BE1031164B1 - SENSOR-BASED EARLY WARNING METHOD FOR CLUSTER FOG FORMATION - Google Patents

Abstract

The present invention provides a sensor-based early warning method for cluster fog formation and relates to the field of traffic control, comprising: obtaining historical cluster fog information of the target area, determining a plurality of locations to be monitored in the target area, establishing a plurality of cluster fog monitoring sites and a plurality of edge computing sites, determining the weather forecast information of the plurality of locations to be monitored based on the weather forecast information of the target area, obtaining the pollutant forecast information of the plurality of locations to be monitored based on the pollutant forecast information of the target area, predicting cluster fog occurrence areas and cluster fog formation information, generating and outputting cluster fog warning information, determining the target location for cluster fog monitoring and the best image acquisition position, determining the target location for edge computing, collecting cluster fog images based on the best image acquisition position, determining the real-time status of the cluster fog based on cluster fog images, updating the Cluster fog occurrence areas and cluster fog formation information, adapting and issuing cluster fog warning information. The advantage is that automatic early warning of cluster fog formation is possible.

Description

DESCRIPTION

SENSOR-BASED EARLY WARNING SYSTEM FOR

CLUSTER FOG FORMATION

Technical field

The invention relates to the field of traffic control and in particular to a sensor-based early warning method for

Cluster fog formation.

State of the art

Cluster Nebula is essentially a nebula. It is formed by the

Radiative cooling of the ground causes the air near the ground to become cooler and the saturated specific humidity to decrease, causing water vapor to condense. Cluster fog differs from ordinary fog in that its area of influence is smaller.

Visibility outside the Cluster Nebula is good, but inside the

Cluster fog makes it hazy. High-altitude highways favor the

Cluster fog formation due to environmental factors such as higher

Road surface temperatures during the day and larger

Temperature differences between day and night. A quarter of the

Traffic accidents that occur on motorways every year are caused by adverse weather conditions. Special meteorological causes such as cluster fog are the biggest obstacles and

Threats to road safety.

In the existing technology, manual monitoring and

Early warning of road weather conditions as part of the

Video surveillance mainly on video patrols. A manual

Analysis is inefficient. Unusual weather conditions such as

Cluster fogs usually occur in the early morning or evening.

Visibility conditions are poor during these two periods and the mere

Eye cannot effectively detect cluster fog information and provide timely warning and control.

Therefore, it is necessary to develop a sensor-based early warning system for

Cluster fog formation to provide an automated

To achieve early warning of cluster fog formation.

Contents of the present invention

The present invention provides a sensor-based early warning method for

Cluster fog formation ready, includes: Obtaining historical

Cluster fog information of the target area, determining multiple locations to be monitored in the target area based on the historical

Cluster nebula information; establishment of multiple

Cluster fog monitoring sites and several

Edge computing locations based on the multiple locations to be monitored; Obtaining the weather forecast information for the target area and determining the weather forecast information of the multiple locations to be monitored based on the weather forecast information of the

target area; Obtaining pollutant forecast information for the

Target area and determination of pollutant prediction information of the multiple sites to be monitored based on the

Pollutant forecast information of the target area; prediction of

Cluster fog occurrence areas and

Cluster nebula formation information based on the

Weather forecast information and pollutant forecast information of the multiple locations to be monitored; generating and outputting

Cluster fog warning information based on the

Cluster fog occurrence areas and the

Cluster nebula formation information; Based on the

Cluster fog occurrence areas and the

Cluster fog formation information, determining the target location for the

Cluster fog monitoring and the best image acquisition position corresponding to the target fog monitoring site from the multitude of

Cluster fog monitoring sites, determining the target location for edge computing from the plurality of edge computing sites, wherein the target fog monitoring site is configured to collect

Cluster nebula images are formed based on the best image acquisition position and the target location for edge computing is

Determine the real-time status of the cluster nebula based on

Cluster Nebula images and to update the

Cluster fog occurrence areas and the

Cluster nebula formation information based on the real-time status of the cluster nebula; adjusting and outputting

Cluster fog warning information based on the updated

Cluster fog occurrence areas and the … cluster fog formation information.

Further, determining the weather forecast information of the plurality of locations to be monitored based on the

Weather forecast information of the target area: Divide the target area into several sub-areas according to the terrain information of the target area;

Obtain historical weather information of multiple sub-areas and

Confirm weather correlation degree of any two sub-areas based on the historical weather information of the multiple

Sub-areas; Obtaining historical weather information of the multiple locations to be monitored and confirming the weather correlation degree of any two locations to be monitored based on the historical weather information of the multiple locations to be monitored

Locations; Creating a weather forecast model; Training the

Weather forecast model based on historical

Weather information of the several sub-areas, the

Weather correlation degree of any two of the sub-areas, the historical weather information of the multiple monitored

Locations and the degree of weather correlation of any two

Monitoring locations; Determining the

Weather forecast information of the multiple monitored

Locations based on the weather forecast information of the

target area by a trained weather forecast model.

Furthermore, the weather forecast model includes a

Sub-area weather forecast unit, a

Sub-area weather forecast verification unit, a

Site weather forecast unit and a

Location weather forecast verification unit; training the

Weather forecast model based on historical

Weather information of the several sub-areas, the

Weather correlation degree of any two of the sub-areas, the historical weather information of the multiple monitored

locations and the degree of weather correlation of any two locations to be monitored includes: Creating several initial

Training patterns based on the historical weather information of the

target area and the historical weather information of the several

sub-areas, where the first training pattern is the historical

Weather information of the target area for a first historical

Time and historical weather information of the several

sub-areas at the first historical time point; creating several second training patterns based on the historical 5 weather information of the several sub-areas and the historical

Weather information of the multiple locations to be monitored, with the second training pattern being the historical weather information of the

target area at a second historical point in time and the historical

Weather information of the several sub-areas for the second historical

Time includes; creating multiple third training patterns based on the historical weather information of the multiple sub-areas, the historical weather information of the multiple sub-areas and the historical weather information of the multiple

locations, with the third training pattern being the historical

Weather information of the target area to a third historical

time, the historical weather information of the several sub-areas at the third historical time and the historical

Weather information of several locations to be monitored at the third historical time point; training the

Weather forecast model based on the first several

training pattern, the multiple second training patterns and the multiple third training patterns.

Furthermore, training the weather forecast model based on the plurality of first training patterns, the plurality of second

Training pattern and the plurality of third training patterns: Inputting the historical weather information of the target area at a first historical time and the weather correlation degree of any two sub-areas included in the first training pattern into the

Sub-area weather forecast unit, where the

Sub-area weather forecasting unit outputs the predicted first historical weather forecast information of the plurality of sub-areas; Calculating a first loss value based on the first historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecasting unit and the historical weather information of the plurality of sub-areas at the first historical time point included in the first training pattern;

Input of the historical weather information of the multiple sub-areas at a first historical time and the weather correlation degree of any two sub-areas into the

Sub-area weather forecast verification unit, wherein the

Sub-area weather forecast verification unit outputs the predicted historical weather forecast information of the target area;

Calculate a second loss value based on the historical

Weather forecast information of the destination area provided by the

Sub-area weather forecast verification unit and the historical weather information of the target area at the first historical time point included in the first training pattern;

Perform a first round of parameter optimization on the

Sub-area weather forecast unit and the

Sub-area weather forecast verification unit based on the first loss value and the second loss value; inputs of the historical

Weather information of the several sub-areas for the second historical

Time point, which include in the second training pattern, and the

Weather correlation degree of any two monitored

locations into the location weather forecast unit after the

Sub-area weather forecast unit and the

Sub-area weather forecast verification unit the first round of

Parameter optimization has been completed, with the

Site weather forecast unit the predicted first historical

Provides weather forecast information for the locations to be monitored;

Calculate a third loss value based on the first historical

Weather forecast information of the multiple locations to be monitored, and the historical weather information of the multiple locations to be monitored

Locations at the second historical point in time, which in the second

Training patterns include; inputs of the historical weather information of the multiple locations to be monitored to the second historical

Time point, which include in the second training pattern, and the

Weather correlation degree of any two monitored

locations into the location weather forecast verification unit, wherein the location weather forecast verification unit outputs the predicted historical weather forecast information of the plurality of sub-areas; calculating a fourth loss value based on the historical weather forecast information of the plurality of sub-areas predicted by the location weather forecast verification unit and the historical weather information of the plurality

Sub-areas at the second historical point in time, which in the second

Training patterns include; performing a second round of

Parameter optimization of the site weather forecast unit and the

Location weather forecast verification unit based on the third

loss value and the fourth loss value; inputs of the historical

Weather information of the target area at the third historical time point, which includes the third training pattern, and the weather correlation degree of any two sub-areas in the

Sub-area weather forecast unit after the

Site weather forecast unit and the

Location Weather Forecast Verification Unit the second round of

Parameter optimization has been completed, with the

Sub-area weather forecasting unit outputs the predicted second historical weather forecast information of the plurality of sub-areas; Calculating a fifth loss value based on the second historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecasting unit and the historical weather information of the plurality of sub-areas at the third historical time point included in the third training pattern;

Inputting the second historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecast unit and the weather correlation degree of any two sub-areas to the location weather forecast unit, wherein the location weather forecast unit outputs the predicted second historical weather forecast information of the plurality of locations to be monitored; calculating a sixth loss value based on the second historical weather forecast information of the plurality of locations to be monitored,

Site weather forecast unit and the historical weather information of the multiple monitored

Locations at the third historical point in time, which in the third

Training patterns include; performing a third round of

Parameter optimization of the sub-area weather forecast unit and the site weather forecast unit based on the fifth

loss value and the sixth loss value, where the trained

Weather forecast model includes a sub-area weather forecast unit and a site weather forecast unit, which is the third round of

have completed parameter optimization.

Furthermore, the first loss value is calculated based on the following first

Loss function calculated: h Ë ik a ; ) > {+ ef orentis Vetra 3 y

L.= s=i 1 P, . Li . Vrixforsenstti where the first loss value is “the value of the k-th meteorological factor in the i-th sub-area, which is determined by the

Vri crea)

Sub-area weather forecast unit is predicted; 25 Jer

is the value of the k-th meteorological factor in the i-th sub-area, which includes K of the first training samples; is the total number of meteorological factors, ° is the total number of sub-areas,

P, B, 8 is the preset parameter and ° ; where the third loss value is based on the following third

Loss function is calculated: ) ) > _ {Fr (formel) Vri real j NE

F mE hi , £ 3 . . 6, [a Forecast)jj where - is the third loss value, °° which is calculated by the

Site weather forecast unit predicted value of the k-th meteorological factor of the j-th site to be monitored, > is the value of the k-th meteorological factor of the j-th site to be monitored, which in the second training sample includes 7 FE. ' is the total number of sites to be monitored, * is the preset parameter, and > >

Furthermore, determining the pollutant prediction information of the multiple sites to be monitored based on the

Pollutant forecast information of the target area: Obtaining historical pollutant information of the multiple sub-areas and

Confirm the pollutant correlation level of any two of the

Sub-areas based on the historical pollutant information of the multiple sub-areas; Obtaining historical pollutant information of the multiple sites to be monitored and confirming the

Pollutant correlation degree of any two of the monitored

Locations based on the historical pollutant information of the multiple sites to be monitored; Creating a

pollutant prediction model; training the

Pollutant prediction model based on historical

Pollutant information of the several sub-areas, the

Pollutant correlation degree of any two sub-areas, the historical pollutant information of the several monitored

locations and the pollutant correlation level of any two of the sites to be monitored; Determine the

Pollutant forecast information of the multiple monitored

Locations based on the pollutant forecast information of the

target area by the trained pollutant prediction model.

Furthermore, the prediction of cluster fog occurrence areas and cluster fog formation information based on the

Weather forecast information and pollutant forecast information of the multiple sites to be monitored: Obtaining historical cluster fog information of the multiple sites to be monitored

locations and confirm the degree of correlation of the cluster fog of any two of the monitored locations based on the historical cluster fog information of the multiple monitored

Locations; for each location to be monitored, determining the initial cluster fog occurrence areas and the

Cluster fog formation information at the location to be monitored based on the weather forecast information and

Pollutant forecast information of the site to be monitored by a local cluster fog forecast model; determining the

Cluster fog occurrence areas and the

Cluster nebula formation information based on the initial

Cluster fog occurrence areas and the

Cluster fog formation information at each of the monitored

locations and the degree of correlation of the cluster fog from any two locations to be monitored by a global

Cluster fog prediction model.

Furthermore, generating the cluster fog warning information based on the cluster fog occurrence areas and the

Cluster nebula formation information: Determine

Traffic warning road based on the

Cluster nebula occurrence areas; determining

Traffic control measures for the traffic warning road based on the cluster fog formation information.

Further, determining the real-time status of the cluster nebula based on cluster nebula images from which the target location for the

Edge Computing: Identifying Cluster Nebula Images by a

Convolutional neural network model to display the real-time status of the

Cluster Nebula.

Furthermore, the update of cluster fog occurrence areas and cluster fog formation information based on the

Real-time status of the cluster nebula: Generate real-time adjusted

Cluster fog occurrence areas and

Cluster fog formation information by a long short-term memory network model based on the real-time status of the

Cluster nebula at several historical times and the predicted cluster nebula occurrence areas and cluster nebula formation information.

Compared to existing technology, the present

Invention provided sensor-based early warning method for

Cluster fog formation at least the following advantageous

Effects: 1. Achieve a targeted early fog warning by monitoring several locations in the target area and avoid the

Processing large amounts of invalid data and monitoring

Cluster fog. By predicting weather forecast information and pollutant forecast information at multiple monitored

locations, it provides accurate and comprehensive data support for the prediction of subsequent cluster fog occurrence areas and

Cluster fog formation information. On this basis, the

Target location for cluster fog monitoring and the target location for edge computing real-time status monitoring of the generated

Cluster fog is realized so that the cluster fog warning information in

can be adjusted in real time. This provides traffic control

Real-time information and more accurate cluster fog warning information. 2. First, the first training sample is used to perform the first round of parameter optimization on the sub-area weather forecast unit and the sub-area weather forecast verification unit based on the first loss value and the second loss value.

Then, the second training sample is used to perform a second round of parameter optimization on the site weather forecast unit and the site weather forecast verification unit based on the third loss value and the fourth loss value. This enables independent training of the

Sub-area weather forecast unit and the

Sub-area weather forecast verification unit, the

Site weather forecast unit and the

Location weather forecast verification unit. In addition, the

Weather forecast model by the previous forecast and subsequent verification of the loss implicitly complex nonlinear

Learning mapping relationships between different sub-areas, different locations to be monitored and different meteorological variables. After completing the independent

Training will be the independently trained regional

weather forecast unit and the location weather forecast unit are extracted for joint training. By first independent training and then joint training, the

The prediction accuracy of the trained weather forecast model is ensured and the efficiency of model training is improved. 3. By confirming the weather correlation degree of any two

sub-areas and the degree of weather correlation of any two monitored locations, multidimensional information is provided to the weather forecast model, which

weather forecast model is guided to produce more accurate

To generate weather forecast information for multiple locations to be monitored. By confirming the degree of correlation of

pollutants in any two sub-areas and the degree of correlation of pollutants in any two monitoring sites, multidimensional information is provided for the

pollutant prediction model provided and the

Pollutant prediction model is guided to provide more accurate

Pollutant forecast information for several monitored

to generate locations.

Brief description of the characters

This specification will be further explained by means of exemplary embodiments, which are described in detail with reference to the accompanying drawings. These exemplary embodiments are not restrictive. In these

In the embodiments, the same numbers represent the same structures, where:

Fig. 1 is a schematic flow diagram of a sensor-based

Early warning procedure for cluster fog formation according to some

embodiments of this specification;

Fig. 2 is a schematic flow diagram for determining

Weather forecast information of the plurality of locations to be monitored according to some embodiments of this specification;

Fig. 3a is a schematic flow diagram of the first round of

Parameter optimization according to some embodiments of this

Specification;

Fig. 3b is a schematic flow diagram of the second round of

Parameter optimization according to some embodiments of this

Specification;

Fig. 3c is a schematic flow diagram of the third round of

Parameter optimization according to some embodiments of this

Specification;

Fig. 4 is a schematic flow diagram for determining

Pollutant forecast information of the multiple monitored

Locations according to some embodiments of this specification.

Detailed embodiments

In order to understand the technical solutions of the embodiments of this

To explain the specification more clearly, the attached drawings are briefly presented below, which are used in the description of the

The examples must be used. Obviously, the

Drawings in the following description are only some examples or

embodiments of this specification. For those of ordinary skill in the art, this description may also apply to other similar scenarios on the

The basis of these drawings can be applied without creative

efforts are required. Unless otherwise indicated by the location, the same represent

Reference symbols in the figures have the same structure or function.

Fig. 1 is a schematic flow diagram of a sensor-based

Early warning procedure for cluster fog formation according to some

Embodiments of this specification. As shown in Fig. 1, the sensor-based early warning method for cluster fog formation may include the following steps.

Step 110: Obtaining historical cluster nebula information from the

Target area, determining several locations to be monitored in the

Target area based on historical cluster nebula information.

The historical cluster nebula information of the target area can

Place where in the target area in the past period (e.g. 3

years, 1 year, etc.) cluster nebula has occurred, and the

Cluster nebula formation information, where the

Cluster nebula formation information which may include nebula size, visibility, etc. of the same nebula at multiple historical times.

The location to be monitored can be a location within the

Target area where there is a possibility of cluster fog occurring.

Step 120: Set up multiple cluster fog monitoring sites and multiple edge computing sites based on the multiple sites to be monitored.

The cluster fog monitoring sites are used to

To obtain cluster nebula formation information in the space it is located in. The edge computing location can be used to determine the cluster nebula occurrence area and

Cluster fog formation information based on the

Cluster fog monitoring sites preserved

To determine cluster nebula formation information.

For example only, multiple cluster fog monitoring sites and multiple edge computing sites can be set up near the site to be monitored.

Step 130: Obtaining the weather forecast information for the target area and determining the weather forecast information of the multiple locations to be monitored based on the

Weather forecast information for the destination area;

Weather forecast information can at least

Forecast information on meteorological factors such as temperature,

humidity, wind speed and wind direction.

Fig. 2 is a schematic flow diagram for determining

Weather forecast information of the plurality of locations to be monitored according to some embodiments of this specification. As shown in Fig. 2, determining the weather forecast information of the plurality of locations to be monitored based on the

Weather forecast information for the destination area:

Dividing the target area into several sub-areas according to the

Terrain information of the target area. In particular, there are differences

In the area between two adjacent sub-areas;

Obtain historical weather information of multiple sub-areas and

Confirm the weather correlation degree of any two sub-areas based on the historical weather information of the multiple

sub-areas;

Obtaining historical weather information of the multiple locations to be monitored and confirming the weather correlation degree of any two locations to be monitored based on the historical weather information of the multiple locations to be monitored

locations;

Creating a weather forecast model. In particular, a

Weather forecast model based on the generative adversarial

network can be created;

Training the weather forecast model based on the historical

Weather information of the several sub-areas, the

Weather correlation degree of any two of the sub-areas, the historical weather information of the multiple monitored

Locations and the degree of weather correlation of any two locations to be monitored;

Determining the weather forecast information of the multiple

Monitoring locations based on the

Weather forecast information of the target area by a trained

Weather forecast model.

In particular, the degree of weather correlation between the two sub-areas can be calculated using the following formula:

Ps

Cash 7 ts QU ee { “° Fe un — Fr ant [ N si Likes) CREAM: 7

Citer where °° ‘is the weather correlation degree between the e-th sub-area (rik eis) and the f-th sub-area, the value of the k-th meteorological factor in the e-th sub-area at the t-th historical

VR

time point, is the value of the k-th meteorological factor in the 4 FT +4 f-th sub-area at the t-th time point, is the total number of recorded ee Ps. . historical time points, ' "is the preset parameter and

FO

The method for calculating the degree of weather correlation between two locations to be monitored is similar to the method for calculating the

Weather correlation degree of two sub-areas and is not described again here.

It is understood that by confirming the weather correlation degree of any two sub-areas and the weather correlation degree of any two locations to be monitored, more dimensional

Information is provided for the weather forecast model.

This guides the weather forecast model to produce more accurate

To generate weather forecast information for multiple locations to be monitored.

In some embodiments, the weather forecast model includes a

Sub-area weather forecast unit, a

Sub-area weather forecast verification unit, a

Site weather forecast unit and a

Site weather forecast verification unit; training the weather forecast model based on the historical weather information of the multiple sub-areas, the

Weather correlation degree of any two of the sub-areas, the historical weather information of the multiple monitored

Locations and the degree of weather correlation of any two locations to be monitored includes:

Creating several initial training patterns based on the historical

Weather information of the destination area and the historical

Weather information of the several sub-areas, the first

Training pattern the historical weather information of the target area at a first historical time and the historical

Weather information of the several sub-areas for the first historical

time includes;

Creating multiple second training patterns based on the historical

Weather information of the several sub-areas and the historical

Weather information of the multiple locations to be monitored, with the second training pattern being the historical weather information of the

target area at a second historical point in time and the historical

Weather information of the several sub-areas for the second historical

time includes;

Creating multiple third training patterns based on the historical

Weather information of the several sub-areas, the historical

Weather information of the several sub-areas and the historical

Weather information of the multiple locations to be monitored, wherein the third training pattern comprises the historical weather information of the target area at a third historical time, the historical

Weather information of the several sub-areas for the third historical

time and the historical weather information of the multiple locations to be monitored at the third historical time;

Training the weather forecast model based on the plurality of first training samples, the plurality of second training samples, and the plurality of third training samples.

As shown in Figures 3a to 3c, training the

Weather forecast model based on the first several

training pattern, the multiple second training patterns and the multiple third training patterns:

Input of the historical weather information of the target area at a first historical time and the weather correlation degree of any two sub-areas included in the first training pattern into the

Sub-area weather forecast unit, where the

Sub-area weather forecast unit outputs the predicted first historical weather forecast information of the plurality of sub-areas;

Calculate a first loss value based on the first historical

Weather forecast information of the several sub-areas provided by the

sub-area weather forecast unit, and the historical weather information of the plurality of sub-areas at the first historical time point included in the first training pattern;

Input of the historical weather information of the multiple sub-areas at a first historical time and the weather correlation degree of any two sub-areas into the

Sub-area weather forecast verification unit, wherein the

Sub-area weather forecast verification unit outputs the predicted historical weather forecast information of the target area;

Calculate a second loss value based on the historical

Weather forecast information of the destination area provided by the

Sub-area weather forecast verification unit and the historical weather information of the target area at the first historical time point included in the first training pattern;

Perform a first round of parameter optimization on the

Sub-area weather forecast unit and the

Sub-area weather forecast verification unit based on the first loss value and the second loss value by the

Gradient descent method;

Inputting the historical weather information of the plurality of sub-areas at the second historical time point included in the second training pattern and the weather correlation degree of any two locations to be monitored to the location weather forecasting unit after the sub-area weather forecasting unit and the

Sub-area weather forecast verification unit the first round of

Parameter optimization has been completed, with the

Site weather forecast unit the predicted first historical

Provides weather forecast information for the locations to be monitored;

Calculate a third loss value based on the first historical

Weather forecast information of the multiple locations to be monitored,

and the historical weather information of the multiple monitored

Locations at the second historical point in time, which in the second

Training patterns include;

Inputs of the historical weather information of the multiple locations to be monitored at the second historical time point, which comprise the second training pattern, and the weather correlation degree of any two locations to be monitored into the

Location weather forecast verification unit, wherein the

The location weather forecast verification unit outputs the predicted historical weather forecast information of the plurality of sub-areas;

Calculate a fourth loss value based on the historical

Weather forecast information of the several sub-areas provided by the

location weather forecast verification unit, and the historical weather information of the plurality of sub-areas at the second historical time point included in the second training pattern;

Performing a second round of parameter optimization on the

Site weather forecast unit and the

Location weather forecast verification unit based on the third

loss value and the fourth loss value by the

Gradient descent method;

Inputs of the historical weather information of the target area at the third historical time point, which includes training patterns in the third, and the weather correlation degree of any two sub-areas into the

Sub-area weather forecast unit after the

Site weather forecast unit and the

Location Weather Forecast Verification Unit the second round of

Parameter optimization has been completed, with the

Sub-area weather forecasting unit outputs the predicted second historical weather forecast information of the plurality of sub-areas;

Calculating a fifth loss value based on the second historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecasting unit and the historical weather information of the plurality of sub-areas at the third historical time point including the third training pattern;

Inputting the second historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecast unit and the weather correlation degree of any two sub-areas to the site weather forecast unit, wherein the site weather forecast unit outputs the predicted second historical weather forecast information of the plurality of sites to be monitored;

Calculating a sixth loss value based on the second historical weather forecast information of the plurality of locations to be monitored predicted by the location weather forecasting unit and the historical weather information of the plurality of locations to be monitored at the third historical time point including training patterns in the third;

Performing a third round of parameter optimization on the

Sub-area weather forecast unit and the

Site weather forecasting unit based on the fifth loss value and the sixth loss value by the gradient descent method; wherein the trained weather forecasting model has a

Sub-area weather forecast unit and a

Site Weather Forecast Unit, which includes the third round of

have completed parameter optimization.

In some embodiments, the first loss value is calculated based on the following first loss function: > Vi persons To Ve ; a #ck=i

Ey = — 7 5 . P, by . Vime RES where the first loss value is," " is the value of the k-th meteorological factor in the i-th sub-area, which is determined by the

CN

Sub-area weather forecast unit; SEES of the

Value of the k-th meteorological factor in the i-th sub-area, which includes in ee K . of the first training samples; the total number of meteorological factors is, ; the total number of sub-areas is,

P, PO is the default parameter and * ;

The second loss value is calculated based on the following second

Loss function calculated: > Vi, Forecast) — Wr krezi) y

Li, = i—i z A ‚ds . Vi A Foren where “the first loss value,” is the value of the k-th meteorological factor in the target area, which is determined by the

Sub-area weather forecast verification unit;

Vele west) . . . Le > is the value of the k-th meteorological factor in the target area,

F, which includes the first training pattern; * the preset

Parameter is and “37 >

The third loss value is calculated based on the following third

Loss function calculated: \ 7 « , z pz where Sa is the third loss value, “(sporsenstts) obtained from the

Location weather forecast unit predicted value of the k-th meteorological factor of the j-th location to be monitored, 'is the value of the k-th meteorological factor of the j-th location to be monitored, which includes in the second training sample,

I + . P is the total number of locations to be monitored, ° is the preset parameter, and PS ‘

The calculation method for the fourth loss value and the fifth

Loss value is similar to the calculation method for the first loss value, and the calculation method for the sixth loss value is similar to the

Calculation method for the third loss value, which is not the case, is described here again.

It is understood that the first training sample is initially used to perform the first round of parameter optimization on the

Sub-area weather forecast unit and the sub-area

Weather forecast verification unit based on the first

loss value and the second loss value. Then the second training sample is used to perform a second round of

Parameter optimization at the site weather forecast unit and the

Location weather forecast verification unit based on the third

loss value and the fourth loss value. This allows independent training of the sub-area weather forecast unit and the sub-area weather forecast verification unit, the

Site weather forecast unit and the

Location weather forecast verification unit. In addition, the

Weather forecast model by the previous forecast and subsequent verification of the loss implicitly complex nonlinear

Learning mapping relationships between different sub-areas, different locations to be monitored and different meteorological variables. After completing the independent

Training will be the independently trained regional

weather forecast unit and the location weather forecast unit are extracted for joint training. By first independent training and then joint training, the

Ensures the forecast accuracy of the trained weather forecast model and improves the efficiency of model training.

Step 140: Obtaining the pollutant forecast information for the

Target area and determination of pollutant prediction information of the multiple sites to be monitored based on the

Pollutant forecast information of the target area;

Fig. 4 is a schematic flow diagram for determining

Pollutant forecast information of the multiple monitored

Locations according to some embodiments of this specification. As shown in Fig. 4, determining the

Pollutant forecast information of the multiple monitored

Locations based on the pollutant forecast information of the

Target area in some embodiments:

Obtaining historical pollutant information of the multiple sub-areas and confirming the pollutant correlation level of any two of the

Sub-areas based on the historical pollutant information of the multiple sub-areas, where the historical pollutant information includes the concentration of coarse dust PM10 and the concentration of

particulate matter PM2.5 at several historical points in time;

Obtaining historical pollutant information of the multiple sites to be monitored and confirming the pollutant correlation degree of any two of the sites to be monitored based on the historical pollutant information of the multiple sites to be monitored

locations;

Creating a pollutant prediction model, where the

Pollutant prediction model can be a Generative Adversarial Networks model;

Training the pollutant prediction model based on the historical pollutant information of the multiple sub-areas, the

Pollutant correlation degree of any two sub-areas, the historical pollutant information of the several monitored

locations and the pollutant correlation level of any two of the sites to be monitored;

Determine the pollutant prediction information of the multiple sites to be monitored based on the

Pollutant prediction information of the target area by the trained

Pollutant prediction model.

Specifically, the pollutant correlation degree of the two sub-areas can be calculated using the following formula:

Crrarum = B

Ag Fa £ —F La > Many - Vian)

[The Soe fie]

Cr (ea where * °° is the pollutant correlation degree between the e-th

Vr far

sub-area and the f-th sub-area, ** the concentration of the y-th type of particles in the e-th sub-area at the t-th historical

Ve

time, the concentration of the y-th type of particulate matter in the

For the fifth sub-area at the t-th historical time point, the total number . Ps . Pa > © of particle types is, the preset parameter is and .

The method for calculating the pollutant correlation degree of two sites to be monitored is similar to the method for calculating the

Pollutant correlation degree of two sub-areas and is not described again here.

It is understood that multidimensional information is essential for

Weather forecast model by confirming the

Weather correlation degree of any two sub-areas and the

Weather correlation degree of any two monitored

locations, thereby guiding the weather forecast model to generate more accurate weather forecast information for multiple locations to be monitored.

In some embodiments, the structure and

Training method of the pollutant prediction model of the structure and

training method of the weather forecast model. Further

Descriptions of the structure and training methods of the

pollutant prediction model can be found in the “corresponding

Description of the structure and training methods of the

weather forecast model, which is not described again here.

Step 150: Prediction of cluster fog occurrence areas and

Cluster nebula formation information based on the

Weather forecast information and pollutant forecast information of the multiple locations to be monitored.

Includes:

Obtaining historical cluster fog information from the multiple sites to be monitored and confirming the degree of correlation of the

Cluster fog from any two of the monitored sites based on the historical cluster fog information of the multiple monitored sites; for each monitored site, determining the initial

Cluster fog occurrence areas and the

Cluster fog formation information at the location to be monitored based on the weather forecast information and

Pollutant forecast information of the site to be monitored by a local cluster fog forecast model;

Determine the cluster nebula occurrence areas and the

Cluster nebula formation information based on the initial

Cluster fog occurrence areas and the

Cluster fog formation information at each of the monitored

locations and the degree of correlation of the cluster fog from any two locations to be monitored by a global

Cluster fog prediction model, where the local

Cluster fog prediction model and the global

Cluster fog prediction model RNN models (Recurrent Neural

Networks).

Step 160: Generate cluster fog warning information based on the cluster fog occurrence areas and the

Cluster nebula formation information

Includes:

Determine traffic warning road based on the

Cluster fog occurrence areas;

Determine traffic control measures for the traffic warning road based on the cluster fog formation information.

The traffic warning road may be a road that passes through an area where cluster fog occurs.

The corresponding traffic control measures for cluster fog with different visibility conditions may vary. For example, if the visibility is greater than 200 meters and less than or equal to 500

meters, the corresponding traffic control measures

Information signs. If visibility is more than 100 meters and less than or equal to 200 meters, the corresponding

Traffic control measures include a speed limit and release. If visibility is more than 50 meters and less than or equal to 100 meters, the corresponding

Traffic control measures, road closures, etc.

Step 170: Based on the cluster fog occurrence areas and the cluster fog formation information, determining the target location for cluster fog monitoring and the best image acquisition position corresponding to the target fog monitoring location from the plurality of

Cluster fog monitoring locations, determining the target location for edge computing from the multitude of edge computing locations.

The target fog monitoring site is for collecting

Cluster nebula images based on the best image acquisition position.

The target location for edge computing is used to determine the

Real-time status of the cluster nebula based on cluster nebula images and to update the cluster nebula occurrence areas and the

Cluster nebula formation information is formed based on the real-time status of the cluster nebula.

In some embodiments, determining the real-time status of the cluster nebula based on cluster nebula images from which the

Target location for edge computing:

Identifying cluster fog images by convolutional neural

Network model to determine the real-time status of the cluster nebula.

Specifically, the core of the convolutional neural network model is trained using a large number of precisely annotated street image sets.

Convolutional neural network implements the extraction of

image features (e.g., visibility, road surface status) and implements image classification via a classifier.

Architecture of a convolutional neural network model can be roughly divided into three

types of layers, including input layer, hidden layer and output layer. The main purpose of the

Input layer consists of receiving annotated input images and passing them to subsequent hidden layers.

Most of the computation takes place in the hidden layer. The hidden layer can be divided into three types of layers:

Convolution layer, excitation layer (ReLU) and

Pooling layer (Pooling). The output layer usually consists of two types of layers: fully connected layer and

Softmax layer.

The total number of verification samples is 7280 images with 27

Cameras and samples cover all fog conditions.

Verification results are shown in Table 1. It was confirmed that the classification of the convolutional neural network model in 6880 images is consistent with the visual judgment and the

Classification success rate reached 94.5%; In low visibility conditions of 200 m, 2670 out of 2880 images passed the test, and the success rate reached 92.7%.

Table 1

Visual Score Visual Classification of Convolutional Neurons

Qualified success rate in the

Quantity for classification

classification

Step 180: Adjusting and outputting

Cluster fog warning information based on the updated

Cluster fog occurrence areas and the

Cluster nebula formation information.

Includes:

Generate real-time customized

Cluster fog occurrence areas and

Cluster fog formation information by a long short-term memory network model based on the real-time status of the

Cluster nebula at several historical times and the predicted cluster nebula occurrence areas and cluster nebula formation information.

Finally, it should be clear that the features described in this specification

Embodiments are intended only to illustrate the principles of

embodiments of this specification. Other

Variations may also fall within the scope of this specification.

Therefore, alternative configurations of the embodiments of this

specification should be considered as exemplary and not limiting, consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to those expressly presented and described in this specification.

Claims (10)

PATENT CLAIMS 1. A sensor-based early warning method for cluster fog formation, characterized in that it comprises: - obtaining historical cluster fog information of the target area, determining a plurality of locations to be monitored in the target area based on the historical cluster fog information; - establishing a plurality of cluster fog monitoring sites and a plurality of edge computing sites based on the plurality of locations to be monitored; - obtaining weather forecast information for the target area and determining the weather forecast information of the plurality of locations to be monitored based on the weather forecast information of the target area; - obtaining pollutant forecast information for the target area and determining pollutant forecast information of the plurality of locations to be monitored based on the pollutant forecast information of the target area; - predicting cluster fog occurrence areas and cluster fog formation information based on the weather forecast information and pollutant forecast information of the plurality of locations to be monitored; - generating and outputting cluster fog warning information based on the cluster fog occurrence areas and the cluster fog formation information; - Based on the cluster fog occurrence areas and the cluster fog formation information, determining the target location for cluster fog monitoring and the best image acquisition position corresponding to the target fog monitoring location from the plurality of cluster fog monitoring locations; determining the target location for edge computing from the plurality of edge computing locations, wherein the target fog monitoring location is configured to collect cluster fog images based on the best image acquisition position and the target location for edge computing is configured to determine the real-time status of the cluster fog based on cluster fog images and to update the cluster fog occurrence areas and the cluster fog formation information based on the real-time status of the cluster fog; adjusting and outputting cluster fog warning information based on the updated cluster fog occurrence areas and the cluster fog formation information. 2. The sensor-based cluster fog early warning method according to claim 1, characterized in that determining the weather forecast information of the plurality of locations to be monitored based on the weather forecast information of the target area comprises: - dividing the target area into a plurality of sub-areas according to the terrain information of the target area; - obtaining historical weather information of the plurality of sub-areas and confirming the weather correlation degree of any two sub-areas based on the historical weather information of the plurality of sub-areas; - obtaining historical weather information of the plurality of locations to be monitored and confirming the weather correlation degree of any two locations to be monitored based on the historical weather information of the plurality of locations to be monitored; - creating a weather forecast model; - training the weather forecast model based on the historical weather information of the plurality of sub-areas, the weather correlation degree of any two of the sub-areas, the historical weather information of the plurality of locations to be monitored, and the weather correlation degree of any two locations to be monitored; - Determining the weather forecast information of the multiple locations to be monitored based on the weather forecast information of the target area by a trained weather forecast model. 3. A sensor-based early warning method for cluster fog formation according to claim 2, characterized in that the weather forecast model comprises a sub-area weather forecast unit, a sub-area weather forecast verification unit, a location weather forecast unit, and a location weather forecast verification unit; - training the weather forecast model based on the historical weather information of the plurality of sub-areas, the weather correlation degree of any two of the sub-areas, the historical weather information of the plurality of locations to be monitored, and the weather correlation degree of any two locations to be monitored comprises: - Creating a plurality of first training patterns based on the historical weather information of the target area and the historical weather information of the plurality of sub-areas, wherein the first training pattern comprises the historical weather information of the target area at a first historical time and the historical weather information of the plurality of sub-areas at the first historical point in time; - creating a plurality of second training patterns based on the historical weather information of the plurality of sub-areas and the historical weather information of the plurality of locations to be monitored, wherein the second training pattern comprises the historical weather information of the target area at a second historical time and the historical weather information of the plurality of sub-areas at the second historical time; - Creating a plurality of third training patterns based on the historical weather information of the plurality of sub-areas, the historical weather information of the plurality of sub-areas, and the historical weather information of the plurality of locations to be monitored, wherein the third training pattern includes the historical weather information of the target area at a third historical time, the historical weather information of the plurality of sub-areas at the third historical time, and the historical weather information of the plurality of locations to be monitored at the third historical time; - Training the weather forecast model based on the plurality of first training samples, the plurality of second training samples and the plurality of third training samples. 4. The sensor-based early warning method for cluster fog formation according to claim 3, characterized in that training the weather forecast model based on the plurality of first training patterns, the plurality of second training patterns, and the plurality of third training patterns comprises: - inputting the historical weather information of the target area at a first historical time and the weather correlation degree of any two sub-areas included in the first training pattern to the sub-area weather forecasting unit, wherein the sub-area weather forecasting unit outputs the predicted first historical weather forecast information of the plurality of sub-areas; - calculating a first loss value based on the first historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecasting unit and the historical weather information of the plurality of sub-areas at the first historical time included in the first training pattern; - Inputting the historical weather information of the plurality of sub-areas at a first historical time and the weather correlation degree of any two sub-areas into the sub-area weather forecast verification unit, wherein the sub-area weather forecast verification unit outputs the predicted historical weather forecast information of the target area; - Calculating a second loss value based on the historical weather forecast information of the target area predicted by the sub-area weather forecast verification unit and the historical weather information of the target area at the first historical time included in the first training sample; - Performing a first round of parameter optimization on the sub-area weather forecast unit and the sub-area weather forecast verification unit based on the first loss value and the second loss value; - Inputting the historical weather information of the plurality of sub-areas at the second historical time point included in the second training sample and the weather correlation degree of any two locations to be monitored to the location weather forecasting unit after the sub-area weather forecasting unit and the sub-area weather forecasting verification unit have completed the first round of parameter optimization, wherein the location weather forecasting unit outputs the predicted first historical weather forecast information of the locations to be monitored; - Calculating a third loss value based on the first historical weather forecast information of the plurality of locations to be monitored and the historical weather information of the plurality of locations to be monitored at the second historical time point included in the second training sample; - Inputting the historical weather information of the plurality of locations to be monitored at the second historical time point included in the second training sample and the weather correlation degree of any two locations to be monitored to the location weather forecasting verification unit, wherein the location weather forecasting verification unit outputs the predicted historical weather forecast information of the plurality of sub-areas; - calculating a fourth loss value based on the historical weather forecast information of the plurality of sub-areas predicted by the location weather forecast verification unit and the historical weather information of the plurality of sub-areas at the second historical time point included in the second training pattern; - Perform a second round of parameter optimization on the location weather forecast unit and the location weather forecast verification unit based on the third loss value and the fourth loss value; - inputting the historical weather information of the target area at the third historical time point included in the third training sample and the weather correlation degree of any two sub-areas to the sub-area weather forecasting unit after the site weather forecasting unit and the site weather forecasting verification unit complete the second round of parameter optimization, wherein the sub-area weather forecasting unit outputs the predicted second historical weather forecast information of the plurality of sub-areas; - calculating a fifth loss value based on the second historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecasting unit and the historical weather information of the plurality of sub-areas at the third historical time point including training patterns in the third; - inputting the second historical weather forecast information of the plurality of sub-areas predicted by the sub-area weather forecasting unit and the weather correlation degree of any two sub-areas to the site weather forecasting unit, wherein the site weather forecasting unit outputs the predicted second historical weather forecast information of the plurality of sites to be monitored; - Calculating a sixth loss value based on the second historical weather forecast information of the plurality of locations to be monitored predicted by the location weather forecast unit and the historical weather information of the plurality of locations to be monitored at the third historical time point, which was obtained in the third Training patterns include; - Perform a third round of parameter optimization on the sub-area weather forecast unit and the Site weather prediction unit based on the fifth loss value and the sixth loss value; - wherein the trained weather prediction model comprises a sub-area weather prediction unit and a site weather prediction unit that have completed the third round of parameter optimization. 5. Sensor-based early warning method for cluster fog formation according to claim 4, characterized in that the first loss value is calculated based on the following first loss function: : EE ml „3 > (Pie forseacet.fs TS era F L; Fe De forens) 4} where the first loss value is, * * is the value of the k-th meteorological factor in the i-th sub-area predicted by the F, gore! sub-area weather forecast unit; © 5 is the value of the k-th meteorological factor in the i-th KK sub-area included in the first training sample; Ï is the total number of meteorological factors, P, is the total number P, of sub-areas, ' is the preset parameter and PE, >= where the third loss value is calculated based on the following third loss function: ) > _, {F, ie Foren) sr Ve kesmiljj 3 : Da = Fa wa 3 P, where Fa is the third loss value, Fick Fersensth?) is the value of the k-th meteorological factor of the j-th location to be monitored predicted by the location weather forecast unit, Vire is the value of the k-th meteorological factor of the j-th location to be monitored, which includes in the second training sample, ; is the total number of locations to be monitored, Fa is the preset parameter, Pf and ° 6. The sensor-based cluster fog early warning method according to any one of claims 2-5, characterized in that determining the pollutant prediction information of the plurality of locations to be monitored based on the pollutant prediction information of the target area comprises: - obtaining historical pollutant information of the plurality of sub-areas and confirming the pollutant correlation degree of any two of the sub-areas based on the historical pollutant information of the plurality of sub-areas; - obtaining historical pollutant information of the plurality of locations to be monitored and confirming the pollutant correlation degree of any two of the locations to be monitored based on the historical pollutant information of the plurality of locations to be monitored; - creating a pollutant prediction model; - Training the pollutant prediction model based on the historical pollutant information of the multiple sub-areas, the pollutant correlation degree of any two sub-areas, the historical pollutant information of the multiple sites to be monitored, and the pollutant correlation degree of any two sites to be monitored; - Obtaining the pollutant prediction information of the multiple sites to be monitored based on the pollutant prediction information of the target area by the trained pollutant prediction model. 7. A sensor-based early warning method for cluster fog formation according to any one of claims 1-5, characterized in that the prediction of cluster fog occurrence areas and cluster fog formation information based on the weather forecast information and pollutant forecast information of the plurality of locations to be monitored comprises the following: - obtaining historical cluster fog information of the plurality of locations to be monitored and confirming the degree of correlation of the cluster fog of any two of the locations to be monitored based on the historical cluster fog information of the plurality of locations to be monitored; - for each location to be monitored, determining the initial cluster fog occurrence areas and cluster fog formation information at the location to be monitored based on the weather forecast information and Pollutant prediction information of the site to be monitored by a local cluster fog prediction model; - Determining the cluster fog occurrence areas and the cluster fog formation information based on the initial cluster fog occurrence areas and the cluster fog formation information at each of the sites to be monitored and the correlation degree of the cluster fog of any two sites to be monitored by a global cluster fog prediction model. 8. The sensor-based cluster fog early warning method according to any one of claims 1-5, characterized in that generating the cluster fog warning information based on the cluster fog occurrence areas and the cluster fog formation information comprises: - determining traffic warning roads based on the cluster fog occurrence areas; - determining traffic control measures for the traffic warning road based on the cluster fog formation information. 9. Sensor-based early warning method for cluster fog formation according to one of claims 1-5, characterized in that determining the real-time status of the cluster fog based on cluster fog images from the target location for edge computing comprises: - Identifying the cluster nebula images through a convolutional neural network model to determine the real-time status of the cluster nebula. 10. The sensor-based cluster fog formation early warning method according to claim 9, characterized in that updating the cluster fog occurrence areas and the cluster fog formation information based on the real-time status of the cluster fog comprises: - generating real-time adjusted cluster fog occurrence areas and cluster fog formation information by a long short-term memory network model based on the real-time status of the cluster fog at multiple historical times and the predicted cluster fog occurrence areas and the cluster fog formation information.