CN117950087B

CN117950087B - Artificial intelligence downscale climate prediction method based on large-scale optimal climate mode

Info

Publication number: CN117950087B
Application number: CN202410327331.XA
Authority: CN
Inventors: 杨修群; 王昱; 孙旭光; 房佳蓓; 陶凌峰; 张志琦; 张昱培
Original assignee: Nanjing University
Current assignee: Nanjing University
Priority date: 2024-03-21
Filing date: 2024-03-21
Publication date: 2024-06-21
Anticipated expiration: 2044-03-21
Also published as: CN117950087A

Abstract

The application provides an artificial intelligent downscale climate prediction method based on a large-scale optimal climate mode, which utilizes a space-time coupling mode decomposition method to extract a contemporaneous large-scale optimal climate mode and a time sequence for determining the abnormal relative tendency of a large-scale climate element based on the large-scale climate element corresponding to a downscale climate prediction target element; training and constructing a downscaled prediction model of a nonlinear relationship between a large-scale optimal climate mode and a regional refined prediction target climate element abnormal relative tendency by using an artificial intelligent model; the contemporaneous large-scale optimal climate modal time coefficient predicted by the global climate dynamic mode is brought into the prediction model, and the abnormal relative tendency of the refined climate elements in the prediction area is predicted; and combining recent background abnormality, realizing artificial intelligent downscaling climate prediction for finely predicting the distance between target climate elements in the region. The method can improve the regional refined climate prediction capability by establishing an efficient and accurate downscaling climate prediction model.

Description

Artificial intelligence downscale climate prediction method based on large-scale optimal climate mode

Technical Field

The application relates to a design and application of an artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode. In the specific area-oriented refined prediction business, the method can realize the refined and quantitative intelligent prediction of key climate elements such as regional precipitation and air temperature based on a large-scale optimal climate mode and an artificial intelligent model, and provide technological support for regional climate disaster early warning and other services.

Background

Because the topography and topography are complex and changeable, the climate conditions of different areas are different, and the factors bring a plurality of challenges to the regional refined climate prediction. With the continuous development of economy and technology, the demands of many industries for refined climate prediction services are increasing, and particularly in the fields of agriculture, traffic, new energy and the like with higher requirements on regional climate conditions, the climate prediction service capability in the related technology is far from meeting the actual demands.

In order to cope with the above-mentioned needs, the main technical means in the related art is a downscaling climate prediction scheme, which aims to convert the output information of global or large-scale climate modes into detailed climate prediction of specific small-scale areas, and mainly comprises two schemes of dynamic downscaling and statistical downscaling. The dynamic downscaling is mainly realized through regional climate modes (Regional Climate Models, RCMs), the models are constructed based on global climate modes and run under the same boundary conditions, and the specific regional output has higher spatial resolution through a nested grid scheme and the like. The regional climate mode is constructed based on aerodynamic equations, can take into account the internal dynamics of a complex climate system, and has higher resolution, so that the regional climate mode can better capture micro-scale climate characteristics and is suitable for regions with complex terrains or special climate characteristics. However, the regional climate mode needs to consume huge amounts of computing resources and computing time, and is limited by the physical framework and boundary conditions of the global climate mode, so that the regional climate mode has very obvious uncertainty. The statistical downscaling mainly establishes a relation between large-scale climate elements and small-scale climate elements through a statistical method and a mathematical method, and further downscaling climate prediction is carried out. However, the effectiveness of the statistical downscaling method is highly dependent on high-quality and long-term historical observation data, and the limitation that unknown conditions cannot be predicted is also more obvious under the large background of global climate change.

Disclosure of Invention

The application provides an artificial intelligent downscale climate prediction method based on a large-scale optimal climate mode.

In a first aspect, an artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode is provided, the method comprising:

Based on large-scale prediction target elements corresponding to the downscaled climate prediction target elements, selecting a large-scale circular flow field for physical statistics relation construction, and respectively determining a corresponding climate abnormal field, an abnormal relative trend field and a recent background abnormal field;

Performing space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element to obtain a large-scale optimal climate mode of the abnormal relative trend field of the large-scale prediction target element and a time sequence corresponding to the large-scale optimal climate mode;

Taking the time sequence as a prediction factor, taking the abnormal relative trend field of the downscaled climate prediction target element as a prediction target, and training the nonlinear prediction model to be trained to obtain the nonlinear prediction model;

Determining a time coefficient corresponding to the synchronous large-scale annular flow field according to the large-scale optimal climate mode and the synchronous large-scale annular flow field, and importing the time coefficient into the nonlinear prediction model to obtain a prediction result of the abnormal relative trend field of the downscaled climate prediction target element;

And obtaining a nonlinear quantitative prediction result of the downscaled climate prediction target element from a flat field based on the prediction results of the recent background abnormal field of the downscaled climate prediction target element and the abnormal relative trend field of the downscaled climate prediction target element.

In the technical scheme, the artificial intelligence downscale climate prediction method based on the large-scale optimal climate mode has the advantages that the demand on computing resources is smaller than that of the regional climate power mode, and meanwhile, compared with the downscale scheme in the related art, the prediction capability of a nonlinear system is improved. The selection of the predictors is based on the space-time coupling relation between large-scale climate elements, the predictors and the prediction targets have strong physical constraint, and meanwhile, the predictors affecting the climate elements of different areas are subjected to nonlinear combination and weighting through an artificial intelligent model, so that a more perfect nonlinear prediction model is constructed. The nonlinear downscaling prediction is carried out on the downscaling climate prediction target elements based on the large-scale optimal mode, so that the integral characteristics of the regional climate large-scale background can be effectively grasped, the nonlinear quantitative intelligent prediction is carried out efficiently and accurately through the nonlinear prediction model, and the accuracy of the downscaling climate prediction is improved.

With reference to the first aspect, in some possible implementation manners, the large-scale optimal climate mode is obtained by extracting a large-scale climate element corresponding to the target element and a large-scale circulation element for determining abnormality of the large-scale climate element based on the downscaling climate prediction by a space-time coupling decomposition method.

With reference to the first aspect, in some possible implementations, the selecting a large-scale toroidal flow field for constructing a physical statistical relationship based on a large-scale prediction target element corresponding to the downscaled climate prediction target element includes:

and selecting the large-scale annular flow field for physical statistical relation construction based on the large-scale prediction target elements corresponding to the downscale climate prediction target elements by adopting a climate dynamics theory.

In the scheme, the large-scale circulation field can be used as a basic climate element for extracting a large-scale optimal climate mode and can be used as a source for extracting a predictive factor in a subsequent process.

With reference to the first aspect, in some possible implementations, the performing space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale predicted target element to obtain a large-scale optimal climate mode of the abnormal relative trend field of the large-scale predicted target element and a time sequence corresponding to the large-scale optimal climate mode includes:

Singular value decomposition is carried out on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element to obtain the large-scale optimal climate mode;

Projecting the abnormal relative trend field of the large-scale annular flow field to the large-scale optimal climate mode to obtain a time sequence corresponding to the large-scale optimal climate mode.

In the scheme, the optimal climate mode capable of predicting the climate condition of the target climate element is obtained through the space-time coupling decomposition method, and the time sequence of the abnormal relative trend field of the large-scale circular flow field is obtained through the projection method, so that the integral characteristics of the large-scale background of the regional climate can be effectively extracted, and the accuracy of the climate prediction result of the regional climate refinement is improved.

With reference to the first aspect, in some possible implementations, the performing singular value decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale predicted target element to obtain the large-scale optimal climate mode includes:

Singular value decomposition is carried out on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element;

and sequencing according to covariance corresponding to the singular value decomposition result to obtain the large-scale optimal climate mode.

In the scheme, the large-scale optimal climate mode of the abnormal relative trend field of the large-scale annular flow field is determined by analyzing the covariance contribution, and further the region refinement target climate elements can be effectively predicted through the large-scale optimal climate mode.

With reference to the first aspect, in some possible implementations, the determining, according to the large-scale optimal climate mode and the contemporaneous large-scale annular flow field, a time coefficient corresponding to the contemporaneous large-scale annular flow field includes:

predicting the circulation field corresponding to the downscaling climate prediction target element by adopting a global climate power mode to obtain the contemporaneous large-scale circulation field;

and projecting the contemporaneous large-scale annular flow field to the large-scale optimal climate mode to obtain a time coefficient corresponding to the contemporaneous large-scale annular flow field.

In the scheme, the synchronous large-scale circulation field is predicted by adopting the global climate power mode, and the synchronous large-scale circulation is projected to the large-scale optimal climate mode, so that the corresponding time coefficient is used as the input of the nonlinear prediction model, and the nonlinear prediction model is obtained through training of target climate elements with different resolutions, so that the obtained prediction result can accurately represent the downscaled climate prediction.

With reference to the first aspect, in some possible implementations, the obtaining a nonlinear quantitative prediction result of the downscaled climate prediction target element from a flat field based on a prediction result of a recent background abnormal field of the downscaled climate prediction target element and an abnormal relative trend field of the downscaled climate prediction target element includes:

and adding the recent background abnormal field of the downscaled climate forecast target element and the abnormal relative trend field of the downscaled climate forecast target element to obtain a nonlinear quantitative forecast result of the downscaled climate forecast target element from a flat field.

According to the scheme, the artificial intelligence downscaling climate prediction method based on the large-scale optimal climate mode can be used for predicting target climate elements of different scales, and further can be used for further completing efficient and accurate artificial intelligence downscaling nonlinear prediction based on a large-scale dynamic mode prediction result of coarse resolution, so that the requirements of actual service application can be effectively met, and an efficient and reliable solution is provided for improving the downscaling climate prediction service level.

With reference to the first aspect, in some possible implementations, the training the nonlinear prediction model to be trained by using the time sequence as a predictor and the abnormal relative trend field of the downscaled climate prediction target element as a prediction target to obtain the nonlinear prediction model includes:

acquiring an artificial intelligent model;

Constructing a nonlinear prediction model to be trained comprising a plurality of decision trees based on the artificial intelligence model; wherein the artificial intelligence model comprises: random forest regression model.

In the scheme, nonlinear combination and weighting are carried out on the prediction factors affecting the climate elements of different areas through a preset random forest regression model, so that a nonlinear prediction model with more accurate prediction results is constructed.

With reference to the first aspect, in some possible implementations, the method further includes:

And uniformly dividing any climatic element abnormal field into an abnormal relative trend field and a near-term background abnormal field.

In the scheme, the prediction target and the prediction factor are decomposed on the time scale through the process, and the prediction is concentrated on the required time scale, so that interference of signals on other time scales on the prediction is reduced.

In a second aspect, an artificial intelligence downscale climate prediction device based on a large-scale optimal climate modality is provided, the device comprising:

the first determining module is used for selecting a large-scale circular flow field for physical statistics relation construction based on a large-scale prediction target element corresponding to a downscale climate prediction target element, and respectively determining a corresponding climate abnormal field, an abnormal relative trend field and a recent background abnormal field;

The first decomposition module is used for carrying out space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element to obtain a large-scale optimal climate mode of the abnormal relative trend field of the large-scale prediction target element and a time sequence corresponding to the large-scale optimal climate mode;

The training module is used for taking the time sequence as a prediction factor, taking the abnormal relative trend field of the downscaled climate prediction target element as a prediction target, and training the nonlinear prediction model to be trained to obtain the nonlinear prediction model;

The second determining module is used for determining a time coefficient corresponding to the synchronous large-scale annular flow field according to the large-scale optimal climate mode and the synchronous large-scale annular flow field, and importing the time coefficient into the nonlinear prediction model to obtain a prediction result of the abnormal relative trend field of the downscaled climate prediction target element;

And the third determining module is used for obtaining a nonlinear quantitative prediction result of the downscaled climate prediction target element from a flat field based on a near-term background abnormal field of the downscaled climate prediction target element and a prediction result of an abnormal relative trend field of the downscaled climate prediction target element.

In a third aspect, an electronic device is provided that includes a memory and a processor. The memory is for storing executable program code and the processor is for calling and running the executable program code from the memory to cause the apparatus to perform the method of the first aspect or any one of the possible implementations of the first aspect.

In a fourth aspect, there is provided a computer program product comprising: computer program code which, when run on a computer, causes the computer to perform the method of the first aspect or any one of the possible implementations of the first aspect.

In a fifth aspect, a computer readable storage medium is provided, the computer readable storage medium storing computer program code which, when run on a computer, causes the computer to perform the method of the first aspect or any one of the possible implementations of the first aspect.

Drawings

FIG. 1 is a schematic flow chart of an artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode provided by an embodiment of the application;

FIG. 2 is a further schematic flow chart of an artificial intelligence downscale climate prediction method based on a large-scale optimal climate modality provided by an embodiment of the present application;

FIG. 3 is another schematic flow chart of an artificial intelligence downscale climate prediction method based on a large-scale optimal climate modality provided by an embodiment of the application;

FIG. 4 is a flowchart of an artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode provided by an embodiment of the application;

FIG. 5 is an operational flow diagram of an artificial intelligence downscale climate prediction method based on a large-scale optimal climate modality provided by an embodiment of the present application;

FIG. 6 is a graph showing the spatial distribution of the rainfall distance between the middle and downstream river basins of the selected 2019 summer 2374 station and 160 station;

FIG. 7 is a block diagram of an artificial intelligence prediction model based on random forest regression constructed in an embodiment of the present application;

FIG. 8 is a graph showing the spatial distribution of the prediction result of the drainage basin rainfall distance flat field service mode in the middle and downstream of the Yangtze river in summer in 2019 versus the weather prediction result provided by the embodiment of the application;

FIG. 9 is a schematic structural diagram of an artificial intelligent downscaled climate prediction device based on a large-scale optimal climate mode, which is provided by the embodiment of the application;

fig. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The technical scheme of the application will be clearly and thoroughly described below with reference to the accompanying drawings. Wherein, in the description of the embodiments of the present application, unless otherwise indicated, "/" means or, for example, a/B may represent a or B: the text "and/or" is merely an association relation describing the associated object, and indicates that three relations may exist, for example, a and/or B may indicate: the three cases where a exists alone, a and B exist together, and B exists alone, and furthermore, in the description of the embodiments of the present application, "plural" means two or more than two.

The terms "first," "second," and the like, are used below for descriptive purposes only and are not to be construed as implying or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature.

The following describes a technical solution provided by the embodiment of the present application, and the embodiment of the present application provides an artificial intelligence downscaling climate prediction method based on a large-scale optimal climate mode, referring to fig. 1, fig. 1 is a schematic flowchart of an artificial intelligence downscaling climate prediction method based on a large-scale optimal climate mode provided by the embodiment of the present application, where the method includes the following steps:

101, selecting a large-scale circular flow field for physical statistics relation construction based on large-scale prediction target elements corresponding to the downscaled climate prediction target elements, and respectively determining a corresponding climate abnormal field, an abnormal relative trend field and a near-term background abnormal field.

The downscaled climate prediction target element may be a region refined climate prediction element, for example, may include: rainfall prediction or temperature prediction with refined region, etc. The large-scale prediction target elements corresponding to the downscaled climate prediction target elements are coarse-granularity target climate elements with resolution larger than that of the downscaled climate prediction target elements. For example, the downscaled climate forecast target element may be precipitation at 2374 and the large scale forecast target element may be precipitation at 160. And selecting the large-scale annular flow field for physical statistical relation construction based on the large-scale prediction target elements corresponding to the downscale climate prediction target elements by adopting a climate dynamics theory. The large-scale circular flow field for physical statistics relation construction is selected from the historical return data set of the climate power mode through the dynamics theory. Thus, the large-scale circulation field can be used as a basic climate element for extracting a large-scale optimal climate mode and used as a source for extracting a predictive factor in a subsequent step. For example, the data of the tropical outward long wave radiation (Outgoing Longwave Radiation, OLR) in the summer of the year from 3 months and 1 day each year, the high latitude 500h Pa potential height (Geopotential Height @500 hPa, Z500) in the northern hemisphere are selected from the historical return data set of the climate power mode, and the abnormal relative trend is calculated. The summer tropical OLR abnormal relative trend field and the northern hemisphere medium-high latitude Z500 abnormal relative trend field are used as large-scale circular flow fields. Therefore, the large-scale circulation field is selected through the dynamics theory, so that the obtained large-scale circulation field can more accurately determine the large-scale prediction target elements corresponding to the downscaled climate prediction target elements.

In some possible implementations, the anomaly relative trend approach refers to bisecting the prediction target distance into two parts in the prediction: the relative trend of anomalies and the corresponding background distance of recent observations are flat. The level is predicted by predicting the abnormal relative tendency of the target season average, and the prediction is focused on the abnormal relative tendency portion determined by the predictable annual variability. In step 101, a climate anomaly field, an anomaly relative trend field, a near-term background anomaly field, and a climate anomaly field, an anomaly relative trend field, a near-term background anomaly field, corresponding to the downscaled climate forecast target element, are calculated. Wherein, the climate abnormal field is the sum of the abnormal relative trend field and the recent background abnormal field. If an anomaly relative trend field is defined as the difference between the predicted target year (e.g., t+1 year) and its two adjacent years (i.e., t year), the corresponding recent background anomaly field is the previous year's apparent range plateau.

In some possible implementations, determining a corresponding anomaly relative trend field and a recent background anomaly field based on the large-scale prediction target elements corresponding to the downscaled climate prediction target elements, respectively; and adding the abnormal relative trend field and the recent background abnormal field to obtain the climate abnormal field. For example, the abnormal relative trend field and the recent background abnormal field are added to obtain the climate abnormal field.

The climate abnormal field, the abnormal relative trend field and the recent background abnormal field can be calculated through the following processes:

Firstly, obtaining climate variable original data; then subtracting the climate state to obtain a distance level (namely a climate abnormal field); finally, combining a formula (climate abnormal field=abnormal relative trend+abnormal relative background), and obtaining an abnormal relative trend field and a near-term background abnormal field. In this way, by calculating the climate abnormal field, the abnormal relative trend field and the recent background abnormal field, the fine climate prediction of the subsequent region can be facilitated.

102, Performing space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element to obtain a large-scale optimal climate mode of the abnormal relative trend field of the large-scale prediction target element and a time sequence corresponding to the large-scale optimal climate mode.

And carrying out space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element by adopting a space-time coupling decomposition method to obtain a large-scale optimal climate mode serving as space information and a time sequence serving as time information.

The summer tropical OLR abnormal relative trend field and the northern hemisphere medium-high latitude Z500 abnormal relative trend field are used as large-scale circular flow fields. And carrying out space-time coupling decomposition on the abnormal relative trend fields of the synchronous large-scale prediction target elements respectively to obtain time sequences corresponding to the large-scale optimal climate mode and the large-scale optimal climate mode.

The large-scale optimal climate mode is obtained by extracting a large-scale climate element corresponding to the target element of the downscaling climate prediction and the large-scale circulation element for determining the abnormality of the large-scale climate element through a space-time coupling decomposition method. What is actually used as a predictor and used for predictive model training is a time series corresponding to the large-scale optimal climate modality. The large-scale optimal climate mode is extracted based on large-scale climate elements corresponding to the regional refinement prediction target elements and large-scale circulation elements for determining large-scale climate element abnormality through a space-time coupling decomposition method, and the prediction factors represent the causes of large-scale climate element abnormality backgrounds corresponding to the regional refinement climate element abnormality. The large-scale circulation factor determining the abnormality of the large-scale climate factor can be a large-scale circulation field.

In some possible implementations, to obtain the large-scale optimal climate mode and the corresponding time sequence more accurately, the step 102 may be implemented by the steps shown in fig. 2:

And 201, performing singular value decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element to obtain the large-scale optimal climate mode.

Singular value decomposition is carried out on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element, so that a large-scale optimal climate mode is obtained. Singular value decomposition (Singular Value Decomposition Analysis, SVD), also known as maximum covariance analysis (Maximum Covariance Analysis, MCA), is a matrix decomposition method used to reduce a matrix to its constituent parts. The method is commonly used for diagnosis and analysis of two meteorological field space-time distribution coupling signals in the meteorological field, and an optimal climate mode for determining summer large-scale atmospheric flow anomaly relative trend of synchronous rainfall anomaly relative trend is extracted from historical observation data through an SVD method.

In some possible implementation manners, the large-scale optimal climate mode is obtained by carrying out singular value decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element and sequencing according to covariance corresponding to a singular value decomposition result. Here, if the singular value decomposition method is used to decompose the abnormal relative tendency field of the large-scale toroidal flow field and the abnormal relative tendency field of the large-scale prediction target element, three matrices, i.e., an S matrix (a matrix composed of eigenvectors), a U matrix (an orthogonal matrix), and a V matrix (an orthogonal matrix), are obtained. Thus, three matrices after singular value decomposition can be taken as singular value decomposition results. The S matrix, the U matrix and the V matrix carry characteristic values, wherein the larger the characteristic value is, the larger the contribution of the matrix to covariance is, and the climate mode represented by the matrix is further described as being capable of representing the climate corresponding to the target climate element; the matrix can be regarded as an optimal climate modality. Therefore, the covariance contribution of the U matrix is maximum, so that the U matrix can be used as a large-scale optimal climate mode, and the climate condition of the target climate element can be represented by the U matrix. For example, the target climate element is precipitation, and the el nino mode can occupy more than 30%, so that the mode is an important mode. Therefore, the large-scale optimal climate mode of the abnormal relative trend field of the large-scale annular flow field is determined by analyzing the covariance contribution, and further the region refinement target climate elements can be effectively predicted through the large-scale optimal climate mode.

202, Projecting the abnormal relative trend field of the large-scale annular flow field to the large-scale optimal climate mode to obtain a time sequence corresponding to the large-scale optimal climate mode.

The time sequence is obtained by calculating a large-scale optimal climate mode through a projection method. Projecting an abnormal relative trend field of the large-scale annular flow field to a large-scale optimal climate mode to obtain a time sequence, wherein the value of the time sequence can represent the similarity; wherein the higher the degree of similarity, the larger the corresponding value, and the lower the degree of similarity, the smaller the corresponding value. For example, if the abnormal relative trend fields of the large-scale annular flow field are the abnormal relative trend fields of the OLR and the Z500, and the large-scale optimal climate mode is the U matrix, then the abnormal relative trend fields of the OLR and the Z500 are projected onto the decomposed U matrix, and a time sequence can be obtained. In this way, the optimal climate mode capable of predicting the climate condition of the target climate element is obtained through a space-time coupling decomposition method, and the time sequence of the abnormal relative trend field of the large-scale circular flow field is obtained through a projection method, so that the integral characteristics of the large-scale background of the regional climate can be effectively extracted, and the accuracy of the climate prediction result of the regional climate refinement is improved.

And 103, training the nonlinear prediction model to be trained by taking the time sequence as a prediction factor and taking the abnormal relative trend field of the downscaled climate prediction target element as a prediction target to obtain the nonlinear prediction model.

The abnormal relative trend field of the target element is predicted based on time sequence and downscaled climate and is obtained through training. The time sequence is used as a prediction factor, the abnormal relative trend field of the downscaled climate prediction target element is used as a training target, and the nonlinear prediction model to be trained is trained to obtain the nonlinear prediction model. In this way, by constructing the artificial intelligent nonlinear prediction model, taking the time sequence of the large-scale optimal climate mode as a prediction factor, taking the abnormal relative trend field of the downscaled climate prediction target element as a prediction target, and training the nonlinear prediction model to be trained by utilizing the artificial intelligent model, the obtained nonlinear prediction model is more perfect.

In some possible implementations, first, an artificial intelligence model is obtained; constructing a nonlinear prediction model to be trained, which comprises a plurality of decision trees, based on the artificial intelligent model; wherein the artificial intelligence model comprises: random forest regression model.

In some possible implementations, firstly, a nonlinear prediction model to be trained, which comprises a plurality of decision trees, is constructed based on a preset random forest regression model; here, the nonlinear prediction model to be trained including a plurality of decision trees may also be constructed based on a classification tree scheme, a support vector machine, a recurrent neural network, and the like. Taking a preset random forest regression model as an example, building a nonlinear prediction model to be trained, which comprises a plurality of decision trees, through the plurality of decision trees in the preset random forest regression model. Wherein the plurality of decision trees may be a first level regression tree, a second level regression tree, and a third level regression tree as shown in FIG. 7. And then, taking the time sequence and the abnormal relative trend field of the downscaled climate prediction target element as a training data set, and carrying out parameter adjustment on the nonlinear prediction model to be trained to obtain the nonlinear prediction model. The time sequence is taken as a prediction factor, the abnormal relative trend field of the downscaled climate prediction target element is taken as a prediction target, and the training data set is divided into a plurality of sub data sets, so that the network parameters such as weights of a plurality of decision trees in the nonlinear prediction model to be trained are respectively adjusted, and the nonlinear prediction model is obtained. Thus, the nonlinear combination and weighting are carried out on the predictors affecting the climate elements of different areas through the artificial intelligent model, so that a nonlinear prediction model with more accurate prediction results is constructed.

After taking the time sequence and the abnormal relative trend field of the downscaled climate prediction target element as a training data set, firstly, sampling the training data set to obtain a plurality of sub-data sets. Here, the training data set is randomly sampled through self-service aggregation to obtain a plurality of sub-data sets, for example, the same number of sub-data sets can be extracted according to the number of decision trees. And then, training each decision tree in the nonlinear prediction model to be trained based on each sub-data set in the plurality of sub-data sets to obtain a plurality of trained decision trees. Here, one decision tree is trained by one sub-data set, and since the number of sub-data sets is the same as the number of decision trees, each decision tree of the plurality of decision trees is trained, resulting in a plurality of trained decision trees. Finally, the nonlinear predictive model is obtained based on the plurality of trained decision trees. Here, the nonlinear predictive model can be obtained by weighted combination of a plurality of trained decision trees. Thus, by training a decision tree by extracting each sub-data set, the average value of all the decision tree predictions is used as a prediction result. Compared with a single decision tree model, the nonlinear prediction model can effectively reduce the overfitting problem, can provide estimation of the importance of each prediction factor, and plays an important role in understanding the causal relationship between the prediction factors and the prediction targets. In the application of downscaling climate prediction, the nonlinear prediction model can analyze the contribution of large-scale optimal climate modes affecting different refined sites, is beneficial to understanding the cause of climate prediction results, and improves the accuracy and the robustness of nonlinear prediction model prediction.

104, Determining a time coefficient corresponding to the synchronous large-scale annular flow field according to the large-scale optimal climate mode and the synchronous large-scale annular flow field, and importing the time coefficient into the nonlinear prediction model to obtain a prediction result of the abnormal relative trend field of the downscaled climate prediction target element.

Wherein, the contemporaneous large-scale annular flow field is obtained by predicting global climate power mode.

Predicting the circulation field corresponding to the downscaling climate prediction target element by adopting a global climate power mode to obtain the contemporaneous large-scale circulation field; and then, projecting the contemporaneous large-scale annular flow field to the large-scale optimal climate mode to obtain a time coefficient corresponding to the contemporaneous large-scale annular flow field.

In some possible implementations, a historical return data set of the climate power mode is obtained, a global climate power mode is adopted, and a circulation field corresponding to a downscaled climate prediction target element is predicted according to the historical return data set of the climate power mode, so that the synchronous large-scale circulation field is obtained. And projecting the contemporaneous large-scale annular flow field to a large-scale optimal climate mode by adopting a projection method, so that the similarity of the contemporaneous large-scale annular flow field and the large-scale optimal climate mode is obtained, and the similarity is used as a time coefficient corresponding to the contemporaneous large-scale annular flow field. In a specific example, if the target climate element is 2019 precipitation, i.e. 2019 precipitation needs to be predicted, selecting the current year of summer tropical OLR, northern hemisphere middle high latitude Z500 data from the historical return data set of the climate power mode; and then predicting the abnormal relative trend fields of the tropical OLR in summer and the high latitude Z500 in the northern hemisphere in 2019 according to the data of the tropical OLR in summer and the high latitude Z500 in the northern hemisphere in the current year, and obtaining the synchronous large-scale annular flow field. And calculating a time coefficient corresponding to the large-scale optimal climate mode in summer in 2019 by a projection method, wherein the time coefficient is used as an actual prediction factor for calculating the relative trend field of the rainfall abnormality in the middle and downstream river of the Yangtze river in summer in 2019. And inputting a time coefficient serving as a prediction factor into a nonlinear prediction model to obtain an artificial intelligent downscale prediction result of the relative trend field of the rainfall anomaly at the downstream site in the Yangtze river in summer in 2019, namely a prediction result of the relative trend field of the anomaly of the downscale climate prediction target element. In this way, the synchronous large-scale circulation flow field is predicted by adopting the global climate power mode, and the synchronous large-scale circulation flow is projected to the large-scale optimal climate mode, so that the corresponding time coefficient is used as the input of the nonlinear prediction model, and the nonlinear prediction model is obtained by training target climate elements with different resolutions, so that the obtained prediction result can accurately represent the downscaled climate prediction.

105, Obtaining a nonlinear quantitative prediction result of the downscaled climate prediction target element from a flat field based on a prediction result of a recent background abnormal field of the downscaled climate prediction target element and an abnormal relative trend field of the downscaled climate prediction target element.

And calculating a distance flat field of the historical observation data through the regional refined climate element historical observation data set. Distance flat field of the historical observation = historical downscaled climate forecast object element anomaly relative trend field + recent background anomaly field. For example, the historical observation data is the sum of the 2374 station precipitation anomaly relative trend field and the 2374 station precipitation recent background anomaly field from the flat field. The recent background abnormal field of the downscaling climate forecast target element is 2374 station precipitation recent background abnormal field. And combining the recent background abnormal field of the downscaled climate forecast target element with the forecast result output by the nonlinear forecast model to obtain a climate forecast result.

In some possible implementations, the step 105 may be implemented by the steps shown in fig. 3:

301, adding a recent background abnormal field of the downscaled climate prediction target element and an abnormal relative trend field of the downscaled climate prediction target element to obtain a nonlinear quantitative prediction result of the downscaled climate prediction target element from a flat field.

Here, because the prediction result indicates the abnormal relative trend field of the downscaled climate prediction target element output by the nonlinear prediction model, the distance flat field of the downscaled climate prediction target element can be accurately obtained by adding all the abnormal relative trend fields of the downscaled climate prediction target element obtained by prediction and the recent background abnormal field of the downscaled climate prediction target element.

The distance flat field of the downscaling climate prediction target element can be used for representing the climate prediction condition of the downscaling climate prediction target element, so that the distance flat field is used as a nonlinear quantitative prediction result of the downscaling climate prediction target element, and the region-refined climate condition can be accurately represented. Therefore, the artificial intelligence downscaling climate prediction method based on the large-scale optimal climate mode can further complete efficient and accurate artificial intelligence downscaling nonlinear prediction based on the large-scale dynamic mode prediction result of coarse resolution aiming at the prediction target climate elements of different scales, can effectively meet the requirements of actual service application, and provides an efficient and reliable solution for improving the downscaling climate prediction service level.

In steps 101, 104 and 105, any one of the abnormal fields of the climate elements is decomposed into an abnormal relative trend field and a near-term background abnormal field, so that the prediction targets and the prediction factors are decomposed on a time scale through the process, and the predictions are concentrated on a required time scale, so that interference of signals on other time scales on the predictions is reduced.

According to the artificial intelligence downscale climate prediction method based on the large-scale optimal climate mode, which is provided by the embodiment of the application, the demand on computing resources is smaller than that of the regional climate power mode, and meanwhile, compared with a downscale scheme in the related art, the prediction capability of a nonlinear system is improved. The selection of the predictors is based on the space-time coupling relation between large-scale climate elements, the predictors and the prediction targets have strong physical constraint, and meanwhile, the predictors affecting the climate elements of different areas are subjected to nonlinear combination and weighting through an artificial intelligent model, so that a more perfect nonlinear prediction model is constructed. The nonlinear downscaling prediction is carried out on the downscaling climate prediction target elements based on the large-scale optimal mode, so that the integral characteristics of the regional climate large-scale background can be effectively grasped, the nonlinear quantitative intelligent prediction is carried out efficiently and accurately through the nonlinear prediction model, and the accuracy of the downscaling climate prediction is improved.

In some embodiments, the climate system is a complex nonlinear system, the relationship between the large-scale climate elements and the small-scale climate elements cannot be accurately described through a simple linear model, and the artificial intelligence technology has strong nonlinear modeling capability, so that the climate system has wide application prospect in downscale climate prediction. The artificial intelligence scheme in the related art generally requires training based on a large amount of observation data and power mode output, and not only consumes a large amount of computing resources, but also has no interpretability.

Based on the above, the embodiment of the application provides an artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode, clear physical relations among large-scale climate mode elements established by using large-scale prediction output of a dynamic mode and a climate statistics method are utilized, and a nonlinear prediction model between the large-scale climate mode and the small-scale climate elements is established by using an artificial intelligence technology, so that nonlinear intelligent prediction of abnormality of the regional refined climate elements is realized.

In the embodiment of the application, aiming at the difficult problem of regional refined climate prediction, in particular to constructing a high-efficiency nonlinear downscaling climate prediction model, an artificial intelligent downscaling climate prediction method based on a large-scale optimal climate mode is provided. Based on the physical statistical relationship between key elements such as a large-scale optimal climate mode and large-scale precipitation, a nonlinear prediction model between the optimal climate mode and the regional refined climate elements is constructed by using an artificial intelligence scheme, the nonlinear intelligent prediction of the regional refined climate element abnormality is finally realized by utilizing a large-scale climate mode prediction result output by a power mode, and the process of artificial intelligent modeling of the climate prediction based on the optimal mode can be realized by the following steps:

Firstly, selecting climate elements with different scales, and calculating a corresponding distance flat field, an abnormal relative trend field and a recent background abnormal field.

According to the climate dynamics theory, a synchronous large-scale circular flow field capable of determining a large-scale prediction target climate element corresponding to a downscale climate prediction target element is selected as a source for extracting and calculating a prediction factor in a subsequent step.

As shown in fig. 4, in a first step, a large scale climate element history observation data pitch plane 401 is determined from a climate element history observation data set 41, the pitch plane comprising: a large scale climate element recent background anomaly 410 and a large scale climate element anomaly relative trend 411; the contemporaneous large scale climate element anomaly relative trend 412 that determines the predicted large scale predicted target element anomaly is selected from the large scale climate element anomaly relative trends. And selecting a large scale predicted target climate element anomaly relative trend 413 from the large scale climate element anomaly relative trends. In the regional refinement climate element historical observation data set 42, a historical observation data pitch plane 402 is determined. The relative trend 47 of regional refinement climate element anomalies and the recent background anomaly 46 of regional refinement climate element can be obtained from the historical observation data distance plane 402. Meanwhile, the regional refinement prediction target climate element anomaly relative trend 48 is determined from the regional refinement climate element anomaly relative trends, and corresponds to the large-scale prediction target climate element anomaly relative trend.

And secondly, extracting a large-scale optimal climate mode by a space-time coupling method and calculating a corresponding time sequence.

Here, the method of extracting the climate modality and the corresponding time series for the large scale optimization is: decomposing the contemporaneous large-scale circular flow field and the large-scale prediction target climate elements by using a space-time coupling decomposition method (such as a singular value decomposition method), sequencing based on covariance contribution, obtaining a large-scale optimal climate mode for determining the large-scale prediction target, and calculating by using a projection method to obtain a time sequence corresponding to the large-scale optimal climate mode.

As shown in fig. 4, a space-time coupled mode decomposition method 403 (for example, a singular value decomposition method) is used to decompose the relative tendency of the contemporaneous large-scale climate element anomaly and the relative tendency of the target climate element anomaly obtained in the first step, so as to obtain an optimal climate mode (SM) 404. And determining an optimal modal time series 405 corresponding to the optimal climate model by using a projection method, and taking the optimal modal time series 405 as a prediction factor.

Thirdly, taking the time sequence as a prediction factor, taking the abnormal relative trend field of the regional refined climate elements as a prediction target, and constructing a downscaled nonlinear prediction model based on an artificial intelligence method.

The time sequence of the large-scale optimal climate mode is taken as a prediction factor, the abnormal relative trend field of the downscaled climate prediction target element is taken as a prediction target, and the artificial intelligent model is utilized to train the nonlinear climate prediction model. In model construction, a variety of artificial intelligence models may be selected, including classification tree schemes, support vector machines, recurrent neural networks, and the like.

As shown in fig. 4, the relative abnormal tendency of the climate elements of the target is predicted through region refinement, the optimal modal time sequence is taken as a prediction target, and a down-scale nonlinear prediction model 406 is built by adopting an artificial intelligence method.

And step four, calculating a time coefficient corresponding to a large-scale optimal climate mode predicted by the contemporaneous global climate power mode, importing the artificial intelligent downscaling prediction model, and carrying out nonlinear prediction on the abnormal relative trend field of the regional refined climate elements.

Here, the specific calculation method for calculating the time coefficient corresponding to the abnormal relative trend field of the synchronous large-scale circular flow field predicted by the global climate power mode comprises the following steps: and (3) projecting the contemporaneous large-scale circular flow field abnormal relative trend field predicted by the global climate dynamic mode corresponding to the downscale climate prediction target element to the large-scale optimal climate mode obtained in the second step by using a projection method to obtain a corresponding time coefficient. And taking the time coefficient as an actual prediction factor, and taking the actual prediction factor into an artificial intelligent nonlinear climate prediction model constructed in the third step to obtain a prediction result of the abnormal relative trend field of the downscaled climate prediction target element.

As shown in fig. 4, based on the climate power pattern historical return data set 43, a contemporaneous power pattern predicted large scale climate element 407 is determined. And determining the abnormal relative trend 408 of the large-scale climate elements predicted by the synchronous dynamic mode through the large-scale climate elements predicted by the synchronous dynamic mode 407. And projecting the abnormal relative tendency of the large-scale climate elements predicted by the synchronous dynamic mode to an optimal climate mode to obtain a synchronous predictor time coefficient 409. The time coefficient is input to the downscaled nonlinear prediction model 406, and the region refinement prediction target climate element anomaly relative tendency prediction result 44 is output.

And fifthly, obtaining a downscaled prediction result of the region refined climate elements from the flat field based on the prediction result of the abnormal relative trend field and the corresponding recent background abnormal field.

Here, the method for calculating the distance between the downscaled climate forecast target elements and the flat field is as follows: according to the corresponding relation, the prediction results of the near-term background abnormal field of the downscaling climate prediction target element in the first step and the abnormal relative trend field of the downscaling climate prediction target element obtained in the fourth step are added, so that the distance between the downscaling climate prediction target element and the flat field is obtained, and finally, the artificial intelligent downscaling prediction based on the large-scale optimal climate mode is realized.

As shown in fig. 4, the region refinement prediction target climate element anomaly relative tendency prediction result 44 is combined with the region refinement climate element recent background anomaly 46 to obtain the downscaled climate prediction target element distance flat field 45.

In some embodiments, taking prediction of a summer precipitation distance flat field in a river basin 2019 in the middle and downstream of the Yangtze river as an example, an artificial intelligent downscaling climate prediction method based on a large-scale optimal climate mode is described in detail.

First, basic information of the embodiment of the present application is described in relation to:

In the embodiment of the application, the regional refined climate prediction target is a summer rainfall distance flat field in a middle and downstream river basin 2019 of the Yangtze river (the data is extracted from 2374 station weather information data issued by a weather bureau CIPAS system, the middle and downstream river basin is within a region of 25-35 DEG N and 110-125 DEG E), and the corresponding large scale elements select regional rainfall (the data is extracted from 160 station weather information data issued by the weather bureau) so as to determine that the synchronous tropical region (30-30 DEG S) with abnormal regional summer rainfall emits external long wave radiation and 500hPa potential altitude data in a middle and high latitude region (90-20 DEG N) of the northern hemisphere as the basis for extracting the large scale optimal climate mode. As shown in fig. 6, where spatial distribution 601 represents the spatial distribution of the drainage basin precipitation distance flat field in the middle and downstream of the Yangtze river at 2374 in summer in 2019, and spatial distribution 602 represents the spatial distribution of the drainage basin precipitation distance flat field in the middle and downstream of the Yangtze river at 160 in summer in 2019.

In the embodiment of the application, a space-time coupling decomposition method is a singular value decomposition method (Singular Value Decomposition, SVD), a large-scale climate mode of a high latitude 500hPa potential highly abnormal relative trend field and 160 stations large-scale precipitation in tropical OLR and northern hemisphere is extracted through the SVD method, the first few modes with the variance contribution ratio sum exceeding 90% are selected as large-scale optimal climate modes according to the maximum covariance principle, time sequences corresponding to the modes are used as forecasting factors, precipitation of a refined station in a river basin in the middle and lower reaches of a corresponding period of Yangtze river is used as a forecasting target to carry out artificial intelligent modeling, and the time dimension of a modeling training set is 1989-2018 (the first 30 years).

In the embodiment of the application, the selected artificial intelligent model is a random forest regression model (Random Forest Regression Model, RFRM), which is an integrated learning method based on Decision Trees (Decision Trees), and the accuracy and the robustness of the prediction are improved by constructing a plurality of Decision Trees and integrating and predicting all prediction results. The model firstly constructs a plurality of different sub-data sets by self-service aggregation (Bootstrap Aggregating) from random sampling in an original training set, trains a decision tree by using each characteristic data set, and finally calculates the average value of all the decision tree predictions to generate a prediction result. Compared with a single decision tree model, the model can effectively reduce the overfitting problem, can provide estimation on the importance of each predictive factor, and plays an important role in understanding the causal relationship between the predictive factors and the predictive targets. In the downscaling climate prediction application, the model not only can construct a nonlinear prediction model, but also can analyze the contribution of large-scale optimal climate modes affecting different refined sites, and is helpful for understanding the cause of climate prediction results. As shown in FIG. 7, the original training set 701 is randomly sampled, constructing set 1, set 2, &. 16, set N-1, and set N. Training decision trees corresponding to the set 1, the set 2, the set N-1 and the set N respectively, training the decision trees respectively through a first layer regression tree, a second layer regression tree and a third layer regression tree, and performing set voting 702 to output a prediction result 703.

In the embodiment of the application, the prediction results of the summer precipitation are given by the main service power mode BCC_CSM1.1 (m) existing in the national climate center as the prediction results of the summer precipitation abnormality in the middle and lower reaches of the Yangtze river, which are used as the comparison, in the tropical OLR predicted by the synchronous power mode and the high latitude region 500hPa potential height in the northern hemisphere (the data are issued by the Beijing climate center), and the reporting time of the power mode prediction in the embodiment of the application is set to be 2019, 3 and 1 day in consideration of the actual seasonal climate prediction service requirements.

As shown in fig. 5, the objective of predicting the water-gap flat field of the refined station in the middle and downstream river in summer in 2019 can be achieved by the following steps:

firstly, selecting OLR field of a tropical region in summer in 1989-2018, Z500 field data of a high latitude region in the northern hemisphere and 160 station precipitation data from a climate element historical observation data set 51, selecting 2374 station precipitation data from a regional refined climate element historical observation data set 52, and respectively calculating a distance flat field, an abnormal relative trend field and a recent background abnormal field corresponding to each data. The method comprises the steps of obtaining historical observation data of the large-scale climate elements, namely, a distance plane 501, a recent background anomaly 502 of the large-scale climate elements, relative tendencies of the large-scale climate elements, namely, 1989-2018, and relative tendencies of precipitation anomalies, namely, 1989-2018, of the historical observation data, namely, a distance plane 1989-2018, and a 2374 station. Then, determining 160-station precipitation abnormal relative tendency 503 from large-scale climate element abnormal relative tendencies 1989-2018; and determining the relative tendency 504 of the precipitation abnormality at the 2374 station from the relative tendencies 1989-2018 of the precipitation abnormality at the 2374 station.

And selecting the current year tropical OLR and the middle and high latitude Z500 data reported on 1 day of 3 months from the power mode history return data set 53 to obtain the current year tropical OLR and the middle and high latitude Z500 (2019) of the contemporaneous power mode prediction, respectively calculating the corresponding distance flat field, the abnormal relative trend field and the recent background abnormal field, selecting the corresponding summer precipitation data, interpolating to 2374 sites to obtain the 2374 site precipitation data set of the power mode history return, and calculating the corresponding distance flat field as a control group of the embodiment.

Secondly, carrying out SVD decomposition on the abnormal relative trend fields of the synchronous 160-station precipitation abnormal relative trend fields by adopting a space-time coupling modal decomposition method 54 respectively, wherein the abnormal relative trend fields of the tropical OLR in summer and the abnormal relative trend fields of the high latitude Z500 in northern hemisphere in 1981-2010, namely the abnormal relative trend fields 505 of the synchronous OLR and Z500 for determining 160-station precipitation abnormal; and then sequencing by a projection method and a covariance maximum principle to obtain a large-scale optimal mode 55 corresponding to the OLR and the Z500 and a time sequence 56 corresponding to the large-scale optimal mode, wherein the time sequence length is 30 (years).

Thirdly, according to the principle that the sum of covariance contribution ratio exceeds 90%, the first 12 OLR modes and the first 13Z 500 modes calculated in the second step are selected, and time sequences corresponding to the modes are used as prediction factors. And (3) taking the site precipitation abnormal relative trend field belonging to the region of the middle and lower river basins in the summer 2374 station in 1989-2018 obtained in the first step as a prediction target, and performing modeling training by using a random forest regression model to obtain a downscale climate prediction model 57 of the middle and lower river basin refined precipitation artificial intelligence based on the large-scale precipitation optimal climate mode. The main parameters of the model are set as follows: the number of decision trees is 10, the maximum depth of the decision tree is 3 layers, the sampling is replaced by sampling to construct a tree, and the rest parameters are set as default values.

Fourth, calculating a time coefficient corresponding to the 2019 summer large-scale optimal climate mode, such as a contemporaneous (2019) predictor time coefficient 507 in fig. 5, by a projection method based on the large-scale optimal climate mode obtained in the second step and the 2019 OLR for the actual prediction target and the high latitude Z500 anomaly relative trend field (such as the large-scale climate element anomaly relative trend field 506 predicted by the contemporaneous dynamic mode in fig. 5) calculated in the first step; the time coefficient is used as an actual prediction factor for calculating the relative trend field of the rainfall abnormality of the middle and lower river in the Yangtze river in summer in 2019, and is input into the downscale climate prediction model 57 obtained in the third step, so that an artificial intelligent downscale prediction result of the relative trend field of the rainfall abnormality of the middle and lower website in the Yangtze river in summer in 2019 can be obtained; for example, 2019 2374 in fig. 5 shows predicted result 508 of precipitation anomaly relative trend.

Fifthly, based on the near-term background abnormal tendency field of the rainfall in the middle and lower river basin in the middle of the Yankee river in the summer in the 2019 year (such as the near-term background abnormal tendency field 510 of the rainfall in the 2374 station in the 2019 year in fig. 5) calculated in the first step and the relative tendency field downscaling prediction result of the rainfall in the middle and lower river basin in the Yankee river in the summer in the 2019 year calculated in the fourth step, according to the corresponding relation in the first step, the prediction result of the rainfall distance flat field of the 2374 station in the middle and lower river in the Yankee river in the summer in the 2019 year can be finally obtained; such as the 2019 summer 2374 stop precipitation level 509 in fig. 5.

Sixth, after all calculation is completed, comparing the prediction result of the water fall distance between the middle and lower river in the summer of 2019 obtained by the prediction method provided by the embodiment of the application with the historical return result of the water fall distance between the middle and lower river in the first step of 2019 in the BCC_CSM1.1 (m) power mode, the spatial distribution is shown as in fig. 8, wherein the spatial distribution 801 represents the prediction result of the water fall distance between the middle and lower river in the summer of 2019 and the prediction result of the service mode of the water fall distance between the middle and lower river in the summer of 2019, and the spatial distribution 802 represents the prediction result of the scheme provided by the embodiment of the application, and it can be found that the artificial intelligent downscale climate prediction method based on the large-scale optimal climate mode in the middle and lower river in the summer of 2019 can grasp the integral characteristics of the precipitation more accurately compared with the power mode, and simultaneously, the prediction of a finer site, especially the regional area of the lower river in the middle and lower river in the middle of the summer is shown better prediction performance.

According to the embodiment of the application, a space-time coupling modal decomposition method is utilized, and a contemporaneous large-scale optimal climate mode and a time sequence corresponding to the abnormal relative tendency of the large-scale climate elements are extracted and determined based on the large-scale climate elements corresponding to the regional refined prediction target climate elements; training and constructing a downscaled prediction model of a nonlinear relationship between a large-scale optimal climate modal time sequence and a regional refined prediction target climate element abnormal relative tendency by using an artificial intelligent model; the contemporaneous large-scale optimal climate modal time coefficient predicted by the global climate dynamic mode is brought into a nonlinear downscaling prediction model, so that the prediction of the abnormal relative tendency of the regional refined climate elements is realized; and combining recent background abnormality of historical observation, and finally realizing artificial intelligent downscaling climate prediction for finely predicting the target climate element distance of the region. Compared with the downscaling climate prediction method in the related art, the embodiment of the application can fully utilize the nonlinear prediction capability of artificial intelligent modeling and the prediction capability of the global climate power mode on the downscaling climate mode according to the physical relationship between the downscaling climate elements corresponding to the upscaling optimal climate mode and the refined prediction target, and simultaneously integrates the advantages of various existing prediction schemes to establish an efficient and accurate downscaling climate prediction model, thereby effectively improving the regional refined climate prediction capability.

The embodiment of the application provides a climate prediction device, and fig. 9 is a schematic structural diagram of an artificial intelligence downscale climate prediction device based on a large-scale optimal climate mode. Illustratively, as shown in FIG. 9, the artificial intelligent downscale climate prediction device 900 based on the large-scale optimal climate mode comprises:

the first determining module 901 is configured to select a large-scale annular flow field for physical statistics construction based on a large-scale prediction target element corresponding to a downscaled climate prediction target element, and determine a corresponding climate abnormal field, an abnormal relative trend field and a recent background abnormal field respectively;

the first decomposition module 902 is configured to perform space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale predicted target element, so as to obtain a large-scale optimal climate mode of the abnormal relative trend field of the large-scale predicted target element and a time sequence corresponding to the large-scale optimal climate mode;

the training module 903 is configured to train the nonlinear prediction model to be trained by using the time sequence as a predictor and the abnormal relative trend field of the downscaled climate prediction target element as a prediction target, so as to obtain the nonlinear prediction model;

A second determining module 904, configured to determine a time coefficient corresponding to the contemporaneous large-scale annular flow field according to the large-scale optimal climate mode and the contemporaneous large-scale annular flow field, and import the time coefficient into the nonlinear prediction model to obtain a prediction result of the abnormal relative trend field of the downscaled climate prediction target element;

And a third determining module 905, configured to obtain a nonlinear quantitative prediction result of the downscaled climate prediction target element from a flat field based on a prediction result of a recent background abnormal field of the downscaled climate prediction target element and an abnormal relative trend field of the downscaled climate prediction target element.

In the device, the large-scale optimal climate mode is obtained by extracting the large-scale climate elements corresponding to the target elements and the large-scale circulation elements for determining the abnormality of the large-scale climate elements based on the downscale climate prediction by a space-time coupling decomposition method.

In the above device, the first determining module is further configured to add the anomaly relative trend field and a recent background anomaly field to obtain the climate anomaly field.

In the above apparatus, the first determining module is further configured to select the large-scale annular flow field for constructing a physical statistical relationship based on the large-scale prediction target element corresponding to the downscaled climate prediction target element by adopting a climate dynamics theory.

In the above device, the first decomposition module is further configured to perform singular value decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element, so as to obtain the large-scale optimal climate mode; projecting the abnormal relative trend field of the large-scale annular flow field to the large-scale optimal climate mode to obtain a time sequence corresponding to the large-scale optimal climate mode.

In the above device, the first decomposition module is further configured to perform singular value decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale prediction target element; and sequencing according to covariance corresponding to the singular value decomposition result to obtain the large-scale optimal climate mode.

In the above device, the second determining module is further configured to predict a circulation field corresponding to the downscaled climate prediction target element by using a global climate power mode, so as to obtain the contemporaneous large-scale circulation field; and projecting the contemporaneous large-scale annular flow field to the large-scale optimal climate mode to obtain a time coefficient corresponding to the contemporaneous large-scale annular flow field.

In the above device, the third determining module is further configured to add a recent background abnormal field of the downscaled climate forecast target element and an abnormal relative trend field of the downscaled climate forecast target element to obtain a nonlinear quantitative prediction result of the downscaled climate forecast target element from a flat field.

In the above device, the training module is further configured to obtain an artificial intelligence model; constructing a nonlinear prediction model to be trained comprising a plurality of decision trees based on the artificial intelligence model; wherein the artificial intelligence model comprises: random forest regression model.

In the above apparatus, the first determining module is further configured to uniformly divide any one of the anomaly fields of the climate element into an anomaly relative trend field and a recent background anomaly field.

It should be noted that: the artificial intelligence downscaling climate prediction device based on the large-scale optimal climate mode provided by the embodiment is only exemplified by the division of the functional modules, and in practical application, the functional allocation can be completed by different functional modules according to the needs, namely, the internal structure of the computer equipment is divided into different functional modules so as to complete all or part of the functions described above. In addition, the category predicting device and the category predicting method provided in the foregoing embodiments belong to the same concept, and detailed implementation processes of the category predicting device and the category predicting method are detailed in the method embodiments, which are not repeated herein.

The embodiment of the application also provides an electronic device, and fig. 10 is a schematic structural diagram of the electronic device provided by the embodiment of the application.

Illustratively, as shown in FIG. 10, the electronic device 1000 includes: memory 1001 and processor 1002, wherein, the executable program code 10011 is stored in memory 1001, and processor 1002 is used for calling and executing the executable program code 10011 and executing an artificial intelligence downscaled climate prediction method based on a large scale optimal climate mode.

In addition, the embodiment of the application also protects a device, which can comprise a memory and a processor, wherein executable program codes are stored in the memory, and the processor is used for calling and executing the executable program codes to execute the artificial intelligence downscaled climate prediction method based on the large-scale optimal climate mode.

In this embodiment, the functional modules of the apparatus may be divided according to the above method example, for example, each functional module may be corresponding to one processing module, or two or more functions may be integrated into one processing module, where the integrated modules may be implemented in a hardware form. It should be noted that, in the embodiment of the present application, the division of the modules is schematic, which is merely a logic function division, and other division manners may be implemented in actual implementation.

In the case of dividing the respective modules by the respective functions, the apparatus may further include a signal uploading module, a determining module, an adjusting module, and the like. It should be noted that, all relevant contents of each step related to the above method embodiment may be cited to the functional description of the corresponding functional module, which is not described herein.

It should be appreciated that the apparatus provided in this embodiment is configured to perform the above-described artificial intelligent downscaled climate prediction method based on a large-scale optimal climate mode, so that the same effects as those of the implementation method can be achieved.

In case of an integrated unit, the apparatus may comprise a processing module, a memory module. When the device is applied to equipment, the processing module can be used for controlling and managing the actions of the equipment. The memory module may be used to support devices executing inter-program code, etc.

Wherein a processing module may be a processor or controller that may implement or execute the various illustrative logical blocks, modules, and circuits described in connection with the present disclosure. A processor may also be a combination of computing functions, including for example one or more microprocessors, digital Signal Processing (DSP) and microprocessor combinations, etc., and a memory module may be a memory.

In addition, the device provided by the embodiment of the application can be a chip, a component or a module, wherein the chip can comprise a processor and a memory which are connected; the memory is used for storing instructions, and when the processor calls and executes the instructions, the chip can be made to execute the artificial intelligent downscaled climate prediction method based on the large-scale optimal climate mode.

The present embodiment also provides a computer-readable storage medium, in which a computer program code is stored, which when run on a computer causes the computer to execute the above-mentioned related method steps to implement an artificial intelligent downscaled climate prediction method based on a large-scale optimal climate mode provided in the above-mentioned embodiment.

The present embodiment also provides a computer program product, which when run on a computer, causes the computer to perform the above-mentioned related steps to implement an artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode provided by the above-mentioned embodiment.

The apparatus, the computer readable storage medium, the computer program product, or the chip provided in this embodiment are used to execute the corresponding method provided above, and therefore, the advantages achieved by the apparatus, the computer readable storage medium, the computer program product, or the chip can refer to the advantages of the corresponding method provided above, which are not described herein.

It will be appreciated by those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional modules is illustrated, and in practical application, the above-described functional allocation may be performed by different functional modules according to needs, i.e. the internal structure of the apparatus is divided into different functional modules to perform all or part of the functions described above.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of modules or units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another apparatus, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application. Therefore, the protection scope of the application is subject to the protection scope of the claims.

Claims

1. An artificial intelligence downscale climate prediction method based on a large-scale optimal climate mode, which is characterized by comprising the following steps:

taking the time sequence as a prediction factor, taking the abnormal relative trend field of the downscaled climate prediction target element as a prediction target, and training an artificial intelligence prediction model to be trained to obtain a nonlinear prediction model based on an artificial intelligence method;

Determining a time coefficient corresponding to the synchronous large-scale annular flow field according to the large-scale optimal climate mode and the synchronous large-scale annular flow field, and importing the time coefficient into the artificial intelligent prediction model to obtain a prediction result of the abnormal relative trend field of the downscaled climate prediction target element;

And obtaining a nonlinear quantitative artificial intelligent prediction result of the downscaled climate prediction target element from a flat field based on the prediction results of the recent background abnormal field of the downscaled climate prediction target element and the abnormal relative trend field of the downscaled climate prediction target element.

2. The method according to claim 1, wherein the large-scale optimal climate mode is obtained by extracting a large-scale climate element corresponding to the target element and a large-scale circulation element for determining abnormality of the large-scale climate element based on the downscaling climate prediction by a space-time coupling decomposition method.

3. The method according to claim 1, wherein the method further comprises:

And adding the abnormal relative trend field and the recent background abnormal field to obtain the climate abnormal field.

4. The method of claim 1, wherein the selecting a large scale toroidal flow field for physical statistical relationship construction based on the large scale predicted target elements corresponding to the downscaled climate predicted target elements comprises:

5. The method according to claim 1, wherein the performing space-time coupling decomposition on the abnormal relative trend field of the large-scale annular flow field and the abnormal relative trend field of the large-scale predicted target element to obtain a time sequence corresponding to a large-scale optimal climate mode of the abnormal relative trend field of the large-scale predicted target element and the large-scale optimal climate mode includes:

6. The method of claim 5, wherein performing singular value decomposition on the outlier relative trend field of the large scale annular flow field and the outlier relative trend field of the large scale predicted target element to obtain the large scale optimal climate modality comprises:

7. The method of claim 1, wherein determining the time coefficient corresponding to the contemporaneous large scale annular flow field according to the large scale optimal climate modality and contemporaneous large scale annular flow field comprises:

8. The method of claim 1, wherein the obtaining a nonlinear quantitative prediction of the downscaled climate predicted target element from a flat field based on a prediction of a recent background anomaly field of the downscaled climate predicted target element and an anomaly relative trend field of the downscaled climate predicted target element comprises:

9. The method according to claim 1, wherein the training the nonlinear prediction model to be trained by using the time sequence as a predictor and the abnormal relative trend field of the downscaled climate prediction target element as a prediction target includes:

acquiring an artificial intelligent model;

10. The method according to claim 1, wherein the method further comprises: