CN117932360A

CN117932360A - Artificial intelligence sub-season prediction method based on optimal climate mode

Info

Publication number: CN117932360A
Application number: CN202410321404.4A
Authority: CN
Inventors: 杨修群; 陶凌峰; 孙旭光; 房佳蓓; 王昱; 张志琦; 张昱培
Original assignee: Nanjing University
Current assignee: Nanjing University
Priority date: 2024-03-20
Filing date: 2024-03-20
Publication date: 2024-04-26

Abstract

The application provides an artificial intelligence sub-season prediction method based on an optimal climate mode, which comprises the following steps: the method comprises the steps of processing data of an OLR field of a tropical region day by day, a 500hPa potential height field of a middle and high latitude region and each meteorological element field in observation to respectively obtain a corresponding weather trend distance flat field; selecting a climate mode which is most closely related to a meteorological element field, and establishing a nonlinear prediction model between the climate mode and a meteorological element climate tendency distance flat field by an artificial intelligence method; and selecting an optimal climate mode and a corresponding prediction model from the optimal climate mode to realize the prediction of future meteorological elements. The application can effectively identify the complex nonlinear relation between the climate mode and the meteorological element on the sub-season scale by using the artificial intelligence method, thereby establishing a nonlinear prediction model between the climate mode and the meteorological element. These ensure the accuracy of the prediction of meteorological elements by artificial intelligence methods through optimal climate modalities.

Description

Artificial intelligence sub-season prediction method based on optimal climate mode

Technical Field

The application relates to the field of sub-season prediction of meteorological elements such as precipitation and air temperature in weather climate prediction business, and in particular relates to an artificial intelligence sub-season prediction method based on an optimal climate mode.

Background

Weather and climate prediction has important significance for weather disaster prevention and reduction, economic and social development and national security. How to improve the prediction accuracy of meteorological elements such as precipitation, air temperature and the like is a difficult problem of current meteorological prediction, and is an important task to be solved urgently in meteorological scientific research. The main weather and climate prediction technology at present mainly comprises two types, namely statistical method prediction and dynamic mode prediction. Whether statistical or dynamic mode prediction, there are currently serious shortcomings to the estimation of the nonlinear relationship between naturally occurring climate modalities and meteorological elements.

For statistical method prediction, a physical statistical relationship is searched mainly based on a historical evolution rule of a strong atmospheric signal, and then a statistical model is established for prediction. The physical relationship based on the statistical prediction model at the present stage is mainly a simple linear relationship; however, the atmosphere has strong nonlinear characteristics, and the simple statistical linear relation cannot accurately describe the climate modal change and the change of meteorological element fields such as precipitation and air temperature mainly determined by the climate modal change. For numerical predictive power models, they describe various motions and interactions in the atmosphere based primarily on physical equations, such as kinetic equations, thermodynamic equations, and the like. The numerical forecasting model divides the atmosphere into three-dimensional grids, and the evolution condition of the atmosphere state is simulated by solving an equation on each grid point. When the numerical forecasting model operates, the initial condition which is consistent with the actual atmospheric state as much as possible is used for calculating the atmospheric change in a time stepping mode, and the model outputs atmospheric state parameters including temperature, humidity, wind speed and the like at the future moment. However, the prediction result output by the numerical prediction model often has a certain deviation, and needs to be adjusted and corrected after verification with the actual observation condition. Especially, the result of the direct output of the numerical forecasting model is insufficient for forecasting meteorological elements such as air temperature, precipitation and the like. The existing adjustment correction method is mainly based on observation statistics of linear relations, and cannot well consider the nonlinear relations which exist naturally.

Therefore, how to better identify the complex relationship between the climate modality and the meteorological element is a problem that needs to be solved to improve the accuracy of meteorological element prediction.

Disclosure of Invention

The application provides an artificial intelligence sub-season prediction method based on an optimal climate mode.

In a first aspect, an artificial intelligence sub-season prediction method based on an optimal climate modality is provided, the method comprising:

Respectively calculating and converting an OLR field of a tropical region day by day, a 500hPa potential height field of a middle and high latitude region and each meteorological element field in observation into a weather trend distance flat field taking each day as a center to obtain a weather trend distance flat field of an early-stage or contemporaneous OLR field and the 500hPa potential height field and a weather trend distance flat field of each meteorological element;

Performing singular value decomposition on the pre-period or contemporaneous OLR field and the 500hPa potential height field, and the weather trend distance flat field of each meteorological element respectively to obtain an OLR field and a 500hPa potential height field which are the pre-period or contemporaneous weather modes most closely related to the meteorological elements, and normalizing the time sequences of the OLR and the 500hPa potential height fields obtained by decomposition to obtain normalized time sequences;

combining the standardized time sequences, and establishing a nonlinear prediction model between the early-stage or contemporaneous weather modes and the weather tendency pitch level of each weather element by using an LSTM or ANN artificial intelligence method;

Carrying out statistics-artificial intelligence combined prediction on future meteorological elements by utilizing the nonlinear prediction model between the early-stage climate mode and the meteorological elements, and carrying out power-statistics-artificial intelligence combined prediction on the future meteorological elements by utilizing the nonlinear prediction model between the contemporaneous climate mode and the meteorological elements and the climate mode predicted by utilizing the power mode;

performing independent sample inspection on the historical return of the nonlinear prediction model with a flat weather tendency, and determining an optimal weather mode and a corresponding statistical-artificial intelligent combined prediction model and dynamic-statistical-artificial intelligent combined prediction model;

And predicting the future meteorological elements by using the optimal climate mode and the corresponding statistical-artificial intelligence combined prediction model and the dynamic-statistical-artificial intelligence combined prediction model.

Compared with the prior art, the technical scheme has the following advantages:

(1) According to the application, the climate mode and the meteorological element are subjected to nonlinear modeling by an artificial intelligence method. Unlike traditional linear methods, artificial intelligence methods can effectively identify nonlinear relationships between climate modes and meteorological elements, which ensure the accuracy of the meteorological elements derived by the nonlinear predictive model using the climate modes.

(2) The application also converts direct prediction of meteorological elements into indirect prediction of weather trend range flatness. The method has the advantages that the sub-seasonal scale signal can be extracted better, and longer-time change is taken as a continuous background to be introduced through the last horizon in observation, so that the long-time change part does not need to be predicted, and the prediction accuracy is improved effectively.

(3) According to the method, weather-by-weather modeling is performed on the daily data sliding, so that weather element prediction results of each day can be given.

With reference to the first aspect, in some possible implementations, the singular value decomposition is performed on the pre-OLR field and the 500hPa potential height field from the flat field and the weather trend distance of each meteorological element, respectively, the obtained OLR field and the 500hPa potential height field are the pre-weather modes most closely related to the meteorological element, and the time sequence of the OLR and the 500hPa potential height field obtained by the decomposition is normalized, so as to obtain normalized time sequences, including: extracting the OLR tendency distance flat field from the t-n-1 to the t-n+1 in each year and the 500hPa potential height tendency distance flat field when predicting the t-th in advance, and the weather trend distance flat field of the weather elements from the t-1 th to the t+1 th weather, obtaining an extended OLR weather trend distance flat field, a 500hPa potential height weather trend distance flat field and a weather trend distance flat field of the weather elements; performing singular value decomposition on the extended OLR weather trend-to-flat field, the 500hPa potential height weather trend-to-flat field and the weather trend-to-flat field of the meteorological elements respectively, and determining the first P OLR modes and the first P500 hPa potential height modes obtained by decomposition as the early weather modes most closely related to the meteorological elements, wherein P is a positive integer;

and respectively normalizing the time sequences corresponding to the first P OLR modes and the first P500 hPa potential height modes obtained through decomposition to obtain each time sequence after normalization.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the performing, with the nonlinear prediction model between the advanced climate mode and the meteorological element, statistical-artificial intelligence combined prediction on a future meteorological element includes: obtaining the weather trend distance of the observation weather mode; obtaining a time coefficient corresponding to the weather trend distance level of the observation climate mode based on the weather trend distance level of the observation climate mode and the early-stage climate mode; substituting the time coefficient into the nonlinear prediction model to predict the future meteorological elements.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the performing a historical return independent sample test of a weather tendency range for the nonlinear prediction model to determine an optimal weather mode and a corresponding statistical-artificial intelligence combined prediction model includes: obtaining a history weather trend distance of the recent year I, wherein I is a positive integer; obtaining the return results of the plurality of nonlinear predictive models on the historical weather tendency distance level, wherein the return results indicate weather element prediction results of the nonlinear predictive models based on the historical weather tendency distance level; comparing the return results of the nonlinear predictive models with the spatial correlation coefficients of the observed data; based on the sequencing result of the space correlation coefficient from large to small, selecting weather mode combinations corresponding to the first M OLR field time sequences and the first N500 hPa potential height field time sequences as the optimal weather modes, and determining a nonlinear prediction model corresponding to the optimal weather modes as the statistical-artificial intelligent combined prediction model, wherein M, N is a positive integer.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the predicting the future meteorological element by using the optimal climate mode and the corresponding statistical-artificial intelligence combined prediction model includes: obtaining a corresponding time coefficient projected onto the optimal climate mode based on an OLR (on-line road) tendency distance flat field of the t-n th day of the predicted year and the optimal climate mode, wherein n is a positive integer, and t-n is a positive integer; substituting the time coefficient corresponding to the projection to the optimal climate mode into the statistical-artificial intelligent combined prediction model to obtain the weather trend distance of the t-th weather of the predicted year; adding the weather trend pitch plane of the t-th weather and the weather trend pitch plane of the t-1 th weather observed in the previous weather to obtain the weather trend plane of the t-th weather; and adding the weather distance of the t-th weather and the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the performing singular value decomposition on the candidate moment flat field of the contemporaneous OLR field and the 500hPa potential height field and the candidate moment flat field of each meteorological element respectively, where the obtained OLR field and the 500hPa potential height field are contemporaneous climate modes most closely related to the meteorological element, and normalizing a time sequence of the decomposed OLR field and the decomposed 500hPa potential height field to obtain normalized time sequences, where the normalized time sequence includes: extracting an OLR (on-line road) weather trend distance flat field from the t-1 th to the t+1 th and a weather trend distance flat field with a potential height of 500hPa, and a weather trend distance flat field of weather elements from the t-1 th to the t+1 th from each year to obtain an extended OLR weather trend distance flat field, a weather trend distance flat field with a potential height of 500hPa and a weather trend distance flat field of the weather elements, wherein t-1 is a positive integer; respectively carrying out singular value decomposition on the extended OLR weather trend-to-flat field, the 500hPa potential height weather trend-to-flat field and the weather trend-to-flat field of the meteorological elements, and determining the first Q OLR modes and the first Q500 hPa potential height modes obtained by decomposition as the contemporaneous weather modes most closely related to the meteorological elements, wherein Q is a positive integer; and respectively normalizing the time sequences corresponding to the first Q OLR modes and the first Q500 hPa potential height modes obtained through decomposition to obtain each time sequence after normalization.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the performing power-statistics-artificial intelligence combined prediction on the future meteorological element by using the nonlinear prediction model and the climate mode predicted between the contemporaneous climate mode and the meteorological element includes: obtaining a predicted output result of the power mode on a future climate mode; obtaining a time coefficient corresponding to the contemporaneous climate mode projected on the basis of the prediction output result and the contemporaneous climate mode; substituting the time coefficient corresponding to the contemporaneous weather mode projected into the nonlinear prediction model to predict the future meteorological elements.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the performing a historical return independent sample test of weather tendency distance leveling on the nonlinear prediction model, determining an optimal weather mode and a corresponding power-statistics-artificial intelligence combined prediction model includes: obtaining a predicted weather trend distance of a power mode for nearly J years, wherein J is a positive integer; obtaining the return results of a plurality of nonlinear predictive models on the predicted weather trend distance level, wherein the return results indicate weather element prediction results of the nonlinear predictive models based on the predicted weather trend distance level; comparing the return results of the nonlinear prediction models with the spatial correlation coefficients of the observed data; and selecting a climate mode combination corresponding to the first X OLR field time sequences and the first Y500 hPa potential height field time sequences as the optimal climate modes based on the sequencing result of the space correlation coefficients from large to small, determining the nonlinear prediction model corresponding to the optimal climate modes as the dynamic-statistics-artificial intelligence combined prediction model, wherein X and Y are positive integers.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the predicting the future meteorological element by using the optimal climate mode and the corresponding power-statistics-artificial intelligence combined prediction model includes: projecting the predicted future atmospheric OLR weather trend distance flat field and the predicted 500hPa potential height weather trend distance flat field of the power mode onto the optimal climate mode to obtain a corresponding time coefficient projected onto the optimal climate mode; substituting the time coefficient corresponding to the projection on the optimal climate mode into the power-statistics-artificial intelligence combined prediction model, and calculating to obtain the weather trend range of the future t weather of the meteorological element; adding the weather trend pitch plane of the t-th weather and the weather trend pitch plane of the t-1 th weather observed in the previous weather to obtain the weather trend plane of the t-th weather; and adding the weather distance level of the t-th weather with the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

With reference to the first aspect and the foregoing implementation manner, in some possible implementation manners, the candidate pitch-flat operation process includes:

wherein, Is a climate modal variable or the meteorological element,/>For the climate modal variable or the meteorological element/>Is equal to the distance between (v)/(v)For the climate modal variable or the meteorological element/>The candidate trend is flat, t is a certain candidate, and t-1 is the last candidate of the certain candidate.

Drawings

FIG. 1 is a flow chart of an artificial intelligence sub-season prediction method based on an optimal climate mode applied to statistical-artificial intelligence combined prediction according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a process of applying the artificial intelligence sub-season prediction method based on the optimal climate mode to the combined prediction of statistics and artificial intelligence according to the embodiment of the application;

FIG. 3 is a flowchart of an artificial intelligence sub-season prediction method based on an optimal climate mode applied to power-statistics-artificial intelligence combined prediction according to an embodiment of the present application;

FIG. 4 is a schematic diagram of a process of applying the artificial intelligence sub-season prediction method based on the optimal climate mode to power-statistics-artificial intelligence combined prediction provided by the embodiment of the application;

FIG. 5 is a flow chart of weather element prediction using an artificial intelligence sub-season prediction method based on optimal climate modalities provided by an embodiment of the present application;

fig. 6 is a schematic structural diagram of a computer system 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.

FIG. 1 is a flowchart of an artificial intelligence sub-season prediction method based on an optimal climate mode applied to statistical-artificial intelligence combined prediction according to an embodiment of the present application. As shown in fig. 1, the artificial intelligence sub-season prediction method (statistical-artificial intelligence combined prediction) based on the optimal climate mode comprises the following steps:

And step 101, preprocessing data.

The method comprises the steps of respectively calculating and converting an OLR (atmospheric outward long wave radiation) field, a 500hPa potential height field and each meteorological element field (such as air temperature and precipitation) in the tropical region of the observation data into weather averages centered on respective days, respectively removing weather states to obtain weather distance levels, and then calculating the difference between the weather average and the weather distance level of the last weather to obtain the weather trend level of the weather.

The specific calculation formula of the weather trend distance is as follows:

wherein, Is a climate modal variable or meteorological element,/>For variables or meteorological elements/>Is equal to the distance between (v)/(v)The weather trend distance of the variable A is flat, t is a certain weather, and t-1 is the last weather of the certain weather.

And 102, observing the extraction of the physical relationship between the early-stage climate mode and the meteorological element.

Step 102A, assuming that the nth weather is predicted in advance of n, expanding the OLR weather trend of the nth weather from the flat field, the 500hPa potential altitude field and the weather element weather trend of the nth weather from the flat field of the data of the last year, which are obtained in step 101, of the last 10 years in total from the last 5 years to the last 15 years, forward and backward on the basis of the respective weather factors, namely extracting the OLR weather trend of the nth weather from the nth weather to the (t-n+1) weather from the flat field and the weather element weather trend of the 500hPa potential altitude weather from the flat field and the (t-1) weather elements from the (t+1) weather from the flat field of each year for 30 days.

Step 102B, performing singular value decomposition (Single Value Decomposition, SVD) decomposition on the extended OLR candidate trend-to-flat field and the 500hPa potential height candidate trend-to-flat field and the meteorological element candidate trend-to-flat field obtained in step 102A, wherein the obtained OLR and 500hPa potential height fields are the earlier stage climate modes most closely related to the meteorological element, and respectively normalizing the time sequences corresponding to the first 20 OLR modes and the first 20 500hPa potential height modes obtained by decomposition.

And 103, building a nonlinear prediction model.

And (3) selecting different numbers of time sequences in the step 102B for combination, and establishing a nonlinear prediction model of the time sequences with a plurality of candidate moment ordinary times by using artificial intelligence methods such as LSTM, ANN and the like.

The nonlinear prediction model is used for predicting the nonlinear relation between the early-stage climate mode and the meteorological elements.

Step 104, selecting an optimal climate mode and establishing a statistical-artificial intelligence combined prediction model.

Step 104A projects the spatial field of the early climate mode in step 102B with the weathered moment of the observed climate mode. And substituting the obtained time sequence into a nonlinear prediction model so as to predict the future meteorological elements.

Step 104B, carrying out the return of the weather trend distance level in the last 5 years, comparing the return results of a plurality of nonlinear prediction models with the spatial correlation coefficient of the observed data, taking the weather mode combinations corresponding to the M OLR field time sequences and the N500 hPa potential altitude field time sequences with larger phase relation numbers as optimal weather mode combinations, and obtaining the corresponding nonlinear prediction model which is the final statistical-artificial intelligent combined prediction model.

Step 105, using the statistical-artificial intelligence combined prediction model to make predictions.

Step 105A, projecting the t-n weather OLR weather trend distance flat field and the 500hPa potential height weather trend distance flat field of the predicted year onto the optimal weather modal space field in step 104B to obtain corresponding time coefficients.

And 105B, substituting the time coefficient obtained in the step 105A into a statistical-artificial intelligence combined prediction model, and calculating to obtain the future weather trend distance of the meteorological element (namely the t weather).

If the weather elements after the t-th weather element is required to be predicted, the future weather trend distance outputted by the statistical-artificial intelligence combined prediction model is flat, namely the weather trend distance of the weather elements at the t-th weather is flat.

Step 105C, adding the candidate moment plane of the t-th candidate obtained in step 105B to the candidate moment plane of the t-1 th candidate observed in the previous step to obtain the candidate moment plane of the t-th candidate. Further, the weather distance level of the t-th weather is added with the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

And carrying out daily sliding modeling prediction by utilizing the steps, so as to obtain the prediction result of the weather element statistics-artificial intelligence combined prediction model of each day.

Fig. 2 is a schematic diagram of a process of applying the artificial intelligence sub-season prediction method based on the optimal climate mode to statistical-artificial intelligence combined prediction according to the embodiment of the application. As shown in fig. 2, in the data preprocessing stage, acquiring a front-stage tropical OLR field, a predicted variable and a front-stage Gao Wei hPa potential height field, and respectively acquiring corresponding candidate trend range levels to obtain a front-stage tropical OLR candidate trend range level, a predicted variable candidate trend range level and a front-stage Gao Wei hPa potential height candidate trend range level; in the stage of extracting the early climate modes, singular value decomposition is carried out on the early tropical OLR weather trend pitch level and the predicted variable weather trend pitch level to obtain the first M OLR modes obtained by decomposition, singular value decomposition is carried out on the early Gao Wei hPa potential height weather trend pitch level and the predicted variable weather trend pitch level to obtain the first N potential height modes obtained by decomposition, and the first M OLR modes and the first N potential height modes obtained by decomposition are determined to be the early climate modes with a relatively tight relation with the predicted variables. In an artificial intelligence modeling stage, a plurality of nonlinear prediction models between an early-stage climate mode and a prediction variable are established, a weather trend range of observation data is adopted, historical return inspection is carried out on the plurality of nonlinear prediction models, so that an optimal climate mode is selected from the plurality of nonlinear prediction models, and the nonlinear prediction model corresponding to the optimal climate mode is used as a final statistical-artificial intelligence combined prediction model; outputting a result in the prediction result: if the predicted variable of the t-th day is predicted, obtaining the OLR (on-line road) weather trend horizon of the n-th day in advance and the middle-high latitude 500hPa potential height weather trend horizon of the n-th day in advance, substituting the OLR weather trend horizon into a statistical-artificial intelligent combined prediction model, outputting Hou Qingxiang horizon of the predicted variable at the t-th day, and further deducing the weather trend horizon and weather average value of the predicted variable at the t-th day according to a calculation formula of the weather trend horizon.

FIG. 3 is a flowchart of an artificial intelligence sub-season prediction method based on an optimal climate mode applied to power-statistics-artificial intelligence combined prediction according to an embodiment of the present application. As shown in fig. 3, the artificial intelligence sub-season prediction method (power-statistics-artificial intelligence combined prediction) based on the optimal climate mode comprises the following steps:

Step 301, preprocessing of data.

The method comprises the steps of respectively calculating and converting an OLR field, a 500hPa potential height field and each meteorological element field (such as air temperature and precipitation) of a tropical region day by day in observation data into weather averages centered on respective days, respectively removing weather states to obtain weather distance levels, and then calculating the difference between the weather mean and the weather distance level of the previous weather to obtain the weather trend distance level of the weather.

The specific calculation formula of the weather trend distance is as follows:

Step 302, observe the extraction of the physical relationship between the contemporaneous climate mode and the meteorological element.

Step 302A, expanding the OLR tendency of the t-th weather from the flat field, the 500hPa potential height field and the weather element tendency of the t-th weather from the flat field of each year of about 5 years to about 15 years, which are obtained in step 301, one weather is extended forward and backward on the basis of each weather, namely, extracting the OLR tendency of the t-1 th weather to the t+1 th weather from the flat field and the weather element tendency of the 500hPa potential height from the flat field of each year, and the weather element tendency of the t-1 weather to the t+1 weather from the flat field of each year, which are 30 weather in total.

And 302B, respectively performing SVD decomposition on the extended OLR weather trend distance flat field and the 500hPa potential height weather trend distance flat field obtained in the 302A and the weather element weather trend distance flat field, wherein the obtained OLR and 500hPa potential height fields are synchronous weather modes which are in closest connection with weather elements, and respectively standardizing time sequences corresponding to the first 20 OLR modes and the first 20 500hPa potential height modes obtained by decomposition.

And step 303, establishing a nonlinear prediction model.

And (3) selecting different numbers of time sequences in the step 302B for combination, and establishing a nonlinear prediction model of a plurality of time sequences with a waiting duration by using artificial intelligence methods such as LSTM, ANN and the like.

Step 304, selecting an optimal climate mode and establishing a power-statistics-artificial intelligence combined prediction model.

Step 304A, the predicted output of the future climate mode using the power mode is projected onto the spatial field of the contemporaneous climate mode in step 302B. And substituting the obtained time sequence into a nonlinear prediction model so as to predict the future meteorological elements.

Step 304B, reporting the weather trend distance in the last 5 years, comparing the spatial correlation coefficients of the reporting results and the observed data of the nonlinear prediction models, and taking the weather mode combinations corresponding to the M OLR field time sequences and the N500 hPa potential altitude field time sequences with the large phase relation number as the optimal weather mode combinations, wherein the corresponding nonlinear prediction model is the final power-statistics-artificial intelligence combined prediction model.

Step 305, using the power-statistics-artificial intelligence in combination with the prediction model to make predictions.

Step 305A, projecting the predicted future atmospheric OLR weather trend distance flat field and the predicted 500hPa potential height weather trend distance flat field onto the space field of the optimal weather mode in step 304B, to obtain the corresponding time coefficient.

And 305B, substituting the time coefficient obtained in the step 305A into a power-statistics-artificial intelligence combined prediction model, and calculating to obtain the future weather tendency range of the meteorological elements.

If the weather elements after the t-th weather element is required to be predicted, the future weather trend distance outputted by combining the power-statistics-artificial intelligence with the prediction model is flat, namely the weather trend distance of the weather elements at the t-th weather is flat.

Step 305C, adding the candidate moment plane of the t-th candidate obtained in step 305B to the candidate moment plane of the t-1 th candidate observed in the previous step to obtain the candidate moment plane of the t-th candidate. Further, the weather distance level of the t-th weather is added with the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

And carrying out daily sliding modeling prediction by utilizing the steps, so as to obtain the weather element power-statistics-artificial intelligence prediction result of each day.

Fig. 4 is a schematic diagram of a process of applying the artificial intelligence sub-season prediction method based on the optimal climate mode to power-statistics-artificial intelligence combined prediction according to the embodiment of the application. As shown in fig. 4, in the data preprocessing stage, an observation synchronous tropical OLR field, a prediction variable and an observation synchronous high latitude 500hPa potential height field are acquired, and the corresponding weather trend range is acquired respectively to obtain an observation synchronous tropical OLR weather trend range, a prediction variable weather trend range and an observation synchronous high latitude 500hPa potential height weather trend range; in the stage of extracting contemporaneous climate modes, singular value decomposition is carried out on the observed contemporaneous tropical OLR weather trend pitch level and the predicted variable weather trend pitch level to obtain first M OLR modes obtained by decomposition, singular value decomposition is carried out on the observed contemporaneous high latitude 500hPa potential height weather trend pitch level and the predicted variable weather trend pitch level to obtain first N potential height modes obtained by decomposition, and the first M OLR modes and the first N potential height modes obtained by decomposition are determined to be contemporaneous climate modes with a compact relation with the predicted variable. In an artificial intelligence modeling stage, a plurality of nonlinear prediction models between contemporaneous climate modes and prediction variables are established, OLR (on-line road) weather trend pitch return and middle-high latitude 500hPa potential height weather trend pitch return predicted by adopting a power mode are adopted, return checking is carried out on the plurality of nonlinear prediction models, so that an optimal climate mode is selected from the plurality of nonlinear prediction models, and the nonlinear prediction model corresponding to the optimal climate mode is used as a final power-statistics-artificial intelligence combined prediction model; outputting a result in the prediction result: if the predicted variable of the t-th day is predicted, the OLR weather trend pitch level and the middle-high latitude 500hPa potential height weather trend pitch level obtained by the power mode prediction of the t-th day are obtained, the power-statistics-artificial intelligence combined prediction model is substituted, the Hou Qingxiang pitch level of the predicted variable at the t-th day is output, and then the weather average value of the predicted variable at the t-th day is deduced according to a calculation formula of the weather trend pitch level.

Application example

The following is to carry out power-statistics-artificial intelligence combined prediction on precipitation and air temperature after No. 15 of 6 months of 2023 by an artificial intelligence sub-season prediction method based on an optimal climate mode, and the main prediction process comprises the following steps:

1. Preprocessing data.

And acquiring an OLR field in a tropical region from 2009 to 2023, calculating weather averages for respective day-to-day data by taking each day as a center, respectively removing weather states to obtain a weather flat, and calculating the difference between the weather flat and the weather flat of the last day to obtain the weather tendency flat of the day.

2. And (5) observing the extraction of the physical relationship.

Taking the physical relationship between the weather mode of 6 months 15 and the precipitation field as an example, expanding the OLR weather trend distance flat field and the precipitation weather trend distance flat field corresponding to 6 months 15 in each year from 2009 to 2018 obtained in the step 1 and the height weather trend distance flat field of 500hPa potential and the precipitation weather trend distance flat field forward and backward on the basis of the weather, namely extracting the OLR weather trend distance flat field and the 500hPa weather trend distance flat field of 6 months 1 to 6 months 30 in each year, and respectively taking 30 weather angles.

2.2, Respectively carrying out SVD decomposition on the extended OLR weather trend distance flat field and the 500hPa potential height weather trend distance flat field obtained in the step 2.1 and the precipitation weather trend distance flat field to obtain the first 20 OLR modes and the first 20 500hPa potential height modes which are the contemporaneous weather modes most closely related to the weather elements, and respectively standardizing time sequences corresponding to the first 20 OLR modes and the first 20 500hPa potential height modes obtained by decomposition.

3. And (6) establishing a nonlinear prediction model.

And (3) selecting different numbers of time sequences in the step (2.2) for combination, and establishing a plurality of nonlinear prediction models of the rainfall tendency by using an LSTM artificial intelligence method.

4. And (3) selecting an optimal climate mode and establishing a power-statistics-artificial intelligence combined prediction model.

4.1, Projecting a predicted output result of a future climate mode by using a power mode closest to No. 15 of 6 months onto a space field of the contemporaneous climate mode in the step 2.2 to obtain a time sequence; and substituting the obtained time sequence into the nonlinear prediction model in the step 3, and predicting the future rainfall tendency pitch.

4.2, Carrying out the return of the weather trend distance level in the last 5 years, comparing the space correlation coefficient between the weather trend distance level return results and the observation condition of a plurality of nonlinear prediction models, taking weather mode combinations corresponding to the first M OLR field time sequences and the first N500 hPa potential altitude field time sequences with larger phase relation numbers as optimal weather mode combinations, wherein the nonlinear prediction model corresponding to the optimal weather mode (or the optimal weather mode combination) is the final power-statistics-artificial intelligence combined prediction model.

5. Prediction was performed using a dynamic-statistical-artificial intelligence combined prediction model.

And 5.1, projecting the predicted future atmospheric OLR weather trend distance flat field and the predicted 500hPa potential height weather trend distance flat field result of the power mode closest to the day of 6 and 15 of 2023 onto the space field of the optimal weather mode in the step 4.2 to obtain the corresponding time coefficient.

And 5.2, substituting the time coefficient obtained in the step 5.1 into a power-statistics-artificial intelligence combined prediction model, and calculating to obtain the future weather tendency range of precipitation.

And 5.3, adding the weather trend flat of the t-th weather obtained in the step 5.2 with the weather flat of the t-1 th weather observed in the last weather to obtain the weather flat of the t-th weather. Further, the weather distance level of the t-th weather is added with the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

And carrying out daily sliding modeling prediction by utilizing the steps, so as to obtain weather element prediction results of each day of 6 months. The maximum predicted time depends on the length of the power mode output predicted time.

In this embodiment, the climate mode and the meteorological element are modeled non-linearly by an artificial intelligence method. Unlike traditional linear methods, the artificial intelligence method can effectively identify nonlinear relationships between climate modes and meteorological elements, and accuracy of the meteorological elements obtained by the nonlinear prediction model through the climate modes is guaranteed. The direct prediction of the meteorological elements is converted into the indirect prediction of the weather trend range flatness. The advantage of this approach is that sub-seasonal scale signals can be extracted better, introducing longer-term changes as a persistent background through the last horizon in the observation. Therefore, prediction is not needed for the long-time change part, so that the prediction accuracy is effectively improved. In addition, the weather element prediction result of each day can be given through sliding the weather-by-weather nonlinear modeling on the day-by-day data.

FIG. 5 is a flow chart of weather element prediction using an artificial intelligence sub-season prediction method based on optimal climate modalities, the method 500 comprising:

Step 501, respectively calculating and converting an OLR field of a tropical region day by day, a 500hPa potential height field of a middle and high latitude region and each meteorological element field in observation into a weather trend distance flat field taking each day as a center, and obtaining a weather trend distance flat field of an early-stage or contemporaneous OLR field and the 500hPa potential height field and a weather trend distance flat field of each meteorological element.

The calculation process of the candidate trend distance flat (or the candidate trend distance flat field) can be as follows:

wherein, Is a climate modal variable or meteorological element,/>Is a climate modal variable or meteorological element/>Is characterized in that the pitch of the pattern is flat,Is a climate modal variable or meteorological element/>The candidate trend distance of (1) is flat, t is a certain candidate, and t-1 is the last candidate of a certain candidate.

To be used forFor precipitation as a meteorological element, for example,/>The weather average of the t weather is obtained by taking average values of precipitation in the t weather for 5 days continuously, for example, when calculating the weather average of the weather of 3 months 16, respectively obtaining precipitation amounts of 3 months 14, 3 months 15, 3 months 16, 3 months 17 and 3 months 18 every day, and taking average values to obtain the weather average of the precipitation amount of the weather of 3 months 16; /(I)Representing the climatic state of precipitation at the t-th day, subtracting the climatic state at the t-th day from the average of the weather at the t-th day to obtain the weather distance at the t-th day, i-The difference between the horizon of the precipitation at the t-1 th day and the horizon of the t-1 th day is the horizon of the t-1 th day.

According to the calculation method, the OLR field weather trend flat of each day, the weather trend flat of the 500hPa potential height field of each day and the weather factor weather trend flat of each day can be obtained.

Note that the term "flat" and the term "flat" mean the same meaning as each other.

Step 502, performing singular value decomposition on the pre-period or contemporaneous OLR field and the 500hPa potential height field, and the pre-period or contemporaneous climate mode in which the obtained OLR field and the 500hPa potential height field are most closely related to the meteorological elements, and normalizing the time series of the decomposed OLR field and the decomposed 500hPa potential height field to obtain normalized time series.

The weather trend pitch flat data corresponding to the observed data obtained by calculation in the step 501 can be respectively used for the prediction of the early-stage weather mode and the extraction of the contemporaneous weather mode, and the weather trend pitch flat data selected in the step 501 have differences when the weather mode is extracted.

Taking weather elements for predicting the t weather as an example, when the early weather mode is extracted, the weather trend pitch flat data of the early observation data is selected from the weather trend pitch flat data corresponding to the observation data, and when the contemporaneous weather mode is extracted, the weather trend pitch flat data of the contemporaneous observation data is selected from the weather trend pitch flat data corresponding to the observation data.

The process of extracting the advanced climate mode may include steps 502A-502C.

Step 502A, when predicting the t-th weather in advance of the n-th weather, extracting the OLR weather trend distance flat field from the t-n-1 th weather to the t-n+1 th weather and the 500hPa potential height weather trend distance flat field each year, and the weather trend distance flat field from the t-1 th weather element to the t+1 th weather element, so as to obtain the extended OLR weather trend distance flat field, the 500hPa potential height weather trend distance flat field and the weather element weather trend distance flat field.

Wherein t-n-1 is a positive integer. For example, if 3 days ahead is predicted for 10 days, the OLR trend of 6,7 and 8 days is extracted every year, along with the 500hPa potential altitude trend of 6,7 and 8 days, and the weather elements of 9, 10 and 11 days.

Step 502B, performing singular value decomposition on the extended OLR weather trend flat field, the 500hPa potential height weather trend flat field and the weather trend flat field, respectively, and determining the top P OLR modes and the top P500 hPa potential height modes obtained by decomposition as the early weather modes most closely related to the weather elements, wherein P is a positive integer.

When singular value decomposition is carried out, combining the expanded OLR weather trend distance flat field and the weather trend distance flat field of the meteorological elements into a matrix, carrying out singular value decomposition on the matrix, and taking the first 20 OLR modes for the OLR modes obtained by decomposition; similarly, the extended 500hPa potential height candidate range flat field and the weather element candidate range flat field are combined to form a matrix, singular value decomposition is carried out on the matrix, the first 20 500hPa potential height modes are taken for the 500hPa potential height modes obtained through decomposition, and then the first 20 500hPa potential height modes and the first 20 OLR modes are determined to be the early weather modes most closely related to the weather element.

The specific process of singular value decomposition of the matrix may refer to the singular value decomposition process in the related art, and this embodiment is not described herein.

Step 502C, respectively normalizing the time sequences corresponding to the first P OLR modes and the first P500 hPa potential height modes obtained by decomposition, and obtaining normalized time sequences.

The time sequence is calculated by the following steps: and respectively projecting the expanded OLR weather trend distance flat field and the 500hPa potential height weather trend distance flat field to the space field of each early weather mode so as to obtain each standardized time sequence. The projection onto the spatial field of each pre-climate mode is to compare the similarity between the extended OLR climate trend and the flat field and the pre-climate mode, and the similarity between the extended 500hPa potential height climate trend and the flat field and the pre-climate mode.

On the basis of the 20 OLR modes and the 20 500hPa potential height modes obtained in step 502B, each OLR mode corresponds to a time sequence, and each 500hPa potential height mode corresponds to a time sequence, so that a time sequence corresponding to the 20 OLR modes and a time sequence corresponding to the 20 500hPa potential height modes can be obtained correspondingly.

Optionally, the process of extracting the contemporaneous climate mode may include steps 502D-502F.

Step 502D, extracting the OLR weather trend distance flat field from the t-1 to the t+1 and the weather trend distance flat field with the potential height of 500hPa, and the weather trend distance flat field of the weather elements from the t-1 to the t+1 in each year, so as to obtain the extended OLR weather trend distance flat field, the weather trend distance flat field with the potential height of 500hPa and the weather trend distance flat field of the weather elements, wherein t-1 is a positive integer.

Wherein t-1 is a positive integer. For example, if t is the 10 th day, the 9 th, 10 th and 11 th day OLR weather trend distance flat field and 500hPa potential height weather trend distance flat field and weather factor weather trend distance flat field are extracted every year.

Step 502E, performing singular value decomposition on the extended OLR weather trend-to-flat field, the 500hPa potential height weather trend-to-flat field and the weather trend-to-flat field of the meteorological element respectively, and determining the first Q OLR modes and the first Q500 hPa potential height modes obtained by decomposition as contemporaneous weather modes most closely related to the meteorological element, wherein Q is a positive integer.

The implementation of step 502F may refer to step 502B, and this embodiment is not described herein.

The value of Q can also be 20, and the first 20 OLR modes and the first 20 500hPa potential height modes obtained by decomposing singular values are correspondingly determined to be contemporaneous climate modes. The contemporaneous or earlier stage climate mode is the climate characteristic obtained by decomposition.

Step 502F, respectively normalizing the time sequences corresponding to the first Q OLR modes and the first Q500 hPa potential height modes obtained by decomposition, and obtaining normalized time sequences.

The time sequence is calculated by the following steps: and respectively projecting the expanded OLR weather trend distance flat field and the 500hPa potential height weather trend distance flat field to the space field of each contemporaneous weather mode to obtain each standardized time sequence. The projection onto the spatial field of each contemporaneous climate mode is to compare the similarity between the extended OLR climate trend and the contemporaneous climate mode and the similarity between the extended 500hPa potential height climate trend and the contemporaneous climate mode.

And step 503, combining the standardized time sequences, and establishing a nonlinear prediction model between the early-stage or contemporaneous weather mode and the weather tendency range of each weather element by using an LSTM or ANN artificial intelligence method.

And combining the standardized time sequences corresponding to the early climate modes to construct a nonlinear prediction model between the early climate modes and the weather trend flat corresponding to the weather elements, and similarly, combining the standardized time sequences corresponding to the contemporaneous climate modes to construct a nonlinear prediction model between the contemporaneous climate modes and the weather trend flat corresponding to the weather elements.

And combining the time sequences corresponding to the early climate modes, namely combining the P time sequences corresponding to the former P OLR modes and the P time sequences corresponding to the former P500 hPa potential height modes. The combined time series may include any number of time series of OLR modes and any number of time series of 500hPa potential altitude modes.

Taking P as 20 as an example, the combined time sequence may include a time sequence corresponding to 1 OLR mode and a time sequence corresponding to 2 500hPa potential altitude modes.

After P time sequences corresponding to the first P OLR modes and P time sequences corresponding to the first P500 hPa potential height modes are combined, a plurality of combined time sequences are obtained, a corresponding nonlinear prediction model is built based on each combined time sequence, and a nonlinear prediction model corresponding to each combined time sequence can be obtained. For example, if 100 combined time sequences are obtained, 100 nonlinear predictive models can be constructed correspondingly.

When a nonlinear prediction model between the early-stage climate mode and the weather trend range corresponding to each weather element is constructed, each combined time sequence corresponding to the early-stage climate mode is input into an initial nonlinear model, and the initial nonlinear model predicts the nonlinear relation between the climate mode and the weather element based on the input time sequence to obtain the nonlinear prediction model corresponding to the early-stage climate mode.

When a nonlinear prediction model between the contemporaneous climate mode and the weather trend range corresponding to each weather element is constructed, each combined time sequence corresponding to the contemporaneous climate mode is input into an initial nonlinear model, and the initial nonlinear model predicts the nonlinear relation between the climate mode and the weather element based on the input time sequence to obtain the nonlinear prediction model corresponding to the contemporaneous climate mode.

Step 504, performing statistics-artificial intelligence combined prediction on the future meteorological elements by using a nonlinear prediction model between the early climate mode and the meteorological elements, and performing power-statistics-artificial intelligence combined prediction on the future meteorological elements by using a nonlinear prediction model between the contemporaneous climate mode and the meteorological elements and a climate mode predicted by using a power mode.

After a plurality of nonlinear prediction models between the early-stage climate mode and the meteorological element and a plurality of nonlinear prediction models between the contemporaneous climate mode and the meteorological element are obtained respectively, the future meteorological element can be subjected to statistics-artificial intelligence combined prediction by using the nonlinear prediction models between the early-stage climate mode and the meteorological element, and the future meteorological element can be subjected to power-statistics-artificial intelligence combined prediction by using the nonlinear prediction models between the contemporaneous climate mode and the meteorological element and the climate mode predicted by using the power mode.

The process of performing statistical-artificial intelligent combined prediction on the future meteorological elements by using a nonlinear prediction model between the early climate mode and the meteorological elements can comprise steps 504A-504C.

In step 504A, the weather dip for the observed climate modality is obtained.

Step 504B, obtaining a time coefficient corresponding to the weather trend distance level of the observed climate mode based on the weather trend distance level of the observed climate mode and the earlier stage climate mode.

And 504C, substituting the time coefficient into a nonlinear prediction model to predict the future meteorological elements.

When predicting the future meteorological elements based on the nonlinear prediction model, firstly, obtaining the weather trend pitch corresponding to the observed climate mode, and exemplarily, obtaining the weather trend pitch of the observed OLR, projecting the weather trend pitch of the observed climate mode onto the space field of the earlier stage climate mode to obtain the time coefficient corresponding to the weather trend pitch of the observed climate mode, and substituting the time coefficient into the nonlinear prediction model, wherein the nonlinear prediction model correspondingly outputs the predicted future meteorological element value, such as precipitation.

The process of performing power-statistics-artificial intelligence combined prediction on the future meteorological elements by using a nonlinear prediction model between the contemporaneous climate modes and the meteorological elements and a climate mode predicted by the power modes may include steps 504D to 504F.

Step 504D, obtaining a predicted output result of the power mode on the future climate mode.

Step 504E, based on the prediction output result and the contemporaneous climate mode, obtaining a time coefficient corresponding to the projection to the contemporaneous climate mode.

And 504F, substituting the time coefficient corresponding to the projection to the contemporaneous weather mode into a nonlinear prediction model to predict the future weather elements.

Because the nonlinear prediction model corresponding to the contemporaneous weather mode predicts the weather elements by adopting the contemporaneous weather mode, the contemporaneous future weather mode corresponding to the future weather elements is firstly required to be predicted based on the power mode to obtain a prediction output result, the prediction output result is projected onto a space field of the contemporaneous weather mode to obtain a time coefficient corresponding to the contemporaneous weather mode, and the time coefficient is substituted into the nonlinear prediction model to obtain the future weather element value output by the nonlinear prediction model.

Step 505, performing independent sample test on the historical return of the nonlinear prediction model with a flat weather trend, and determining an optimal weather mode and a corresponding statistical-artificial intelligent combined prediction model and dynamic-statistical-artificial intelligent combined prediction model.

And carrying out return checking on a plurality of nonlinear prediction models of the early-stage climate mode and the meteorological elements so as to select an optimal climate mode from the nonlinear prediction models, and determining the nonlinear prediction model corresponding to the optimal climate mode as a final statistical-artificial intelligent combined prediction model. And similarly, carrying out return inspection on a plurality of nonlinear prediction models of the contemporaneous climate modes and the meteorological elements to select an optimal climate mode from the nonlinear prediction models, and determining the nonlinear prediction model corresponding to the optimal climate mode as a final power-statistics-artificial intelligence combined prediction model.

The process of determining the statistical-artificial intelligence combined prediction model may include steps 505A-505D.

Step 505A, obtain the history weather trend flat of the last I years, I is a positive integer.

Step 505B, obtaining the return results of the plurality of nonlinear prediction models on the historical weather dip flat, where the return results indicate weather element prediction results of the nonlinear prediction models based on the historical weather dip flat.

And checking the prediction accuracy of the plurality of nonlinear prediction models through the existing historical observation data. For example, the historical weather trend pitch of the last year I is obtained, the historical weather trend pitch is projected onto a space field corresponding to a previous weather mode to obtain a historical time coefficient, the historical time coefficient is input into a plurality of nonlinear prediction models, the nonlinear prediction models are used for weather element prediction, and a return result of each nonlinear prediction model on the historical weather trend pitch is obtained.

The process of predicting the weather element by the nonlinear prediction model may refer to the process of performing statistical-artificial intelligent combined prediction on the future weather element by using the nonlinear prediction model between the early weather mode and the weather element in step 504, which is not described in detail herein.

By way of example, the historical weather pattern may be the daily OLR weather pattern in the recent I-year observation, and the 500hPa potential high field.

For example, the value of I may be 5 years, that is, the last 5 years of historical observation data is used to verify the nonlinear predictive model.

Step 505C, comparing the reported results of the nonlinear prediction models with the spatial correlation coefficients of the observed data.

Wherein, the observation data is a weather element (or a weather element value) corresponding to the history weather trend distance of the recent I year. Comparing the return results of the nonlinear predictive models based on the historical weather tendency with the actual observed data to obtain the similarity between the return results and the observed data, namely obtaining the space correlation coefficients of the return results of the nonlinear predictive models and the observed data.

Step 505D, based on the sequencing result of the spatial correlation coefficient from large to small, selecting the climate mode combination corresponding to the first M OLR field time sequences and the first N500 hPa potential altitude field time sequences as the optimal climate mode, and determining the nonlinear prediction model corresponding to the optimal climate mode as a statistical-artificial intelligent combined prediction model, wherein M, N is a positive integer.

The larger the spatial correlation coefficient is, the higher the prediction accuracy of the nonlinear prediction model is, and the smaller the spatial correlation coefficient is, the lower the prediction accuracy of the nonlinear prediction model is. After the spatial correlation coefficients corresponding to the nonlinear prediction models are obtained, the nonlinear prediction models can be ranked according to the spatial correlation coefficients to obtain a ranking result, then the climate mode combination corresponding to the first M OLR field time sequences and the first N500 hPa potential height field time sequences with larger spatial correlation coefficients is taken as an optimal climate mode, and the nonlinear prediction model corresponding to the optimal climate mode is determined as a statistical-artificial intelligent combined prediction model. That is, the nonlinear prediction model with the largest space coefficient is taken and is determined as the final statistical-artificial intelligence combined prediction model, and the earlier stage climate mode corresponding to the nonlinear prediction model with the largest space coefficient is the optimal climate mode.

The process of determining the power-statistics-artificial intelligence combined prediction model may include steps 505E-505H.

Step 505E, the predicted candidate distance of the power mode for the last J years is obtained, and J is a positive integer.

In step 505F, a plurality of return results of the nonlinear prediction models on the predicted weather trend flat are obtained, where the return results indicate weather element prediction results of the nonlinear prediction models based on the predicted weather trend flat.

The difference between the process of verifying the nonlinear prediction model between the contemporaneous climate modes and the meteorological elements and the process of verifying the nonlinear prediction model between the contemporaneous climate modes and the meteorological elements is that the predicted climate trend pitch of the power mode for the near J years is obtained, the predicted climate trend pitch is projected onto the space field of the contemporaneous climate modes to obtain a predicted time coefficient, and the predicted time coefficient is substituted into each nonlinear prediction model to obtain the return result of each nonlinear prediction model on the predicted climate trend pitch, namely the meteorological element prediction result of the nonlinear prediction model based on the predicted climate trend pitch.

Illustratively, J may have a value of 5.

Step 505G compares the reported results of the nonlinear predictive models with the spatial correlation coefficients of the observed data.

The observation data is a weather element (or a weather element value) corresponding to the predicted weather tendency distance level of the last J years. Comparing the return results of the nonlinear predictive models based on the predicted weather tendency distance with the actual observed data to obtain the similarity between the return results and the observed data, namely obtaining the space correlation coefficients of the return results of a plurality of nonlinear predictive models and the observed data.

Step 505H, based on the sequencing result of the spatial correlation coefficient from large to small, selecting the climate mode combination corresponding to the first X OLR field time sequences and the first Y500 hPa potential altitude field time sequences as the optimal climate mode, and determining the nonlinear prediction model corresponding to the optimal climate mode as a dynamic-statistical-artificial intelligent combined prediction model, wherein X and Y are positive integers.

The larger the spatial correlation coefficient is, the higher the prediction accuracy of the nonlinear prediction model is, and the smaller the spatial correlation coefficient is, the lower the prediction accuracy of the nonlinear prediction model is. And after the spatial correlation coefficients corresponding to the nonlinear prediction models are obtained, the nonlinear prediction models can be ranked according to the spatial correlation coefficients to obtain a ranking result, then the climate mode combination corresponding to the first X OLR field time sequences and the first Y500 hPa potential height field time sequences with larger spatial correlation coefficients is taken as an optimal climate mode, and the nonlinear prediction model corresponding to the optimal climate mode is determined as a power-statistics-artificial intelligence combined prediction model. That is, the nonlinear prediction model with the largest space coefficient is taken and determined as the final power-statistics-artificial intelligence combined prediction model, and the contemporaneous climate mode corresponding to the nonlinear prediction model with the largest space coefficient is the optimal climate mode (or the optimal climate mode combination).

Step 506, predicting the future meteorological elements by using the optimal climate mode and the corresponding statistical-artificial intelligence combined prediction model and dynamic-statistical-artificial intelligence combined prediction model.

Through the steps 501-505, an optimal climate mode and an optimal statistics-artificial intelligence combined prediction model in the early stage climate modes and an optimal climate mode and final power-statistics-artificial intelligence combined prediction model in the contemporaneous climate modes can be obtained respectively, future meteorological elements can be predicted by using the optimal climate mode and the corresponding statistics-artificial intelligence combined prediction model correspondingly, and future meteorological elements can be predicted by using the optimal climate mode and the power-statistics-artificial intelligence combined prediction model correspondingly.

The process of predicting the future meteorological elements by using the optimal climate mode and the corresponding statistical-artificial intelligence combined prediction model may include steps 506A to 506D.

Step 506A, obtaining a corresponding time coefficient projected onto the optimal climate mode based on the OLR climate trend distance flat field of the t-n th day of the predicted year and the climate trend distance flat field and the climate mode of the 500hPa potential height field, wherein n is a positive integer, and t-n is a positive integer.

In the process of constructing the statistical-artificial intelligent combined prediction model, weather elements of the t weather are predicted by using a weather mode of n weather in advance. Therefore, in the process of predicting the meteorological elements by combining the statistical-artificial intelligence with the prediction model, if the meteorological elements of the t-th weather of the prediction year are required to be predicted, firstly, the OLR weather trend distance flat field of the t-n-th weather of the prediction year and the weather trend distance flat field of the 500hPa potential height field are required to be obtained, and the weather trend distance flat field is projected onto the space field corresponding to the optimal weather mode, so that the corresponding time coefficient is obtained.

And 506B, substituting the corresponding time coefficient projected onto the optimal climate mode into a statistical-artificial intelligent combined prediction model to obtain the weather trend flat of the t-th weather of the predicted year.

Further, the time coefficient obtained in step 506A is substituted into a statistical-artificial intelligence combined prediction model to predict the weather trend distance of the weather element at the t-th day of the predicted year.

Step 506C, adding the candidate trend pitch plane of the t-th candidate to the candidate trend pitch plane of the t-1 th candidate observed in the previous candidate to obtain the candidate pitch plane of the t-th candidate.

Step 506D, adding the weather distance of the t-th day and the weather state of the t-th day to obtain the predicted weather average value of the t-th day.

After the weather trend flat of the weather elements of the t weather is obtained, the weather average value of the weather elements of the t weather can be further deduced according to the calculation process of the weather trend flat. Correspondingly, the weather trend pitch level of the t weather factor and the weather trend pitch level of the t-1 weather of the previous weather factor can be added to obtain the weather trend level of the t weather, and then the weather trend level of the t weather and the weather state of the t weather are added to obtain the predicted weather average value of the t weather. By analogy, the daily forecast meteorological element value can be obtained.

Similarly, the process of predicting future meteorological elements using the optimal climate modality and the corresponding dynamic-statistical-artificial intelligence combined prediction model may include steps 506E-506H.

Step 506E, projecting the predicted future atmospheric OLR weather trend distance flat field and the predicted 500hPa potential height weather trend distance flat field to the optimal climate mode to obtain the corresponding time coefficient projected to the optimal climate mode.

In the process of constructing the dynamic-statistical-artificial intelligent combined prediction model, the weather elements of the t weather are predicted by using the weather mode of the t weather, namely, the contemporaneous weather mode is used for predicting the contemporaneous weather elements. Therefore, in the process of predicting the meteorological elements by combining the power-statistics-artificial intelligence with the prediction model, if the meteorological elements of the t-th day of the prediction year need to be predicted, the future OLR weather trend distance flat field and the weather trend distance flat field of 500hPa potential height field of the t-th day need to be obtained based on the power mode prediction, and the weather trend distance flat field is projected onto the space field corresponding to the optimal weather mode, so as to obtain the corresponding time coefficient.

And 506F, substituting the corresponding time coefficient projected to the optimal climate mode into a power-statistics-artificial intelligence combined prediction model, and calculating to obtain the weather trend flat of the future t weather of the meteorological element.

Further, the time coefficient obtained in step 506E is substituted into the power-statistics-artificial intelligence combined prediction model to predict the weather trend flat of the weather element at the t-th day of the predicted year.

Step 506G, adding the candidate trend pitch plane of the t-th candidate to the candidate trend pitch plane of the t-1 th candidate observed in the previous candidate to obtain the candidate pitch plane of the t-th candidate.

Step 506H, adding the candidate distance level of the t-th day to the climate state of the t-th day to obtain the predicted average value of the t-th day.

In the embodiment, an artificial intelligent sub-season prediction method based on an optimal climate mode is provided, and a physical relationship between the climate modes which are most closely related with a meteorological element field on a sub-season scale is extracted by converting the climate modes in the early period and the same period of observation and the meteorological element field into a climate trend distance flat field. On the basis, a prediction model with a nonlinear relation is established by using an artificial intelligence method, so that the future state of the meteorological element is predicted.

Fig. 6 is a schematic structural diagram of a computer system according to an embodiment of the present application, as shown in fig. 6, the computer system 600 includes: the memory 610 and the processor 620 and the computer program 630 stored on the memory 610 and executable on the processor 620 implement the aforementioned artificial intelligence sub-season prediction method based on the optimal climate modality when the computer program 630 is loaded into the processor 620.

By way of example, the memory 610 may be used to store a program related to the artificial intelligence sub-season prediction method based on the optimal climate modality provided in the embodiments of the present application; the processor 620 may invoke a program stored in the memory 610 associated with the optimal climate modality based artificial intelligence sub-season prediction method to perform the optimal climate modality based artificial intelligence sub-season prediction method of embodiments of the present application.

In this embodiment, the functional modules of the system 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 this embodiment, the division of the system is schematic, only one logic function is divided, and another division manner may be implemented in practice.

In the case of dividing each functional module by corresponding each function, the system may further include a processing module, a determining module, a first predicting module, a building module, a checking module, a second predicting 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 system provided in this embodiment is configured to perform an artificial intelligence sub-season prediction method based on an optimal climate mode as described above, so that the same effects as those of the implementation method described above can be achieved.

In case of an integrated unit, the system may comprise a processing module, a memory module. Wherein the processing module may be a processor or controller that may implement or perform the various exemplary logic blocks, modules and circuits described in connection with this 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 system 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 caused to execute the artificial intelligence sub-season prediction method based on the optimal climate mode.

The present application also provides a computer readable storage medium having stored therein computer program code which, when run on a computer, causes the computer to perform the above-mentioned related method steps to implement an artificial intelligence sub-season prediction method based on an optimal climate modality provided by the above-mentioned embodiments. The computer readable storage medium may include, among other things, any type of disk including floppy disks, optical disks, digital versatile disks (Digital Video Disc, DVD), compact disk Read-Only Memory (CD-ROM), micro-drives, and magneto-optical disks, read-Only Memory (ROM), random access Memory (Random Access Memory, RAM), erasable programmable Read-Only Memory (Erasable Programmable Read-Only Memory, EPROM), electrically erasable programmable Read-Only Memory (ELECTRICALLY ERASABLE PROGRAMMABLE READ ONLY MEMORY, EEPROM), dynamic random access Memory (Dynamic Random Access Memory, DRAM), image random access Memory (Video Random Access Memory, VRAM), flash Memory devices, magnetic or optical cards, nanosystems (including molecular Memory ICs), or any type of medium or device suitable for storing instructions and/or data.

The present application also provides a computer program product which, when run on a computer, causes the computer to perform the above-described related steps to implement an artificial intelligence sub-season prediction method based on an optimal climate modality provided by the above-described embodiments.

The computer readable storage medium, the computer program product or the chip provided by the present application are used for executing the corresponding method provided above, and therefore, the advantages achieved by the present application can refer to the advantages in the corresponding method provided above, and will not be 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 sub-season prediction method based on an optimal climate mode, the method comprising:

2. The method according to claim 1, wherein the singular value decomposition is performed on the pre-OLR field and the 500hPa potential height field from the flat field and the weather element from the flat field, respectively, the resulting OLR field and the 500hPa potential height field are the pre-weather modes most closely related to the weather element, and the time series of the decomposed OLR and 500hPa potential height fields are normalized, to obtain normalized time series, respectively, comprising:

Extracting the OLR tendency distance flat field from the t-n-1 to the t-n+1 in each year and the 500hPa potential height tendency distance flat field when predicting the t-th in advance, and the weather trend distance flat field of the weather elements from the t-1 th to the t+1 th weather, obtaining an extended OLR weather trend distance flat field, a 500hPa potential height weather trend distance flat field and a weather trend distance flat field of the weather elements;

performing singular value decomposition on the extended OLR weather trend-to-flat field, the 500hPa potential height weather trend-to-flat field and the weather trend-to-flat field of the meteorological elements respectively, and determining the first P OLR modes and the first P500 hPa potential height modes obtained by decomposition as the early weather modes most closely related to the meteorological elements, wherein P is a positive integer;

3. The method of claim 2, wherein said utilizing the nonlinear predictive model between the advanced climate modality and the meteorological elements to make a combined statistical-artificial intelligence prediction of future meteorological elements comprises:

Obtaining the weather trend distance of the observation weather mode;

obtaining a time coefficient corresponding to the weather trend distance level of the observation climate mode based on the weather trend distance level of the observation climate mode and the early-stage climate mode;

substituting the time coefficient into the nonlinear prediction model to predict the future meteorological elements.

4. The method of claim 2, wherein said performing a historical return independent sample test of the nonlinear predictive model for weather dip leveling, determining an optimal climate modality and corresponding combined statistical-artificial intelligence predictive model, comprises:

Obtaining a history weather trend distance of the recent year I, wherein I is a positive integer;

obtaining the return results of the plurality of nonlinear predictive models on the historical weather tendency distance level, wherein the return results indicate weather element prediction results of the nonlinear predictive models based on the historical weather tendency distance level;

comparing the return results of the nonlinear predictive models with the spatial correlation coefficients of the observed data;

Based on the sequencing result of the space correlation coefficient from large to small, selecting weather mode combinations corresponding to the first M OLR field time sequences and the first N500 hPa potential height field time sequences as the optimal weather modes, and determining a nonlinear prediction model corresponding to the optimal weather modes as the statistical-artificial intelligent combined prediction model, wherein M, N is a positive integer.

5. The method of claim 4, wherein predicting the future meteorological element using the optimal climate modality and the corresponding statistical-artificial intelligence combined prediction model comprises:

Obtaining a corresponding time coefficient projected onto the optimal climate mode based on an OLR (on-line road) tendency distance flat field of the t-n th day of the predicted year and the optimal climate mode, wherein n is a positive integer, and t-n is a positive integer;

Substituting the time coefficient corresponding to the projection to the optimal climate mode into the statistical-artificial intelligent combined prediction model to obtain the weather trend distance of the t-th weather of the predicted year;

Adding the weather trend pitch plane of the t-th weather and the weather trend pitch plane of the t-1 th weather observed in the previous weather to obtain the weather trend plane of the t-th weather;

And adding the weather distance of the t-th weather and the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

6. The method according to claim 1, wherein the singular value decomposition is performed on the contemporaneous OLR field and the 500hPa potential height field from the flat field and the weather element from the flat field, respectively, the resulting OLR field and the 500hPa potential height field are the contemporaneous weather modes most closely related to the weather element, and the time series of the decomposed OLR field and the 500hPa potential height field are normalized, to obtain normalized time series, respectively, comprising:

Extracting an OLR (on-line road) weather trend distance flat field from the t-1 th to the t+1 th and a weather trend distance flat field with a potential height of 500hPa, and a weather trend distance flat field of weather elements from the t-1 th to the t+1 th from each year to obtain an extended OLR weather trend distance flat field, a weather trend distance flat field with a potential height of 500hPa and a weather trend distance flat field of the weather elements, wherein t-1 is a positive integer;

respectively carrying out singular value decomposition on the extended OLR weather trend-to-flat field, the 500hPa potential height weather trend-to-flat field and the weather trend-to-flat field of the meteorological elements, and determining the first Q OLR modes and the first Q500 hPa potential height modes obtained by decomposition as the contemporaneous weather modes most closely related to the meteorological elements, wherein Q is a positive integer;

And respectively normalizing the time sequences corresponding to the first Q OLR modes and the first Q500 hPa potential height modes obtained through decomposition to obtain each time sequence after normalization.

7. The method of claim 6, wherein the performing power-statistics-artificial intelligence combined prediction of the future meteorological elements using the nonlinear prediction model and a power mode predicted climate modality between the contemporaneous climate modality and the meteorological elements comprises:

obtaining a predicted output result of the power mode on a future climate mode;

Obtaining a time coefficient corresponding to the contemporaneous climate mode projected on the basis of the prediction output result and the contemporaneous climate mode;

substituting the time coefficient corresponding to the contemporaneous weather mode projected into the nonlinear prediction model to predict the future meteorological elements.

8. The method of claim 6, wherein said performing a historical return independent sample test of the nonlinear predictive model for weather dip leveling, determining an optimal climate modality and corresponding combined dynamic-statistical-artificial intelligence predictive model, comprises:

Obtaining a predicted weather trend distance of a power mode for nearly J years, wherein J is a positive integer;

obtaining the return results of a plurality of nonlinear predictive models on the predicted weather trend distance level, wherein the return results indicate weather element prediction results of the nonlinear predictive models based on the predicted weather trend distance level;

comparing the return results of the nonlinear prediction models with the spatial correlation coefficients of the observed data;

And selecting a climate mode combination corresponding to the first X OLR field time sequences and the first Y500 hPa potential height field time sequences as the optimal climate modes based on the sequencing result of the space correlation coefficients from large to small, determining the nonlinear prediction model corresponding to the optimal climate modes as the dynamic-statistics-artificial intelligence combined prediction model, wherein X and Y are positive integers.

9. The method of claim 8, wherein predicting the future meteorological element using the optimal climate modality and the corresponding dynamic-statistical-artificial intelligence in combination with a prediction model comprises:

projecting the predicted future atmospheric OLR weather trend distance flat field and the predicted 500hPa potential height weather trend distance flat field of the power mode onto the optimal climate mode to obtain a corresponding time coefficient projected onto the optimal climate mode;

Substituting the time coefficient corresponding to the projection on the optimal climate mode into the power-statistics-artificial intelligence combined prediction model, and calculating to obtain the weather trend range of the future t weather of the meteorological element;

And adding the weather distance level of the t-th weather with the weather state of the t-th weather to obtain the predicted weather average value of the t-th weather.

10. The method according to any one of claims 1 to 9, wherein the candidate pitch-flattening operation comprises: