CN113204061B - Method and device for constructing lattice point wind speed correction model - Google Patents
Method and device for constructing lattice point wind speed correction model Download PDFInfo
- Publication number
- CN113204061B CN113204061B CN202110759432.0A CN202110759432A CN113204061B CN 113204061 B CN113204061 B CN 113204061B CN 202110759432 A CN202110759432 A CN 202110759432A CN 113204061 B CN113204061 B CN 113204061B
- Authority
- CN
- China
- Prior art keywords
- historical
- wind speed
- grid point
- time period
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
Abstract
The invention provides a method and a device for constructing a grid point wind speed correction model, wherein the method comprises the following steps: acquiring historical meteorological data and historical landform data of each grid point in a historical period; inputting historical meteorological data and historical landform data of the same historical period into a meteorological feature extraction model aiming at each grid point to obtain meteorological features of the historical period; aiming at each grid point, acquiring a grid point historical wind speed forecasting result forecasted by the grid point wind speed forecasting model in each historical time period; acquiring a grid point historical wind speed forecast result and an actual observed historical wind speed and calculating a grid point historical time interval error aiming at each historical time interval; and aiming at each historical time period, inputting the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period, and training the grid point wind speed correction training models to obtain the grid point wind speed correction model. The accuracy of wind speed value prediction can be improved.
Description
Technical Field
The invention relates to the technical field of meteorological forecasting, in particular to a method and a device for constructing a grid point wind speed correction model.
Background
The grid point wind speed is the result of calculating the longitude and latitude wind speed condition of a fixed stride according to real-time observation wind speed data and an atmospheric physical model. The grid point wind speed prediction is an important branch in meteorological prediction and has very important influence on daily life of people, industrial and agricultural production and the like. However, since the grid point wind speed prediction relates to an atmospheric physics model and numerous Meteorological factors, and the interaction relationship between each physical quantity and the geographic environment is very complex, at present, for the grid point wind speed prediction of the atmospheric physics model, for example, the grid point wind speed prediction models of National Meteorological Center (NMC), European Meteorological Center (EC), etc., due to the different Meteorological factors considered by each atmospheric physics model, the adopted algorithms are different, and there are such or similar defects, so that the accuracy of the grid point wind speed predicted by a single atmospheric physics model is low.
Disclosure of Invention
In view of the above, the present invention provides a method and an apparatus for constructing a grid point wind speed correction model to improve the accuracy of forecasting the wind speed value.
In a first aspect, an embodiment of the present invention provides a method for constructing a grid point wind speed correction model, including:
acquiring historical meteorological data and historical landform data of each grid point in a preset historical time period;
inputting historical meteorological data and historical geomorphic data of the same historical period into a meteorological feature extraction model trained in advance aiming at each grid point to obtain the meteorological features of the historical period;
aiming at each grid point, acquiring grid point historical wind speed forecasting results forecasted by grid point wind speed forecasting models with preset numbers in various historical time periods;
acquiring a grid point historical wind speed forecast result and an actual observation historical wind speed of each historical time period, and calculating a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observation historical wind speed;
and aiming at each historical time period, inputting the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period, and training the grid point wind speed correction training models to obtain the grid point wind speed correction model.
With reference to the first aspect, an embodiment of the present invention provides a first possible implementation manner of the first aspect, where the grid wind speed correction training model includes an action network submodel and a value network submodel, and for each historical period, the grid wind speed correction training model is input into a historical period meteorological features of the historical period and grid historical errors of each grid wind speed prediction model in the historical period, and is trained to obtain the grid wind speed correction model, including:
splicing historical time period meteorological features of the historical time period and grid point historical errors of each grid point wind speed prediction model in the historical time period to obtain splicing data;
inputting the splicing data into a full-connection characteristic layer in the action network submodel to obtain the training weight of the wind speed prediction model of each lattice point;
inputting the splicing data and the training weight of the historical period into a full-connection characteristic layer in the value network submodel to obtain a grid point wind speed forecasting weighting result of each grid point wind speed forecasting model on a preset historical forecasting period;
and calculating the value loss according to the grid point wind speed forecast weighting result and the grid point wind speed measured value corresponding to the historical forecast time period, updating the parameters of the full-connection characteristic layer if the value loss is greater than a preset value loss threshold, and training the next historical time period until the value loss is not greater than the value loss threshold to obtain a grid point wind speed correction model.
With reference to the first possible implementation manner of the first aspect, an embodiment of the present invention provides a second possible implementation manner of the first aspect, where the method further includes:
acquiring current meteorological data and current landform data of forecast grid points in a current time period;
splicing the current meteorological data and the current landform data, and inputting a pre-trained meteorological feature extraction model to obtain meteorological features of the current time period;
aiming at each historical time period of the forecast grid points, acquiring forecast grid point historical wind speed forecasting results of forecasting each historical time period by the grid point wind speed forecasting model with the preset number;
calculating a model error of each grid point wind speed prediction model based on the prediction grid point historical wind speed prediction result and the actual historical wind speed of each historical time period;
inputting the meteorological features and model errors in the current time period into a full-connection feature layer of the grid point wind speed correction model to obtain the prediction weight of the grid point wind speed prediction model;
acquiring grid point future wind speed forecasting results of the grid point wind speed forecasting models with the preset number respectively forecasting based on the current time period;
and aiming at each grid point wind speed prediction model, calculating the product of the grid point future wind speed prediction result of the grid point wind speed prediction model and the prediction weight, and obtaining the grid point future wind speed prediction correction result of the predicted grid point based on the products.
With reference to the second possible implementation manner of the first aspect, an embodiment of the present invention provides a third possible implementation manner of the first aspect, where the calculating a model error of the wind speed prediction model at each grid point based on the prediction result of the predicted grid point historical wind speed and the actual historical wind speed at each historical time period includes:
calculating a forecast grid point historical wind speed forecast result in a historical time period and a wind speed difference value of an actual historical wind speed aiming at each grid point wind speed forecast model;
acquiring the sum of squares of the wind speed difference;
and carrying out weighted average on the square sum of each historical time period to obtain the model error of the grid point wind speed prediction model.
With reference to the first aspect and any one possible implementation manner of the first to third possible implementation manners of the first aspect, an embodiment of the present invention provides a fourth possible implementation manner of the first aspect, where the meteorological feature extraction model includes: the meteorological feature extraction network and the global pooling network are used for inputting historical meteorological data and historical landform data of the same historical period into a meteorological feature extraction model trained in advance to obtain the meteorological features of the historical period, and the meteorological feature extraction network comprises the following steps:
inputting historical meteorological data and historical geomorphic data of the historical time period into a meteorological feature extraction network trained in advance to obtain initial historical meteorological features; and carrying out global pooling on the historical meteorological initial characteristics according to the global pooling network to obtain the meteorological characteristics in the historical time period.
With reference to the first aspect and any one possible implementation manner of the first to third possible implementation manners of the first aspect, an embodiment of the present invention provides a fifth possible implementation manner of the first aspect, where the method further includes:
carrying out bilinear interpolation processing on the historical landform data so as to enable the size of the historical landform data to be the same as that of the historical meteorological data;
and respectively carrying out standardization and regularization processing on the historical meteorological data and the historical landform data after interpolation processing.
With reference to the first aspect and any one possible implementation manner of the first to third possible implementation manners of the first aspect, an embodiment of the present invention provides a sixth possible implementation manner of the first aspect, where the historical meteorological data includes: the ground temperature data, the ground humidity data, the wind speed U component data and the wind speed V component data, the historical landform data comprises: terrain data, and solar elevation angle data.
In a second aspect, an embodiment of the present invention further provides an apparatus for constructing a grid point wind speed correction model, including:
the data acquisition module is used for acquiring historical meteorological data and historical landform data of each grid point in a preset historical time period;
the characteristic extraction module is used for inputting historical meteorological data and historical landform data of the same historical period into a meteorological characteristic extraction model trained in advance aiming at each grid point to obtain the meteorological characteristics of the historical period;
the forecasting result acquiring module is used for acquiring a grid point historical wind speed forecasting result forecasted by a grid point wind speed forecasting model with preset number in each historical time period aiming at each grid point;
the error calculation module is used for acquiring a grid point historical wind speed forecast result and an actual observation historical wind speed of each historical time period, and calculating a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observation historical wind speed;
and the wind speed correction module is used for inputting the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period into the grid point wind speed correction training model and training the grid point wind speed correction training model to obtain the grid point wind speed correction model.
In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, performs the steps of the method described above.
According to the method and the device for constructing the grid point wind speed correction model, historical meteorological data and historical geomorphic data of each grid point in a preset historical time period are obtained; inputting historical meteorological data and historical geomorphic data of the same historical period into a meteorological feature extraction model trained in advance aiming at each grid point to obtain the meteorological features of the historical period; aiming at each grid point, acquiring grid point historical wind speed forecasting results forecasted by grid point wind speed forecasting models with preset numbers in various historical time periods; acquiring a grid point historical wind speed forecast result and an actual observation historical wind speed of each historical time period, and calculating a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observation historical wind speed; and aiming at each historical time period, inputting the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period, and training the grid point wind speed correction training models to obtain the grid point wind speed correction model. Therefore, by constructing the grid point wind speed correction model and performing weight analysis on the plurality of grid point wind speed prediction models, the output results of the grid point wind speed prediction models are subjected to fusion correction, and the wind speed value prediction accuracy can be effectively improved.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a schematic flow chart illustrating a method for constructing a grid wind speed correction model according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of an apparatus for constructing a grid wind speed correction model according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a computer device 300 according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention provides a method and a device for constructing a grid point wind speed correction model, which are described by the following embodiments.
Fig. 1 shows a flow chart of a method for constructing a grid point wind speed correction model according to an embodiment of the present invention. As shown in fig. 1, the method includes:
in the embodiment of the present invention, a grid point refers to an area including a certain range, and in the area, one or more observation sites may be included, and if the grid point includes a plurality of observation sites, as an optional embodiment, the historical weather data of the grid point may be a weighted average of the historical weather data of each observation site.
In the embodiment of the invention, for each grid point, each historical time interval corresponds to one piece of historical meteorological data, and for the historical geomorphic data, a plurality of historical time intervals correspond to one piece of historical geomorphic data, or one historical time interval corresponds to one piece of historical geomorphic data. As an alternative embodiment, the historical time interval is consistent with the prediction time interval of the current grid point wind speed prediction model, for example, if the prediction time interval is three points or six points, three points or six points are also selected in the historical time interval.
In this embodiment of the present invention, as an optional embodiment, the historical meteorological data includes: the ground temperature data, the ground humidity data, the wind speed U component data and the wind speed V component data, the historical landform data comprises: terrain data, and solar elevation angle data.
In the embodiment of the invention, the historical meteorological data and the historical geomorphic data can be acquired by different meteorological data acquisition equipment, and the sizes of the meteorological data and the historical geomorphic data can be different. Thus, as an alternative embodiment, the method further comprises:
carrying out bilinear interpolation processing on the historical landform data so as to enable the size of the historical landform data to be the same as that of the historical meteorological data;
and respectively carrying out standardization and regularization processing on the historical meteorological data and the historical landform data after interpolation processing.
In the embodiment of the invention, as an optional embodiment, the dimension of the historical meteorological data is taken as a reference, and bilinear interpolation is performed on the historical geomorphic data with the dimension different from that of the historical meteorological data, so that the dimension of the historical geomorphic data after interpolation is the same as that of the historical meteorological data. For example, for a size ofH hh xW h The size of the historical landform data is converted into the size of the historical meteorological data by a bilinear interpolation methodH lh xW l 。
In the embodiment of the present invention, since the data types included in the data are different, in order to facilitate uniform processing of the various types of data, it is necessary to perform normalization and regularization processing on the various types of data, respectively. Taking the landform data as an example, the standardization and the regularization operation are respectively carried out on the landform data, the ground feature data and the solar altitude angle data.
In this embodiment of the present invention, as an optional embodiment, the method further includes:
and performing data fusion on the historical meteorological data and the historical geomorphic data of each grid point in each historical time period of the grid point to obtain a historical data set.
In the embodiment of the invention, for each grid point, historical meteorological data of ground multimodalities in the same historical time period and historical geomorphic data are fused to obtain a historical data set, and each historical time period of each grid point corresponds to one historical data set. For example, for a certain historical period of a certain point, the historical meteorological data with the size of TxHxW and the historical geomorphic data with the size of CxHxW are superposed to obtain a historical data set with the size of (T + C) xHxW, wherein T is the number of data types contained in the historical meteorological data, H is the width of the historical meteorological data, W is the length of the historical meteorological data,Cfor the number of data types contained in the historical geomorphic data,HxWis the size of the data. For example, the historical geomorphic data includes topographic data, surface feature data and solar elevation angle data, and the number of corresponding data types is 3.
102, inputting historical meteorological data and historical geomorphic data of the same historical period into a meteorological feature extraction model trained in advance aiming at each grid point to obtain the meteorological features of the historical period;
in the embodiment of the invention, for the historical data set obtained by data fusion, the historical data set is input into the meteorological feature extraction model, and for the training of the meteorological feature extraction model, reference can be specifically made to related technical documents, and the detailed description is omitted here.
In the embodiment of the invention, the historical period meteorological features comprise historical period meteorological features of various types of data. For example, for historical geomorphic data, the corresponding historical period meteorological features include: topographic data meteorological features, surface feature data meteorological features, solar altitude angle data meteorological features.
In the embodiment of the invention, historical meteorological data and historical geomorphic data are input into a meteorological feature extraction model by taking a historical time interval as a unit.
In the embodiment of the present invention, as an optional embodiment, the meteorological feature extraction model includes: the meteorological feature extraction network and the global pooling network input historical meteorological data and historical landform data of the same historical period into a meteorological feature extraction model trained in advance to obtain the meteorological features of the historical period, and the meteorological feature extraction network comprises the following steps:
inputting historical meteorological data and historical geomorphic data of the historical time period into a meteorological feature extraction network trained in advance to obtain initial historical meteorological features; and carrying out global pooling on the historical meteorological initial characteristics according to the global pooling network to obtain the meteorological characteristics in the historical time period.
In the embodiment of the invention, the meteorological feature extraction network is a deep convolution network, and the deep convolution network is utilized to extract the historical meteorological initial features and perform pooling treatment on the historical data set (historical meteorological data and historical geomorphic data).
In this embodiment of the present invention, as an optional embodiment, the global pooling operation is as follows:
the input characteristics are set as follows:X∈R CxWxH the output size is:
the calculation method is as follows:
wherein the content of the first and second substances,
M i is as followsiHistorical time interval meteorological features of the species data;
X i，m，n is as followsiClass data ofmGo to the firstnMeteorological features of the pixels of the column.
103, acquiring a grid point historical wind speed forecasting result forecasted by a preset number of grid point wind speed forecasting models in each historical time period aiming at each grid point;
in the embodiment of the present invention, the preset number may be set according to actual requirements, as an optional embodiment, if the grid wind speed prediction models are EC, NMC, and DL, respectively, and correspondingly, the preset number is 3, then for each grid point, the grid point historical wind speed prediction results of three different grid point wind speed prediction models in each historical time period are obtained. As an alternative embodiment, the grid point historical wind speed forecast result comprises: wind speed U component data, wind speed V component data and time interval data.
In the embodiment of the invention, the grid point historical wind speed forecasting result of the grid point wind speed forecasting model in the historical time period refers to a wind speed forecasting result obtained by forecasting the historical time period by the grid point wind speed forecasting model according to historical meteorological data before the historical time period.
104, acquiring a grid point historical wind speed forecast result and an actual observation historical wind speed of each historical time period, and calculating a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observation historical wind speed;
in the embodiment of the invention, the grid point historical time period errors of different grid point wind speed prediction models in the same historical time period may be the same or different. Assume a grid size ofNxMThe lattice point includesNxMThe horizontal and vertical coordinates of each observation station are respectivelyx，yThe actual observed historical wind speed of the observed station isR(x，y)The grid point historical wind speed forecast result of the observation station isP(x，y)The grid point historical time interval error adopts mean square error, and a grid point wind speed prediction model is calculated by using the following formulaGrid historical period error for that grid:
in the formula (I), the compound is shown in the specification,
Eis the grid historical period error.
And 105, inputting the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period into a grid point wind speed correction training model for each historical time period, and training the grid point wind speed correction training model to obtain the grid point wind speed correction model.
In the embodiment of the present invention, the grid wind speed correction training model includes an action network submodel and a value network submodel, as an optional embodiment, for each historical time period, inputting the historical time period meteorological features of the historical time period and the grid point historical errors of each grid point wind speed prediction model in the historical time period into the grid point wind speed correction training model, and training the grid point wind speed correction training model to obtain the grid point wind speed correction model, including:
splicing historical time period meteorological features of the historical time period and grid point historical errors of each grid point wind speed prediction model in the historical time period to obtain splicing data;
inputting the splicing data into a full-connection characteristic layer in the action network submodel to obtain the training weight of the wind speed prediction model of each lattice point;
inputting the splicing data and the training weight of the historical period into a full-connection characteristic layer in the value network submodel to obtain a grid point wind speed forecasting weighting result of each grid point wind speed forecasting model on a preset historical forecasting period;
and calculating the value loss according to the grid point wind speed forecast weighting result and the grid point wind speed measured value corresponding to the historical forecast time period, updating the parameters of the full-connection characteristic layer if the value loss is greater than a preset value loss threshold, and training the next historical time period until the value loss is not greater than the value loss threshold to obtain a grid point wind speed correction model.
In the embodiment of the invention, as an optional embodiment, the sum of the training weights of the wind speed prediction models at each grid point is 1.
In the embodiment of the invention, the grid point wind speed correction training model with the value loss not greater than the preset value loss threshold value is a grid point wind speed correction model. The grid point wind speed measured value can be extracted from the acquired historical meteorological data. If the value loss is larger than a preset value loss threshold value, updating parameters of the full-connection feature layer according to a back propagation algorithm, then selecting historical time period meteorological features of another historical time period and grid point historical errors of each grid point wind speed prediction model in the historical time period to be spliced, training the action network submodel and the value network submodel, and repeating the steps.
In the embodiment of the invention, the parameters of the fully-connected characteristic layer in the action network submodel are the same as the parameters of the fully-connected characteristic layer in the value network submodel.
In the embodiment of the invention, under the reinforcement learning framework, the output of the value network submodel is the reward value, and the reward value evaluates the output of the action network submodel. As an alternative embodiment, the reward value outputs a weighted value according to the action function, and the error between the calculated final prediction result and the actual prediction result is negative.
In the embodiment of the invention, as an optional embodiment, when the action network submodel and the value network submodel are trained, a Deep Deterministic Policy Gradient (DDPG) algorithm is used, a random Gradient Descent (SGD) optimizer is used for optimization, the learning rates of the action network submodel and the value network submodel are respectively 0.001 and 0.05, and the update weights of the action network submodel and the value network submodel are 0.9.
In this embodiment of the present invention, as an optional embodiment, the method further includes:
a11, acquiring current meteorological data and current landform data of forecast grid points in a current time period;
in the embodiment of the invention, if the forecast grid point contains a plurality of observation stations, various types of data in the current meteorological data measured by the observation stations are respectively averaged, and various types of data in the current geomorphic data measured by the observation stations are respectively averaged, so that the current meteorological data and the current geomorphic data of the forecast grid point in the current time period are obtained.
A12, splicing the current meteorological data and the current landform data, and inputting a meteorological feature extraction model trained in advance to obtain meteorological features in the current time period;
a13, acquiring forecast grid point historical wind speed forecasting results of forecasting each historical time period by the grid point wind speed forecasting model with the preset number aiming at each historical time period of the forecast grid point;
in the embodiment of the invention, each grid point wind speed prediction model corresponds to a prediction grid point historical wind speed prediction result obtained by predicting the historical time period in each historical time period.
A14, calculating a model error of a wind speed prediction model of each grid point based on a prediction result of the historical wind speed of the grid point and the actual historical wind speed at each historical time period;
in the embodiment of the present invention, as an optional embodiment, calculating a model error of a wind speed prediction model at each grid point based on a prediction result of a predicted grid point historical wind speed and an actual historical wind speed at each historical time period includes:
calculating a forecast grid point historical wind speed forecast result in a historical time period and a wind speed difference value of an actual historical wind speed aiming at each grid point wind speed forecast model;
acquiring the sum of squares of the wind speed difference;
and carrying out weighted average on the square sum of each historical time period to obtain the model error of the grid point wind speed prediction model.
In the embodiment of the invention, the model error of the grid point wind speed prediction model is calculated by the following formula:
in the formula (I), the compound is shown in the specification,
A15, inputting the meteorological features and model errors in the current time period into a full-connection feature layer of a grid point wind speed correction model to obtain the prediction weight of each grid point wind speed prediction model;
a16, obtaining grid point future wind speed forecasting results of the grid point wind speed forecasting models with the preset number respectively forecasting based on the current time period;
and A17, calculating the product of the grid point future wind speed forecast result of the grid point wind speed forecasting model and the forecasting weight aiming at each grid point wind speed forecasting model, and obtaining the grid point future wind speed forecast correction result of the forecast grid point based on each product.
In the embodiment of the invention, the sum of the prediction weights of the grid point wind speed prediction models is 1, products are summed to obtain a grid point future wind speed prediction correction result, and if the sum of the prediction weights of the grid point wind speed prediction models is other values, the products are summed and then averaged to obtain the grid point future wind speed prediction correction result.
In the embodiment of the present invention, assume thattIndividual grid wind speed predictionThe model, containing the forecast lattice points3x3Setting a wind speed prediction model of each grid point to predict based on the current time interval at each observation station, wherein the future wind speed prediction results of the corresponding grid points are respectivelyM 1 ,M 2 ,…,M t ,The corresponding prediction weights are respectivelyr 1 ,r 2 ,…,r t ,The matrix form of the grid point future wind speed forecast result corresponding to a certain grid point wind speed prediction model can be expressed as follows:
in the formula (I), the compound is shown in the specification,M i is as followsiGrid point future wind speed forecasting results corresponding to the grid point wind speed forecasting models;
In the embodiment of the invention, the grid point future wind speed forecast result can be the average of the future wind speed forecast results of all observation sites.
In the embodiment of the invention, the grid point future wind speed forecast results of different grid point wind speed forecasting models are subjected to weighted summation, so that the grid point future wind speed forecast correction result of the forecast grid point can be obtained:
in the embodiment of the invention, on the basis of a plurality of existing lattice point wind speed prediction models, a lattice point wind speed correction model containing a deep convolutional network is constructed, a training set is constructed by utilizing multi-modal meteorological data and lattice point historical errors in various historical time periods, a reinforcement learning method is utilized to train the lattice point wind speed training model on the training set, and after the training is completed, prediction weight analysis can be carried out on the plurality of lattice point wind speed prediction models, so that the output results of the existing lattice point wind speed prediction models are subjected to fusion correction.
In order to illustrate the correcting effect of the embodiment of the invention, the method of the embodiment of the invention is utilized to predict the grid point wind speed of 3 hours by 3 hours in the middle ten days of 7 and 7 months of 2020 in 2019 on the ground in south China.
Fig. 2 shows a schematic structural diagram of an apparatus for constructing a grid point wind speed correction model according to an embodiment of the present invention. As shown in fig. 2, the apparatus includes:
the data acquisition module 201 is configured to acquire historical meteorological data and historical geomorphic data of each grid point in a preset historical time period;
in the embodiment of the present invention, as an optional embodiment, the meteorological data includes: ground temperature data, ground humidity data, wind speed U component data and wind speed V component data, the landform data includes: terrain data, and solar elevation angle data.
The feature extraction module 202 is configured to, for each grid point, input historical meteorological data and historical geomorphic data of the same historical period into a meteorological feature extraction model trained in advance, and obtain historical time period meteorological features of the historical period;
in the embodiment of the invention, the historical period meteorological features comprise historical period meteorological features of various types of data.
In the embodiment of the present invention, as an optional embodiment, the meteorological feature extraction model includes: the meteorological feature extraction network and the global pooling network are used for inputting historical meteorological data and historical landform data of the same historical period into a meteorological feature extraction model trained in advance to obtain the meteorological features of the historical period, and the meteorological feature extraction network comprises the following steps:
inputting historical meteorological data and historical geomorphic data of the historical time period into a meteorological feature extraction network trained in advance to obtain initial historical meteorological features; and carrying out global pooling on the historical meteorological initial characteristics according to the global pooling network to obtain the meteorological characteristics in the historical time period.
A forecast result obtaining module 203, configured to obtain, for each grid point, a grid point historical wind speed forecast result forecasted by a preset number of grid point wind speed forecasting models in each historical time period;
an error calculation module 204, configured to obtain a grid point historical wind speed forecast result and an actual observed historical wind speed in each historical time period, and calculate a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observed historical wind speed;
in the embodiment of the invention, the mean square error is adopted for the grid point historical time interval error.
And the wind speed correction module 205 is configured to, for each historical time period, input the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period into the grid point wind speed correction training model, and train the grid point wind speed correction training model to obtain the grid point wind speed correction model.
In this embodiment of the present invention, as an optional embodiment, the wind speed correction module 205 includes:
the splicing unit (not shown in the figure) is used for splicing the historical period meteorological features of the historical period and the grid point historical errors of the grid point wind speed prediction models in the historical period to obtain splicing data;
the weight parameter acquisition unit is used for inputting the splicing data into a full-connection characteristic layer in the action network submodel to obtain the training weight of the wind speed prediction model of each lattice point;
the weighting result acquisition unit is used for inputting the splicing data and the training weight of the historical period into a full-connection characteristic layer in the value network submodel to obtain a grid point wind speed forecasting weighting result of each grid point wind speed forecasting model to the preset historical forecasting period;
and the model determining unit is used for calculating the value loss according to the grid point wind speed forecast weighting result and the grid point wind speed measured value corresponding to the historical forecast time period, updating the parameters of the full-connection characteristic layer if the value loss is greater than a preset value loss threshold value, and training the next historical time period until the value loss is not greater than the value loss threshold value to obtain a grid point wind speed correction model.
In this embodiment of the present invention, as an optional embodiment, the apparatus further includes:
a wind speed forecasting module (not shown in the figure) for acquiring current meteorological data and current landform data of the forecast grid points in the current time period;
splicing the current meteorological data and the current landform data, and inputting a pre-trained meteorological feature extraction model to obtain meteorological features of the current time period;
aiming at each historical time period of the forecast grid points, acquiring forecast grid point historical wind speed forecasting results of forecasting each historical time period by the grid point wind speed forecasting model with the preset number;
calculating a model error of each grid point wind speed prediction model based on the prediction grid point historical wind speed prediction result and the actual historical wind speed of each historical time period;
inputting the meteorological features and model errors in the current time period into a full-connection feature layer of the grid point wind speed correction model to obtain the prediction weight of the grid point wind speed prediction model;
acquiring grid point future wind speed forecasting results of the grid point wind speed forecasting models with the preset number respectively forecasting based on the current time period;
and aiming at each grid point wind speed prediction model, calculating the product of the grid point future wind speed prediction result of the grid point wind speed prediction model and the prediction weight, and obtaining the grid point future wind speed prediction correction result of the predicted grid point based on the products.
In the embodiment of the present invention, as an optional embodiment, calculating a model error of a wind speed prediction model at each grid point based on a prediction result of a predicted grid point historical wind speed and an actual historical wind speed at each historical time period includes:
calculating a forecast grid point historical wind speed forecast result in a historical time period and a wind speed difference value of an actual historical wind speed aiming at each grid point wind speed forecast model;
acquiring the sum of squares of the wind speed difference;
and carrying out weighted average on the square sum of each historical time period to obtain the model error of the grid point wind speed prediction model.
In this embodiment, as another optional embodiment, the apparatus further includes:
the consistency processing module is used for carrying out bilinear interpolation processing on the historical landform data so as to enable the size of the historical landform data to be the same as that of the historical meteorological data;
and respectively carrying out standardization and regularization processing on the historical meteorological data and the historical landform data after interpolation processing.
In the embodiment of the invention, various types of data are respectively subjected to standardization and regularization processing. Taking the landform data as an example, the standardization and the regularization operation are respectively carried out on the landform data, the ground feature data and the solar altitude angle data.
As shown in fig. 3, an embodiment of the present application provides a computer device 300 for executing the method for constructing a grid point wind speed correction model in fig. 1, the device includes a memory 301, a processor 302, and a computer program stored on the memory 301 and executable on the processor 302, wherein the processor 302 implements the steps of the method for constructing a grid point wind speed correction model when executing the computer program.
Specifically, the memory 301 and the processor 302 can be general-purpose memories and processors, and are not limited to specific ones, and the processor 302 can execute the above method for constructing the grid wind speed correction model when executing the computer program stored in the memory 301.
Corresponding to the method for constructing the grid point wind speed correction model in fig. 1, the embodiment of the present application further provides a computer readable storage medium, on which a computer program is stored, which, when being executed by a processor, performs the steps of the above method for constructing the grid point wind speed correction model.
In particular, the storage medium can be a general-purpose storage medium, such as a removable disk, a hard disk, or the like, and when executed, the computer program on the storage medium can execute the above method for constructing the grid wind speed correction model.
In the embodiments provided in the present application, it should be understood that the disclosed system and method may be implemented in other ways. The above-described system embodiments are merely illustrative, and for example, the division of the units is only one logical functional division, and there may be other divisions in actual implementation, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of systems or units through some communication interfaces, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus once an item is defined in one figure, it need not be further defined and explained in subsequent figures, and moreover, the terms "first", "second", "third", etc. are used merely to distinguish one description from another and are not to be construed as indicating or implying relative importance.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the present disclosure, which should be construed in light of the above teachings. Are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (8)
1. A method for constructing a grid point wind speed correction model is characterized by comprising the following steps:
acquiring historical meteorological data and historical landform data of each grid point in a preset historical time period;
inputting historical meteorological data and historical geomorphic data of the same historical period into a meteorological feature extraction model trained in advance aiming at each grid point to obtain the meteorological features of the historical period;
aiming at each grid point, acquiring grid point historical wind speed forecasting results forecasted by grid point wind speed forecasting models with preset numbers in each historical time period, wherein each grid point wind speed forecasting model corresponds to one grid point historical wind speed forecasting result in each historical time period;
acquiring a grid point historical wind speed forecast result and an actual observation historical wind speed of each historical time period, and calculating a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observation historical wind speed;
aiming at each historical time period, inputting the historical time period meteorological features of the historical time period and the grid point historical errors of each grid point wind speed prediction model in the historical time period into a grid point wind speed correction training model, and training the grid point wind speed correction training model to obtain a grid point wind speed correction model;
the lattice point wind speed correction training model comprises an action network submodel and a value network submodel, and the method comprises the following steps of inputting the historical time period meteorological features of the historical time period and the lattice point historical errors of each lattice point wind speed prediction model in the historical time period into the lattice point wind speed correction training model aiming at each historical time period, and training the lattice point wind speed correction training model to obtain the lattice point wind speed correction model, wherein the method comprises the following steps:
splicing historical time period meteorological features of the historical time period and grid point historical errors of each grid point wind speed prediction model in the historical time period to obtain splicing data;
inputting the splicing data into a full-connection characteristic layer in the action network submodel to obtain the training weight of the wind speed prediction model of each lattice point;
inputting the splicing data and the training weight of the historical period into a full-connection characteristic layer in the value network submodel to obtain a grid point wind speed forecasting weighting result of each grid point wind speed forecasting model on a preset historical forecasting period;
calculating value loss according to the grid point wind speed forecast weighting result and grid point wind speed measured values corresponding to historical forecast time periods, if the value loss is larger than a preset value loss threshold, updating parameters of the full-connection characteristic layer, and training the next historical time period until the value loss is not larger than the value loss threshold, so as to obtain a grid point wind speed correction model;
the method further comprises the following steps:
acquiring current meteorological data and current landform data of forecast grid points in a current time period;
splicing the current meteorological data and the current landform data, and inputting a pre-trained meteorological feature extraction model to obtain meteorological features of the current time period;
aiming at each historical time period of the forecast grid points, acquiring forecast grid point historical wind speed forecasting results of forecasting each historical time period by the grid point wind speed forecasting model with the preset number;
calculating a model error of each grid point wind speed prediction model based on the prediction grid point historical wind speed prediction result and the actual historical wind speed of each historical time period;
inputting the meteorological features and model errors in the current time period into a full-connection feature layer of the grid point wind speed correction model to obtain the prediction weight of the grid point wind speed prediction model;
acquiring grid point future wind speed forecasting results of the grid point wind speed forecasting models with the preset number respectively forecasting based on the current time period;
and aiming at each grid point wind speed prediction model, calculating the product of the grid point future wind speed prediction result of the grid point wind speed prediction model and the prediction weight, and obtaining the grid point future wind speed prediction correction result of the predicted grid point based on the products.
2. The method of claim 1, wherein calculating a model error of each grid point wind speed prediction model based on the predicted grid point historical wind speed prediction result and the actual historical wind speed of each historical time period comprises:
calculating a forecast grid point historical wind speed forecast result in a historical time period and a wind speed difference value of an actual historical wind speed aiming at each grid point wind speed forecast model;
acquiring the sum of squares of the wind speed difference;
and carrying out weighted average on the square sum of each historical time period to obtain the model error of the grid point wind speed prediction model.
3. The method of claim 1 or 2, wherein the meteorological feature extraction model comprises: the meteorological feature extraction network and the global pooling network are used for inputting historical meteorological data and historical landform data of the same historical period into a meteorological feature extraction model trained in advance to obtain the meteorological features of the historical period, and the meteorological feature extraction network comprises the following steps:
inputting historical meteorological data and historical geomorphic data of the historical time period into a meteorological feature extraction network trained in advance to obtain initial historical meteorological features; and carrying out global pooling on the historical meteorological initial characteristics according to the global pooling network to obtain the meteorological characteristics in the historical time period.
4. The method according to claim 1 or 2, characterized in that the method further comprises:
carrying out bilinear interpolation processing on the historical landform data so as to enable the size of the historical landform data to be the same as that of the historical meteorological data;
and respectively carrying out standardization and regularization processing on the historical meteorological data and the historical landform data after interpolation processing.
5. The method of claim 1 or 2, wherein the historical meteorological data comprises: the ground temperature data, the ground humidity data, the wind speed U component data and the wind speed V component data, the historical landform data comprises: terrain data, and solar elevation angle data.
6. An apparatus for constructing a grid point wind speed correction model, comprising:
the data acquisition module is used for acquiring historical meteorological data and historical landform data of each grid point in a preset historical time period;
the characteristic extraction module is used for inputting historical meteorological data and historical landform data of the same historical period into a meteorological characteristic extraction model trained in advance aiming at each grid point to obtain the meteorological characteristics of the historical period;
the forecasting result acquiring module is used for acquiring a grid point historical wind speed forecasting result forecasted by a grid point wind speed forecasting model with preset number in each historical time period aiming at each grid point;
the error calculation module is used for acquiring a grid point historical wind speed forecast result and an actual observation historical wind speed of each historical time period, and calculating a grid point historical time period error according to the grid point historical wind speed forecast result and the actual observation historical wind speed;
the wind speed correction module is used for inputting the historical time period meteorological features of the historical time period and the grid point historical errors of the grid point wind speed prediction models in the historical time period into the grid point wind speed correction training model aiming at each historical time period, and training the grid point wind speed correction training model to obtain the grid point wind speed correction model;
the wind speed correction module comprises:
the splicing unit is used for splicing the historical period meteorological features of the historical period and the grid point historical errors of the grid point wind speed prediction models in the historical period to obtain splicing data;
the weight parameter acquisition unit is used for inputting the splicing data into a full-connection characteristic layer in the action network submodel to obtain the training weight of the wind speed prediction model of each lattice point;
the weighting result acquisition unit is used for inputting the splicing data and the training weight of the historical period into a full-connection characteristic layer in the value network submodel to obtain a grid point wind speed forecasting weighting result of each grid point wind speed forecasting model to the preset historical forecasting period;
the model determining unit is used for calculating the value loss according to the grid point wind speed forecast weighting result and the grid point wind speed measured value corresponding to the historical forecast time period, updating the parameters of the full-connection characteristic layer if the value loss is larger than a preset value loss threshold value, and training the next historical time period until the value loss is not larger than the value loss threshold value to obtain a grid point wind speed correction model;
the wind speed forecasting module is used for acquiring current meteorological data and current landform data of the forecasting grid points in the current time period;
splicing the current meteorological data and the current landform data, and inputting a pre-trained meteorological feature extraction model to obtain meteorological features of the current time period;
aiming at each historical time period of the forecast grid points, acquiring forecast grid point historical wind speed forecasting results of forecasting each historical time period by the grid point wind speed forecasting model with the preset number;
calculating a model error of each grid point wind speed prediction model based on the prediction grid point historical wind speed prediction result and the actual historical wind speed of each historical time period;
inputting the meteorological features and model errors in the current time period into a full-connection feature layer of the grid point wind speed correction model to obtain the prediction weight of the grid point wind speed prediction model;
acquiring grid point future wind speed forecasting results of the grid point wind speed forecasting models with the preset number respectively forecasting based on the current time period;
and aiming at each grid point wind speed prediction model, calculating the product of the grid point future wind speed prediction result of the grid point wind speed prediction model and the prediction weight, and obtaining the grid point future wind speed prediction correction result of the predicted grid point based on the products.
7. A computer device, comprising: a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory communicating over the bus when a computer device is run, the machine-readable instructions when executed by the processor performing the steps of the method of constructing a lattice point wind speed correction model according to any of claims 1 to 5.
8. A computer-readable storage medium, having stored thereon a computer program for performing, when being executed by a processor, the steps of the method for constructing a grid wind speed correction model according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110759432.0A CN113204061B (en) | 2021-07-06 | 2021-07-06 | Method and device for constructing lattice point wind speed correction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110759432.0A CN113204061B (en) | 2021-07-06 | 2021-07-06 | Method and device for constructing lattice point wind speed correction model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113204061A CN113204061A (en) | 2021-08-03 |
| CN113204061B true CN113204061B (en) | 2021-10-08 |
Family
ID=77022741
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110759432.0A Active CN113204061B (en) | 2021-07-06 | 2021-07-06 | Method and device for constructing lattice point wind speed correction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113204061B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516320B (en) * | 2021-09-14 | 2021-12-10 | 国能日新科技股份有限公司 | Wind speed correction and predicted wind speed optimization method and device based on multi-objective genetic algorithm |
| CN114545528B (en) * | 2022-03-09 | 2024-02-06 | 北京墨迹风云科技股份有限公司 | Machine learning-based correction method and device after meteorological numerical mode element forecast |
| CN116663432B (en) * | 2023-07-28 | 2023-10-27 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Hundred-meter height wind speed forecast correction downscaling method and device |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109358381A (en) * | 2018-09-12 | 2019-02-19 | 浙江省气象科学研究所 | A kind of website forecast correction wind method |
| CN110187727B (en) * | 2019-06-17 | 2021-08-03 | 武汉理工大学 | Glass melting furnace temperature control method based on deep learning and reinforcement learning |
| CN111723929A (en) * | 2020-05-08 | 2020-09-29 | 广州华工中云信息技术有限公司 | Numerical prediction product correction method, device and system based on neural network |
| CN111929748B (en) * | 2020-09-17 | 2022-12-09 | 南京浦蓝大气环境研究院有限公司 | Meteorological element forecasting method and system |
| CN112668235B (en) * | 2020-12-07 | 2022-12-09 | 中原工学院 | Robot control method based on off-line model pre-training learning DDPG algorithm |
| CN112748483B (en) * | 2020-12-24 | 2022-08-19 | 北京思湃德信息技术有限公司 | Air temperature forecast deviation correction method and device based on deep learning |
| CN112684520B (en) * | 2020-12-30 | 2022-12-13 | 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 |
| CN113011106B (en) * | 2021-04-15 | 2021-12-03 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Multi-value mode integrated forecasting method and device |
| CN112990354B (en) * | 2021-04-15 | 2021-08-13 | 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 deep convolution regression network for wind speed prediction |
-
2021
- 2021-07-06 CN CN202110759432.0A patent/CN113204061B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113204061A (en) | 2021-08-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113204061B (en) | Method and device for constructing lattice point wind speed correction model | |
| CN113253364B (en) | Method and device for correcting station wind speed forecast result | |
| CN112200358B (en) | Landslide level prediction method, landslide level prediction device, landslide level prediction equipment and storage medium | |
| CN110232471B (en) | Rainfall sensor network node layout optimization method and device | |
| EP3794386B1 (en) | Determination of location-specific weather information for agronomic decision support | |
| CN109919515B (en) | Ecological environment quality evaluation method and device | |
| CN107247690B (en) | Estimate the method and service terminal of temperature | |
| CN112163375A (en) | Long-time sequence near-surface ozone inversion method based on neural network | |
| Tesfamichael et al. | Investigating the impact of discrete-return lidar point density on estimations of mean and dominant plot-level tree height in Eucalyptus grandis plantations | |
| CN112418498A (en) | Temperature prediction method and system for intelligent greenhouse | |
| CN109472283B (en) | Dangerous weather prediction method and device based on multiple incremental regression tree model | |
| CN112990354B (en) | Method and device for constructing deep convolution regression network for wind speed prediction | |
| CN112417734A (en) | Wind speed correction method and device based on geographic information of wind power plant | |
| CN117131758A (en) | Training method and device of irrigation water estimation model and irrigation water estimation method | |
| CN116341841A (en) | Runoff forecast error correction method, apparatus, device, medium and program product | |
| CN114926040B (en) | Comprehensive evaluation method and system for natural capital of drainage basin and electronic equipment | |
| CN112183292B (en) | Method for extracting drought phenotype and evaluating drought resistance of field crops based on low-altitude remote sensing | |
| CN113159426A (en) | Weather type similarity judgment method and device, electronic equipment and readable storage medium | |
| Le Toumelin et al. | A two-folds deep learning strategy to correct and downscale winds over mountains | |
| CN117272182B (en) | Daily air temperature prediction method, device, medium and equipment | |
| CN115620171B (en) | Forest storage amount evaluation method, device, equipment and storage medium | |
| Xia et al. | A Markov-Kalman model of land-use change prediction in XiuHe Basin, China | |
| CN117592005B (en) | PM2.5 concentration satellite remote sensing estimation method, device, equipment and medium | |
| CN116449460B (en) | Regional month precipitation prediction method and system based on convolution UNet and transfer learning | |
| CN110929225B (en) | Near-surface sulfur dioxide concentration estimation method based on geographical weighted regression model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |