CN115857060B - Short-term precipitation prediction method and system based on layered generation countermeasure network - Google Patents
Short-term precipitation prediction method and system based on layered generation countermeasure network Download PDFInfo
- Publication number
- CN115857060B CN115857060B CN202310132177.6A CN202310132177A CN115857060B CN 115857060 B CN115857060 B CN 115857060B CN 202310132177 A CN202310132177 A CN 202310132177A CN 115857060 B CN115857060 B CN 115857060B
- Authority
- CN
- China
- Prior art keywords
- image
- radar echo
- reflectivity
- radar
- countermeasure network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention provides a short-term precipitation prediction method and a system based on a layered generation countermeasure network, which belong to the technical field of weather prediction, wherein the short-term precipitation prediction method based on the layered generation countermeasure network comprises the following steps of S1, receiving an input website as a target webpage, and acquiring a plurality of radar echo reflectivity image diagrams at the current moment in interval time based on the target webpage; s2, detecting a plurality of radar echo reflectivity image graphs at the current moment, generating image graph detection information, and correspondingly processing the image graph detection information to obtain a plurality of available radar echo reflectivity image graphs; the global generator generates a future time radar echo reflectivity image, the local discriminator discriminates that the future time radar echo reflectivity image is a predicted image or an observed image, the two images are mutually game and jointly optimized, and the finally obtained model can generate a sufficiently clear future radar echo sequence close to reality.
Description
Technical Field
The invention belongs to the technical field of weather forecast, and particularly relates to a short-term precipitation forecast method and system based on a layered generation countermeasure network.
Background
The short-term rainfall forecast has an extremely important role in weather disaster prevention and reduction, and according to the definition of WMO in 1985, the short-term rainfall forecast generally predicts the rainfall or strong convection weather and other weather phenomena in a certain area within 0-2 h. However, establishing an effective model for short-term precipitation prediction faces a significant challenge. Firstly, precipitation is a very complex nonlinear problem, which relates to the exchange problem of ground and air moisture, heat, momentum and the like in the water circulation process; secondly, the precipitation system has various forms, and can be divided into cyclone rain, convection rain, frontal rain and the like, so that the forecasting difficulty is increased; the strong convection precipitation occurs in a medium-small scale weather system, has certain burstiness, and has the characteristics of high speed, small space and the like, so that the strong convection precipitation has certain difficulty in forecasting.
The radar echo map has higher space-time resolution, a large number of radar echo map sequences appear along with the rapid development of radar technology in recent years, and the space-time change of the whole evolution process of the radar echo is reflected in a visual mode, so that the radar echo map becomes a main tool for forecasting the short-term precipitation.
However, the image at the time of the current short-term precipitation prediction is not clear enough, and compared with the observed image, many details are ignored, and the position of the future echo can not be predicted accurately with the lapse of prediction time.
Disclosure of Invention
The embodiment of the invention provides a short-term precipitation prediction method and a short-term precipitation prediction system based on a layered generation countermeasure network, which aim to solve the problem that an image map generated at present is not clear enough.
In view of the above problems, the technical scheme provided by the invention is as follows:
in a first aspect, a method for forecasting short-term precipitation based on a stratified generation countermeasure network includes the steps of:
s1, receiving an input website as a target webpage, and acquiring a plurality of radar echo reflectivity image graphs at the current moment in interval time based on the target webpage;
s2, detecting a plurality of radar echo reflectivity image graphs at the current moment, generating image graph detection information, and correspondingly processing the image graph detection information to obtain a plurality of available radar echo reflectivity image graphs;
s3, converting a plurality of available radar echo reflectivity image maps into a plurality of gray level maps;
s4, arranging a plurality of gray level images in time sequence to form a time sequence image atlas;
and S5, inputting the time sequence image atlas into a future prediction layering generation countermeasure network model for calculation, and generating a plurality of future time radar echo reflectivity image maps.
As a preferable technical solution of the present invention, the detection conditions for detecting the plurality of radar echo reflectivity image maps at the current time include: webpage image graph publishing time, image graph acquisition time and radar echo reflectivity image graph at the current moment.
As a preferred technical solution of the present invention, the method for detecting the radar echo reflectivity image map at the current time includes:
and comparing the radar echo reflectivity image graph at the current moment with the image graph published by the target webpage, and finally detecting information by using the image graph.
As a preferable technical scheme of the invention, the method for comparing the radar echo reflectivity image map at each current moment with the image map published by the target webpage specifically comprises the following steps:
determining a first detection area on the radar echo reflectivity image graph at the current moment, and determining a second detection area corresponding to the first detection area on the image graph published by the target webpage;
determining a first characteristic pixel unit in the first detection area, determining a second characteristic pixel unit on the second detection area, and respectively comparing coordinates and pixel areas between the first characteristic pixel unit and the second characteristic pixel unit to give a first approximate value; if the first approximation value is smaller than a first preset approximation value, executing the next step;
determining a third detection area on the radar echo reflectivity image graph at the current moment, and determining a fourth detection area corresponding to the third detection area on the image graph published by the target webpage;
determining a third characteristic pixel unit in the third detection area, determining a fourth characteristic pixel unit on the fourth detection area, and respectively comparing coordinates and pixel areas between the third characteristic pixel unit and the fourth characteristic pixel unit to give a second approximate value;
synthesizing the first approximation value and the second approximation value to obtain a synthesized approximation value, and comparing the synthesized approximation value with a second preset approximation value; and if the comprehensive approximation value is larger than the second preset approximation value, determining the radar echo reflectivity image graph at the current moment as the defected radar echo reflectivity image graph at the current moment.
As a preferable technical scheme of the invention, the processing method of the image map detection information comprises the following steps:
and re-acquiring according to the image acquisition time of the defective radar echo reflectivity image at the current moment.
As a preferable technical scheme of the invention, the training method for generating the countermeasure network model by the future prediction layering comprises the following steps:
acquiring a plurality of historical radar echo basic reflectivity image graphs and a plurality of historical radar echo basic reflectivity live image graphs, and simultaneously extracting a plurality of typical precipitation image graphs from the plurality of historical radar echo basic reflectivity image graphs and the plurality of historical radar echo basic reflectivity live image graphs;
filtering out the basic reflectivity with less than a preset value in the typical precipitation image graphs to form a required initial sample set;
reading a typical precipitation image graph of the initial sample set, taking the current moment as the moment t, selecting the moment t and N typical precipitation image graphs in interval time before the moment t to construct 1 sample, and obtaining a plurality of available sample sets according to the mode;
randomly scrambling a plurality of available sample sets, and dividing the available sample sets into a training set and a testing set according to a proportion;
and establishing an architecture of an initial layering generation countermeasure network model, inputting the training set into the initial layering generation countermeasure network model for training, and obtaining the future prediction layering generation countermeasure network model.
As a preferred technical solution of the present invention, the future prediction layering generation countermeasure network model includes a global generator and a local discriminator;
the global generator is configured to generate a future time radar basic reflectivity map, and the local discriminator is configured to distinguish the future time radar basic reflectivity map as a predicted image map or an observed image map;
as a preferred technical solution of the present invention, the local discriminator differentiating method is as follows: dividing the radar basic reflectivity image map at the future time into a plurality of local areas, calculating the proportion occupied by radar echoes of each local area, giving weight according to the proportion, calculating the first probability that each local area is the observed image map, multiplying the weight of each local area by the first probability to obtain the second probability that the radar basic reflectivity image map at the future time is the observed image map, and if the second probability is larger than the preset probability, the radar basic reflectivity image map at the future time is the observed image map.
As a preferred embodiment of the present invention, the local discriminator includes a buffer configured to store the predictive image map of the history.
On the other hand, the embodiment of the invention also provides a short-term precipitation prediction system based on a layered generation countermeasure network, which comprises the following steps:
the definition module is configured to receive an input website as a target webpage and acquire a plurality of radar echo reflectivity image graphs at the current moment in an interval time based on the target webpage;
the detection processing module is configured to detect the radar echo reflectivity image graphs at the current moment, generate image graph detection information, and correspondingly process the radar echo reflectivity image graphs based on the image graph detection information to obtain a plurality of available radar echo reflectivity image graphs;
a conversion module configured to convert the plurality of available radar echo reflectivity image maps into a plurality of gray scale maps;
a sorting module configured to chronologically arrange the plurality of gray scale images to form a time-series image atlas;
a generation module configured to generate a plurality of future time radar echo reflectivity image maps using the time series image atlas input to a future prediction hierarchy generation countermeasure network model for calculation.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
(1) According to the method, after the target webpage captures the radar echo reflectivity image at the current moment, the radar echo reflectivity image at the current moment is detected, the defected radar echo reflectivity image at the current moment is found and determined, and therefore the accuracy of radar echo reflectivity image generation at the future moment can be guaranteed.
(2) The future prediction layering generates an antagonism network model, and the image quality reduction caused by downsampling is avoided in an upsampling mode, meanwhile, the evolution trend of radar echo can be captured, and the generation of a clear future radar echo diagram is facilitated. The local discriminator distinguishes the predicted image from the observed image according to the local region, so that the local discriminator discriminates the predicted image or the observed image in combination according to the probability of the local region. And meanwhile, a buffer area mechanism is introduced, a historical predicted image sequence is stored, and in the identification process, the final result is identified according to the current predicted image and the duration predicted image, so that the final predicted result accords with the time sequence. The two are trained in a countermeasure mode, the global generator generates a future time radar echo reflectivity image, the local discriminator discriminates that the future time radar echo reflectivity image is a predicted image or an observed image, the two images are mutually game and jointly optimized, and the finally obtained model can generate a sufficiently clear future radar echo sequence close to reality.
The foregoing description is only an overview of the present invention, and is intended to be implemented in accordance with the teachings of the present invention in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present invention more readily apparent.
Drawings
FIG. 1 is a flow chart of the disclosed method for short-term precipitation prediction based on a stratified generation countermeasure network;
FIG. 2 is a schematic diagram of an initial hierarchical generation countermeasure network model based on a method for forecasting short-term precipitation of a hierarchical generation countermeasure network according to the present invention;
FIG. 3 is a schematic diagram of a process of a local discriminator of the short-term precipitation prediction method based on a stratified generation countermeasure network disclosed in the present invention;
FIG. 4 is a schematic diagram of a buffer area of a method for forecasting short-term precipitation based on a stratified generation countermeasure network according to the present invention;
FIG. 5 is a schematic diagram of a system for forecasting short-term precipitation based on a stratified generation countermeasure network in accordance with the present invention.
Reference numerals illustrate: 100. defining a module; 200. a detection processing module; 300. a conversion module; 400. a sequencing module; 500. and generating a module.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the invention.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the invention.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
Examples
Referring to the accompanying drawings 1-4, the invention provides a technical scheme that: the short-term precipitation forecasting method based on the layered generation countermeasure network comprises the following steps:
s1, receiving an input website as a target webpage, and acquiring a plurality of radar echo reflectivity image graphs at the current moment in interval time based on the target webpage; wherein the target webpage is
(i.e., a central weather station typhoon net); the interval time ranges from 1 to 20 minutes, and is set according to practical conditions.
S2, detecting a plurality of radar echo reflectivity image graphs at the current moment, generating image graph detection information, and correspondingly processing the image graph detection information to obtain a plurality of available radar echo reflectivity image graphs;
s3, converting a plurality of available radar echo reflectivity image maps into a plurality of gray level maps;
s4, arranging a plurality of gray level images in time sequence to form a time sequence image atlas;
and S5, inputting the time sequence image atlas into a future prediction layering generation countermeasure network model for calculation, and generating a plurality of future time radar echo reflectivity image maps.
Specifically, after the radar echo reflectivity image map at the current moment is captured by the target webpage, the radar echo reflectivity image map at the current moment is detected, and the defected radar echo reflectivity image map at the current moment is found and determined, so that the accuracy of generating the radar echo reflectivity image map at the future moment can be ensured.
Meanwhile, in the process of generating the countermeasure network model training by the future prediction layering, the local discriminator distinguishes the prediction image graph from the observation image graph according to the local area, so that the local discriminator judges and distinguishes the prediction image graph or the observation image graph according to the probability of the local area. And a buffer mechanism is introduced to store the historical prediction sequence. During the identification, the final result is identified according to not only the predicted image but also the duration predicted image, so that the final predicted result accords with the time sequence.
In some embodiments, the detection conditions for detecting the plurality of radar echo reflectivity image maps at the current moment include: webpage image graph publishing time, image graph acquisition time and radar echo reflectivity image graph at the current moment.
In some embodiments, the method for detecting the radar echo reflectivity image map at the current time includes:
and comparing the radar echo reflectivity image graph at the current moment with the image graph published by the target webpage, and finally detecting information by using the image graph.
Specifically, in each capturing process, due to the influence of the jump speed of the webpage image map, the information in the radar echo reflectivity image map obtained at the current moment is occasionally lost, so that the prediction of the antagonism network model is not generated in the future prediction layering. Therefore, in order to obtain a plurality of radar echo reflectivity image graphs at the current moment with perfect information, the image graphs are compared to find the defective radar echo reflectivity image graph at the current moment.
In some embodiments, the comparing the radar echo reflectivity image map at the current time with the image map published by the target web page specifically includes:
determining a first detection area on the radar echo reflectivity image graph at the current moment, and determining a second detection area corresponding to the first detection area on the image graph published by the target webpage;
determining a first characteristic pixel unit in the first detection area, determining a second characteristic pixel unit on the second detection area, and respectively comparing coordinates and pixel areas between the first characteristic pixel unit and the second characteristic pixel unit to give a first approximate value; if the first approximation value is smaller than a first preset approximation value, executing the next step;
determining a third detection area on the radar echo reflectivity image graph at the current moment, and determining a fourth detection area corresponding to the third detection area on the image graph published by the target webpage;
determining a third characteristic pixel unit in the third detection area, determining a fourth characteristic pixel unit on the fourth detection area, and respectively comparing coordinates and pixel areas between the third characteristic pixel unit and the fourth characteristic pixel unit to give a second approximate value;
synthesizing the first approximation value and the second approximation value to obtain a synthesized approximation value, and comparing the synthesized approximation value with a second preset approximation value; and if the comprehensive approximation value is larger than the second preset approximation value, determining the radar echo reflectivity image graph at the current moment as the defected radar echo reflectivity image graph at the current moment.
Specifically, in the comparison process, specific areas are selected from the radar echo reflectivity image map and the webpage image map at the current moment to perform comparison among pixel units, so that the comparison is more significant, and the evaluation approximation value can be more approximate to reality.
In some embodiments, the method for processing the image detection information includes:
and re-acquiring according to the image acquisition time of the defective radar echo reflectivity image at the current moment.
In some embodiments, the training method for generating the countermeasure network model by the future prediction hierarchy is as follows:
acquiring a plurality of historical radar echo basic reflectivity image graphs and a plurality of historical radar echo basic reflectivity live image graphs, and simultaneously extracting a plurality of typical precipitation image graphs from the plurality of historical radar echo basic reflectivity image graphs and the plurality of historical radar echo basic reflectivity live image graphs; for example, in 2019, 2020, and 2021, a plurality of historical radar echo basic reflectance image maps and a plurality of historical radar echo basic reflectance live image maps of the preceding three years are acquired.
Filtering out the basic reflectivity with less than a preset value in the typical precipitation image graphs to form a required initial sample set; wherein the preset value is 1/10.
Reading a typical precipitation image graph of the initial sample set, taking the current moment as the moment t, selecting the moment t and N typical precipitation image graphs in interval time before the moment t to construct 1 sample, and obtaining a plurality of available sample sets according to the mode; wherein the interval time is 10 minutes, and N is 10.
Randomly scrambling a plurality of available sample sets, and dividing the available sample sets into a training set and a testing set according to a proportion; wherein the ratio is 8:2.
And establishing an architecture of an initial layering generation countermeasure network model, inputting the training set into the initial layering generation countermeasure network model for training, and obtaining the future prediction layering generation countermeasure network model.
In some embodiments, the future predictive stratification generation countermeasure network model includes a global generator and a local discriminator;
specifically, the global generator is composed of numerous subnets in a hierarchical structure. Each subnet generates a radar base reflectivity image map of a different resolution scale. The small scale resolution radar basic reflectance image generated from the upper subnetwork will be used as input to the lower subnetwork.
Further, reference is made to fig. 2. G 0 And G 1 Are two sub-networks of G and are both convolutional neural networks.
The input X is the training set obtained above, and shows the structure and evolution trend of radar echo. X comprises X 0 And X 1 Two subsequences, which are sequences of radar base reflectivity maps at the same two sampling scales. During training, X is taken as 1 Input to G 1 In the subnetwork, generate and X 1 Future time radar basic reflectivity image graph G with same resolution 1 (X 1 ). Lower layer subnetwork G 0 X is to be 0 And up-sampled G 1 (X 1 ) Radar basic reflectivity image is taken as input, and generation and X are carried out 0 Future time radar basic reflectivity image graph G with same resolution 0 (X 0 )。
The above operation is repeated, and the image generated by the lowest subnet will be used as the predicted image map generated by the global generator G.
The global generator is configured to generate a future time radar basic reflectivity map, and the local discriminator is configured to distinguish the future time radar basic reflectivity map as a predicted image map or an observed image map;
referring to fig. 3, the local discriminator distinguishing method is as follows: dividing the radar basic reflectivity image map at the future time into a plurality of local areas, calculating the proportion occupied by radar echoes of each local area, giving weight according to the proportion, calculating the first probability that each local area is the observed image map, multiplying the weight of each local area by the first probability, combining and distinguishing to obtain the second probability that the radar basic reflectivity image map at the future time is the observed image map, and if the second probability is larger than the preset probability, the radar basic reflectivity image map at the future time is the observed image map.
Wherein the expression of the second probability is as follows:
wherein:
a first probability that the local region is the observation image map is represented;
In particular, combining the predictive probabilities of all local regions to make a decision, the local discriminator may better characterize radar echoes moving within the local region relative to the entire stationary base map.
In some embodiments, referring to FIG. 4, since the global generator generates the future time radar base reflectivity map during the future prediction hierarchy generation of the future time radar base reflectivity map against the network model, the local discriminator discriminates that the future time radar base reflectivity map is a predicted or observed map, however when the local discriminator focuses only on the predicted map, the local discriminator is easily confused by the current predicted map, thereby ignoring the timing of the predicted radar echo map. Thus, to enhance the ability of a local discriminator to distinguish between predicted image maps, the local discriminator includes a buffer configured to store historical predicted image maps.
Specifically, the local discriminator will be updated in conjunction with the predicted image map and the randomly sampled images from the buffer in each iteration of training the discriminator. After each training iteration, the predicted image is put into the buffer area and the predicted image which enters the buffer area earliest is removed according to the first-in first-out principle, so that the buffer area is updated.
It should be noted that the size of the buffer is fixed; at the same time, the size of the trim buffer does not affect the performance of the model. By introducing the buffer area, the radar echo sequence generated by the predictive image map is more consistent with the time sequence of the data.
In the training process, after each batch of training sets is used for training the initial layering generation countermeasure network model, the gradient descent method is used for updating the parameters of the initial layering generation countermeasure network model, so that a loss function is reduced, and the method is used for predicting a high-quality radar basic reflectivity image map at the future time. The specific loss function is set as follows:
by training the global generator and the local discriminator using the resistance penalty, the two compete with each other. The global generator and the local discriminator are updated alternately according to the countermeasure policy.
Assuming that the global generator has M-layer subnetworks, the countermeasures loss through the sum of the M-layer subnetworks, the countermeasures loss function of the training global generator is as follows:
wherein G is m (X m ) Representing a radar basic reflectivity image map generated by a global generator subnet;
T m an observation image map corresponding to the radar basic reflectivity image map at the future moment is shown;
d (x) represents the probability that the incoming available radar base reflectivity map is the observation map.
Then, through L 1 The loss function is used for punishing the distance difference between each pixel of the predicted image and each pixel of the observed image, so that the global generator can generate a radar basic reflectivity image with enough reality at the future moment, and the specific formula is as follows:
wherein: w represents the width of the observed image in pixels;
h represents the height of the observed image in pixels;
(p, q) represents pixel coordinates.
Further, the gradient difference between the predicted image and the corresponding observed image is calculated to improve the quality of generating the radar basic reflectivity image at the future time, and the specific gradient difference loss function is as follows:
specifically, when the global generator is trained, the gradient difference loss function directly penalizes the difference in the image gradient, and then the radar basic reflectivity image map generated by the global generator at the future time is sharpened.
Further, the total variation loss is introduced into a loss function of the global generator to restrict noise and prevent the generated image from being influenced by artifacts. Total variation loss is defined as follows:
further, the overall loss function of the global generator is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
、/>
、/>
and->
Are all empirically weighted parameters.
Further, the challenge loss function of training the local discriminator is as follows:
referring to fig. 5, the embodiment of the invention further discloses a short-term precipitation prediction system based on a layered generation countermeasure network, which comprises:
the definition module 100 is configured to receive an input website as a target webpage, and acquire a plurality of radar echo reflectivity image graphs at the current moment in an interval time based on the target webpage;
the detection processing module 200 is configured to detect the radar echo reflectivity image maps at the current moment, generate image map detection information, and perform corresponding processing based on the image map detection information to obtain a plurality of available radar echo reflectivity image maps;
a conversion module 300 configured to convert the plurality of available radar echo reflectivity image maps into a plurality of gray scale maps;
a sorting module 400 configured to chronologically arrange the plurality of gray scale images to form a time-series image atlas;
a generation module 500 configured to generate a plurality of future time radar echo reflectivity image maps using the time series image atlas input to a future prediction hierarchy generation countermeasure network model for calculation.
In the foregoing detailed description, various features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the claimed embodiments of the subject matter require more features than are expressly recited in each claim. Rather, as the following claims reflect, invention lies in less than all features of a single disclosed embodiment. Thus the following claims are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate preferred embodiment of this invention.
Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. The processor and the storage medium may reside as discrete components in a user terminal.
For a software implementation, the techniques described herein may be implemented with modules (e.g., procedures, functions, and so on) that perform the functions described herein. These software codes may be stored in memory units and executed by processors. The memory unit may be implemented within the processor or external to the processor, in which case it can be communicatively coupled to the processor via various means as is known in the art.
The foregoing description includes examples of one or more embodiments. It is, of course, not possible to describe every conceivable combination of components or methodologies for purposes of describing the aforementioned embodiments, but one of ordinary skill in the art may recognize that many further combinations and permutations of various embodiments are possible. Accordingly, the embodiments described herein are intended to embrace all such alterations, modifications and variations that fall within the scope of the appended claims. Furthermore, as used in the specification or claims, the term "comprising" is intended to be inclusive in a manner similar to the term "comprising," as interpreted when employed as a transitional word in a claim. Furthermore, any use of the term "or" in the specification of the claims is intended to mean "non-exclusive or".
Claims (6)
1. The short-term precipitation prediction method based on the layered generation countermeasure network is characterized by comprising the following steps of:
s1, receiving an input website as a target webpage, and acquiring a plurality of radar echo reflectivity image graphs at the current moment in interval time based on the target webpage;
s2, detecting a plurality of radar echo reflectivity image graphs at the current moment, generating image graph detection information, and correspondingly processing the image graph detection information to obtain a plurality of available radar echo reflectivity image graphs; wherein, the detection conditions include: webpage image graph publishing time, image graph acquisition time and radar echo reflectivity image graph at the current moment; the detection method comprises the following steps: comparing the radar echo reflectivity image graph at the current moment with the image graph published by the target webpage to finally obtain image graph detection information; the method specifically comprises the following steps:
determining a first detection area on the radar echo reflectivity image graph at the current moment, and determining a second detection area corresponding to the first detection area on the image graph published by the target webpage;
determining a first characteristic pixel unit in the first detection area, determining a second characteristic pixel unit on the second detection area, and respectively comparing coordinates and pixel areas between the first characteristic pixel unit and the second characteristic pixel unit to give a first approximate value; if the first approximation value is smaller than a first preset approximation value, executing the next step;
determining a third detection area on the radar echo reflectivity image graph at the current moment, and determining a fourth detection area corresponding to the third detection area on the image graph published by the target webpage;
determining a third characteristic pixel unit in the third detection area, determining a fourth characteristic pixel unit on the fourth detection area, and respectively comparing coordinates and pixel areas between the third characteristic pixel unit and the fourth characteristic pixel unit to give a second approximate value;
synthesizing the first approximation value and the second approximation value to obtain a synthesized approximation value, and comparing the synthesized approximation value with a second preset approximation value; if the comprehensive approximation value is larger than the second preset approximation value, determining the radar echo reflectivity image graph at the current moment as the defective radar echo reflectivity image graph at the current moment;
the processing method of the image map detection information comprises the following steps: re-acquiring according to the image acquisition time of the defective radar echo reflectivity image at the current moment;
s3, converting a plurality of available radar echo reflectivity image maps into a plurality of gray level maps;
s4, arranging a plurality of gray level images in time sequence to form a time sequence image atlas;
and S5, inputting the time sequence image atlas into a future prediction layering generation countermeasure network model for calculation, and generating a plurality of future time radar echo reflectivity image maps.
2. The short-term precipitation prediction method based on a layered generation countermeasure network according to claim 1, wherein the training method of the future prediction layered generation countermeasure network model is as follows:
acquiring a plurality of historical radar echo basic reflectivity image graphs and a plurality of historical radar echo basic reflectivity live image graphs, and simultaneously extracting a plurality of typical precipitation image graphs from the plurality of historical radar echo basic reflectivity image graphs and the plurality of historical radar echo basic reflectivity live image graphs;
filtering out the basic reflectivity with less than a preset value in the typical precipitation image graphs to form a required initial sample set;
reading a typical precipitation image graph of the initial sample set, taking the current moment as the moment t, selecting the moment t and N typical precipitation image graphs in interval time before the moment t to construct 1 sample, and obtaining a plurality of available sample sets according to the mode;
randomly scrambling a plurality of available sample sets, and dividing the available sample sets into a training set and a testing set according to a proportion;
and establishing an architecture of an initial layering generation countermeasure network model, inputting the training set into the initial layering generation countermeasure network model for training, and obtaining the future prediction layering generation countermeasure network model.
3. The method of claim 2, wherein the future prediction hierarchy generation countermeasure network model comprises a global generator and a local discriminator;
the global generator is configured to generate a future time radar basic reflectivity map, and the local discriminator is configured to distinguish the future time radar basic reflectivity map as a predicted or observed image map.
4. A method of short-term precipitation prediction based on a stratified generation countermeasure network as claimed in claim 3, wherein the local discriminator distinguishing method is: dividing the radar basic reflectivity image map at the future time into a plurality of local areas, calculating the proportion occupied by radar echoes of each local area, giving weight according to the proportion, calculating the first probability that each local area is the observed image map, multiplying the weight of each local area by the first probability to obtain the second probability that the radar basic reflectivity image map at the future time is the observed image map, and if the second probability is larger than the preset probability, the radar basic reflectivity image map at the future time is the observed image map.
5. The method of claim 4, wherein the local discriminator comprises a buffer configured to store the predictive image of history.
6. The short-term precipitation prediction system based on the layered generation countermeasure network, which is applied to the short-term precipitation prediction method based on the layered generation countermeasure network as claimed in any one of claims 1 to 5, is characterized by comprising the following steps:
the definition module is configured to receive an input website as a target webpage and acquire a plurality of radar echo reflectivity image graphs at the current moment in an interval time based on the target webpage;
the detection processing module is configured to detect the radar echo reflectivity image graphs at the current moment, generate image graph detection information, and correspondingly process the radar echo reflectivity image graphs based on the image graph detection information to obtain a plurality of available radar echo reflectivity image graphs;
a conversion module configured to convert the plurality of available radar echo reflectivity image maps into a plurality of gray scale maps;
a sorting module configured to chronologically arrange the plurality of gray scale images to form a time-series image atlas;
a generation module configured to generate a plurality of future time radar echo reflectivity image maps using the time series image atlas input to a future prediction hierarchy generation countermeasure network model for calculation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310132177.6A CN115857060B (en) | 2023-02-20 | 2023-02-20 | Short-term precipitation prediction method and system based on layered generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310132177.6A CN115857060B (en) | 2023-02-20 | 2023-02-20 | Short-term precipitation prediction method and system based on layered generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115857060A CN115857060A (en) | 2023-03-28 |
| CN115857060B true CN115857060B (en) | 2023-05-09 |
Family
ID=85658360
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310132177.6A Active CN115857060B (en) | 2023-02-20 | 2023-02-20 | Short-term precipitation prediction method and system based on layered generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115857060B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116228046B (en) * | 2023-05-09 | 2023-07-18 | 成都信息工程大学 | Mountain area space precipitation estimation method based on satellite remote sensing and geographic data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6018307A (en) * | 1998-03-30 | 2000-01-25 | Mitsubishi Denki Kabushiki Kaisha | Meteorological radar system |
| CN115113301A (en) * | 2022-08-23 | 2022-09-27 | 南京信息工程大学 | Emergency short-term forecasting method and system based on multi-source data fusion |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH1114760A (en) * | 1997-04-30 | 1999-01-22 | Nippon Telegr & Teleph Corp <Ntt> | Meteorological image prediction method and device, and recording medium with recorded meteorological image prediction program |
| JP3476750B2 (en) * | 2000-06-19 | 2003-12-10 | 三菱電機株式会社 | Competitive radio wave transmission method and device |
| CN108364097B (en) * | 2018-02-07 | 2019-01-08 | 国家海洋局北海预报中心 | Based on the typhoon cloud system prediction technique for generating confrontation network |
| CN110765878A (en) * | 2019-09-20 | 2020-02-07 | 苏州大圜科技有限公司 | Short-term rainfall prediction method |
| CN110568442B (en) * | 2019-10-15 | 2021-08-20 | 中国人民解放军国防科技大学 | Radar echo extrapolation method based on confrontation extrapolation neural network |
| CN111242372A (en) * | 2020-01-10 | 2020-06-05 | 上海眼控科技股份有限公司 | Weather radar echo map prediction method and device, computer equipment and storage medium |
| CN111814961A (en) * | 2020-07-03 | 2020-10-23 | 深圳市赑玄阁科技有限公司 | Rainfall forecast method based on generative confrontation network |
| CN113239722B (en) * | 2021-03-31 | 2022-08-30 | 成都信息工程大学 | Deep learning based strong convection extrapolation method and system under multi-scale |
-
2023
- 2023-02-20 CN CN202310132177.6A patent/CN115857060B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6018307A (en) * | 1998-03-30 | 2000-01-25 | Mitsubishi Denki Kabushiki Kaisha | Meteorological radar system |
| CN115113301A (en) * | 2022-08-23 | 2022-09-27 | 南京信息工程大学 | Emergency short-term forecasting method and system based on multi-source data fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115857060A (en) | 2023-03-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115857060B (en) | Short-term precipitation prediction method and system based on layered generation countermeasure network | |
| CN112183432A (en) | Building area extraction method and system based on medium-resolution SAR image | |
| Jing et al. | HPRNN: A hierarchical sequence prediction model for long-term weather radar echo extrapolation | |
| Jing et al. | AENN: A generative adversarial neural network for weather radar echo extrapolation | |
| CN112597815A (en) | Synthetic aperture radar image ship detection method based on Group-G0 model | |
| CN110765878A (en) | Short-term rainfall prediction method | |
| CN113496104A (en) | Rainfall forecast correction method and system based on deep learning | |
| CN115223063B (en) | Deep learning-based unmanned aerial vehicle remote sensing wheat new variety lodging area extraction method and system | |
| CN110532918B (en) | Off-shore wind farm space-time attribute determination method based on time series remote sensing images | |
| CN115063786A (en) | High-order distant view fuzzy license plate detection method | |
| CN114627387A (en) | Regional carbon emission prediction method, device and medium based on Beidou positioning and remote sensing image detection | |
| CN115965862A (en) | SAR ship target detection method based on mask network fusion image characteristics | |
| CN115841629A (en) | SAR image ship detection method based on convolutional neural network | |
| Wang et al. | Revisiting SLIC: Fast superpixel segmentation of marine SAR images using density features | |
| Zhong et al. | Spatiotemporal convolutional LSTM for radar echo extrapolation | |
| Zhou et al. | A lightweight SE-YOLOv3 network for multi-scale object detection in remote sensing imagery | |
| CN114565824A (en) | Single-stage rotating ship detection method based on full convolution network | |
| CN114898206A (en) | Short-time heavy rainfall forecasting method, computer equipment and storage medium | |
| CN112766108A (en) | SAR image target detection method based on context information | |
| CN116704366A (en) | Storm drop zone identification early warning method and device based on transform deep learning model | |
| Luo et al. | Robust precipitation bias correction through an ordinal distribution autoencoder | |
| CN112446256A (en) | Vegetation type identification method based on deep ISA data fusion | |
| CN114913504A (en) | Vehicle target identification method of remote sensing image fused with self-attention mechanism | |
| CN112149536B (en) | Squall line wind speed prediction method | |
| Zhao et al. | Advancing Realistic Precipitation Nowcasting with a Spatiotemporal Transformer-Based Denoising Diffusion Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |