driving safety and traffic efficiency. Furthermore, to address the adverse effects of small-scale, locally occurring low-visibility events, station-by-station visibility forecasting is necessary. However, mature technologies in this area are currently lacking and require urgent research and development. Summary of the Invention

[0003] The purpose of this invention is to provide a method for short-term visibility forecasting and correction on highways, so as to improve highway traffic safety and meet the development needs of safe highway operation.

[0004] The objective of this invention is achieved as follows:

[0005] A method for short-term visibility forecasting and correction on highways includes the following steps:

[0006] S1. Obtain at least one year's worth of historical meteorological observation data from traffic meteorological observation stations along the measured section of the highway, including visibility, temperature, relative humidity, average wind speed over the past 2 minutes, temperature change over the past 24 hours, highest temperature over the past 10 minutes, and lowest temperature over the past 10 minutes. Among these, the elements other than visibility constitute meteorological factors related to visibility prediction. The starting point of each past time is the time point for each 10-minute interval.

[0007] S2. Calculate the Pearson correlation coefficient between visibility and various meteorological factors using correlation analysis and perform a significance test; select meteorological factors with α = 0.01 in the significance test and use them as meteorological variables for visibility prediction.

[0008] S3. Obtain real-time meteorological observation data along the measured section of the highway every hour, and introduce the visibility changes over 10 minutes, 1 hour, 3 hours and 6 hours as time series trend characteristic variables; introduce the minimum and standard deviation of visibility over the most recent 6 hours as time series aggregation characteristic variables.

[0009] S4. After standardizing the three types of variables—meteorological variables, time series trend characteristic variables, and time series aggregate characteristic variables—they are used together as variables for visibility prediction to train the LSTM prediction model and obtain visibility prediction data for the location of traffic meteorological observation points along the measured section of the highway in 10-minute increments over the next 2 hours. Spatial interpolation is then used to perform spatial difference on the visibility prediction data to obtain visibility prediction results for the measured section of the highway in 10-minute increments and by station number over the next 2 hours.

[0010] S5. Obtain visibility observation data from automatic weather stations within a 5km radius of the measured section of the highway at the time corresponding to the visibility prediction result. Obtain visibility data retrieved from video images at each station number along the measured section of the highway at the time corresponding to the visibility prediction result. After filtering out invalid data from the above two sets of data, first use the retained visibility observation data to perform probability correction on the visibility prediction result to obtain the correction probability. Then use the retained visibility data retrieved from video images to perform trend correction on the visibility prediction result to obtain the visibility change trend. After using the correction probability and visibility change trend to correct the newly obtained visibility prediction results respectively, the visibility forecast results for each station number and each 10-minute interval along the measured section of the highway within a short-term near-term timeframe of 0-2 hours are obtained.

[0011] Furthermore, the Pearson correlation coefficient r between visibility and various meteorological factors in step S2 is calculated as follows:

[0012]

[0013] Where X and Y are two variables, and is the average of the two sets of variables, and n is the number of meteorological factors.

[0014] Furthermore, the significance test in step S2 is performed as follows:

[0015] First, calculate using the following formula:

[0016]

[0017] Where t is the deviation statistic between the sample mean and the population mean. σ is the sample mean, μ is the population mean, and σ is the population mean. Z Where m is the sample standard deviation and m is the sample size;

[0018] Then, by consulting the t-value table, determine whether the significance test with α = 0.01 has been passed.

[0019] Furthermore, the standardization process for the three types of variables in step S4 is as follows:

[0020]

[0021] Where V is the standardized value of the variable, g is the number of variables, and C g Let g be the sequence of variables. Let g be the average value of the g-th variable.

[0022] Furthermore, in step S5, the filtering conditions for invalid data removal of the acquired visibility observation data and the visibility data inverted from the video images are one of the following: ① The change in visibility between two consecutive minutes is not greater than 500m; ② The minimum duration of visibility is not less than 3min; ③ The relative humidity is greater than 80%; ④ The wind speed is not greater than 4.5m / s.

[0023] Furthermore, the method for probabilistically correcting the visibility prediction results in step S5 is to use the Gmama probability function for correction.

[0024] Furthermore, the method for trend correction of the visibility prediction results in step S5 is to correct the trend of the visibility prediction results using visibility data retrieved from video images. Specifically, this involves establishing a statistical model:

[0025] y = Ax + B

[0026] Where y is the visibility value after trend correction, x is the visibility prediction result, and A and B are coefficients.

[0027] This invention enables refined visibility forecasting along highways at each station number and every 10 minutes. It can accurately forecast fog and visibility on highways, especially providing relatively accurate forecasts of visibility on highways within the next 2 hours. This is of great significance for improving highway driving safety and traffic efficiency. Furthermore, its prediction results and corrected forecast results are superior to those of existing visibility prediction methods. Detailed Implementation

[0028] The present invention will be further described in detail below with reference to the embodiments.

[0029] This invention uses a 20-kilometer section of the Beijing-Hong Kong-Macau Expressway (Beijing-Shijiazhuang section) as the test section to predict visibility every 10 minutes and by station number for the next 2 hours. The specific prediction method is as follows:

[0030] The first step involves acquiring 10-minute historical meteorological observation data from traffic meteorological observation stations along the measured section of the highway, covering a period from October 2023 to September 2024. This includes 10-minute data on visibility, temperature, relative humidity, average wind speed over the past 2 minutes, temperature changes over the past 24 hours, the highest temperature over the past 10 minutes, and the lowest temperature over the past 10 minutes. The starting points for the past 2 minutes, past 10 minutes, and past 24 hours are all 10-minute data collection points. Except for visibility, the other elements constitute the meteorological factors related to visibility prediction.

[0031] The second step is to first calculate the Pearson correlation coefficient between visibility and the aforementioned meteorological factors using correlation analysis. The calculation method for the Pearson correlation coefficient r between visibility and each meteorological factor is as follows:

[0032]

[0033] Where X and Y are two variables, and is the average of the two sets of variables, and n is the number of meteorological factors.

[0034] The results of the Pearson correlation coefficient calculation are shown in Table 1.

[0035] Table 1 shows the Pearson correlation coefficients between visibility and various meteorological factors.

[0036]

[0037] Note: Marked with ** indicates that the significance test was passed at α = 0.01.

[0038] Secondly, a significance test is performed, which includes the following two sub-steps:

[0039] ① Calculate according to the following formula:

[0040]

[0041] Where t is the deviation statistic between the sample mean and the population mean. σ is the sample mean, μ is the population mean, and σ is the population mean. Z denoted as the sample standard deviation, and m as the sample size.

[0042] ② By consulting the t-value table, determine whether the significance test of α = 0.01 has been passed.

[0043] Meteorological factors with a significance level of α = 0.01 in the significance test were selected and used as meteorological variables for visibility prediction. As shown in Table 1, all six listed meteorological factors passed the significance test with α = 0.01 and can be used as meteorological variables for visibility prediction.

[0044] The third step involves obtaining hourly real-time meteorological observation data from traffic meteorological observation stations along the measured section of the highway in October 2024 for testing and verification. Visibility changes over 10 minutes, 1 hour, 3 hours, and 6 hours are introduced as time-series trend characteristic variables. The minimum and standard deviation of visibility over the most recent 6 hours are introduced as time-series aggregate characteristic variables.

[0045] The fourth step is to standardize the three types of variables: meteorological variables, time series trend characteristic variables, and time series aggregate characteristic variables. The standardization methods for these three types of variables are as follows:

[0046]

[0047] Where V is the standardized value of the variable, g is the number of variables, and C g Let g be the sequence of variables. Let g be the average value of the g-th variable.

[0048] After standardization, these three types of variables are used together as visibility prediction variables to train the LSTM prediction model. The LSTM prediction model is a commonly used deep learning neural network for time series prediction, which contains three LSTM layers, several hidden layers, and one fully connected layer.

[0049] This invention sets the time step of the LSTM prediction model to 10 minutes, the number of hidden layers to 1, the number of neurons in the hidden layers to 64, the dropout rate to 0.01, the training batch size to 128, and the number of iterations to 500. Using the visibility prediction variables and the LSTM prediction model, the visibility prediction data for the next two hours at the locations of two sets of traffic meteorological observation points along the measured section of the highway is calculated in 10-minute increments. The predicted data is then compared with the actual observation data as follows:

[0050] Visibility is categorized into five levels: less than or equal to 50 meters, greater than 50 meters but less than or equal to 200 meters, greater than 200 meters but less than or equal to 500 meters, greater than 500 meters but less than or equal to 1000 meters, and greater than 1000 meters. The first four levels are considered foggy, and the fifth level is considered fog-free. Forecast levels matching the observed levels are considered accurate. Forecasts of fog but actual observations of fog-free conditions are considered false alarms, while forecasts of fog-free conditions but actual observations of fog-stricken conditions are considered missed alarms. The verification criteria are shown in the table below.

[0051]

[0052] Using spline interpolation, spatial interpolation is performed on the visibility prediction data from the third step to obtain the visibility prediction results for the measured section of the highway in 10-minute increments and by station number for the next 2 hours.

[0053] The fifth step involves acquiring visibility observation data from two automatic weather stations within a 5km radius of the measured section of the highway at the time corresponding to the visibility prediction results. It also involves acquiring station-by-station visibility data retrieved from video images at the time corresponding to the visibility prediction results along the measured section of the highway. The invalid data selection criteria for the acquired visibility observation data and the visibility data retrieved from video images are one of the following: ① The change in visibility between two consecutive minutes does not exceed 500m; ② The minimum duration of visibility is not less than 3min; ③ Relative humidity is greater than 80%; ④ Wind speed is not greater than 4.5m / s.

[0054] After filtering out invalid data from the two sets of data mentioned above, the visibility prediction results obtained in step four are first probabilistically corrected using the retained visibility observation data to obtain the corrected probability. The probability correction method uses the Gmama probability function, which is:

[0055]

[0056] Here, Γ(α) is the Gmama function, where α is the shape parameter of the Gmama function, representing the shape of the distribution curve. The larger the value, the flatter the curve, and the lower the probability of small values. β is the scaling parameter of the Gmama function. The larger the value, the wider the range of the probability function and the more dispersed the distribution.

[0057] α and β were estimated using the maximum likelihood method:

[0058]

[0059] Where u is the mean of the sequence, σ 2 Let be the variance of the sequence.

[0060] Then, the visibility data retrieved from video images along the highway at each station number is used to perform trend correction on the visibility prediction results obtained in step four, thus obtaining the visibility change trend. The trend correction method is to establish a statistical model:

[0061] y = Ax + B

[0062] Where y is the visibility value after trend correction, x is the visibility prediction result, and A and B are coefficients.

[0063] Finally, using the correction probability and visibility change trend obtained from the above calculations, the newly obtained visibility prediction results are corrected respectively. The result is the visibility forecast result along the measured section of the highway within a short-term near-term of 0-2 hours, at each station number and every 10-minute interval.

[0064] After probability correction, the visibility forecast results of this invention reduce the false alarm rate by 23% to 13.4% and the false alarm rate by 45% to 12.2%. After trend correction, the fog forecast accuracy below 1000 meters is 63.2%, and the root mean square error is reduced by 12 meters.

[0065] The visibility forecast performance, as reported in the June 2022 issue of the Journal of Meteorology and Environment, Volume 38, Issue 3, titled "Research on Low Visibility Classification in Hebei Province Based on CART Decision Tree," was as follows: underreporting rate: 51.4% in summer and 15.7% in winter; accuracy rate: 42.1% in summer and 49.8% in winter; false alarm rate: 24.1% in summer and 45.1% in winter.

[0066] As can be seen from the comparison, the present invention has significant advantages in both false negative rate and accuracy.

Claims

1. A method for short-term visibility forecasting and correction on highways, characterized in that, Includes the following steps: S1. Obtain at least one year's worth of 10-minute meteorological observation historical data from traffic meteorological observation stations along the measured section of the highway, including visibility, temperature, relative humidity, average wind speed over the past 2 minutes, temperature change over the past 24 hours, highest temperature over the past 10 minutes, and lowest temperature over the past 10 minutes; among these, the elements other than visibility constitute meteorological factors related to visibility prediction. S2. Calculate the Pearson correlation coefficient between visibility and various meteorological factors using correlation analysis, and perform a significance test; select meteorological factors with α = 0.01 in the significance test, and use them as meteorological variables for visibility prediction; S3. Obtain real-time meteorological observation data along the measured section of the highway every hour, and introduce the visibility changes over 10 minutes, 1 hour, 3 hours and 6 hours as time series trend characteristic variables; introduce the minimum and standard deviation of visibility over the most recent 6 hours as time series aggregation characteristic variables. S4. After standardizing the three types of variables—meteorological variables, time series trend characteristic variables, and time series aggregate characteristic variables—they are used together as variables for visibility prediction to train the LSTM prediction model and obtain visibility prediction data for the location of traffic meteorological observation points along the measured section of the highway in 10-minute increments over the next 2 hours. Spatial interpolation is then used to perform spatial difference on the visibility prediction data to obtain visibility prediction results for the measured section of the highway in 10-minute increments and by station number over the next 2 hours. S5. Obtain visibility observation data from automatic weather stations within a 5km radius of the measured section of the highway at the time corresponding to the visibility prediction result. Obtain visibility data retrieved from video images at each station number along the measured section of the highway at the time corresponding to the visibility prediction result. After filtering out invalid data from the above two sets of data, first use the retained visibility observation data to perform probability correction on the visibility prediction result to obtain the correction probability. Then use the retained visibility data retrieved from video images to perform trend correction on the visibility prediction result to obtain the visibility change trend. After using the correction probability and visibility change trend to correct the newly obtained visibility prediction results respectively, the visibility forecast results for each station number and each 10-minute interval along the measured section of the highway within a short-term near-term timeframe of 0-2 hours are obtained.

2. The method for short-term visibility forecasting and correction on highways according to claim 1, characterized in that, The Pearson correlation coefficient r between visibility and various meteorological factors in step S2 is calculated as follows: Where X and Y are two variables, and is the average of the two sets of variables, and n is the number of meteorological factors.

3. The method for short-term visibility forecasting and correction on highways according to claim 1, characterized in that, The significance test in step S2 is performed as follows: First, calculate using the following formula: Where t is the deviation statistic between the sample mean and the population mean. σ is the sample mean, μ is the population mean, and σ is the population mean. Z Where m is the sample standard deviation and m is the sample size; Then, by consulting the t-value table, determine whether the significance test with α = 0.01 has been passed.

4. The method for short-term visibility forecasting and correction on highways according to claim 1, characterized in that, The standardization process for the three types of variables in step S4 is as follows: Where V is the standardized value of the variable, g is the number of variables, and C g Let g be the sequence of variables. Let g be the average value of the g-th variable.

5. The method for short-term visibility forecasting and correction on highways according to claim 1, characterized in that, The invalid data screening conditions for the acquired visibility observation data and the visibility data inverted from video images in step S5 are one of the following: ① The change in visibility between two consecutive minutes is not greater than 500m; ② The minimum duration of visibility is not less than 3min; ③ The relative humidity is greater than 80%; ④ The wind speed is not greater than 4.5m / s.

6. The method for short-term visibility forecasting and correction on highways according to claim 1, characterized in that, The method for probabilistically correcting the visibility prediction results in step S5 is to use the Gmama probability function for correction.

7. The method for short-term visibility forecasting and correction on highways according to claim 1, characterized in that, The method for trend correction of the visibility prediction results in step S5 is to correct the trend of the visibility prediction results using visibility data retrieved from video images. Specifically, this involves establishing a statistical model: y = Ax + B Where y is the visibility value after trend correction, x is the visibility prediction result, and A and B are coefficients.

Images