CN114545528A - Meteorological numerical model element forecasting and post-correcting method and device based on machine learning - Google Patents
Meteorological numerical model element forecasting and post-correcting method and device based on machine learning Download PDFInfo
- Publication number
- CN114545528A CN114545528A CN202210222310.2A CN202210222310A CN114545528A CN 114545528 A CN114545528 A CN 114545528A CN 202210222310 A CN202210222310 A CN 202210222310A CN 114545528 A CN114545528 A CN 114545528A
- Authority
- CN
- China
- Prior art keywords
- scene
- data
- forecast
- model
- samples
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 85
- 238000010801 machine learning Methods 0.000 title description 9
- 238000012937 correction Methods 0.000 claims abstract description 95
- 238000012549 training Methods 0.000 claims abstract description 36
- 230000001932 seasonal effect Effects 0.000 claims abstract description 15
- 238000012216 screening Methods 0.000 claims abstract description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 47
- 238000003066 decision tree Methods 0.000 claims description 16
- 238000001556 precipitation Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000007405 data analysis Methods 0.000 claims description 4
- 238000007667 floating Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000012950 reanalysis Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000000638 solvent extraction Methods 0.000 claims description 4
- 238000013480 data collection Methods 0.000 claims description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 13
- 230000000694 effects Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 8
- 230000007547 defect Effects 0.000 description 6
- 238000007637 random forest analysis Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 238000011049 filling Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000013277 forecasting method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000000462 isostatic pressing Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 239000003513 alkali Substances 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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)
- Environmental & Geological Engineering (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Environmental Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Atmospheric Sciences (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to a weather forecast method and a weather forecast device, which comprise the following steps: collecting historical data: respectively screening historical pattern forecast data and truth value data from data in a historical time period; scene division: dividing various forecast scenes according to terrain height, landform and seasonal factors, and classifying all samples according to each forecast scene; model training: respectively carrying out fitting learning on the samples in each forecast scene to obtain a correction model in the scene; model reasoning: acquiring real-time meteorological pattern forecast data, determining a certain forecast scene, inputting a real-time pattern forecast result into a corresponding correction model, and acquiring a corrected result. The content related to the application can improve the accuracy of weather forecast.
Description
Technical Field
The application relates to the field of computer information processing, in particular to a meteorological numerical model element forecast post-correction technology and system based on machine learning.
Background
At present, the direct output forecast of the meteorological numerical mode often deviates from the actual situation in practical application, and effective correction of the mode forecast is an important means for improving the forecast accuracy. In recent years, various numerical forecast post-correction technologies are formed, and the following two main types are mainly formed:
(1) statistical correction methods, such as a Model Output Statistics (MOS) method, a kalman filtering method, and the like;
(2) machine learning correction methods, such as ridge regression, random forest, etc.
The method uses the value of the numerical forecast variable as a forecast factor, uses the actual meteorological element value at the corresponding moment of forecast aging as a forecast amount, and establishes a statistical relationship according to the physical quantity field output by numerical forecast and the actual meteorological element field. However, this method is less effective in correcting when the weather is changed drastically or when extreme weather occurs.
The Kalman filtering correction method directly corrects the ensemble prediction result by estimating the decreasing average deviation at the correction time. For state estimates that are statistically inaccurate, state estimates obtained by kalman filtering methods using weight averaging are typically closer to true values than other estimation methods. The correction method has the characteristics of less required computing resources, smaller historical sample amount and the like, is not influenced by a forecasting mode and frequent upgrading of an assimilation system, and is suitable for practical application of meteorological services. Since the correction method assumes that the error change is stable and continuous, the correction effect of the method is not ideal when the deviation rule is discontinuous due to the turning change of weather.
The ridge regression method is an improvement on the traditional multiple linear regression, is a regularization method which is most frequently used when regression analysis is carried out on an ill-posed problem, is suitable for data research with a collinearity problem, and a ridge regression model learns the distribution rule of pattern prediction errors through linear combination of features, so that the fluctuation range of estimation parameters is reduced and the estimation parameters are more stable. However, there is also a disadvantage that part of the information is lost when learning the features, thereby reducing the accuracy of the correction model.
The random forest method is to create a forest in a random mode, and each number in the forest is a small decision tree. The random forest can be used for classification and regression. When creating each decision tree, there are mainly two steps: 1) and randomly sampling the sample data with the sample data put back to form a new sample set, wherein the size of the new sample set is the same as that of the original sample set. Assuming that there are N input sample data, the number of sampled samples is also N, that is, there may be repeated samples in the obtained sample set; 2) random drawn features without putting back. If there are a total of m features, n features without repetition are randomly extracted from the m features. An error correction model for forecasting a training mode by a random forest method is one of machine learning correction methods. It should be noted that if there is noise in the training data for some classification/regression problems, the data set in the random forest will appear to be overfit. In addition, the random forest algorithm is more complex than the decision tree algorithm, the calculation cost is higher, and more time is relatively needed for training the model.
Disclosure of Invention
At present, a meteorological numerical pattern element forecasting and post-correction technology based on machine learning mainly extracts element variable values from a numerical pattern according to longitude and latitude information of an observation station to serve as characteristics, matches observation station data truth values, each group of characteristics and corresponding truth values form a sample, and the samples are input into a machine learning frame such as XGboost, LightGBM and the like to be fitted to obtain a numerical pattern element forecasting and post-correction model, wherein the model training mode mainly comprises the following two modes:
1. samples from different observers were input indiscriminately into the trainer to obtain 1 post-correction model.
2. Respectively training and correcting models for samples from different observation stations, namely training 1 post-correcting model for 1 observation station.
Since the correction model after numerical model element prediction is essentially the learning of the distribution rule of the model prediction errors, the two methods have the defect of poor model correction effect.
For the method 1, as the mode error distributions of different observation stations have larger differences, all samples are input into a trainer without distinguishing for fitting learning, so that the correction effect of the model is reduced;
for the method 2, although the samples from different observation stations are distinguished for model training, and the model correction effect can be improved to a certain extent, many models are generated when the number of observation stations is extremely large, for example, 2000 observation stations will train 2000 correction models, which brings certain difficulties for service deployment and later maintenance. In addition, the model training is carried out by distinguishing each observation station, so that the number of samples which can be obtained by each observation station during model training is also obviously reduced, and the correction effect of the model is reduced.
In view of the above problems, the present application provides a method for clustering similar samples using a forecast scenario to solve the above problems in the prior art. Based on the clustering result, a correction model is trained for each class, the characteristics of each class can be more prominent aiming at the defects of the method 1, and the number of finally generated models can be reduced and the number of samples participating in training the models can be correspondingly increased aiming at the defects of the method 2. The method can improve the correction effect and reduce the number of models.
Furthermore, the samples are clustered according to the forecast scenes, and different forecast scenes are corrected through machine learning models such as LightGBM, and the like, so that the correction effect is improved.
The application provides a weather forecast method and a weather forecast device, which are used for solving the problems in the related technology, and the technical scheme is as follows:
in a first aspect, the present application provides a weather forecasting method, including:
collecting historical data: respectively screening historical pattern forecast data and truth value data from data in a historical time period;
scene division: dividing various forecast scenes according to terrain height, landform and seasonal factors, and classifying all samples according to each forecast scene;
model training: respectively carrying out fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
model reasoning: acquiring real-time meteorological pattern forecast data, determining a certain forecast scene, inputting a real-time pattern forecast result into a corresponding correction model, and acquiring a corrected result.
Preferably, the landform refers to an underlying land type.
Further, in one embodiment, the collecting the historical data further comprises:
extracting characteristic variables from the historical numerical model forecast by adopting a nearest distance method or an interpolation method;
collecting truth value data of a corresponding time period, wherein the truth value data can be historical observation data, historical live data or reanalysis data;
sample data combination: matching each 1 group of characteristic variables with a true value data at a corresponding moment to form sample data;
preferably, the collecting the historical data further includes missing value processing, which specifically includes:
if discrete data such as precipitation, discarding samples containing missing values; if the continuous variables are continuous variables such as wind speed and air temperature, interpolation filling is carried out by using data at each of front and rear 2 moments, and if the data at the front and rear two moments are not available, the sample is discarded;
time processing: and (3) respectively performing sin treatment and cos treatment on months and hours to enable the result to be between-1 and 1, and having periodicity.
Preferably, the characteristic variables refer to 1 group of meteorological elements and geographic information; the meteorological elements refer to: 2m air temperature, 10 m wind speed, ground air pressure, 2m relative humidity, precipitation, visibility, air temperature of an equal pressure surface, wind speed, potential height, relative humidity and the like extracted from the mode; the geographic information refers to: longitude, latitude, altitude, land type, ground albedo, etc.
Preferably, the isostatic pressing surface comprises 1000, 925, 850, 700, 500 hPa.
The nearest distance method is that N grid points around each station are found, then the distances between the N grid points and the station are calculated by using the grid points and the longitude and latitude of the station respectively, then the grid point with the minimum distance is selected, and the characteristic variable is obtained from the grid point; the interpolation method is to utilize N points around a station to obtain a characteristic variable by bilinear interpolation to the station;
preferably, the N is a natural number greater than 2; as a preferable mode, N is 4.
Further, in an embodiment, the method further comprises: default handling, which means discarding samples containing missing values if they are discrete data such as precipitation; if the continuous variable is a continuous variable such as wind speed and air temperature, interpolating and filling the continuous variable by using data at each of front and back 2 moments, and if the data at the front and back two moments are lack, discarding the sample;
furthermore, because the obvious characteristics of the daily change and the monthly change period exist in the meteorological data, compared with the method of directly using linear values of the hours and the months, the method can obtain the characteristics of the periodic change after respectively performing sin treatment and cos treatment on the hours and the months, thereby being more suitable for the daily change characteristics of meteorological elements. Therefore, in a preferred embodiment, the method may further include: and time processing, namely respectively performing sin and cos processing on the months and the hours to enable the result to be between-1 and 1, and the time processing has periodicity.
Further, in an embodiment, the scene division further includes:
dividing a basic scene, wherein the dividing of the basic scene comprises the basic dividing according to a terrain scene, a land type scene and/or a seasonal scene;
scene combination, wherein the scene combination refers to scene combination division based on terrain scenes, land type scenes and/or seasonal scenes to obtain combined scenes covering the whole country;
preferably, the scene may be further combined for the region of interest and the element of interest;
preferably, the area of interest is the kyojin Ji area; the concerned elements refer to air temperature, wind speed and precipitation elements.
Preferably, in an embodiment, the scene division further includes a wind speed correction scene division, specifically including:
dividing all samples into two according to a cold season (11 months-4 months of next year) and a warm season (5 months-10 months);
dividing the sample into a low land and a non-low land according to the altitude; subdividing the low earth into a first low earth and a second low earth;
the non-lowland comprises a mountain land and a highland; the first kind of lowland comprises towns and forest lands; the second type of low land comprises dry land, wet land, bare land and land water area;
preferably, in an embodiment, the scene division further includes an air temperature correction scene division, specifically including:
dividing all samples into two according to a cold season (11 months-4 months of next year) and a warm season (5 months-10 months);
dividing the sample into a low land and a non-low land according to the altitude; the low land is subdivided into 3 categories of a third category of low land, a fourth category of low land and a land water area;
the non-lowland comprises a mountain land and a highland; the third type of lowland comprises towns and bare lands; the fourth type of low lands comprise woodland, dry land and wet land.
Preferably, in an embodiment, the scene division further includes precipitation correction scene division, specifically including:
dividing all samples into two according to rainy season (7-8 months) and dry season (other months);
dividing the sample into a low land and a non-low land according to the altitude; the low earth is further subdivided into a fifth low earth and a sixth low earth.
The non-lowland comprises a mountain land and a highland; the fifth category of lowland comprises towns; the sixth type of low land comprises bare land, forest land, dry land, wet land and land water area.
Further, the model training further comprises:
correcting the predicted data to the observed data by adopting a residual error analysis method through information provided by a residual error; the residual error is the difference between an actual observed value and a regression estimated value;
the residual analysis is carried out by adopting a Boosting framework, and specifically comprises the following steps: and fitting the residual error by using the partial derivative of the cost function to the model function f trained in the previous round, and stopping when the residual error is small enough or reaches the set maximum iteration number.
The Boosting framework reduces the weight of the correct sample in the previous training round and increases the weight of the wrong sample. That is, the pair residuals are small and the error residuals are large. The Boosting framework is therefore suitable for residual analysis, i.e. post-correction of numerical predictions.
Further, the Boosting algorithm framework mainly comprises:
step S1, a decision tree algorithm, wherein the decision tree algorithm is to firstly determine the number of boxes required for each characteristic and allocate an integer to each box; dividing the range of the floating point number into a plurality of intervals, wherein the number of the intervals is equal to that of the boxes, and updating sample data belonging to the boxes into values of the boxes; finally, representing by a histogram;
step S2, a Leaf-Wise algorithm, wherein the Leaf-Wise algorithm is that one Leaf with the maximum splitting gain is found out from all current leaves every time, then splitting is carried out, and the steps are circulated;
step S3, a single-side gradient sampling algorithm, which is an algorithm that balances reduction of data amount and assurance of accuracy, based on the reduced samples, excludes most of samples with small gradients, and calculates information gain using only the remaining samples.
Preferably, the Boosting framework selects a Light Gradient Boosting Machine (LightGBM) framework training model;
correspondingly, aiming at the LightGBM framework training model, the decision tree algorithm adopts a decision tree algorithm based on Histopram;
the Leaf-Wise algorithm adopts a Leaf-Wise algorithm with depth limitation;
the single-side gradient sampling algorithm adopts a GOSS algorithm.
Further, in an embodiment, the model reasoning further comprises:
acquiring real-time service data, and inputting a corresponding scene model according to scene division to obtain a correction result, wherein the method specifically comprises the following steps:
the forecasting scene refers to the combined scene further clustered into a plurality of main scenes by comprehensively considering terrain height, land utilization types and seasonal factors and combining each meteorological element.
The land utilization type of the underlying surface is that the land surface is divided into a plurality of land types according to the vegetation coverage condition, the meaning of the land types is the same as the landform in the application, and each land type corresponds to corresponding land surface buildings, vegetation and physical parameters.
Clustering refers to dividing a data set into different classes or clusters according to a certain criterion (such as distance), so that the similarity of data objects in the same cluster is as large as possible, and the difference of data objects not in the same cluster is also as large as possible. After clustering, the data of the same class are gathered together as much as possible, and the data of different classes are separated as much as possible.
Further, the present application also provides a weather forecast apparatus, including:
the historical data collection module is used for respectively screening historical pattern forecast data and truth value data from data in a historical time period;
the scene division module is used for dividing various forecast scenes according to the terrain height, the landform and the seasonal factors and classifying all samples according to each forecast scene;
the model training module is used for respectively performing fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
and the model reasoning module is used for acquiring real-time meteorological model forecast data, determining a certain forecast scene, inputting the real-time meteorological model forecast result into the corresponding correction model and acquiring the corrected result.
Further, the present application also provides a computer device, including: processor and memory, and a computer program stored in the memory and executable in the processor, wherein execution of the program by the processor enables implementation of the steps of the method as claimed in any one of the preceding claims.
Further, the present application provides a computer-readable storage medium, which includes a computer program stored in the computer-readable storage medium, and the computer program is used for implementing the method according to any one of the above.
The method can solve the problems of large space consumption, large time overhead, unfriendly cache optimization and the like of the traditional weather prediction method based on the neural network algorithm.
Compared with the traditional method, the method has the advantages of higher speed and smaller used memory. The method is characterized in that a training sample set is divided into a plurality of forecasting scenes, a full amount of samples are not used when training is carried out on a certain forecasting scene, the used memory is reduced, and the model is converged more quickly due to the fact that the samples belonging to the certain forecasting scene have more common change characteristics. In addition, the LightGBM model adopts a histogram algorithm to convert the traversal samples into the traversal histograms, so that the time complexity is greatly reduced; a unilateral gradient algorithm is adopted in the training process to filter out samples with small gradients, so that a large amount of calculation is reduced; a growth strategy based on a Leaf-wise algorithm is adopted to construct the tree, so that a large amount of unnecessary calculation is reduced; the optimized feature parallel and data parallel methods are adopted to accelerate calculation, and a voting parallel strategy can be adopted when the data volume is very large; the cache is optimized, and the cache hit rate is increased;
meanwhile, aiming at the problem that the decision tree of the Boosting frame model algorithm is deep and overfitting is possibly generated, the data volume of each training subset is reduced through proper scene segmentation, and overfitting is prevented while the high efficiency of the whole system is ensured.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings. The drawings described below are only some embodiments of the present application, and other drawings may be derived from those drawings by those skilled in the art without inventive effort.
Fig. 1 is a flow diagram of the present application.
FIG. 2 is a schematic diagram of an inference model flow.
Detailed Description
Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The same reference numerals denote the same or similar parts in the drawings, and thus, a repetitive description thereof will be omitted.
Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that the subject matter of the present application can be practiced without one or more of the specific details, or with other methods, components, devices, steps, and so forth. In other instances, well-known methods, devices, implementations, or operations have not been shown or described in detail to avoid obscuring aspects of the application.
The block diagrams shown in the figures are functional entities only and do not necessarily correspond to physically separate entities. I.e. these functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor means and/or microcontroller means.
The flow charts shown in the drawings are merely illustrative and do not necessarily include all of the contents and operations/steps, nor do they necessarily have to be performed in the order described. For example, some operations/steps may be decomposed, and some operations/steps may be combined or partially combined, so that the actual execution sequence may be changed according to the actual situation.
It will be understood that, although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one element from another. Thus, a first component discussed below may be termed a second component without departing from the teachings of the present concepts. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
It will be appreciated by those skilled in the art that the drawings are merely schematic representations of exemplary embodiments, and that the blocks or processes shown in the drawings are not necessarily required to practice the present application and are, therefore, not intended to limit the scope of the present application.
The present application provides a solution to clustering similar samples using a forecast scenario. Based on the clustering result, a correction model is trained for each class, the characteristics of each class can be more prominent aiming at the defects of the method 1, and the number of finally generated models can be reduced and the number of samples participating in training the models can be correspondingly increased aiming at the defects of the method 2. The method can improve the correction effect and reduce the number of models.
Further, the samples are clustered according to the forecast scenes, and different forecast scenes are corrected through machine learning models such as LightGBM, so that the correction effect is improved.
The application provides a weather forecasting method and a weather forecasting device, which are used for solving the problems in the related technology, and the technical scheme is as follows:
in a first aspect, the present application provides a weather forecasting method, including:
collecting historical data: respectively screening historical mode forecast data and truth value data from data in a historical time period;
scene division: dividing various forecast scenes according to terrain height, landform and seasonal factors, and classifying all samples according to each forecast scene;
model training: respectively carrying out fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
model reasoning: acquiring real-time meteorological pattern forecast data, determining a certain forecast scene, inputting a real-time pattern forecast result into a corresponding correction model, and acquiring a corrected result.
Preferably, the landform refers to an underlying land type.
Further, in one embodiment, the collecting the historical data further comprises:
extracting characteristic variables from the historical numerical model forecast by adopting a nearest distance method or an interpolation method;
collecting truth value data of a corresponding time period, wherein the truth value data can be historical observation data, live data or reanalysis data;
sample data combination: matching each 1 group of characteristic variables with a true value data at a corresponding moment to form sample data;
preferably, the collecting the historical data further includes missing value processing, which specifically includes:
discarding samples containing missing values if they are discrete data such as precipitation; if the continuous variables are continuous variables such as wind speed and air temperature, interpolation filling is carried out by using data at each of front and rear 2 moments, and if the data at the front and rear two moments are not available, the sample is discarded;
time processing: and (3) respectively performing sin treatment and cos treatment on months and hours to enable the result to be between-1 and 1, and having periodicity.
Preferably, the characteristic variables refer to 1 group of meteorological elements and geographic information; the meteorological elements refer to: 2m air temperature, 10 m wind speed, ground air pressure, 2m relative humidity, precipitation, visibility, air temperature of an equal pressure surface, wind speed, potential height, relative humidity and the like extracted from the mode; the geographic information is as follows: longitude, latitude, altitude, land type, ground albedo, etc.
Preferably, the isostatic pressing surface comprises 1000, 925, 850, 700, 500 hPa.
The nearest distance method is that N grid points around each station are found, then the distances between the N grid points and the station are calculated by using the grid points and the longitude and latitude of the station respectively, then the grid point with the minimum distance is selected, and the characteristic variable is obtained from the grid point; the interpolation method is to utilize N points around a station to obtain a characteristic variable by bilinear interpolation to the station;
preferably, the N is a natural number greater than 2; as a preferable mode, N is 4.
Further, in an embodiment, the scene division further includes:
dividing a basic scene, wherein the dividing of the basic scene comprises the basic dividing according to a terrain scene, a land type scene and/or a season scene;
scene combination, wherein the scene combination refers to scene combination division based on terrain scenes, land type scenes and/or seasonal scenes to obtain combined scenes covering the whole country;
preferably, the scene may be further combined for the region of interest and the element of interest;
preferably, the area of interest is the kyojin Ji area; the concerned elements refer to air temperature, wind speed and precipitation elements;
preferably, in an embodiment, the scene division further includes a wind speed correction scene division, specifically including:
dividing all samples into two according to a cold season (11 months-4 months of next year) and a warm season (5 months-10 months);
dividing the sample into 2 types of low land, (mountain land and high land) according to the altitude; the low lands are subdivided into 2 categories (cities and towns, forest lands), (dry lands, wetlands, bare lands and land water areas);
preferably, in an embodiment, the scene division further includes an air temperature correction scene division, specifically including:
dividing all samples into two according to a cold season (11 months-4 months of next year) and a warm season (5 months-10 months);
dividing the sample into 2 types of low land, (mountain land and high land) according to the altitude; the low land is subdivided into 3 categories (town, bare land), (forest land, dry land, wet land) and land water area;
preferably, in an embodiment, the scene division further includes precipitation correction scene division, specifically including:
dividing all samples into two according to rainy season (7-8 months) and dry season (other months);
dividing the sample into 2 types of low land, (mountain land and high land) according to the altitude; the low lands are further divided into 2 categories of cities and towns, (bare lands, forest lands, dry lands, wetlands and land water areas).
Further, in an embodiment, the model training further comprises:
the Boosting algorithm framework mainly comprises:
a decision tree algorithm, wherein the decision tree algorithm is to determine how many boxes are needed for each characteristic and allocate an integer to each box; dividing the range of the floating point number into a plurality of intervals, wherein the number of the intervals is equal to that of the boxes, and updating sample data belonging to the boxes into values of the boxes; finally, representing by a histogram;
and (3) a Leaf-Wise algorithm, wherein the Leaf-Wise algorithm is to find one Leaf with the maximum splitting gain from all current leaves at a time, split the Leaf and loop.
The single-side gradient sampling algorithm is an algorithm which is balanced in the aspects of reducing data volume and ensuring precision, wherein the single-side gradient sampling algorithm is an algorithm which excludes most samples with small gradients from the perspective of reducing samples and only calculates information gain by using the rest samples.
Preferably, the Boosting framework model selects a Light Gradient Boosting Machine (LightGBM) framework training model;
the decision tree algorithm adopts a decision tree algorithm based on Histogram;
the Leaf-Wise algorithm adopts a Leaf-Wise algorithm with depth limitation;
the single-side gradient sampling algorithm adopts a GOSS algorithm.
Further, in an embodiment, the model reasoning further comprises:
acquiring real-time service data, and inputting a corresponding scene model according to scene division to obtain a correction result, wherein the method specifically comprises the following steps:
in a preferred embodiment, to further illustrate the method adopted in the present application, the kyojin Ji area is taken as an example to further illustrate the implementation manner of the present application, which specifically includes:
1. taking the correction of wind speed in Jingjin Ji area as an example, all samples are firstly classified into 2 types according to the cold season (11 months-4 months in the next year) and the warm season (5 months-10 months); then, clustering the samples into 2 types according to the altitude into low lands, (mountain lands and high lands); the low land is subdivided into 2 categories (town, forest land), (dry land, wetland, bare land and land water area), and a correction model is trained for each scene.
2. Taking the air temperature correction in Jingjin Ji area as an example, firstly, dividing all samples into 2 types according to the cold season (11 months-4 months in the next year) and the warm season (5 months-10 months); then, clustering the samples into 2 types according to the altitude into lowland, (mountain land and highland); the low land is subdivided into 3 classes (town, bare land), (forest land, dry land, wet land) and land water area, and a correction model is trained according to each scene.
3. Taking the precipitation correction in Jingjin Ji area as an example, firstly, dividing all samples into 2 types according to the rainy season in North China (7-8 months) and the dry season in North China (other months); then, clustering the samples into 2 types according to the altitude into low lands, (mountain lands and high lands); the low lands are subdivided into 2 categories of cities and towns, (bare land, forest land, dry land, wetland and land water), and a correction model is trained according to each scene.
4. And performing similar clustering on the forecast scene based on terrain height, underlying surface land type and seasonal factors, and training an adaptive post-correction model based on the clustered sample classification.
Taking wind speed correction for 24 hours in a cold season in Jingjin Ji area as an example, comparing the wind speed before correction, the corrected wind speed without clustering and the accuracy rate of the corrected wind speed for clustering (the error threshold is 2m/s), and finding that the accuracy rate of correction after clustering can be improved by 2-10 percent compared with the accuracy rate of correction without clustering.
Taking temperature correction in cold seasons in Jingjin Ji area as an example, the accuracy rates (the error threshold is 2 ℃) of the temperature before correction, the correction temperature without clustering and the correction temperature with clustering are compared, and the accuracy rate of correction after clustering can be improved by 2-5% compared with the correction without clustering.
Taking the rainfall correction in the Jingjin Ji area in cold seasons as an example, the weather accuracy before correction, weather accuracy without clustering correction and weather accuracy with clustering correction are compared, and the weather accuracy with clustering correction after clustering correction can be improved by 0.2-2 percent compared with the weather accuracy without clustering correction.
The specific operation of the device of the present application is described in detail below with reference to fig. 1-2, it being noted that the embodiment shown in fig. 1-2 is only an exemplary embodiment of the present application and does not represent a specific restriction on the scope of protection of the present application.
Fig. 1 shows a specific working scenario of the present application. The method specifically comprises the following steps:
1) historical data is collected. Including historical pattern forecast data and historical observation data (truth labels);
2) and (5) dividing the scene. Dividing the samples into a plurality of forecast scenes according to the terrain height, the landform (underlying surface land type) and seasonal factors, and classifying all the samples according to each forecast scene;
3) and (5) training a model. Respectively carrying out fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
4) and (4) model reasoning. Acquiring real-time meteorological pattern forecast data, determining a certain forecast scene, inputting a real-time pattern forecast result into a corresponding correction model, and acquiring a corrected result.
Further, the specific operation steps comprise:
step S1: collecting historical data
Extracting characteristic variables (generally 1 group of meteorological elements) from the historical numerical model forecast, wherein the extraction method can be that the characteristic variables are obtained according to lattice points closest to a site or by interpolating the characteristic variables to the site;
true value data is collected for a corresponding time period, typically observation data, live data, or reanalysis data. Matching each 1 group of characteristic variables with truth value data at a corresponding moment, and calling the truth value data as a sample;
and (3) processing a defect value: discarding samples containing missing values if they are discrete data such as precipitation; if the continuous variables are continuous variables such as wind speed and air temperature, interpolation filling is carried out by using data at each of front and rear 2 moments, and if the data at the front and rear two moments are not available, the sample is discarded;
time processing: and (3) respectively performing sin treatment and cos treatment on months and hours to enable the result to be between-1 and 1, and having periodicity.
Step S2: scene division, wherein, further include:
step S2.1: 4, the terrain scene division specifically comprises:
the terrain of China is generally in a step-shaped distribution form with high west and low east, the altitude of the Qinghai-Tibet plateau of the west is basically more than 4000 meters, and the altitude of the North-China plain of the east is mainly less than 200 meters. In order to highlight the main characteristics of terrain height variation, the Chinese terrains are divided into the following 4 types: low ground (below 200 m in altitude), mountain land (200-800 m in altitude), high ground (800-2500 m in altitude), and plateau (Qinghai-Tibet) with an altitude above 2500 m (as shown in Table 1).
TABLE 1 terrain type partitioning
| Serial number | Altitude range | Type of terrain |
| 1 | <200 m | Low ground |
| 2 | 200-800 m | Mountain land |
| 3 | 800-2500 m | Highland of high land |
| 4 | >2500 m | (Qinghai-Tibet) plateau |
Step S2.2: 6 soil type scene
The land utilization types of China are mainly divided into 24 types of woodland, shrub forest land and the like, and the 24 types are combined into the following 6 types (table 2) by considering the material composition, building height and physical properties of different underlying surface utilization types: a) woodland (including 4 woodland, shrub woodland, sparse woodland and other woodland), b) dry land (including rural residential sites, plain dry land, high coverage grassland, medium coverage grassland and low coverage grassland), c) land water area (including 3 river channels, lakes, reservoirs and pools), d) wetland (including 4 plain paddy field, shoal, beach land and marshland land), e) town (including 2 town and traffic construction land) and f) bare land (including 6 sand land, gobi, saline-alkali land, glacier and permanent snow, bare land and gravel land).
TABLE 2 land use types and merged type partitioning
Step S2.3: 4 seasonal scenes
From the view of temperature change, the product is divided into a cold season (11 months-4 months of next year) and a warm season (5 months-10 months); from the viewpoint of precipitation, the method is divided into dry seasons and rainy seasons.
Step S2.4: 96 combination scenarios
Based on the scene division combination in steps S2.1, S2.2, and S2.3, a nationwide combined scene 4 × 6 × 4 — 96 can be finally obtained. The scene may be further combined for the region of interest and the element of interest.
Step S2.5: wind speed correction scene division
Taking the correction of wind speed in Jingjin Ji area as an example, firstly dividing all samples into two according to the cold season (11 months-4 months in the next year) and the warm season (5 months-10 months); then dividing the sample into 2 types of low land, (mountain land and high land) according to the altitude; the low lands are further subdivided into 2 categories (towns, woodlands), (dry lands, wetlands, bare lands and land water areas).
Step S2.6: air temperature correction scene partitioning
Taking the air temperature correction in Jingjin Ji area as an example, firstly dividing all samples into two according to the cold season (11 months-4 months in the next year) and the warm season (5 months-10 months); then dividing the sample into 2 types of low land, (mountain land and high land) according to the altitude; the low land is further divided into 3 categories (town, bare land), (forest land, dry land, wet land) and land water area.
Step S2.7: precipitation correction scene division
Taking the precipitation correction in Jingjin Ji area as an example, firstly dividing all samples into two according to the North China rainy season (7-8 months) and the North China dry season (other months); then dividing the sample into 2 types of low land, (mountain land and high land) according to the altitude; the low lands are further subdivided into 2 categories of towns, (bare land, woodland, dry land, wetland, land water).
Step S3: model training, wherein the model training refers to training a model by using a Light Gradient Boosting Machine (Light weight Gradient Boosting learning Machine) framework, and specifically includes:
step S3.1: the decision tree algorithm of Histopram comprises the following steps:
firstly, determining how many boxes (bins) are needed for each feature and allocating an integer to each box; dividing the range of the floating point number into a plurality of intervals, wherein the number of the intervals is equal to that of the boxes, and updating sample data belonging to the boxes into values of the boxes; finally, it is represented by histograms (# bins).
Step S3.2: a Leaf-wise algorithm with depth limitation, comprising:
LightGBM adopts a Leaf-wise growing strategy, which finds one Leaf with the largest splitting gain from all the current leaves at a time, then splits, and so on. Thus compared to Level-wise, a Leaf-wise has the following advantages: under the condition of the same splitting times, the Leaf-wise can reduce more errors and obtain better precision; the disadvantages of Leaf-wise are: a deeper decision tree may be grown resulting in an overfitting. LightGBM therefore adds a maximum depth limit above the Leaf-wise, preventing overfitting while ensuring high efficiency.
Step S3.3: a single-sided gradient sampling algorithm, comprising:
the GOSS algorithm is an algorithm that balances reduction of data amount and assurance of accuracy by excluding most of samples with small gradients and calculating information gain using only the remaining samples from the viewpoint of sample reduction.
Step S4: the model reasoning specifically comprises the following steps:
and acquiring real-time service data, and inputting a corresponding scene model according to scene division to obtain a correction result. Preferably, fig. 2 shows a specific implementation manner of the model inference process, which includes:
after acquiring real-time forecast data, extracting the features of all grid points (sites) to form a feature sample set; then, classifying and judging all the characteristic samples according to three factors of terrain, surface type and season: scene 1, scene 2, … … and scene N, and then inputting a corresponding correction model for the feature sample of each scene to obtain a correction result set.
For example, a feature sample belonging to scene 1 is input to the correction model 1 to obtain a correction result set 1, a feature sample belonging to scene 2 is input to the correction model 2 to obtain a correction result set 2, … …, and a feature sample belonging to scene N is input to the correction model N to obtain a correction result set N.
Preferably, other frameworks that can also implement the Boosting method can be adopted to implement the method of the present application instead of the LightGBM framework.
Preferably, alternative 1, the training data set is partitioned into a plurality of forecast scenarios according to the above-mentioned scenario classification method; and then, respectively training a correction model for a certain forecast scene by adopting a Catboost framework (as well as a LightGBM framework, which is an implementation of a Boosting method).
Preferably, there is another alternative 2, according to the above-mentioned scene classification method, the training data set is divided into a plurality of forecast scenes; then, an XGboost framework (as well as a LightGBM framework, which is an implementation of a Boosting method) is adopted to train and correct a model respectively for a certain forecast scene.
Further, the present application also provides a weather forecast apparatus, including:
the historical data collection module is used for respectively screening historical pattern forecast data and truth value data from data in a historical time period;
the scene division module is used for dividing various forecast scenes according to the terrain height, the landform and the seasonal factors and classifying all samples according to each forecast scene;
the model training module is used for respectively performing fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
and the model reasoning module is used for acquiring real-time meteorological model forecast data, determining a certain forecast scene, inputting the real-time meteorological model forecast result into the corresponding correction model and acquiring the corrected result.
Further, the present application also provides a computer device, including: processor and memory, and a computer program stored in the memory and executable in the processor, wherein execution of the program by the processor enables implementation of the steps of the method as claimed in any one of the preceding claims.
Further, the present application provides a computer-readable storage medium, which includes a computer program stored in the computer-readable storage medium, and the computer program is used for implementing the method according to any one of the above.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, but any modifications or equivalent variations made according to the technical spirit of the present invention are within the scope of the present invention as claimed.
Claims (13)
1. A weather forecast method, comprising:
respectively screening historical pattern forecast data and truth value data from data in a historical time period;
dividing various forecast scenes according to terrain height, landform and seasonal factors, and classifying all samples according to each forecast scene;
respectively carrying out fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
acquiring real-time meteorological pattern forecast data, determining a certain forecast scene, inputting a real-time pattern forecast result into a corresponding correction model, and acquiring a corrected result.
2. The method of claim 1, wherein the collecting historical data further comprises:
extracting characteristic variables from the historical numerical model forecast by adopting a nearest distance method or an interpolation method;
collecting truth value data of a corresponding time period, wherein the truth value data can be historical observation data, historical live data or reanalysis data;
matching each 1 group of characteristic variables with a true value data at a corresponding moment to form sample data.
3. The method of claim 2, wherein the collecting historical data further comprises an absence value processing, comprising:
discarding samples containing missing values if they are discrete data such as precipitation; if continuous variables such as wind speed and air temperature are adopted, interpolation of data at each 2 moments before and after the continuous variables is adopted, and if data at both the two moments before and after the continuous variables are lacked, the sample is discarded.
4. The method of claim 2, wherein: the nearest distance method is that N grid points around each station are found, then the distances between the N grid points and the station are calculated by using the grid points and the longitude and latitude of the station respectively, then the grid point with the minimum distance is selected, and the characteristic variable is obtained from the grid point; the interpolation method is to utilize N points around a station to obtain characteristic variables by bilinear interpolation to the station.
5. The method of claim 1, wherein the scene division further comprises:
dividing a basic scene, wherein the dividing of the basic scene comprises the basic dividing according to a terrain scene, a land type scene and/or a season scene;
and scene combination, wherein the scene combination refers to scene combination division based on terrain scenes, land type scenes and/or seasonal scenes to obtain combined scenes covering the whole country.
6. The method of claim 1, wherein the base scene partitioning further comprises:
dividing all samples into two according to the cold season and the warm season;
dividing the sample into a low land and a non-low land according to the altitude; the low earth is subdivided into a first low earth and a second low earth.
7. The method of claim 1, wherein the model training further comprises:
correcting the predicted data to the observed data by adopting a residual error analysis method through information provided by a residual error; the residual refers to the difference between the actual observed value and the regression estimated value.
8. The method according to claim 7, wherein the residual analysis is performed using a Boosting framework, specifically comprising: and fitting the residual error by using the partial derivative of the cost function to the model function f trained in the previous round, and stopping when the residual error is small enough or reaches the set maximum iteration number.
9. The method of claim 7, wherein the Boosting algorithm framework consists essentially of:
step S1, a decision tree algorithm, wherein the decision tree algorithm is to firstly determine the number of boxes required for each characteristic and allocate an integer to each box; dividing the range of the floating point number into a plurality of intervals, wherein the number of the intervals is equal to that of the boxes, and updating sample data belonging to the boxes into values of the boxes; finally, representing by a histogram;
step S2, a Leaf-Wise algorithm, wherein the Leaf-Wise algorithm is that one Leaf with the maximum splitting gain is found out from all current leaves every time, then splitting is carried out, and the steps are circulated;
step S3, a single-side gradient sampling algorithm, which is an algorithm that balances reduction of data amount and assurance of accuracy, based on the reduced samples, excludes most of samples with small gradients, and calculates information gain using only the remaining samples.
10. The method of claim 1, wherein: the model reasoning further comprises:
and acquiring real-time service data, and inputting a corresponding scene model according to scene division to obtain a correction result.
11. A weather forecast apparatus, comprising:
the historical data collection module is used for respectively screening historical pattern forecast data and truth value data from data in a historical time period;
the scene division module is used for dividing various forecast scenes according to the terrain height, the landform and the seasonal factors and classifying all samples according to each forecast scene;
the model training module is used for respectively performing fitting learning on the samples in each forecast scene to obtain a correction model in the scene;
and the model reasoning module is used for acquiring real-time meteorological model forecast data, determining a certain forecast scene, inputting the real-time meteorological model forecast result into the corresponding correction model and acquiring the corrected result.
12. A computer device, comprising: processor and memory, and a computer program stored in the memory and executable in the processor, wherein execution of the program by the processor enables implementation of the steps of the method of any one of claims 1 to 10.
13. A computer-readable storage medium, comprising a computer program stored in the computer-readable storage medium, the program being configured to implement the method of any one of claims 1 to 10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210222310.2A CN114545528B (en) | 2022-03-09 | 2022-03-09 | Machine learning-based correction method and device after meteorological numerical mode element forecast |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210222310.2A CN114545528B (en) | 2022-03-09 | 2022-03-09 | Machine learning-based correction method and device after meteorological numerical mode element forecast |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114545528A true CN114545528A (en) | 2022-05-27 |
| CN114545528B CN114545528B (en) | 2024-02-06 |
Family
ID=81663271
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210222310.2A Active CN114545528B (en) | 2022-03-09 | 2022-03-09 | Machine learning-based correction method and device after meteorological numerical mode element forecast |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114545528B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115755220A (en) * | 2022-10-21 | 2023-03-07 | 中国民用航空飞行学院 | Airport gust forecasting and correcting method based on combination of numerical simulation and deep learning |
| CN116205138A (en) * | 2023-01-16 | 2023-06-02 | 中国气象局广州热带海洋气象研究所（广东省气象科学研究所） | Wind speed forecast correction method and device |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480818A (en) * | 2017-08-09 | 2017-12-15 | 中国热带农业科学院科技信息研究所 | A kind of method that rapid evaluation human activities of vegetation covering change influences |
| US20190057171A1 (en) * | 2017-02-10 | 2019-02-21 | Guangxi Institute of Meteorlogical Disaster- Reducing Research | Novel nonlinear method for area-wide near surface air temperature precision retrieval |
| CN110824477A (en) * | 2019-09-27 | 2020-02-21 | 兰州大方电子有限责任公司 | Method for correcting rainfall on rainfall surface by terrain |
| CN110824586A (en) * | 2019-10-23 | 2020-02-21 | 上海理工大学 | Rainfall prediction method based on improved decision tree algorithm |
| CN112684519A (en) * | 2020-12-30 | 2021-04-20 | 北京墨迹风云科技股份有限公司 | Weather forecasting method and device, computer equipment and storage medium |
| CN112684520A (en) * | 2020-12-30 | 2021-04-20 | 北京墨迹风云科技股份有限公司 | Weather forecast correction method and device, computer equipment and storage medium |
| CN113204061A (en) * | 2021-07-06 | 2021-08-03 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Method and device for constructing lattice point wind speed correction model |
-
2022
- 2022-03-09 CN CN202210222310.2A patent/CN114545528B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190057171A1 (en) * | 2017-02-10 | 2019-02-21 | Guangxi Institute of Meteorlogical Disaster- Reducing Research | Novel nonlinear method for area-wide near surface air temperature precision retrieval |
| CN107480818A (en) * | 2017-08-09 | 2017-12-15 | 中国热带农业科学院科技信息研究所 | A kind of method that rapid evaluation human activities of vegetation covering change influences |
| CN110824477A (en) * | 2019-09-27 | 2020-02-21 | 兰州大方电子有限责任公司 | Method for correcting rainfall on rainfall surface by terrain |
| CN110824586A (en) * | 2019-10-23 | 2020-02-21 | 上海理工大学 | Rainfall prediction method based on improved decision tree algorithm |
| CN112684519A (en) * | 2020-12-30 | 2021-04-20 | 北京墨迹风云科技股份有限公司 | Weather forecasting method and device, computer equipment and storage medium |
| CN112684520A (en) * | 2020-12-30 | 2021-04-20 | 北京墨迹风云科技股份有限公司 | Weather forecast correction method and device, computer equipment and storage medium |
| CN113204061A (en) * | 2021-07-06 | 2021-08-03 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Method and device for constructing lattice point wind speed correction model |
Non-Patent Citations (2)
| Title |
|---|
| 王咏薇: "土地利用变化对城市气象环境影响的数值研究", 《南京大学学报（自然科学）》, vol. 42, no. 6, pages 562 - 581 * |
| 郭涛;: "基于人工智能技术预测正逐渐成为现实", 互联网天地, no. 01 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115755220A (en) * | 2022-10-21 | 2023-03-07 | 中国民用航空飞行学院 | Airport gust forecasting and correcting method based on combination of numerical simulation and deep learning |
| CN116205138A (en) * | 2023-01-16 | 2023-06-02 | 中国气象局广州热带海洋气象研究所（广东省气象科学研究所） | Wind speed forecast correction method and device |
| CN116205138B (en) * | 2023-01-16 | 2023-11-03 | 中国气象局广州热带海洋气象研究所（广东省气象科学研究所） | Wind speed forecast correction method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114545528B (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shorabeh et al. | A decision model based on decision tree and particle swarm optimization algorithms to identify optimal locations for solar power plants construction in Iran | |
| Ramesh et al. | Analysis of crop yield prediction using data mining techniques | |
| CN112800634B (en) | Rainfall estimation method and system coupling dry-wet state identification and multi-source information fusion | |
| CN114545528B (en) | Machine learning-based correction method and device after meteorological numerical mode element forecast | |
| CN105184423B (en) | A kind of wind power plant cluster wind speed forecasting method | |
| CN112465332A (en) | Method for evaluating stability of ecological geological environment of urban artificial wetland park | |
| CN111879915B (en) | High-resolution monthly soil salinity monitoring method and system for coastal wetland | |
| Malede et al. | Analysis of land use/land cover change trends over Birr River Watershed, Abbay Basin, Ethiopia | |
| Verant et al. | Sensitivity of the continental hydrological cycle to the spatial resolution over the Iberian Peninsula | |
| Padulano et al. | Future rainfall scenarios for the assessment of water availability in Italy | |
| CN115630567A (en) | Coastal zone soil organic carbon reserve simulation and prediction method | |
| CN109657988B (en) | Tobacco leaf quality partitioning method based on HASM and Euclidean distance algorithm | |
| CN115063437B (en) | Mangrove canopy visible light image index feature analysis method and system | |
| CN115420690A (en) | Near-surface trace gas concentration inversion model and inversion method | |
| Ohba et al. | Projected future changes in water availability and dry spells in Japan: Dynamic and thermodynamic climate impacts | |
| CN113191553A (en) | Population space distribution estimation method and system based on building scale | |
| Kalayci | Settlement sizes and agricultural production territories: a remote sensing case study for the early Bronze Age in Upper Mesopotamia | |
| Arlisa et al. | Flood vulnerability analysis using random forest method in Gresik Regency, Indonesia | |
| Ayalew | A geographic information system based soil loss estimation in Lalen watershed for soil conservation planning, highlands of Ethiopia | |
| Strachan et al. | Reference Model Supporting Documentation for CABIN Analytical Tools: Columbia Basin 2020 | |
| Bac et al. | Modeling effects of abiotic and anthropogenic factors to rice production-A case study in Sapa district, Lao Cai province, Vietnam | |
| CN117556695B (en) | Crop root soil water content simulation method based on deep learning | |
| CN117828312B (en) | Method for managing watershed hydrologic environment and related equipment | |
| Santarsiero | SPATIAL ANALYSES AND REMOTE SENSING FOR LAND COVER CHANGE DYNAMICS: ASSESSING IN A SPATIAL PLANNING | |
| Liu | Supplement of Estimating ground-level CO concentrations across China based on the national monitoring network and MOPITT: potentially overlooked CO hotspots in the Tibetan Plateau |
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 |