A method, system, computer device and storage medium for predicting shipping visibility

Through deep learning methods, a visibility prediction model is constructed by combining meteorological characteristic factors and boundary layer atmospheric circulation conditions, which solves the short-term and short-term prediction problems of low visibility prediction in water shipping, achieves more accurate prediction results, and reduces the water traffic Influence.

CN118656682BActive Publication Date: 2024-12-03SUZHOU METEOROLOGICAL BUREAU +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411132246.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-12-03
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

The existing visibility prediction method cannot be effectively applied to short-term early warning and short-term prediction of water shipping, especially in the Yangtze River waters due to the special meteorological conditions, the low visibility weather has a huge impact on waterway transportation. The existing method has a computing power and modeling. High cost and unsatisfactory results.

Method used

Deep learning method is used to combine meteorological characteristic factors and boundary layer atmospheric circulation conditions, and to build a library of low-visibility events and use long and short-term memory neural networks to predict visibility, extract historical meteorological, boundary layer microwave radiometer and wind profile radar data. Determine the weight coefficient of meteorological factors and establish a visibility prediction model.

Benefits of technology

Improve the accuracy and timeliness of low-visibility predictions, reduce the impact of low-visibility weather on water traffic, respond to sudden disasters in advance, reduce personnel and property losses, improve the safety of water transportation and meteorological disaster warning ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118656682B_ABST
    Figure CN118656682B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, system, computer equipment and storage medium for predicting shipping visibility, comprising: obtaining historical meteorological data, historical boundary layer microwave radiometer data, historical wind profile radar data and historical high-altitude circulation diagram; constructing a low-visibility event case library; classifying the low-visibility event case library to obtain multiple low-visibility event case library; adding the boundary layer microwave radiometer data and wind profile radar data on the day when the low-visibility event occurs to the corresponding low-visibility event case library; determining the weight coefficient of each meteorological factor in each low-visibility event case library, and reconstructing each low-visibility event case library using each meteorological factor and its weight coefficient; training a long-short-term memory neural network using each reconstructed low-visibility event case library to obtain a visibility prediction model under each category; and obtaining a visibility prediction result using the visibility prediction model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological forecasting, and specifically relates to a shipping visibility forecasting method, system, computer equipment and storage medium. Background Art

[0002] Atmospheric visibility is an indicator of atmospheric transparency. In meteorology, weather phenomena with horizontal visibility less than 1 km caused by precipitation, fog, haze, dust and other processes are called low-visibility weather. Low-visibility weather seriously affects air, ground and waterway traffic safety every year, causing accidents and traffic blockages. In the past, visibility prediction methods were mainly concentrated on statistical methods and numerical model prediction methods. Among them, traditional statistical models are mostly based on empirical models and linear regression models. Empirical models are mainly based on subjective experience, and the overall objective prediction degree is low. Linear regression models are mostly based on the linear relationship between meteorological factors and visibility factors. Their biggest disadvantage is that they cannot fit nonlinear processes, and most atmospheric low visibility processes are nonlinear. Among them, traditional visibility prediction methods based on numerical models mainly use parameterization schemes established by corresponding physical and chemical mechanisms to construct meteorological numerical model systems. Visibility forecasts based on numerical models are highly dependent on the accuracy of meteorological factors predicted by the model, resulting in low visibility forecasts that still show relatively unstable and poor performance. The system of numerical models is large, the construction cost is high, and the server computing power requirements are high. The accuracy of the model initial value is often affected by the observation process and system errors, and the model does not fully describe its physical and dynamic processes.

[0003] With the rapid development of information technology, artificial intelligence technology based on machine learning has occupied an important position in different fields of society. It has been widely and effectively applied to prediction and supervised learning problems in many fields. Among them, visibility prediction based on support vector machine method, BP neural network visibility prediction model and convolutional neural network model are widely studied. Among them, the visibility prediction based on support vector machine method is mainly a supervised learning algorithm based on statistical learning theory. Its goal is to consider the minimum structural risk and seek an optimal classification. In the prediction model, multiple meteorological elements are generally used as input vectors and visibility is the output vector. However, it is difficult to find a suitable kernel function for the visibility prediction based on support vector machine method for nonlinear problems. In addition, the computer memory consumption is large, and the operation and parameter adjustment are also time-consuming and laborious. Among them, the BP neural network visibility prediction model is a gradient descent supervised algorithm, which includes two processes: forward propagation of feature information and back propagation of error. However, the performance of the BP neural network visibility prediction model depends more on the number of hidden layers and the number of neurons in each layer in the neural network model. However, there is a certain contradiction between the two. Good convergence requires as few hidden layer neurons as possible, but it will significantly increase the error of the test sample. Therefore, it is difficult to find an optimal state to build a superior visibility prediction model. Among them, the convolutional neural network (CNN) is a multi-layer feedforward neural network containing a special hidden layer. It needs to define a convolution kernel. The convolution layer uses the kernel for convolution, and the value of the kernel is a trainable parameter. The typical architecture includes a convolution layer for feature extraction, followed by a fully connected layer for prediction based on the potential highly nonlinear relationship between the feature and the target class. However, the convolutional neural network is mainly based on the visibility value inverted after image or video recognition, which still has a certain deviation from the actual observed value. The accuracy and timeliness need to be improved. Atmospheric low visibility weather often occurs suddenly and disappears rapidly, which is often not conducive to quickly capturing the evolution characteristics of its visibility.

[0004] In addition, various prediction methods are currently mainly concentrated in the fields of highway and aviation, while waterway shipping has higher requirements for visibility and a wider impact. At the same time, due to the evaporation of water bodies in the Yangtze River waters, rich water vapor and small temperature dew point difference, the probability of low visibility weather is higher. This type of weather has a greater adverse impact on the transportation of bulk cargo on the Yangtze River waterway, and the economic losses caused are greater. In particular, low visibility weather in autumn and winter has a huge impact on waterway transportation. However, the current visibility prediction model (mode) consumes high computing power and modeling costs, the visibility forecast effect is not ideal, and there are still shortcomings in the simulation and inversion of low atmospheric visibility. Summary of the invention

[0005] Purpose of the invention: To solve the problem that existing visibility prediction methods cannot be applied to short-term warning and short-term prediction of water shipping visibility, the present invention discloses a shipping visibility prediction method, system, computer equipment and storage medium. Based on meteorological characteristic factors, high and low altitude and boundary layer atmospheric circulation conditions, deep learning methods are combined with meteorological elements to realize visibility prediction along water shipping routes.

[0006] Technical solution: A shipping visibility prediction method, comprising the following steps:

[0007] Step 1: Obtain historical meteorological data along the shipping route, historical boundary layer microwave radiometer data along the shipping route, historical wind profile radar data along the shipping route, and historical upper air circulation maps along the shipping route;

[0008] Step 2: Based on the set first low visibility threshold, the historical meteorological data is screened for individual cases, and the individual cases on the day when the low visibility event occurred are screened out, wherein the individual cases include one or more meteorological factor data affecting the low visibility; the individual cases are sorted into a long sequence low visibility event case library; the meteorological factor data includes meteorological factors and their corresponding values;

[0009] Step 3: Classify the low visibility event case library according to the historical upper air circulation map to obtain multiple low visibility event case libraries, where one low visibility event case library corresponds to one category;

[0010] Step 4: Filter out the boundary layer microwave radiometer data on the day when the low visibility event occurred from the historical boundary layer microwave radiometer data, and add it to the corresponding low visibility event case database; filter out the wind profiler radar data on the day when the low visibility event occurred from the historical wind profiler radar data, and add it to the corresponding low visibility event case database;

[0011] Step 5: using the set second low visibility threshold, determine the weight coefficient of each meteorological factor in each low visibility event case database, and reconstruct each low visibility event case database using each meteorological factor and its weight coefficient to obtain multiple reconstructed low visibility event case databases;

[0012] Step 6: Use each reconstructed low visibility event case database to train the long short-term memory neural network to obtain a visibility prediction model for each category;

[0013] Step 7: Obtain the meteorological factor data at the current moment, the boundary layer microwave radiometer data at the current moment, the wind profiler radar data at the current moment, and the upper-altitude circulation diagram of the numerical model at the current moment, and determine the category to which the meteorological factor data at the current moment belongs according to the upper-altitude circulation diagram of the numerical model at the current moment, and input the meteorological factor data at the current moment, the boundary layer microwave radiometer data at the current moment, and the wind profiler radar data at the current moment into the visibility prediction model under the corresponding category to obtain the visibility prediction result.

[0014] Furthermore, the historical meteorological data includes historical meteorological data observed by meteorological stations located along the shipping route and historical meteorological data collected by transportation stations located along the shipping route.

[0015] Furthermore, the historical upper-air circulation map is drawn from the European Center ERA5 reanalysis data; the European Center ERA5 reanalysis data is grid data with the same regional scope as the shipping line and longitude and latitude close to the national basic meteorological station.

[0016] Furthermore, step 2 specifically includes:

[0017] Based on the set first low visibility threshold, all meteorological factors and their values ​​on the day when the low visibility event occurs are screened out from the historical meteorological data, wherein the low visibility event is an event in which the minimum visibility value on that day is lower than the first low visibility threshold;

[0018] For any meteorological factor among all meteorological factors on the day when the low visibility event occurs, calculate the relationship between visibility and the meteorological factor according to the following formula: spearman Relevance:

[0019]

[0020] In the formula, for spearman Correlation coefficient, and They represent the visibility value at the i-th moment on the day when the low visibility event occurs and the meteorological factor value at the i-th moment on the day when the low visibility event occurs, and They respectively represent the average value of the visibility value on the day when the low visibility event occurs and the average value of the meteorological factor value on the day when the low visibility event occurs;

[0021] According to the relationship between visibility and various meteorological factors spearman Correlation, screening out one or more meteorological factor data that affect low visibility; the meteorological factor data includes meteorological factors and their corresponding values;

[0022] The data of one or more meteorological factors that affect low visibility are sorted into individual cases on the day when the low visibility event occurs, and a long-sequence low visibility event case library is obtained.

[0023] Furthermore, step 5 specifically includes:

[0024] For any low visibility event example library , perform the following steps:

[0025] According to the second low visibility threshold, the low visibility event cases are collected into the database. Divided into and Two parts;

[0026] According to the correlation fitting, the maximum possible value of each meteorological factor under the second lowest visibility threshold is determined;

[0027] Using the maximum possible value of each meteorological factor in turn, and The two parts are further divided; then the Gini index corresponding to each meteorological factor is calculated according to the following formula, and the Gini index is used as a database of meteorological factors in the low visibility event case The weight coefficient on :

[0028]

[0029]

[0030]

[0031] In the formula, represents the meteorological factor, Take 2, ,for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor value; for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor;

[0032] Based on various meteorological factors in the low visibility event case database The weight coefficient on the low visibility event case database Refactoring

[0033] In this way, each low-visibility event case database is reconstructed to obtain multiple reconstructed low-visibility event case databases.

[0034] The present invention also discloses a shipping visibility prediction system, comprising: a data acquisition module, used to acquire historical meteorological data along the shipping route, historical boundary layer microwave radiometer data along the shipping route, historical wind profile radar data along the shipping route, and historical high-altitude circulation diagrams along the shipping route;

[0035] The case screening module is used to screen the historical meteorological data based on the first low visibility threshold, and screen out the cases on the day when the low visibility event occurs, wherein the cases include one or more meteorological factor data affecting the low visibility; and organize the cases into a long sequence of low visibility event case libraries; the meteorological factor data includes meteorological factors and their corresponding values;

[0036] The low visibility event case library construction module is used to classify the low visibility event case library according to the historical upper air circulation map to obtain multiple low visibility event case libraries, and one low visibility event case library corresponds to one category; the boundary layer microwave radiometer data on the day when the low visibility event occurred is screened out from the historical boundary layer microwave radiometer data, and added to the corresponding low visibility event case library; the wind profiler radar data on the day when the low visibility event occurred is screened out from the historical wind profiler radar data, and added to the corresponding low visibility event case library;

[0037] A low visibility event case library reconstruction module is used to determine the weight coefficient of each meteorological factor in each low visibility event case library by using the set second low visibility threshold, and reconstruct each low visibility event case library by using each meteorological factor and its weight coefficient to obtain multiple reconstructed low visibility event case libraries;

[0038] A visibility prediction model training module is used to train the long short-term memory neural network using each reconstructed low visibility event case library to obtain a visibility prediction model for each category;

[0039] The visibility prediction result output module is used to obtain the meteorological factor data at the current moment and the high-altitude circulation diagram of the numerical model at the current moment, determine the category to which the meteorological factor data at the current moment belongs according to the high-altitude circulation diagram of the numerical model at the current moment, and input the meteorological factor data at the current moment into the visibility prediction model under the corresponding category to obtain the visibility prediction result.

[0040] Furthermore, the historical meteorological data includes historical meteorological data observed by meteorological stations located along the shipping route and historical meteorological data collected by traffic stations located along the shipping route;

[0041] The historical upper-air circulation map is drawn from the European Center ERA5 reanalysis data; the European Center ERA5 reanalysis data is grid data with the same regional scope as the shipping line and longitude and latitude close to the national basic meteorological station.

[0042] Furthermore, the second low visibility threshold value is set to determine the weight coefficient of each meteorological factor in each low visibility event case library, and each meteorological factor and its weight coefficient are used to reconstruct each low visibility event case library to obtain multiple reconstructed low visibility event case libraries, specifically including:

[0043] For any low visibility event example library , perform the following steps:

[0044] According to the second low visibility threshold, the low visibility event cases are collected into the database. Divided into and Two parts;

[0045] According to the correlation fitting, the maximum possible value of each meteorological factor under the second lowest visibility threshold is determined;

[0046] Using the maximum possible value of each meteorological factor in turn, and The two parts are further divided; then the Gini index corresponding to each meteorological factor is calculated according to the following formula, and the Gini index is used as a database of meteorological factors in the low visibility event case The weight coefficient on :

[0047]

[0048]

[0049]

[0050] In the formula, represents the meteorological factor, Take 2, ,for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor value; for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor;

[0051] Based on various meteorological factors in the low visibility event case database The weight coefficient on the low visibility event case database Refactoring

[0052] In this way, each low-visibility event case database is reconstructed to obtain multiple reconstructed low-visibility event case databases.

[0053] The invention discloses a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a shipping visibility prediction method are implemented.

[0054] The invention discloses a storage medium storing a visibility prediction program. When the visibility prediction program is executed by at least one processor, the steps of a shipping visibility prediction method are realized.

[0055] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0056] (1) The method of the present invention constructs a low visibility event case database, and uses the upper air circulation diagram to subjectively classify the low visibility event cases, and then inputs them into the visibility prediction model based on the long short-term memory neural network, so as to better grasp the prediction accuracy of similar low visibility processes in the future;

[0057] (2) The method of the present invention first determines the meteorological factors associated with low visibility, and then combines the characteristics of atmospheric circulation and meteorological factors under vertical observation, and attempts to study the formation mechanism of different low visibility events based on the interaction and influence of multiple key meteorological factors. For the occurrence of high-impact low visibility weather processes, the change characteristics of meteorological factors before and after the occurrence of low visibility weather processes are inverted, and the case library is classified according to the formation and evolution mechanism;

[0058] (3) The method of the present invention uses the deep learning technology of long short-term memory neural network to carry out modeling and model prediction for low visibility, and can obtain visibility prediction results in advance. Based on the accurate visibility prediction results, it will effectively reduce the direct impact of low visibility weather on water transportation, respond to sudden disaster weather in advance, reduce unnecessary casualties and property losses, ensure the safety and normal operation of transportation, accelerate the development of water transportation, effectively improve the meteorological disaster warning capabilities of waterways, and achieve significant social and economic benefits in meteorological disaster prevention, disaster reduction, and disaster defense for water transportation;

[0059] (4) Based on the prediction results of the method of the present invention, warning levels can be divided to ensure the safety and normal operation of traffic. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of a flow chart of a shipping visibility prediction method of the present invention;

[0061] Figure 2 This is a schematic diagram of the construction, training and prediction process of the visibility prediction model;

[0062] Figure 3 This is a schematic diagram of the structure of a long short-term memory neural network;

[0063] Figure 4 This is a schematic diagram of visibility forecast and warning service;

[0064] Figure 5 These are the meteorological data of Zhangjiagang Station when low visibility events occurred along the river from November 20 to November 30, 2018;

[0065] Figure 6 The visibility forecast effect is 5 days in advance;

[0066] Figure 7 The visibility forecast effect is 4 days in advance;

[0067] Figure 8 The visibility forecast effect is 3 days in advance;

[0068] Fig. 9 The visibility forecast effect is 2 days in advance;

[0069] Fig.10 The visibility forecast effect for one day in advance;

[0070] Fig.11 To provide visibility forecast effect 12 hours in advance;

[0071] Fig.12 The visibility forecast effect is 6 hours in advance;

[0072] Fig.13 The visibility forecast effect is 3 hours in advance. DETAILED DESCRIPTION

[0073] The technical solution of the present invention is now further described in conjunction with the accompanying drawings and embodiments.

[0074] Example 1: A method for predicting shipping visibility disclosed in this example is used to predict visibility along the Yangtze River. Figure 1 As shown, it mainly includes the following steps:

[0075] Step 1: Obtain historical meteorological data along the Yangtze River, historical boundary layer microwave radiometer data along the Yangtze River, historical wind profiler radar data along the Yangtze River, and historical high-altitude circulation maps along the Yangtze River. The purpose of taking these four types of data in this embodiment is that in the past, sounding data were mostly used to obtain vertical observations of meteorological factors in the high-altitude-boundary layer, but because their spatial resolution and temporal resolution are very rough, and microwave radiometers and wind profiler radars have higher spatiotemporal resolutions, in order to theoretically explain and more finely classify the formation mechanisms of different low-visibility processes that occur, microwave radiometer data and wind profiler radar data can observe meteorological characteristics (including temperature, humidity and wind) in the vertical direction, and high-altitude circulation maps are mainly used to judge atmospheric circulation fluctuations in the horizontal direction. Since the high altitude is less affected by human factors, the atmospheric circulation is relatively stable.

[0076] Specifically, the historical meteorological data include long-sequence historical meteorological data observed by 20 national basic meteorological stations along the Yangtze River and historical meteorological data collected by 60 transportation stations along the Yangtze River; in this embodiment, the historical meteorological data collected by the transportation stations include but are not limited to near-ground 2m air pressure values, air temperature values, dew point temperature values, water vapor pressure values, relative humidity values, evaporation values, wind direction and speed values, and minimum visibility values.

[0077] Specifically, the historical boundary layer microwave radiometer data located along the Yangtze River include but are not limited to temperature values, liquid water content values, relative humidity values ​​and water vapor density values ​​in the vertical direction of 0-3km.

[0078] Specifically, the historical wind profile radar data along the Yangtze River include but are not limited to wind direction and wind speed values ​​in the vertical direction of 0-3km in the boundary layer (below 1500 meters).

[0079] Specifically, the historical high-altitude circulation map along the Yangtze River is drawn from the European Center ERA5 reanalysis data. The European Center ERA5 reanalysis data used here is the grid data of the same area along the Yangtze River (100°E-125°E east longitude, 25°N-35°N north latitude) close to 20 national basic meteorological stations. The grid data includes but is not limited to 500hPa / 700hPa pressure values, temperature values, humidity values ​​and wind fields. The purpose of collecting grid data close to 20 national basic meteorological stations is to ensure that the spatial scale matches the historical meteorological data.

[0080] The collection time periods of historical meteorological data along the Yangtze River, historical boundary layer microwave radiometer data along the Yangtze River, historical wind profile radar data along the Yangtze River, and historical high-altitude circulation maps along the Yangtze River must be consistent. The time resolution of the long-series historical meteorological data observed by the national basic meteorological stations is 5 minutes, and the time resolution of the historical meteorological data collected by the transportation stations is 1 minute; the time resolution of the historical wind profile radar data is 5 minutes. The spatial resolution of the ERA5 reanalysis data of the European Center is 0.1°×0.1°, and the time resolution is 1 hour. Due to the differences in the time resolution of various types of data, it is necessary to match the data on a time scale.

[0081] In order to ensure data quality, it is necessary to preprocess the historical meteorological data along the Yangtze River, the historical boundary layer microwave radiometer data along the Yangtze River, the historical wind profiler radar data along the Yangtze River, and the historical high-altitude circulation maps along the Yangtze River. The data preprocessing includes but is not limited to: data quality control and data normalization.

[0082] Specifically, data quality control includes: filling in missing values ​​in the above four types of data and eliminating outliers or erroneous values ​​in the above four types of data; filling in missing values ​​includes: for a single missing value, filling in the gaps with the previous and next values ​​of the time series; for missing values ​​in a long series period, filling in the gaps with the spatial nearest neighbor interpolation method.

[0083] Specifically, the data normalization process includes: using a discrete normalization method to perform data normalization, which performs a linear transformation on the data after data quality control processing, maps the data value to between [0,1], and uses the following formula:

[0084] (1)

[0085] In the formula, represents discrete normalized data, Represents the data after data quality control processing. Indicates the maximum value of the data type; Indicates the minimum value of the data type.

[0086] Step 2: Based on the first low visibility threshold, perform case screening on the historical meteorological data to screen out cases on the day when the low visibility event occurred. The cases in this embodiment refer to one or more meteorological factor data that affect low visibility; organize the cases into a long-sequence low visibility event case library; the meteorological factor data in this embodiment includes meteorological factors and their corresponding numerical values.

[0087] The step of performing case-by-case screening of historical meteorological data based on the first low visibility threshold is now further described.

[0088] Based on the pre-set first low visibility threshold, the historical meteorological data along the Yangtze River after data quality control and data normalization are screened for individual cases, and all meteorological factors and their values ​​on the day (0:00-24:00) when the low visibility event for shipping along the Yangtze River occurs are screened out. The low visibility event for shipping along the Yangtze River is defined as an event in which the minimum visibility value is lower than the first low visibility threshold; individual cases are further screened out from all meteorological factors and their values ​​on the day (0:00-24:00) when the low visibility event for shipping along the Yangtze River occurs. The composition of the individual cases includes one or more meteorological factors affecting low visibility and their specific values, and the individual cases are organized into a long-sequence low visibility event case library.

[0089] In this embodiment, the meteorological factors affecting low visibility are determined according to the following steps:

[0090] For any meteorological factor, the relationship between visibility and the meteorological factor is calculated according to formula (2). spearman Relevance:

[0091] (2)

[0092] In the formula, for spearman Correlation coefficient, and They represent the visibility value at the i-th moment on the day when the low visibility event occurs and the meteorological factor value at the i-th moment on the day when the low visibility event occurs, and They respectively represent the average value of the visibility value on the day when the low visibility event occurs and the average value of the meteorological factor value on the day when the low visibility event occurs;

[0093] Based on the relationship between visibility and different meteorological factors spearmanThe meteorological factors affecting low visibility are determined by the size of the correlation coefficient. In this embodiment, the meteorological factors affecting low visibility are determined to be air pressure, temperature, relative humidity, and wind direction and speed through the above steps. Due to different setting values ​​of the first low visibility threshold, the meteorological factors affecting low visibility determined through the above steps are not necessarily limited to air pressure, temperature, relative humidity, and wind direction and speed.

[0094] Step 3: Because the time resolution of historical meteorological data and the European Center ERA5 reanalysis data has been matched in step 1, the time resolution of the low visibility event case library also matches the time resolution of the European Center ERA5 reanalysis data. Most studies have shown that the occurrence of low visibility processes has its own specific meteorological circulation characteristics. Different from the ground circulation map, the atmospheric fluctuations in the high-altitude circulation map are less affected by human interference and surface buildings, and the atmospheric circulation field is more stable. In order to seek historical similar processes for the formation of low visibility processes, the historical high-altitude circulation map can be used to classify the formation and evolution mechanisms of the low visibility event case library, and obtain a low visibility event case library under multiple categories. One low visibility event case library corresponds to one category; the classification in this step is artificial classification.

[0095] Step 4: In the past, sounding data were mostly used to obtain vertical observations of meteorological factors in the upper air-boundary layer. However, due to their very coarse spatial and temporal resolutions, they had a poor grasp of the rapidly evolving low visibility process. After using microwave radiometers and wind profiler radars for observations, the temporal and spatial resolutions of meteorological characteristics in the vertical direction are higher, which can better and more accurately capture the formation and evolution of low visibility processes in the vertical direction. Therefore, it is necessary to screen out the boundary layer microwave radiometer data on the day when the low visibility event occurred from the historical boundary layer microwave radiometer data, and add it to the corresponding low visibility event case database; screen out the wind profiler radar data on the day when the low visibility event occurred from the historical wind profiler radar data, and add it to the corresponding low visibility event case database;

[0096] This embodiment is based on the high-altitude circulation diagram (transmission mechanism on the horizontal plane) and vertical observation data, including microwave radiometers and wind profiler radars (transmission mechanism in the vertical direction). The differences in different atmospheric circulation fields and vertical meteorological elements lead to different types of low visibility weather. That is to say, when low visibility weather occurs on the ground, its formation mechanism is different. Some are caused by low clouds propagating vertically downward and low visibility weather after the low clouds touch the ground. Some are caused by transmission from land to the surface of the Yangtze River. Some are caused by precursors such as aerosols being transmitted downstream by the prevailing wind direction on the ground and gathering on the downstream river surface to form low visibility weather. Therefore, a case library classification is made based on the formation and evolution mechanism.

[0097] Step 5: Using the set second low visibility threshold, calculate the weight coefficient of each meteorological factor in each low visibility event case database, reconstruct each low visibility event case database based on the weight coefficient of each meteorological factor, and obtain multiple reconstructed low visibility event case databases; the specific operations include:

[0098] For any low visibility event example library , perform the following steps:

[0099] According to the second low visibility threshold, the low visibility event cases are collected into the database. Divided into and Two parts; in this embodiment, the second low visibility threshold is set according to the Yangtze River shipping visibility warning, such as 2000m, 1500m, 1000m and 500m.

[0100] According to the correlation fitting, the maximum possible value of each meteorological factor under the second lowest visibility threshold is determined; for example, when the second lowest visibility threshold is 500, according to the correlation fitting, the maximum possible value of relative humidity is determined to be 95%;

[0101] For the maximum possible value of any meteorological factor, use the maximum possible value of the meteorological factor to and The two parts are further divided; then the meteorological factor is calculated according to the following formula The corresponding Gini index , the meteorological factors Gini index As the meteorological factor In this low visibility event example library The weight coefficient on :

[0102] (3)

[0103] (4)

[0104] (5)

[0105] In the formula, represents the meteorological factor, Take 2, ,for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor value; for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor;

[0106] Similarly, is the air pressure, is the temperature, is the relative humidity, is the wind speed, and for air pressure, its weight coefficient in the low visibility event case database is , for temperature, its weight coefficient on the low visibility event case database is , for relative humidity, its weight coefficient on the low visibility event case database is , for wind speed, its weight coefficient in the low visibility event case database is .

[0107] Based on the weight coefficients obtained above, a regression mapping for the minimum visibility is established, and a database of low visibility event cases is reconstructed:

[0108] in, , , , , Respectively t Minimum visibility at the time, t The air pressure at the time, t The temperature at the time, t Relative humidity at the time, t Wind speed at the moment.

[0109] In this way, each low-visibility event case database is reconstructed to obtain multiple reconstructed low-visibility event case databases.

[0110] Step 6: Construct a visibility prediction model based on the long short-term memory neural network, use the low visibility event case database reconstructed in step 5 to train the visibility prediction model, use the loss function to converge the visibility prediction model, and finally obtain the converged visibility prediction model. The model training process is as follows: Figure 2 .

[0111] The long short-term memory neural network is now further explained.

[0112] like Figure 3As shown, the basic building block of the long short-term memory neural network is a neuron unit, which uses three gates to protect and adjust the state of the neuron unit, including an input gate that controls the current input information, a forget gate that controls the historical input information, and an output gate that controls the output information. This embodiment chooses to use sigmoid The function selects (0, 1) information. When the information is completely discarded sigmoid The function output is 0 when the information is completely retained. sigmoid The function output is 1. Each gate is actually a fully connected layer, which keeps the characteristic factors in the input data and discards other factors. The characteristic factors are repeatedly trained and modeled, and finally output a more accurate prediction result. The structure of the long short-term memory neural network is shown in Figure 3 . Figure 3 middle , , They represent input gate, output gate and forget gate respectively. is the state vector of the storage unit at the current moment, Represents the state vector of the storage unit at the previous moment, represents the information candidate state vector, represents the output vector of a single neuron at the current moment, and Represents the weight matrix and bias vector.

[0113] The input gate, output gate, forget gate, weight matrix, and bias vector are calculated based on the following formula:

[0114] (6)

[0115] (7)

[0116] (8)

[0117] (9)

[0118] (10)

[0119] (11)

[0120] in, , .

[0121] In the formula, express sigmoid Function, the output is 0~1, and represents the weight matrix and bias vector of the forget gate, and Represents the weight matrix and bias vector of the input gate; and Represents the weight matrix and bias vector of the output gate; represents the output vector of a single neuron at the previous moment, Indicates the input variable, indicates the variable input at the current moment, represents the tangent function, Represents the information candidate state vector.

[0122] Meteorological data has strong spatiotemporal continuity, and historical meteorological data in the same time period or the same season may have great similarities. The loss function is used for model training and convergence, and the L1 norm regularization is used to prevent the loss function from overfitting to achieve better generalization ability. At the same time, the adaptive learning rate Adam algorithm is used for model optimization to accelerate model convergence and obtain an optimized model.

[0123] This embodiment uses the mean square error function as the loss function:

[0124] (12)

[0125] In the formula, represents the true value, represents the predicted value, Indicates the sample size.

[0126] use Norm regularization prevents overfitting of the loss function. The specific operations include:

[0127] Adding in the supervised learning process of long short-term memory neural network After norm regularization (i.e. minimum absolute error):

[0128] (13)

[0129] Therefore, the final loss function is expressed as:

[0130] (14)

[0131] In the formula, represents the regularization term, represents the regularization coefficient, which is used to control the strength of regularization. Represents the weight vector of the input i-th sample; Represents the original loss function. In this embodiment, the original loss function is the mean square error function.

[0132] because Regularization will cause more parameters to become 0. In this embodiment, features with parameters set to 0 are discarded and features are selected.

[0133] Multiple model evaluation indicators are used to evaluate the performance of the visibility prediction model, including:

[0134] The root mean square error of the indicator based on the visibility prediction value ( RMSE ):

[0135] (15)

[0136] Mean absolute error (MAE ):

[0137] (16)

[0138] Fitting coefficients ( ):

[0139] (17)

[0140] In the formula, is the actual observation data set, Predict data sets for the model, is the mean value of the observed data set.

[0141] Adoption accuracy:

[0142] (18)

[0143] Accuracy:

[0144] (19)

[0145] Recall:

[0146] (20)

[0147] F-score:

[0148] (twenty one)

[0149] In the formula, It means that both the actual observation value and the model prediction value meet the set visibility warning standard. It means that the actual observation value meets the set visibility warning standard and the model prediction value does not meet the set visibility warning standard. It means that the actual observation value does not meet the set visibility warning standard and the model prediction value meets the set visibility warning standard. It means that both the actual observation value and the model prediction value do not meet the set visibility warning standard.

[0150] Since a low visibility event example database corresponds to a category, a visibility prediction model is trained for each category according to the above process.

[0151] Step 7: Obtain the meteorological factor data at the current moment, the boundary layer microwave radiometer data at the current moment, the wind profiler radar data at the current moment, and the upper-altitude circulation diagram of the numerical model at the current moment, and determine the category to which the meteorological factor data at the current moment belongs according to the upper-altitude circulation diagram of the numerical model at the current moment, and input the meteorological factor data at the current moment, the boundary layer microwave radiometer data at the current moment, and the wind profiler radar data at the current moment into the visibility prediction model under the corresponding category to obtain the visibility prediction result.

[0152] like Figure 4 As shown, in this embodiment, classification is performed using visibility prediction results and warning level classification standards.

[0153] The warning level classification standards adopted in this embodiment include: red warning level (0≤vis≤500m); orange warning level (500<vis≤1000m); yellow warning level (1000<vis≤1500m); blue warning level (1500<vis≤2000m). Vis represents the visibility prediction result.

[0154] A visibility warning level map of the Yangtze River shipping route is drawn by classification based on visibility forecast results. The visibility warning level map of the Yangtze River shipping route can be used as a reference for decision-making by meteorological, emergency, water affairs and transportation departments to reduce casualties and property losses in shipping and transportation.

[0155] The technical effect of this embodiment is now illustrated by taking the low visibility incident that occurred along the river from November 20 to November 30, 2018 as an example.

[0156] Figure 5 The figure shows the meteorological data of Zhangjiagang Station when low visibility events occurred along the river from November 20 to November 30, 2018, among which: Figure 5 (a) in the figure is the air pressure data. Figure 5 (b) in the figure is the temperature data. Figure 5 (c) in the figure is the relative humidity data. Figure 5 (d) in the figure is the temperature dew point difference data. Figure 5 (e) in is the wind speed data, Figure 5 (f) in the figure is the 10-minute average visibility data. Figure 5 The X-axis coordinate in is one element every 5 minutes, and 12 data are 1 hour. Figure 5The box in (f) indicates the period of low visibility. The characteristics of the low visibility event along the river are: the low visibility process lasts for a long time, is strong in intensity and wide in range, showing a trend of strengthening at night and weakening during the day, with the minimum visibility less than 50 meters. The low visibility area appeared along the Yangtze River in the early stage, and later expanded to most of Jiangsu Province. The strongest and concentrated period of low visibility was from the afternoon of the 26th to the 27th. In the early stage of the process, the air pressure rose first due to the diffusion of weak cold air, and later turned to calm and stable weather (the air pressure dropped to between 1020 and 1024 hpa), the temperature was between 5 and 20 °C, the relative humidity was above 90% when low visibility occurred, the temperature dew point difference (i.e. temperature minus dew point temperature) was below 1 °C, and the wind speed was less than 2 m·s -1 .

[0157] The correlation analysis between the key meteorological factors and visibility during the low visibility process along the river found that the meteorological factors with a high correlation with visibility are mainly 2-meter wind speed, relative humidity, temperature, air pressure and temperature dew point difference, all of which meet the significant correlation of 99% confidence interval. The absolute values ​​of the correlation coefficients of relative humidity and temperature dew point difference are both above 0.8. The 2-meter wind speed and visibility are significantly positively correlated, and the temperature and visibility are significantly positively correlated, with a correlation coefficient close to 0.8 (the Pearson correlation coefficient is 0.788 and the spearman correlation coefficient is 0.816). The air pressure and visibility are significantly negatively correlated, with a Pearson correlation coefficient of -0.411.

[0158] Table 1 Correlation between different meteorological factors and visibility

[0159]

[0160] In the table, ** indicates that the correlation is significant when the confidence level (double test) is 0.01.

[0161] The visibility prediction model proposed in this embodiment is used to perform visibility prediction. Figure 6 The visibility forecast effect is shown 5 days in advance. Figure 7 The visibility forecast effect of 4 days in advance is shown. Figure 8 The visibility forecast effect 3 days in advance is shown. Fig. 9 The visibility forecast effect 2 days in advance is shown. Fig.10 The visibility forecast effect is shown one day in advance. Fig.11 The visibility forecast effect is shown 12 hours in advance. Fig.12 The visibility forecast effect is shown 6 hours in advance. Fig.13 The visibility prediction effect 3 hours in advance is shown. The visibility unit in each figure is meter (m). Figure 6~Figure 13The "actural" in the function represents the actual value, and the "pridict" represents the predicted value.

[0162] The model evaluation results are shown in Table 2, which indicates: The fitting degree is above 0.6, which is significantly improved compared with the previous objective prediction results, and the prediction results are 12 hours in advance. The fitting degree reaches 0.985, indicating that the low visibility prediction effect in the short-term period is very ideal; The fitting degree is above 0.9, and the prediction effect is good. The fitting degree decreases when the time is more than 72 hours or less than 6 hours. The fitting degree is between 0.641 and 0.880. In addition, as the prediction time is shortened, MAE, RMSE and MSE All have declined.

[0163] Table 2 Evaluation results of visibility prediction models under different prediction time periods

[0164]

[0165] Embodiment 2: This embodiment discloses a shipping visibility prediction system, comprising:

[0166] A data acquisition module, used to acquire historical meteorological data along the shipping route, historical boundary layer microwave radiometer data along the shipping route, historical wind profile radar data along the shipping route, and historical high-altitude circulation maps along the shipping route;

[0167] The case screening module is used to screen the historical meteorological data based on the first low visibility threshold, and screen out the cases on the day when the low visibility event occurs, wherein the cases include one or more meteorological factor data affecting the low visibility; and organize the cases into a long sequence of low visibility event case libraries; the meteorological factor data includes meteorological factors and their corresponding values;

[0168] The low visibility event case library construction module is used to classify the low visibility event case library according to the historical upper air circulation map to obtain multiple low visibility event case libraries, and one low visibility event case library corresponds to one category; the boundary layer microwave radiometer data on the day when the low visibility event occurred is screened out from the historical boundary layer microwave radiometer data, and added to the corresponding low visibility event case library; the wind profiler radar data on the day when the low visibility event occurred is screened out from the historical wind profiler radar data, and added to the corresponding low visibility event case library;

[0169] A low visibility event case library reconstruction module is used to determine the weight coefficient of each meteorological factor in each low visibility event case library by using the set second low visibility threshold, and reconstruct each low visibility event case library by using each meteorological factor and its weight coefficient to obtain multiple reconstructed low visibility event case libraries;

[0170] A visibility prediction model training module is used to train the long short-term memory neural network using each reconstructed low visibility event case library to obtain a visibility prediction model for each category;

[0171] The visibility prediction result output module is used to obtain the meteorological factor data at the current moment and the high-altitude circulation diagram of the numerical model at the current moment, determine the category to which the meteorological factor data at the current moment belongs according to the high-altitude circulation diagram of the numerical model at the current moment, and input the meteorological factor data at the current moment into the visibility prediction model under the corresponding category to obtain the visibility prediction result.

[0172] Furthermore, the historical meteorological data includes historical meteorological data observed by meteorological stations located along the shipping route and historical meteorological data collected by traffic stations located along the shipping route;

[0173] The historical upper-air circulation map is drawn from the European Center ERA5 reanalysis data; the European Center ERA5 reanalysis data is grid data with the same regional scope as the shipping line and longitude and latitude close to the national basic meteorological station.

[0174] Furthermore, the weight coefficient of each meteorological factor in each low visibility event case database is determined by using the set second low visibility threshold, and each low visibility event case database is reconstructed by using each meteorological factor and its weight coefficient to obtain multiple reconstructed low visibility event case databases, specifically including:

[0175] For any low visibility event example library , perform the following steps:

[0176] According to the second low visibility threshold, the low visibility event cases are collected into the database. Divided into and Two parts;

[0177] According to the correlation fitting, the maximum possible value of each meteorological factor under the second lowest visibility threshold is determined;

[0178] Using the maximum possible value of each meteorological factor in turn, and The two parts are further divided; then the Gini index corresponding to each meteorological factor is calculated according to the following formula, and the Gini index is used as a database of meteorological factors in the low visibility event case The weight coefficient on :

[0179] (3)

[0180] (4)

[0181] (5)

[0182] In the formula, represents the meteorological factor, Take 2, ,for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor value; for In terms of Indicated in Medium or greater than the meteorological factor The maximum possible value of the meteorological factor, Indicated in Smaller than meteorological factors The maximum possible value of the meteorological factor;

[0183] Based on various meteorological factors in the low visibility event case database The weight coefficient on the low visibility event case database Refactoring

[0184] In this way, each low-visibility event case database is reconstructed to obtain multiple reconstructed low-visibility event case databases.

[0185] Embodiment 3: This embodiment discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps disclosed in any of the above embodiments are implemented.

[0186] Embodiment 4: This embodiment discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps disclosed in any of the above embodiments are implemented.

[0187] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0188] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0189] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for predicting shipping visibility, characterized in that: The following steps are involved: Step 1: Obtain historical meteorological data along the shipping route, historical boundary layer microwave radiometer data along the shipping route, historical wind profile radar data along the shipping route, and historical upper air circulation maps along the shipping route; Step 2: Based on the set first low visibility threshold, the historical meteorological data is screened for individual cases, and the individual cases on the day when the low visibility event occurred are screened out, wherein the individual cases include one or more meteorological factor data affecting the low visibility; the individual cases are sorted into a long sequence low visibility event case library; the meteorological factor data includes meteorological factors and their corresponding values; Step 3: Classify the low visibility event case library according to the historical upper air circulation map to obtain multiple low visibility event case libraries, where one low visibility event case library corresponds to one category; Step 4: Filter out the boundary layer microwave radiometer data on the day when the low visibility event occurred from the historical boundary layer microwave radiometer data, and add it to the corresponding low visibility event case database; filter out the wind profiler radar data on the day when the low visibility event occurred from the historical wind profiler radar data, and add it to the corresponding low visibility event case database; Step 5: using the set second low visibility threshold, determine the weight coefficient of each meteorological factor in each low visibility event case database, and reconstruct each low visibility event case database using each meteorological factor and its weight coefficient to obtain multiple reconstructed low visibility event case databases; Step 6: Use each reconstructed low visibility event case database to train the long short-term memory neural network to obtain a visibility prediction model for each category; Step 7: Obtain the meteorological factor data at the current moment, the boundary layer microwave radiometer data at the current moment, the wind profiler radar data at the current moment, and the upper air circulation diagram of the numerical model at the current moment, determine the category to which the meteorological factor data at the current moment belongs according to the upper air circulation diagram of the numerical model at the current moment, input the meteorological factor data at the current moment, the boundary layer microwave radiometer data at the current moment, and the wind profiler radar data at the current moment into the visibility prediction model under the corresponding category to obtain the visibility prediction result; Step 5 specifically includes: For any low visibility event case database D, perform the following steps: According to the second low visibility threshold, the low visibility event case database D is divided into two parts, D1 and D2; According to the correlation fitting, the maximum possible value of each meteorological factor under the second lowest visibility threshold is determined; The maximum possible value of each meteorological factor is used in turn to further divide the two parts D1 and D2; then the Gini index corresponding to each meteorological factor is calculated according to the following formula, and the Gini index is used as the weight coefficient of the meteorological factor in the case database D of the low visibility event: In the formula, A represents the meteorological factor, K is 2, k=1, 2, for D1, C1 represents the meteorological factor value greater than or equal to the maximum possible value of meteorological factor A in D1, C2 represents the meteorological factor value less than the maximum possible value of meteorological factor A in D1; for D2, C1 represents the meteorological factor value greater than or equal to the maximum possible value of meteorological factor A in D2, C2 represents the meteorological factor value less than the maximum possible value of meteorological factor A in D2; Reconstructing the low visibility event case database D based on the weight coefficients of each meteorological factor on the low visibility event case database D; In this way, each low-visibility event case database is reconstructed to obtain multiple reconstructed low-visibility event case databases.

2. A method for predicting shipping visibility according to claim 1, characterized in that: The historical meteorological data include historical meteorological data observed by meteorological stations located along the shipping route and historical meteorological data collected by traffic stations located along the shipping route.

3. A method for predicting shipping visibility according to claim 1, characterized in that: The historical upper-air circulation map is drawn from the European Center ERA5 reanalysis data; the European Center ERA5 reanalysis data is grid data with the same regional scope as the shipping line and longitude and latitude close to the national basic meteorological station.

4. A method for predicting shipping visibility according to claim 1, characterized in that: Step 2 specifically includes: Based on the set first low visibility threshold, all meteorological factors and their values ​​on the day when the low visibility event occurs are screened out from the historical meteorological data, wherein the low visibility event is an event in which the minimum visibility value on that day is lower than the first low visibility threshold; For any meteorological factor among all meteorological factors on the day when the low visibility event occurs, the spearman correlation between visibility and the meteorological factor is calculated according to the following formula: In the formula, ρ cor is the spearman correlation coefficient, a i and b i They represent the visibility value at the i-th moment on the day when the low visibility event occurs and the meteorological factor value at the i-th moment on the day when the low visibility event occurs, and They respectively represent the average value of the visibility value on the day when the low visibility event occurs and the average value of the meteorological factor value on the day when the low visibility event occurs; According to the spearman correlation between visibility and various meteorological factors, one or more meteorological factor data affecting low visibility are screened out; the meteorological factor data includes meteorological factors and their corresponding values; The data of one or more meteorological factors that affect low visibility are sorted into individual cases on the day when the low visibility event occurs, and a long-sequence low visibility event case library is obtained.

5. A shipping visibility prediction system, characterized in that: include: A data acquisition module is used to acquire historical meteorological data along the shipping route, historical boundary layer microwave radiometer data along the shipping route, historical wind profile radar data along the shipping route, and historical high-altitude circulation maps along the shipping route; The case screening module is used to screen the historical meteorological data based on the first low visibility threshold, and screen out the cases on the day when the low visibility event occurs, wherein the cases include one or more meteorological factor data affecting the low visibility; and organize the cases into a long sequence of low visibility event case libraries; the meteorological factor data includes meteorological factors and their corresponding values; The low visibility event case library construction module is used to classify the low visibility event case library according to the historical upper air circulation map to obtain multiple low visibility event case libraries, and one low visibility event case library corresponds to one category; the boundary layer microwave radiometer data on the day when the low visibility event occurred is screened out from the historical boundary layer microwave radiometer data, and added to the corresponding low visibility event case library; the wind profiler radar data on the day when the low visibility event occurred is screened out from the historical wind profiler radar data, and added to the corresponding low visibility event case library; A low visibility event case library reconstruction module is used to determine the weight coefficient of each meteorological factor in each low visibility event case library by using the set second low visibility threshold, and reconstruct each low visibility event case library by using each meteorological factor and its weight coefficient to obtain multiple reconstructed low visibility event case libraries; A visibility prediction model training module is used to train the long short-term memory neural network using each reconstructed low visibility event case library to obtain a visibility prediction model for each category; The visibility prediction result output module is used to obtain the meteorological factor data at the current moment and the upper air circulation diagram of the numerical model at the current moment, determine the category to which the meteorological factor data at the current moment belongs according to the upper air circulation diagram of the numerical model at the current moment, input the meteorological factor data at the current moment into the visibility prediction model under the corresponding category, and obtain the visibility prediction result; The method uses the set second low visibility threshold to determine the weight coefficient of each meteorological factor in each low visibility event case library, and uses each meteorological factor and its weight coefficient to reconstruct each low visibility event case library to obtain multiple reconstructed low visibility event case libraries, specifically including: For any low visibility event case database D, perform the following steps: According to the second low visibility threshold, the low visibility event case database D is divided into two parts, D1 and D2; According to the correlation fitting, the maximum possible value of each meteorological factor under the second lowest visibility threshold is determined; The maximum possible value of each meteorological factor is used in turn to further divide the two parts D1 and D2; then the Gini index corresponding to each meteorological factor is calculated according to the following formula, and the Gini index is used as the weight coefficient of the meteorological factor in the case database D of the low visibility event: In the formula, A represents the meteorological factor, K is 2, k=1, 2, for D1, C1 represents the meteorological factor value greater than or equal to the maximum possible value of meteorological factor A in D1, C2 represents the meteorological factor value less than the maximum possible value of meteorological factor A in D1; for D2, C1 represents the meteorological factor value greater than or equal to the maximum possible value of meteorological factor A in D2, C2 represents the meteorological factor value less than the maximum possible value of meteorological factor A in D2; Reconstructing the low visibility event case database D based on the weight coefficients of each meteorological factor on the low visibility event case database D; In this way, each low-visibility event case database is reconstructed to obtain multiple reconstructed low-visibility event case databases.

6. A shipping visibility prediction system according to claim 5, characterized in that: The historical meteorological data include historical meteorological data observed by meteorological stations located along the shipping route and historical meteorological data collected by traffic stations located along the shipping route; The historical upper-air circulation map is drawn from the European Center ERA5 reanalysis data; the European Center ERA5 reanalysis data is grid data with the same regional scope as the shipping line and longitude and latitude close to the national basic meteorological station.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a shipping visibility prediction method as claimed in any one of claims 1 to 4 when executing the computer program.

8. A storage medium, characterized in that: The storage medium stores a visibility prediction program, and when the visibility prediction program is executed by at least one processor, the steps of a shipping visibility prediction method according to any one of claims 1 to 4 are implemented.