CN114418211A - Method and device for correcting precipitation - Google Patents
Method and device for correcting precipitation Download PDFInfo
- Publication number
- CN114418211A CN114418211A CN202210051053.0A CN202210051053A CN114418211A CN 114418211 A CN114418211 A CN 114418211A CN 202210051053 A CN202210051053 A CN 202210051053A CN 114418211 A CN114418211 A CN 114418211A
- Authority
- CN
- China
- Prior art keywords
- data
- precipitation
- model
- normalized
- training
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Environmental & Geological Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- Game Theory and Decision Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a device for correcting precipitation, wherein the method comprises the following steps: acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data; preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data; acquiring a precipitation-generation type confrontation model; analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate a precipitation correction value; and performing precipitation correction on the initial precipitation prediction data based on the precipitation correction value to generate corrected precipitation. Initial precipitation prediction data are optimized according to target ground condition data of a target position corresponding to the precipitation to be corrected, and data correction is performed by adopting an intelligent antagonistic learning model according to the optimized data, so that the precipitation is accurately corrected, the accuracy of precipitation data is improved, and the actual requirements of technicians are met.
Description
Technical Field
The invention relates to the technical field of weather data processing, in particular to a precipitation correction method, a precipitation correction device and a computer-readable storage medium.
Background
In the daily life of people, a large number of natural phenomena exist, which have various influences on the daily life of people, for example, precipitation, which is a phenomenon that water vapor in the atmosphere is condensed and then falls to the ground as liquid water or solid water, is a general term for rain, snow, dew, frost, aragonite, hail and other phenomena occurring in nature, the size of precipitation is comprehensively influenced by factors such as geographical position, atmospheric circulation, weather system conditions and the like, and the nonuniformity of precipitation in spatial distribution and the instability of precipitation in time variation are main causes of disasters.
In the prior art, various Weather forecasting technologies exist, and with the continuous development of scientific technologies, the forecasting accuracy of the numerical Weather forecasting on future Weather meteorological elements is gradually improved, for example, the European Centre for Medium-Range Weather Forecasts (EC) releases the middle-term numerical Weather which is the currently more accurate numerical Weather forecasting.
However, in the actual application process, the occurrence of precipitation has the characteristics of discontinuity, instantaneity and the like, and compared with other meteorological elements, the precipitation forecast still has a larger lifting space, so that the existing precipitation prediction method and the prediction result are still inaccurate, and the actual requirements of people cannot be met.
Disclosure of Invention
In order to solve the technical problems in the prior art, embodiments of the present invention provide a method for correcting precipitation, in which initial precipitation prediction data is optimized according to target ground condition data of a target position corresponding to the precipitation to be corrected, and an intelligent countermeasure type learning model is used to correct data according to the optimized data, so that the precipitation is accurately corrected, the accuracy of precipitation data is improved, and actual requirements of technicians are met.
In order to achieve the above object, an embodiment of the present invention provides a method for correcting precipitation, where the method includes: acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data; preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data; acquiring a precipitation-generation type confrontation model; analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate a precipitation correction value; and performing precipitation correction on the initial precipitation prediction data based on the precipitation correction value to generate corrected precipitation.
Preferably, the preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data includes: performing normalization processing on the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain normalized data, wherein the normalized data comprises normalized precipitation prediction data, normalized real-time precipitation data and normalized ground condition data; acquiring at least one preset forecasting factor; screening each preset forecast factor based on a preset characteristic calculation algorithm to obtain at least one precipitation forecast factor; generating pre-processed data based on the normalized data and the at least one precipitation forecast factor.
Preferably, the precipitation-generated confrontation model comprises a first sub-model and a second sub-model, and the obtaining the precipitation-generated confrontation model comprises: determining a first initial parameter of the first submodel and a second initial parameter of the second submodel; acquiring a preset data set, wherein the preset data set comprises a training data set and a testing data set; sequentially taking the first sub-model and the second sub-model as target models; training the target model based on the preset data set to obtain a trained model, wherein the trained model comprises a trained first model and a trained second model; generating a precipitation-generative confrontation model based on the trained first model and the trained second model.
Preferably, the training the target model based on the preset data set to obtain a trained model includes: s341) inputting the training data set into the target model to obtain a corresponding output value; s342) determining loss assessment information for the target model based on the training data set, the output values, and the test data set; s343) optimizing initial parameters of the target model based on the loss evaluation information to obtain optimized parameters; s344) generating a first training model based on the optimized parameters and the target model; s345) judging whether the first training model meets the prediction requirement or not based on a preset prediction evaluation index; s3461), if yes, taking the first training model as a trained model; s3462) if not, taking the first training model as a new target model, and jumping to the step S341).
Preferably, the determining loss evaluation information of the target model based on the training data set, the output value and the test data set comprises: acquiring a cross entropy function and an average absolute error function; generating first loss assessment information for a first sub-model based on a cross entropy function, the mean absolute error function, the training data set, the output values, and the test data set, if the target model is the first sub-model; generating second loss assessment information for a second sub-model based on the cross-entropy function, the training data set, the output values, and the test data set if the target model is the second sub-model; and using the first loss evaluation information or the second loss evaluation information as the loss evaluation information of the target model.
Preferably, the analyzing the preprocessed data based on the precipitation-generative confrontation model to generate precipitation correction values includes: processing the normalized precipitation prediction data and the normalized ground condition data based on the first submodel to generate first processed data; and analyzing the normalized real-time precipitation data and the first processed data based on the second submodel to generate precipitation correction values.
Preferably, the processing the normalized precipitation prediction data and the normalized terrain data based on the first submodel to generate first processed data includes: acquiring preset processing times; performing convolution down-sampling operation on the normalized rainfall prediction data and the normalized ground condition data based on the first sub-model and the preset processing times to obtain down-sampled data; performing deconvolution operation on the down-sampled data based on the second sub-model and the preset processing times to obtain deconvoluted data; and executing normalization operation corresponding to the normalized precipitation prediction data and the normalized ground condition data on the deconvolution data to obtain first processed data.
Preferably, the analyzing the normalized real-time precipitation data and the first processed data based on the second submodel to generate precipitation correction values includes: performing dimensionality reduction on the initial precipitation prediction data to obtain dimensionality reduced data, wherein the dimensionality reduced data is one-dimensional data; generating input data based on the data after the dimensionality reduction, the real-time precipitation data after the normalization and the data after the first processing; and analyzing the input data based on the second submodel to generate a precipitation correction value.
Correspondingly, an embodiment of the present invention further provides a device for correcting precipitation, where the device includes: the data acquisition unit is used for acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data; the preprocessing unit is used for preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data; a model acquisition unit for acquiring a precipitation-generation type countermeasure model; the analysis unit is used for analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate a precipitation correction value; and the correcting unit is used for executing precipitation correction on the initial precipitation prediction data based on the precipitation correction value and generating corrected precipitation.
Preferably, the preprocessing unit includes: the normalization module is used for performing normalization processing on the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain normalized data, and the normalized data comprises normalized precipitation prediction data, normalized real-time precipitation data and normalized ground condition data; the forecasting factor acquiring module is used for acquiring at least one preset forecasting factor; the screening module is used for screening each preset forecasting factor based on a preset characteristic calculation algorithm to obtain at least one precipitation forecasting factor; a preprocessing module for generating preprocessed data based on the normalized data and the at least one precipitation forecast factor.
Preferably, the precipitation-generating countermeasure model includes a first sub-model and a second sub-model, and the model obtaining unit includes: a parameter determining module for determining a first initial parameter of the first submodel and a second initial parameter of the second submodel; the device comprises a data set acquisition module, a data processing module and a data processing module, wherein the data set acquisition module is used for acquiring a preset data set, and the preset data set comprises a training data set and a test data set; the target model determining module is used for sequentially taking the first sub-model or the second sub-model as a target model; the training module is used for training the target model based on the preset data set to obtain a trained model, and the trained model comprises a trained first model and a trained second model; and the model establishing module is used for generating a precipitation-generation type confrontation model based on the trained first model and the trained second model.
Preferably, the training the target model based on the preset data set to obtain a trained model includes: s341) inputting the training data set into the target model to obtain a corresponding output value; s342) determining loss assessment information for the target model based on the training data set, the output values, and the test data set; s343) optimizing initial parameters of the target model based on the loss evaluation information to obtain optimized parameters; s344) generating a first training model based on the optimized parameters and the target model; s345) judging whether the first training model meets the prediction requirement or not based on a preset prediction evaluation index; s3461), if yes, taking the first training model as a trained model; s3462) if not, taking the first training model as a new target model, and jumping to the step S341).
Preferably, the determining loss evaluation information of the target model based on the training data set, the output value and the test data set comprises: acquiring a cross entropy function and an average absolute error function; generating first loss assessment information for a first sub-model based on a cross entropy function, the mean absolute error function, the training data set, the output values, and the test data set, if the target model is the first sub-model; generating second loss assessment information for a second sub-model based on the cross-entropy function, the training data set, the output values, and the test data set if the target model is the second sub-model; and using the first loss evaluation information or the second loss evaluation information as the loss evaluation information of the target model.
Preferably, the analysis unit comprises: the first analysis module is used for processing the normalized precipitation prediction data and the normalized ground condition data based on the first sub-model to generate first processed data; and the second analysis module is used for analyzing the normalized real-time precipitation data and the first processed data based on the second sub-model to generate precipitation correction values.
Preferably, the first analysis module is specifically configured to: acquiring preset processing times; performing convolution down-sampling operation on the normalized rainfall prediction data and the normalized ground condition data based on the first sub-model and the preset processing times to obtain down-sampled data; performing deconvolution operation on the down-sampled data based on the second sub-model and the preset processing times to obtain deconvoluted data; and executing normalization operation corresponding to the normalized precipitation prediction data and the normalized ground condition data on the deconvolution data to obtain first processed data.
Preferably, the second analysis module is specifically configured to: performing dimensionality reduction on the initial precipitation prediction data to obtain dimensionality reduced data, wherein the dimensionality reduced data is one-dimensional data; generating input data based on the data after the dimensionality reduction, the real-time precipitation data after the normalization and the data after the first processing; and analyzing the input data based on the second submodel to generate a precipitation correction value.
In another aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to implement the method provided by the embodiment of the present invention.
Through the technical scheme provided by the invention, the invention at least has the following technical effects:
on one hand, when the precipitation is predicted, firstly, the public precipitation prediction data is optimized through the target ground condition data of the target position so as to improve the accuracy of the precipitation prediction data and further improve the accuracy of the subsequent precipitation correction;
on the other hand, the generation countermeasure network based on the deep neural network model is adopted to intelligently analyze and judge the correction process of the precipitation data, so that the correction accuracy of the precipitation data is effectively improved, and the actual requirements of technicians are met.
Additional features and advantages of embodiments of the invention will be set forth in the detailed description which follows.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the embodiments of the invention without limiting the embodiments of the invention. In the drawings:
FIG. 1 is a flowchart illustrating a specific implementation of a method for correcting precipitation according to an embodiment of the present invention;
fig. 2 is a flowchart illustrating a specific implementation of acquiring preprocessed data in the method for correcting precipitation according to the embodiment of the present invention;
FIG. 3 is a flowchart illustrating an implementation of a precipitation-generative confrontation model in the method for correcting precipitation according to an embodiment of the present invention;
FIG. 4 is a flowchart illustrating a specific implementation of obtaining a trained model in a method for correcting precipitation according to an embodiment of the present invention;
fig. 5 is a schematic diagram illustrating correction of uncorrected precipitation data based on a precipitation-generated countermeasure model in the precipitation correction method according to the embodiment of the present invention;
fig. 6 is a schematic structural diagram of a precipitation correction device according to an embodiment of the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating embodiments of the invention, are given by way of illustration and explanation only, not limitation.
The terms "system" and "network" in embodiments of the present invention may be used interchangeably. The "plurality" means two or more, and in view of this, the "plurality" may also be understood as "at least two" in the embodiments of the present invention. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" generally indicates that the preceding and following related objects are in an "or" relationship, unless otherwise specified. In addition, it should be understood that the terms first, second, etc. in the description of the embodiments of the invention are used for distinguishing between the descriptions and are not intended to indicate or imply relative importance or order to be construed.
Referring to fig. 1, an embodiment of the present invention provides a method for correcting precipitation, where the method includes:
s10) acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data;
s20) preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data;
s30) acquiring a precipitation-generation type confrontation model;
s40) analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate precipitation correction values;
s50) performing precipitation correction on the initial precipitation prediction data based on the precipitation correction value, generating a corrected precipitation.
In a possible embodiment, since the existing weather forecast data has a certain accuracy, the existing precipitation forecast data may be obtained as the initial precipitation forecast data, for example, the EC forecast data issued by the european mid-term weather forecast center may be obtained, and the real-time precipitation data may be obtained, for example, the real-time observation data may be obtained from the data sources such as the satellite, the radar, and the station as the real-time precipitation data, and in order to realize the accurate forecast of the precipitation at the destination, since the actual precipitation at one location has a correlation with the terrain distribution thereof, the target terrain data may need to be further obtained, so as to improve the accuracy of the subsequent precipitation forecast.
In the embodiment of the invention, the prediction data is further optimized according to the target ground condition data of the position to be predicted on the basis of the existing precipitation prediction method, so that the prediction accuracy of the precipitation data is further improved.
After the data are acquired, the data are all original data and cannot be directly used, so that the data are preprocessed to obtain preprocessed data. For example, referring to fig. 2, in an embodiment of the present invention, the preprocessing the initial precipitation prediction data, the real-time precipitation data, and the ground condition data to obtain preprocessed data includes:
s21) performing normalization processing on the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain normalized data, wherein the normalized data comprises the normalized precipitation prediction data, the normalized real-time precipitation data and the normalized ground condition data;
s22) obtaining at least one preset forecasting factor;
s23) screening each preset forecast factor based on a preset feature calculation algorithm to obtain at least one precipitation forecast factor;
s24) generating pre-processed data based on the normalized data and the at least one precipitation forecast factor.
In a possible embodiment, in order to implement uniform processing of the multiple data, a normalization operation may be performed on the data to process the data into uniform decimal numbers between (0,1), for example, by traversing each of the input different elements, recording a maximum value (Max) and a minimum value (Min), and then performing linear transformation on the data so that Max is 1 and Min is 0 after transformation, so that all the data are distributed between 0 and 1, thereby implementing the normalization processing on all the data, and obtaining corresponding normalized precipitation prediction data, normalized real-time precipitation data, and normalized ground condition data. At this time, at least one forecasting factor is further obtained, for example, a plurality of forecasting factors can be extracted through the physical characteristics of the atmospheric circulation corresponding to the normalized data, however, because the data is general weather forecasting data, the influence of part of the forecasting factors on precipitation correction is small, and therefore the forecasting factors with small influence can be eliminated, so that the data amount is reduced, and the subsequent calculation efficiency and calculation accuracy are improved.
For example, a random forest method can be adopted as a preset feature calculation method to screen each piece of preset forecast factor information to obtain at least one precipitation forecast factor, feature importance is calculated for each feature input into the model through the random forest method, and features with high feature importance are retained and features with low feature importance are removed to achieve the purpose of feature selection. At this time, preprocessed data can be generated according to the normalized data and the at least one precipitation forecast factor, and the preprocessed data can be analyzed through an intelligent learning model.
However, the existing general intelligent learning models have corresponding technical problems when applied to precipitation forecast, for example, a linear regression model is used for processing a linear correlation relationship, and a relationship between actual precipitation distribution and numerical weather forecast precipitation distribution is a nonlinear correlation relationship, so that accurate processing cannot be performed; some intelligent algorithms can be only used for rainfall prediction of a single station, cannot be generalized to other stations or lattice points, and cannot correct regional lattice point rainfall; for precipitation with different magnitudes, the precipitation correction methods are often different greatly, so that the existing precipitation correction method has large deviation and cannot meet the actual requirement. In order to solve the above technical problem, an embodiment of the present invention provides a precipitation-generation type confrontation model.
For example, in the embodiment of the present invention, the precipitation-generated confrontation model includes a first sub-model and a second sub-model, and referring to fig. 3, the obtaining the precipitation-generated confrontation model includes:
s31) determining first initial parameters of the first submodel and second initial parameters of the second submodel;
s32), acquiring a preset data set, wherein the preset data set comprises a training data set and a testing data set;
s33) sequentially taking the first sub-model and the second sub-model as target models;
s34) training the target model based on the preset data set to obtain a trained model, wherein the trained model comprises a trained first model and a trained second model;
s35) generating a precipitation-generating confrontation model based on the trained first model and the trained second model.
In a possible embodiment, a precipitation-generative confrontation model is first created, which includes a first sub-model and a second sub-model, for example, in an embodiment of the present invention, the first sub-model employs a generative model, in particular, a U-net model, the second sub-model is a discriminative model, in particular, a deep neural network model. Before the two models are trained, initial parameters of the generated model and the discriminant model are determined, for example, initial values of parameters to be fitted in the generated model and the discriminant model may be randomly assigned, that is, the initial parameters are random values, then a preset data set is obtained, the preset data set includes a training data set and a testing data set, and at this time, the generated model and the discriminant model are trained according to the preset data set, in the embodiment of the present invention, the two models are trained by using a gradient descent algorithm.
Referring to fig. 4, in the embodiment of the present invention, the training the target model based on the preset data set to obtain a trained model includes:
s341) inputting the training data set into the target model to obtain a corresponding output value;
s342) determining loss assessment information for the target model based on the training data set, the output values, and the test data set;
s343) optimizing initial parameters of the target model based on the loss evaluation information to obtain optimized parameters;
s344) generating a first training model based on the optimized parameters and the target model;
s345) judging whether the first training model meets the prediction requirement or not based on a preset prediction evaluation index;
s3461), if yes, taking the first training model as a trained model;
s3462) if not, taking the first training model as a new target model, and jumping to the step S341).
Firstly, inputting a training data set into a target model to be trained, and obtaining a corresponding output value, for example, the output value is a predicted value y _ pred, at this time, a true value y _ real corresponding to the predicted value y _ pred is obtained from a test data set, and loss evaluation information of the target model can be determined according to training data in the training data set, the predicted value y _ pred and the true value y _ real. For example, in the process of determining the loss evaluation information of the target model, the cross entropy function L is firstly obtainedbceAnd an average absolute error function MAE, e.g. in an embodiment of the invention, a cross entropy function LbceCharacterized in that: the mean absolute error function MAE is characterized by:
and generating loss evaluation information corresponding to the submodel, specifically, generating a first loss evaluation message for the first submodel based on the cross entropy function, the mean absolute error function, the training data set, the output value and the test data set under the condition that the target model is the first submodelFor example, the first loss evaluation information is characterized by: loss G ═ Lbce(D(x,ypred),0)+MAE(yreal,ypred) Where x is an input value, for example, in an actual prediction process, x may be an initial precipitation value in an EC value weather forecast, y _ real is an actual precipitation distribution, y _ pred ═ G (x, z) represents the corrected precipitation, and z represents an air pressure forecast factor other than precipitation elements; generating second loss assessment information for a second sub-model based on the cross-entropy function, the training data set, the output values and the test data set, e.g., characterized by a loss D ═ L, if the target model is the second sub-modelbce(D(x,yreal),1)+Lbce(D(x,ypred) 0); and using the first loss evaluation information or the second loss evaluation information as the loss evaluation information of the target model.
After determining the loss evaluation information of the target model, the initial parameters of the target model may be optimized according to the loss evaluation information, for example, the gradient of the loss function may be calculated according to the loss evaluation information, the initial parameters of the target model may be optimized according to the gradient direction that the loss function is decreased, the optimized parameters may be obtained, the optimized target model, that is, the first training model, may be determined according to the optimized parameters, at this time, it is determined whether the first training model needs to be trained and optimized further, for example, in the embodiment of the present invention, the training effect of the first training model may be determined based on the preset evaluation indexes, specifically, first, the preset evaluation indexes such as the weather accuracy, the preset TS score, and the like are obtained, then the training data set is input into the first training model, and the weather accuracy prediction value of the precipitation of the first training model is predicted according to the test data set, Comparing the TS scores and the like with a preset evaluation index, when all the indexes of the first training model meet the preset evaluation index, determining that the first training model meets the prediction requirement, at this time, the first training model may be used as a trained model, if the evaluation value of the first training model does not meet the preset evaluation index, the first training model is used as a new target model, and step S341) is skipped to repeat the model training until the obtained first training model meets the preset evaluation index, at this time, the trained first model and the trained second model that minimize the loss of the test set are obtained, and a precipitation-generated countermeasure model can be generated according to the trained first model and the trained second model, please refer to fig. 5, which is a schematic diagram of correcting the uncorrected precipitation data based on the precipitation-generated countermeasure model provided by the embodiment of the present invention, firstly, inputting unregistered precipitation data and terrain parameters into a generator based on a u-net model, performing first correction treatment, then inputting the generated first corrected data and real precipitation data into a discriminator based on a deep neural network to correct the precipitation data, and generating a correction score for the precipitation data.
After the precipitation-generating type confrontation model is generated, in the process of correcting or correcting precipitation data, the precipitation-generating type confrontation model can be called and the preprocessed data can be analyzed to generate precipitation correction values. In an embodiment of the present invention, the analyzing the preprocessed data based on the precipitation-generated confrontation model to generate a precipitation correction value includes: processing the normalized precipitation prediction data and the normalized ground condition data based on the first submodel to generate first processed data; and analyzing the normalized real-time precipitation data and the first processed data based on the second submodel to generate precipitation correction values.
In a possible embodiment, the normalized precipitation prediction data and the normalized geological data are processed based on the first sub-model, and first processed data are generated, specifically, a preset processing time is obtained, for example, the preset processing time is determined by a technician according to experience or actual conditions, then a convolution down-sampling operation is performed on the normalized precipitation prediction data and the normalized geological data for a corresponding time based on the first sub-model and the preset processing time, so as to obtain down-sampled data, the characteristic latitude of the original input data can be effectively increased through the convolution down-sampling operation (the characteristic dimension is multiplied by 2 on the basis of the previous data every time the down-sampling operation is performed), and then the deconvolution operation is performed on the down-sampled data for the same time based on the second sub-model and the preset processing time, and obtaining the deconvoluted data, wherein each time the deconvolution operation is executed, the horizontal resolution of the characteristic dimension of the previous data can be multiplied by 2, and meanwhile, the number of the data is halved, so that the compression effect on the data is realized.
And finally, performing normalization operation corresponding to the normalized rainfall prediction data and the normalized ground condition data on the deconvolution data to obtain first processed data, for example, cutting the characteristics of a compression path corresponding to the data in the convolution and down-sampling operation process to be consistent with the characteristics of an expansion path corresponding to the deconvolution operation process to realize normalization operation on the data, wherein the size of the finally obtained first processed data is the same as the latitude and longitude size of the originally input rainfall data.
In the embodiment of the invention, the input preprocessed data is processed by adopting the first submodel based on the u-net model, so that the correction capability of the original precipitation data can be effectively improved, and in order to realize the antagonistic learning effect, the precipitation data needs to be further corrected according to the second submodel based on the deep neural network.
In an embodiment of the present invention, the analyzing the normalized real-time precipitation data and the first processed data based on the second submodel to generate a precipitation correction value includes: performing dimensionality reduction on the initial precipitation prediction data to obtain dimensionality reduced data, wherein the dimensionality reduced data is one-dimensional data; generating input data based on the data after the dimensionality reduction, the real-time precipitation data after the normalization and the data after the first processing; and analyzing the input data based on the second submodel to generate a precipitation correction value.
In a possible embodiment, the initial precipitation prediction data is first subjected to dimensionality reduction to obtain data after dimensionality reduction, for example, when the input initial precipitation prediction data is EC forecast data, two-dimensional EC forecast data is tiled into one-dimensional data, and data with a corresponding dimensionality of (2, n) is obtained, for example, the first-dimensional data is precipitation correction data and actual precipitation data respectively, and the second dimension n represents the number of grid points of the tiled picture. At this time, the multidimensional data are input into a second submodel, the second submodel analyzes the data after the dimensionality reduction, the real-time precipitation data after the normalization and the data after the first processing to generate corresponding output data, for example, the output data is a numerical value between 0 and 1, the larger the numerical value is, the smaller the difference between the precipitation correction data and the actual precipitation distribution is, for example, the output data is a correction score for correcting the precipitation data, and the initial precipitation prediction data is corrected according to the correction score, so that an accurate corrected precipitation value can be obtained.
In the embodiment of the invention, the input accurate precipitation prediction data is corrected by adopting the antagonistic intelligent learning model based on the deep neural network, so that an accurate precipitation correction value is obtained, the accuracy of predicting the precipitation value is effectively improved, and the actual requirements of technicians are met.
The following describes a precipitation correction device according to an embodiment of the present invention with reference to the drawings.
Referring to fig. 6, based on the same inventive concept, an embodiment of the present invention provides a device for correcting precipitation, including: the data acquisition unit is used for acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data; the preprocessing unit is used for preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data; a model acquisition unit for acquiring a precipitation-generation type countermeasure model; the analysis unit is used for analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate a precipitation correction value; and the correcting unit is used for executing precipitation correction on the initial precipitation prediction data based on the precipitation correction value and generating corrected precipitation.
Preferably, the preprocessing unit includes: the normalization module is used for performing normalization processing on the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain normalized data, and the normalized data comprises normalized precipitation prediction data, normalized real-time precipitation data and normalized ground condition data; the forecasting factor acquiring module is used for acquiring at least one preset forecasting factor; the screening module is used for screening each preset forecasting factor based on a preset characteristic calculation algorithm to obtain at least one precipitation forecasting factor; a preprocessing module for generating preprocessed data based on the normalized data and the at least one precipitation forecast factor.
Preferably, the precipitation-generating countermeasure model includes a first sub-model and a second sub-model, and the model obtaining unit includes: a parameter determining module for determining a first initial parameter of the first submodel and a second initial parameter of the second submodel; the device comprises a data set acquisition module, a data processing module and a data processing module, wherein the data set acquisition module is used for acquiring a preset data set, and the preset data set comprises a training data set and a test data set; the target model determining module is used for sequentially taking the first sub-model or the second sub-model as a target model; the training module is used for training the target model based on the preset data set to obtain a trained model, and the trained model comprises a trained first model and a trained second model; and the model establishing module is used for generating a precipitation-generation type confrontation model based on the trained first model and the trained second model.
Preferably, the training the target model based on the preset data set to obtain a trained model includes: s341) inputting the training data set into the target model to obtain a corresponding output value; s342) determining loss assessment information for the target model based on the training data set, the output values, and the test data set; s343) optimizing initial parameters of the target model based on the loss evaluation information to obtain optimized parameters; s344) generating a first training model based on the optimized parameters and the target model; s345) judging whether the first training model meets the prediction requirement or not based on a preset prediction evaluation index; s3461), if yes, taking the first training model as a trained model; s3462) if not, taking the first training model as a new target model, and jumping to the step S341).
Preferably, the determining loss evaluation information of the target model based on the training data set, the output value and the test data set comprises: acquiring a cross entropy function and an average absolute error function; generating first loss assessment information for a first sub-model based on a cross entropy function, the mean absolute error function, the training data set, the output values, and the test data set, if the target model is the first sub-model; generating second loss assessment information for a second sub-model based on the cross-entropy function, the training data set, the output values, and the test data set if the target model is the second sub-model; and using the first loss evaluation information or the second loss evaluation information as the loss evaluation information of the target model.
Preferably, the analysis unit comprises: the first analysis module is used for processing the normalized precipitation prediction data and the normalized ground condition data based on the first sub-model to generate first processed data; and the second analysis module is used for analyzing the normalized real-time precipitation data and the first processed data based on the second sub-model to generate precipitation correction values.
Preferably, the first analysis module is specifically configured to: acquiring preset processing times; performing convolution down-sampling operation on the normalized rainfall prediction data and the normalized ground condition data based on the first sub-model and the preset processing times to obtain down-sampled data; performing deconvolution operation on the down-sampled data based on the second sub-model and the preset processing times to obtain deconvoluted data; and executing normalization operation corresponding to the normalized precipitation prediction data and the normalized ground condition data on the deconvolution data to obtain first processed data.
Preferably, the second analysis module is specifically configured to: performing dimensionality reduction on the initial precipitation prediction data to obtain dimensionality reduced data, wherein the dimensionality reduced data is one-dimensional data; generating input data based on the data after the dimensionality reduction, the real-time precipitation data after the normalization and the data after the first processing; and analyzing the input data based on the second submodel to generate a precipitation correction value.
Further, the embodiment of the present invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor implements the method according to the embodiment of the present invention.
Although the embodiments of the present invention have been described in detail with reference to the accompanying drawings, the embodiments of the present invention are not limited to the details of the above embodiments, and various simple modifications can be made to the technical solutions of the embodiments of the present invention within the technical idea of the embodiments of the present invention, and the simple modifications all belong to the protection scope of the embodiments of the present invention.
It should be noted that the various features described in the above embodiments may be combined in any suitable manner without departing from the scope of the invention. In order to avoid unnecessary repetition, the embodiments of the present invention do not describe every possible combination.
Those skilled in the art will understand that all or part of the steps in the method according to the above embodiments may be implemented by a program, which is stored in a storage medium and includes several instructions to enable a single chip, a chip, or a processor (processor) to execute all or part of the steps in the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In addition, any combination of various different implementation manners of the embodiments of the present invention is also possible, and the embodiments of the present invention should be considered as disclosed in the embodiments of the present invention as long as the combination does not depart from the spirit of the embodiments of the present invention.
Claims (10)
1. A method for correcting precipitation, the method comprising:
acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data;
preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data;
acquiring a precipitation-generation type confrontation model;
analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate a precipitation correction value;
and performing precipitation correction on the initial precipitation prediction data based on the precipitation correction value to generate corrected precipitation.
2. The method of modifying of claim 1, wherein said pre-processing said initial precipitation forecast data, real-time precipitation data, and terrain data to obtain pre-processed data comprises:
performing normalization processing on the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain normalized data, wherein the normalized data comprises normalized precipitation prediction data, normalized real-time precipitation data and normalized ground condition data;
acquiring at least one preset forecasting factor;
screening each preset forecast factor based on a preset characteristic calculation algorithm to obtain at least one precipitation forecast factor;
generating pre-processed data based on the normalized data and the at least one precipitation forecast factor.
3. The method of modifying of claim 1, wherein said precipitation-generative confrontation model comprises a first sub-model and a second sub-model, and said obtaining a precipitation-generative confrontation model comprises:
determining a first initial parameter of the first submodel and a second initial parameter of the second submodel;
acquiring a preset data set, wherein the preset data set comprises a training data set and a testing data set;
sequentially taking the first sub-model and the second sub-model as target models;
training the target model based on the preset data set to obtain a trained model, wherein the trained model comprises a trained first model and a trained second model;
generating a precipitation-generative confrontation model based on the trained first model and the trained second model.
4. The method of claim 3, wherein the training the target model based on the predetermined data set to obtain a trained model comprises:
s341) inputting the training data set into the target model to obtain a corresponding output value;
s342) determining loss assessment information for the target model based on the training data set, the output values, and the test data set;
s343) optimizing initial parameters of the target model based on the loss evaluation information to obtain optimized parameters;
s344) generating a first training model based on the optimized parameters and the target model;
s345) judging whether the first training model meets the prediction requirement or not based on a preset prediction evaluation index;
s3461), if yes, taking the first training model as a trained model;
s3462) if not, taking the first training model as a new target model, and jumping to the step S341).
5. The method of revising as defined in claim 4, wherein determining loss assessment information for the target model based on the training data set, the output values, and the test data set comprises:
acquiring a cross entropy function and an average absolute error function;
generating first loss assessment information for a first sub-model based on a cross entropy function, the mean absolute error function, the training data set, the output values, and the test data set, if the target model is the first sub-model;
generating second loss assessment information for a second sub-model based on the cross-entropy function, the training data set, the output values, and the test data set if the target model is the second sub-model;
and using the first loss evaluation information or the second loss evaluation information as the loss evaluation information of the target model.
6. The method of correcting of claim 3, wherein analyzing the pre-processed data based on the precipitation-generated confrontation model to generate precipitation correction values comprises:
processing the normalized precipitation prediction data and the normalized ground condition data based on the first submodel to generate first processed data;
and analyzing the normalized real-time precipitation data and the first processed data based on the second submodel to generate precipitation correction values.
7. The method of modifying of claim 6, wherein said processing said normalized precipitation prediction data and said normalized terrain data based on said first submodel to generate first processed data comprises:
acquiring preset processing times;
performing convolution down-sampling operation on the normalized rainfall prediction data and the normalized ground condition data based on the first sub-model and the preset processing times to obtain down-sampled data;
performing deconvolution operation on the down-sampled data based on the second sub-model and the preset processing times to obtain deconvoluted data;
and executing normalization operation corresponding to the normalized precipitation prediction data and the normalized ground condition data on the deconvolution data to obtain first processed data.
8. The method of modifying of claim 6, wherein said analyzing the normalized real-time precipitation data and the first processed data based on the second submodel to generate precipitation modification values comprises:
performing dimensionality reduction on the initial precipitation prediction data to obtain dimensionality reduced data, wherein the dimensionality reduced data is one-dimensional data;
generating input data based on the data after the dimensionality reduction, the real-time precipitation data after the normalization and the data after the first processing;
and analyzing the input data based on the second submodel to generate a precipitation correction value.
9. A precipitation correction apparatus, characterized in that the correction apparatus comprises:
the data acquisition unit is used for acquiring initial precipitation prediction data, real-time precipitation data and target ground condition data;
the preprocessing unit is used for preprocessing the initial precipitation prediction data, the real-time precipitation data and the ground condition data to obtain preprocessed data;
a model acquisition unit for acquiring a precipitation-generation type countermeasure model;
the analysis unit is used for analyzing the preprocessed data based on the precipitation-generation type confrontation model to generate a precipitation correction value;
and the correcting unit is used for executing precipitation correction on the initial precipitation prediction data based on the precipitation correction value and generating corrected precipitation.
10. A computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, is adapted to carry out the method of any one of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210051053.0A CN114418211A (en) | 2022-01-17 | 2022-01-17 | Method and device for correcting precipitation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210051053.0A CN114418211A (en) | 2022-01-17 | 2022-01-17 | Method and device for correcting precipitation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114418211A true CN114418211A (en) | 2022-04-29 |
Family
ID=81273165
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210051053.0A Pending CN114418211A (en) | 2022-01-17 | 2022-01-17 | Method and device for correcting precipitation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114418211A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114648181A (en) * | 2022-05-24 | 2022-06-21 | 国能大渡河大数据服务有限公司 | Rainfall forecast correction method and system based on machine learning |
CN116228046A (en) * | 2023-05-09 | 2023-06-06 | 成都信息工程大学 | Mountain area space precipitation estimation method based on satellite remote sensing and geographic data |
CN117910658A (en) * | 2024-03-15 | 2024-04-19 | 北京和利时系统工程有限公司 | Precipitation prediction method, model training and correction method and device |
-
2022
- 2022-01-17 CN CN202210051053.0A patent/CN114418211A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114648181A (en) * | 2022-05-24 | 2022-06-21 | 国能大渡河大数据服务有限公司 | Rainfall forecast correction method and system based on machine learning |
CN114648181B (en) * | 2022-05-24 | 2022-11-18 | 国能大渡河大数据服务有限公司 | Rainfall forecast correction method and system based on machine learning |
CN116228046A (en) * | 2023-05-09 | 2023-06-06 | 成都信息工程大学 | Mountain area space precipitation estimation method based on satellite remote sensing and geographic data |
CN117910658A (en) * | 2024-03-15 | 2024-04-19 | 北京和利时系统工程有限公司 | Precipitation prediction method, model training and correction method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114418211A (en) | Method and device for correcting precipitation | |
US20170261646A1 (en) | Self-correcting multi-model numerical rainfall ensemble forecasting method | |
KR102468316B1 (en) | Time series prediction method and apparatus based on past prediction data | |
CN110414738B (en) | Crop yield prediction method and system | |
CN112200362B (en) | Landslide prediction method, landslide prediction device, landslide prediction equipment and storage medium | |
CN113344956B (en) | Ground feature contour extraction and classification method based on unmanned aerial vehicle aerial photography three-dimensional modeling | |
CN111027629A (en) | Power distribution network fault outage rate prediction method and system based on improved random forest | |
CN111598460B (en) | Method, device, equipment and storage medium for monitoring heavy metal content of soil | |
CN114501530B (en) | Method and device for determining antenna parameters based on deep reinforcement learning | |
AU2019100631A4 (en) | Self-correcting multi-model numerical rainfall ensemble forecasting method | |
CN114490905A (en) | Clear sky surface net long wave radiation integrated inversion method and system | |
CN114419430A (en) | Cultivated land plot extraction method and device based on SE-U-Net +model | |
CN113408895A (en) | Ecological quality index construction method and system based on pixel scale | |
CN117116096A (en) | Airport delay prediction method and system based on multichannel traffic image and depth CNN | |
CN109636194B (en) | Multi-source cooperative detection method and system for major change of power transmission and transformation project | |
CN111582035A (en) | Method, device and equipment for identifying tree age of fruit tree and storage medium | |
CN115546157A (en) | Method, device and storage medium for evaluating radiation quality of satellite image | |
CN113240340B (en) | Soybean planting area analysis method, device, equipment and medium based on fuzzy classification | |
CN116415841B (en) | Method and device for determining information of rich areas of biodiversity | |
CN116520457B (en) | Regional set forecasting method and forecasting device for major weather activities under complex terrain | |
CN117011719B (en) | Water resource information acquisition method based on satellite image | |
CN117950088B (en) | Multi-mode-based precipitation prediction data fusion correction method | |
CN116106856B (en) | Identification model establishment method and identification method for thunderstorm strong wind and computing equipment | |
CN117725523B (en) | Thunderstorm weather early warning method and early warning system based on physical constraint | |
CN118051878B (en) | Extreme precipitation sub-season forecasting method based on multi-mode fusion improved deep learning |
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 |