CN-122020555-A - Sea fog visibility level forecasting method based on visibility meter and mesoscale mode

CN122020555ACN 122020555 ACN122020555 ACN 122020555ACN-122020555-A

Abstract

The application discloses a sea fog visibility level forecasting method based on a visibility meter and a mesoscale mode, which relates to the technical field of visibility forecasting, the method preprocesses the original observation data of the visibility meter into the visibility level observation data consistent with the time step of the output data of the mesoscale numerical mode, and extracting mode forecast data of various meteorological elements and derivative elements thereof with each timestamp related to visibility in a local area where the visibility meter is located from output data of the intermediate scale numerical mode, screening out mode forecast data of a plurality of feature variables with the strongest relationship with the visibility, and finally training a visibility forecast model for sea fog visibility level forecast by combining a machine learning method. According to the method, the visibility meter observation data and the continuous forecast data of the mesoscale numerical mode are fused, so that the seamless forecast of the middle and long periods is realized, the precision and the efficiency of forecast and early warning are improved, and the method has extremely strong flexibility and adaptability.

Inventors

ZHANG HONGJIE
LIU KEBANG
XU YUNJIA
WANG YANG
GUAN YATING
FU WENZHUO
KONG LONGSHI
CHU YIQI
WANG SIZHENG

Assignees

航天新气象科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. The sea fog visibility level forecasting method based on the visibility meter and the mesoscale mode is characterized by comprising the following steps of: preprocessing original observation data acquired by a visibility meter in a historical scene to obtain visibility level observation data consistent with the output data time step of a mesoscale numerical mode; performing feature extraction on a historical mode data set of multiple-height-layer meteorological elements in a local area where a visibility meter is located, which is output by a medium-scale numerical mode in a historical scene, obtaining mode forecast data of multiple meteorological elements and derivative elements thereof, wherein each timestamp is related to the visibility, in the local area where the visibility meter is located, and screening out mode forecast data of multiple feature variables which are most strongly related to the visibility; after a characteristic sample set taking mode forecast data of each characteristic variable as input and visibility level observation data of the same time stamp as a sample label is obtained, training the characteristic sample set based on a machine learning algorithm to obtain a visibility forecast model; and extracting multi-height-layer meteorological elements of the region to be forecasted output in the real-time operation scene by the middle-scale numerical mode to obtain mode forecast data of each characteristic variable, and inputting the mode forecast data into a visibility forecast model obtained by training to obtain a sea fog visibility level forecast result of the region to be forecasted.
2. The method of claim 1, wherein the derived elements include model diagnostic atmospheric visibility and other derived elements, and obtaining model forecast data for a plurality of meteorological elements and derived elements thereof related to visibility in a local area where the visibility meter is located comprises: extracting, for each of a meteorological element related to visibility and other derivative elements other than the pattern-diagnosed atmospheric visibility, element values of the element at a plurality of different typical heights within a local area where the visibility meter is located, and an average value of element values of the element at different grid points within the local area where the visibility meter is located, respectively; Atmospheric extinction coefficient is calculated based on hydrogel information in historical mode dataset of multiple-height-layer meteorological elements in local area where visibility meter is located And obtain the atmospheric visibility of the mode diagnosis 。
3. The sea fog visibility level forecasting method of claim 2, wherein the atmospheric extinction coefficient is calculated based on the water condensate information in the historical pattern dataset of the multi-level weather elements in the local area where the visibility meter is located Comprising the following formula: Wherein, the The concentration of the cloud water particles is represented, The concentration of Yun Bingli molecules is indicated, The concentration of the rainwater particles is indicated, Represents the snow water particle concentration and: Wherein, the Indicating the mixing ratio of the cloud water, Indicating the mixing ratio of the cloud ice, Indicating the mixing ratio of the rainwater, Indicating the mixing ratio of snow water; represents the wet air density and , The pressure of the atmosphere is indicated as such, Indicating the specific gas constant of the dry air, Indicating the water-vapor mixing ratio, Indicating the absolute temperature of the air.
4. The method of forecasting sea fog visibility level according to claim 2, wherein the weather elements related to visibility include temperature, relative humidity, dew point temperature, temperature dew point difference, sea level air pressure, wind direction, wind speed, and the derivative elements other than the model-diagnosed atmospheric visibility include relative vorticity, divergence, temperature advection, and temperature gradient in the vertical direction: extracting element values of the meteorological elements at typical ground heights of a low-altitude area in a local area where a visibility meter is located and extracting element values of the meteorological elements at different typical isobaric surface heights of the high-altitude area in the local area where the visibility meter is located for any meteorological element of temperature, relative humidity, dew point temperature, temperature dew point difference, sea level air pressure, wind direction and wind speed; and extracting element values of any derivative element in relative vorticity, divergence, temperature advection and temperature gradient in the vertical direction at a plurality of different typical isobaric surface heights of an overhead area in a local area where a visibility meter is positioned.
5. The method of claim 1, wherein the screening out the pattern forecast data of the feature variables that are most strongly associated with visibility further comprises: The method comprises the steps of completing time alignment of mode forecast data of candidate features and visibility level observation data to obtain an initial sample set which takes the mode forecast data of a plurality of candidate features as input and the visibility level observation data of the same time stamp as a sample label, wherein the candidate features comprise a plurality of meteorological elements and derivative elements thereof which are extracted and related to visibility; the method comprises the steps of attributing a sample with visibility level observation data smaller than a visibility level threshold value in an initial sample set as a low-visibility subset, attributing a sample with the visibility level observation data reaching the visibility level threshold value as a high-visibility subset, performing oversampling treatment on the low-visibility subset, and performing undersampling treatment on the high-visibility subset; And merging the low-visibility subset and the high-visibility subset which are respectively subjected to sampling processing to obtain an extended sample set, screening out mode forecast data of a plurality of feature variables which are most strongly associated with the visibility based on the extended sample set, and obtaining a feature sample set.
6. The sea fog visibility level forecast method of claim 5, wherein screening out pattern forecast data for a plurality of feature variables that are most strongly associated with visibility based on the expanded sample set includes, for each candidate feature; Taking mode forecast data of the candidate features as a horizontal axis value and corresponding visibility level observation data as a vertical axis value, and constructing a scatter diagram of each sample in the extended sample set relative to the current candidate features; Grid dividing the scatter diagram according to different grid dividing modes under the constraint that the total number of grids does not exceed B (n), and respectively calculating mutual information values of mode forecast data and visibility level observation data of the candidate features in each grid dividing mode, wherein B (n) is a function of the number n of samples in the expanded sample set; and when the maximum value of the mutual information values under different grid division modes reaches a contribution threshold, selecting the current candidate feature as a feature variable.
7. The method of claim 1, wherein obtaining visibility level observations consistent with output data time steps of the mesoscale numerical mode comprises: and after performing Butterworth low-pass filtering processing on the original observed data acquired by the visibility meter in the historical scene, extracting the minimum value of the filtered original observed data in each time step of the output data of the mesoscale numerical mode, and mapping the minimum value into the corresponding visibility level observed data according to the visibility influence level classification standard.
8. The sea fog visibility level forecasting method of claim 7, wherein performing Butterworth low pass filtering processing on raw observed data acquired by a visibility meter in a historical scene comprises: Adopting a bilinear discretization Butterworth low-pass filter to forward filter original observed data obtained by a visibility meter in a historical scene to obtain a forward intermediate sequence; And after reversing the forward intermediate sequence, performing backward filtering by adopting a discretized Butterworth low-pass filter with bilinear variation to obtain a backward intermediate sequence, and reversing the backward intermediate sequence to obtain zero-phase filtering output.
9. The method of claim 1, wherein training the feature sample set to obtain a visibility prediction model based on a machine learning algorithm comprises: Constructing a plurality of base learner models based on a plurality of different machine learning algorithms respectively, constructing a proxy model by using a Bayesian method, and training the super parameters of each base learner model by using a characteristic sample set; after a plurality of basic learner models are obtained through pre-training, initializing a meta model into a logistic regression model, taking the output of each basic learner model as the input of the meta model to establish a stacking classifier, and training the feature sample set to obtain the stacking classifier with optimal prediction accuracy to obtain a visibility prediction model.
10. A sea fog visibility level forecasting method according to claim 1, wherein when a trained visibility forecasting model is used to forecast sea fog visibility level, the effective forecasting time length of the visibility forecasting model is dynamically changed along with the time length of mode forecasting data of input characteristic variables.

Description

Sea fog visibility level forecasting method based on visibility meter and mesoscale mode Technical Field The application relates to the technical field of visibility forecast, in particular to a sea fog visibility level forecast method based on a visibility meter and a mesoscale mode. Background Sea fog is a weather phenomenon in which the visibility in the horizontal direction of the atmosphere is lower than 1 km due to the presence of a large number of water droplets or ice crystals in the ocean atmospheric boundary layer, and is one of the major ocean meteorological disasters. When the sea fog occurs, the sea navigation and harbor operation can be affected, and the coastal highway transportation is blocked or even closed. According to statistics, 60% -70% of collision accidents occurring at sea are related to sea fog, and the economic loss caused by the collision accidents is comparable with disasters such as typhoons. It follows that the low visibility events of sea fog have a significant impact on coastal production safety and economic development, and therefore the need for their advanced prediction is highly urgent. Current sea fog forecasting mainly relies on the following three types of methods and combinations thereof: (1) And the numerical mode diagnosis method is used for qualitatively judging the sea fog influence area based on physical quantities such as the atmospheric liquid water content and the like output by the numerical mode. The challenge of the method is that the sea fog forecast itself has complexity that the formation mechanism is influenced by local climate and weather background, and the space-time variability is strong. Meanwhile, the numerical mode has obvious uncertainty on parameterization schemes of key processes such as turbulence activity in an atmosphere boundary layer and cloud microphysics, and the sea fog forecast is a recognized problem in the field of numerical weather forecast together with errors of input data such as an initial field and sea surface temperature. (2) And the statistical model method is to establish a statistical model to carry out qualitative forecast by screening effective forecasting factors. The method is highly dependent on massive observation data to calibrate and optimize the forecast threshold, the existing method is extremely lack of atmospheric element observation information of the upper air of the low-layer sea surface, the performance of the statistical model is easily affected by local characteristics and seasonal changes, and the generalization capability is limited. (3) The machine learning method is to train a machine learning model by utilizing historical observation data so as to realize the approach forecast of the visibility. The method is also strongly dependent on high quality, long time-series observations. The key limitation is that if the input precursor factors fail to capture the change trend of low visibility, huge deviation of the model forecasting result occurs, and the forecasting stability is challenging. Disclosure of Invention Aiming at the problems and the technical requirements, the application provides a sea fog visibility level forecasting method based on a visibility meter and a mesoscale mode, and the technical scheme of the application is as follows: a sea fog visibility level forecasting method based on a visibility meter and a mesoscale mode comprises the following steps: preprocessing original observation data acquired by a visibility meter in a historical scene to obtain visibility level observation data consistent with the output data time step of a mesoscale numerical mode; performing feature extraction on a historical mode data set of multiple-height-layer meteorological elements in a local area where a visibility meter is located, which is output by a medium-scale numerical mode in a historical scene, obtaining mode forecast data of multiple meteorological elements and derivative elements thereof, wherein each timestamp is related to the visibility, in the local area where the visibility meter is located, and screening out mode forecast data of multiple feature variables which are most strongly related to the visibility; after a characteristic sample set taking mode forecast data of each characteristic variable as input and visibility level observation data of the same time stamp as a sample label is obtained, training the characteristic sample set based on a machine learning algorithm to obtain a visibility forecast model; and extracting multi-height-layer meteorological elements of the region to be forecasted output in the real-time operation scene by the middle-scale numerical mode to obtain mode forecast data of each characteristic variable, and inputting the mode forecast data into a visibility forecast model obtained by training to obtain a sea fog visibility level forecast result of the region to be forecasted. The method for obtaining the model forecast data of the multiple meteorological elem