El Nino-southern oscillation prediction method and device

Abstract

The invention discloses an El Nino-southern oscillation prediction method and device, and the method comprises the steps: inputting a meteorological feature sequence of a target sea area in a target historical time period into a self-adaptive multi-scale space hypergraph module, and obtaining an intermediate feature sequence, the intermediate feature sequence comprises a spatial dependency relationship between nodes of each hypergraph structure in a plurality of hypergraph structures, and each hypergraph structure is obtained based on a feature sequence of the meteorological feature sequence under a corresponding grid scale in a plurality of grid scales; based on the intermediate feature sequence, obtaining a marine Nitro index of the target historical time period; and the ocean Nino index of the target historical time period and the target future time period are input into a frequency domain multi-scale time coding and decoding module, the ocean Nino index of the target sea area in the target future time period is obtained, and the ocean Nino index of the target future time period represents the degree of the El Nino-southern oscillation phenomenon occurring in the target future time period.

Description

Technical Field

[0001] The present disclosure generally relates to the field of meteorological prediction technologies, and more specifically, to an El Niño-Southern Oscillation prediction method and apparatus. Background Art

[0002] The El Niño–Southern Oscillation (ENSO) cycle is a fluctuation between anomalously warm (El Niño) and cold (La Niña) conditions in the tropical Pacific Ocean, and is the most prominent interannual climate change on Earth. The term El Niño refers to the warming of the tropical Pacific Ocean that occurs every 2-7 years, and the opposite cold phase is called La Nina. The anomalous warming or cooling conditions are related to a large-scale east-west sea-level pressure seesaw, called the Southern Oscillation, which represents the atmospheric manifestation of the coupled ENSO phenomenon. ENSO originates in the tropical Pacific Ocean, but its influence is not limited to regional climate, and can also trigger large-scale weather and climate anomalies globally. For example, ENSO can trigger droughts in Southeast Asia, heavy rains on the west coast of South America, frequent extreme weather events in the Northern Hemisphere, etc., directly affecting agricultural yields, energy demand, ecosystem balance and the sustainable development of human society. ENSO prediction provides an opportunity for humans to prevent climate anomalies, potentially reducing the impact of this natural phenomenon on society and the economy, and assisting in the management of natural resources and the environment.

[0003] Current ENSO prediction methods generally perform poorly in long-term prediction scenarios, with the prediction accuracy decreasing as the prediction duration increases, and the "Spring Prediction Barrier (SPB)" problem results in the prediction accuracy in spring months being generally lower than that in other months. Summary of the Invention

[0004] Embodiments of the present disclosure provide an El Niño-Southern Oscillation prediction method and apparatus, which can effectively solve the problem that the prediction accuracy of ENSO prediction in the prior art decreases as the prediction duration increases.

[0005] In one general aspect, an El Niño - Southern Oscillation prediction method is provided, including: inputting a meteorological feature sequence of a target sea area in a target historical time period into an adaptive multi - scale spatial hypergraph module to obtain an intermediate feature sequence, where the target historical time period is a time period of a first predetermined duration before the current moment, and the intermediate feature sequence includes the spatial dependence relationships between the nodes of each hypergraph structure in multiple hypergraph structures, and each hypergraph structure is obtained based on the feature sequence at the corresponding grid scale among multiple grid scales of the meteorological feature sequence; obtaining the Ocean Nino Index of the target historical time period based on the intermediate feature sequence; inputting the Ocean Nino Index of the target historical time period and a target future time period into a frequency - domain multi - scale time encoding - decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period, where the target future time period is a time period of a second predetermined duration after the current moment, and the Ocean Nino Index of the target future time period characterizes the degree of occurrence of the El Niño - Southern Oscillation phenomenon in the target future time period.

[0006] Optionally, the adaptive multi - scale spatial hypergraph module includes a multi - scale data processing unit and multiple parallel network blocks, and multiple grid scales correspond one - to - one with the multiple parallel network blocks. Wherein, inputting the meteorological feature sequence of the target sea area in the target historical time period into the adaptive multi - scale spatial hypergraph module to obtain an intermediate feature sequence includes: inputting the meteorological feature sequence of the target historical time period into the multi - scale data processing unit to obtain the feature sequences at multiple grid scales of the meteorological feature sequence; for the feature sequence at each grid scale among the multiple grid scales, inputting the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence, where the hidden feature includes the spatial dependence relationships between the nodes in the hypergraph structure of the current feature sequence; and determining the hidden features of the feature sequences at multiple grid scales as the intermediate feature sequence.

[0007] Optionally, the network block includes a spatial graph construction unit and a hypergraph information aggregation unit. Wherein, inputting the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence includes: inputting the current feature sequence into the spatial graph construction unit to obtain the hypergraph structure of the current feature sequence, where the hypergraph structure of the current feature sequence includes the hypergraph structures of multiple preset durations in the target historical time period; and inputting the hypergraph structure into the hypergraph information aggregation unit to obtain the hidden feature of the current feature sequence.

[0008] Optionally, the frequency-domain multi-scale time encoding and decoding module includes a frequency-domain embedding unit, a frequency-domain encoding unit, a frequency-domain decoding unit, and a frequency-domain fusion unit. Among them, inputting the Ocean Nino Index of the target historical time period and the target future time period into the frequency-domain multi-scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period includes: inputting the Ocean Nino Index of the target historical time period into the frequency-domain embedding unit to obtain time features in different frequency bands; inputting the time features in different frequency bands into the frequency-domain encoding unit to obtain encoded features; inputting the encoded features and the target future time period into the frequency-domain decoding unit to obtain predicted features in the future frequency band; inputting the predicted features in the future frequency band into the frequency-domain fusion unit to obtain the Ocean Nino Index of the target sea area in the target future time period.

[0009] Optionally, the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module are trained in the following manner: based on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature among various meteorological features of the global sea surface, obtain a training sample set, where each training sample includes a meteorological feature sequence of a first predetermined duration and the true Ocean Nino Index of a second predetermined duration after the first predetermined duration; for each training sample, perform the following processing: input the meteorological feature sequence of the first predetermined duration in the current training sample into the adaptive multi-scale spatial hypergraph module to obtain a predicted intermediate feature sequence; based on the predicted intermediate feature sequence, obtain a first predicted Ocean Nino Index, where the first predicted Ocean Nino Index is the predicted Ocean Nino Index of the first predetermined duration; input the first predicted Ocean Nino Index and the second predetermined duration after the first predetermined duration into the frequency-domain multi-scale time encoding and decoding module to obtain a second predicted Ocean Nino Index, where the second predicted Ocean Nino Index is the predicted Ocean Nino Index of the second predetermined duration after the first predetermined duration; based on the second predicted Ocean Nino Index and the true Ocean Nino Index, adjust the parameters of the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module.

[0010] Optionally, obtaining a training sample set based on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature among various meteorological features of the global sea surface includes: obtaining various meteorological features of the global sea surface; preprocessing various meteorological features of the global sea surface to obtain the meteorological feature sequences and the Ocean Nino Index of each meteorological feature; performing sequential sliding sampling on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature; based on the sampling results, obtain a training sample set.

[0011] In another general aspect, an El Niño - Southern Oscillation prediction device is provided, including: an extraction unit configured to input a meteorological feature sequence of a target sea area in a target historical time period into an adaptive multi - scale spatial hypergraph module to obtain an intermediate feature sequence, where the target historical time period is a time period of a first predetermined duration before the current moment, and the intermediate feature sequence includes the spatial dependence relationships between the nodes of each hypergraph structure in a plurality of hypergraph structures, and each hypergraph structure is obtained based on the feature sequence of the corresponding grid scale among a plurality of grid scales of the meteorological feature sequence; an acquisition unit configured to obtain an Ocean Nino Index of the target historical time period based on the intermediate feature sequence; a prediction unit configured to input the Ocean Nino Index of the target historical time period and a target future time period into a frequency - domain multi - scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period, where the target future time period is a time period of a second predetermined duration after the current moment, and the Ocean Nino Index of the target future time period represents the degree of occurrence of the El Niño - Southern Oscillation phenomenon in the target future time period.

[0012] Optionally, the adaptive multi - scale spatial hypergraph module includes a multi - scale data processing unit and a plurality of parallel network blocks, and the plurality of grid scales correspond one - to - one with the plurality of parallel network blocks. Among them, the extraction unit is further configured to input the meteorological feature sequence of the target historical time period into the multi - scale data processing unit to obtain the feature sequences of the meteorological feature sequence at a plurality of grid scales; for the feature sequence at each grid scale among the plurality of grid scales, input the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence, where the hidden feature includes the spatial dependence relationships between the nodes in the hypergraph structure of the current feature sequence; determine the hidden features of the feature sequences at the plurality of grid scales as the intermediate feature sequence.

[0013] Optionally, the network block includes a spatial graph construction unit and a hypergraph information aggregation unit. Among them, the extraction unit is further configured to input the current feature sequence into the spatial graph construction unit to obtain the hypergraph structure of the current feature sequence, where the hypergraph structure of the current feature sequence includes the hypergraph structures of a plurality of preset durations in the target historical time period; input the hypergraph structure into the hypergraph information aggregation unit to obtain the hidden feature of the current feature sequence.

[0014] Optionally, the frequency-domain multi-scale time encoding and decoding module includes a frequency-domain embedding unit, a frequency-domain encoding unit, a frequency-domain decoding unit, and a frequency-domain fusion unit. Among them, the prediction unit is further configured to input the Ocean Nino Index of the target historical time period into the frequency-domain embedding unit to obtain time features in different frequency bands; input the time features in different frequency bands into the frequency-domain encoding unit to obtain encoded features; input the encoded features and the target future time period into the frequency-domain decoding unit to obtain predicted features in the future frequency band; and input the predicted features in the future frequency band into the frequency-domain fusion unit to obtain the Ocean Nino Index of the target sea area in the target future time period.

[0015] Optionally, the training unit is configured to train the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module in the following manner: Based on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature among various meteorological features of the global sea surface, obtain a training sample set, where each training sample includes a meteorological feature sequence of the first predetermined duration and the true Ocean Nino Index of the second predetermined duration after the first predetermined duration; for each training sample, perform the following processing: input the meteorological feature sequence of the first predetermined duration in the current training sample into the adaptive multi-scale spatial hypergraph module to obtain an estimated intermediate feature sequence; based on the estimated intermediate feature sequence, obtain a first estimated Ocean Nino Index, where the first estimated Ocean Nino Index is the estimated Ocean Nino Index of the first predetermined duration; input the first estimated Ocean Nino Index and the second predetermined duration after the first predetermined duration into the frequency-domain multi-scale time encoding and decoding module to obtain a second estimated Ocean Nino Index, where the second estimated Ocean Nino Index is the estimated Ocean Nino Index of the second predetermined duration after the first predetermined duration; and adjust the parameters of the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module based on the second estimated Ocean Nino Index and the true Ocean Nino Index.

[0016] Optionally, the training unit is further configured to obtain various meteorological features of the global sea surface; preprocess the various meteorological features of the global sea surface to obtain the meteorological feature sequences and the Ocean Nino Index of each meteorological feature; perform sequential sliding sampling on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature; and obtain a training sample set based on the sampling results. In another general aspect, there is provided a computer-readable storage medium storing instructions, where when the instructions are run by at least one computing device, at least one computing device is caused to execute any of the above El Nino-Southern Oscillation prediction methods.

[0017] In another general aspect, there is provided a system including at least one computing device and at least one storage device storing instructions, where when the instructions are run by at least one computing device, at least one computing device is caused to execute any of the above El Nino-Southern Oscillation prediction methods.

[0018] In another general aspect, there is provided a computer program product including computer instructions which, when executed by a processor, are configured as the El Niño - Southern Oscillation prediction method as described above.

[0019] According to the El Niño - Southern Oscillation prediction method and apparatus of embodiments of the present disclosure, by adapting the multi - scale spatial hypergraph module, based on the meteorological feature sequence of the target sea area within the target historical time period, an intermediate feature sequence containing spatial information of the target sea area within the target historical time period is obtained. That is, the multi - scale spatial hypergraph module dynamically constructs a hypergraph structure of meteorological feature sequences with different grid scales, which can accurately depict the spatio - temporal evolution pattern of meteorological variables in the target sea area in the time dimension, realize the efficient aggregation of spatial information, and fully extract the multi - grid scale spatial features in the ENSO evolution process. Moreover, through the frequency - domain multi - scale time encoding and decoding module, based on the intermediate feature sequence containing spatial information, the Oceanic Nino Index of the target sea area within the target future time period is obtained. That is, the frequency - domain multi - scale time encoding and decoding module performs frequency - domain modeling on the above - mentioned intermediate feature sequence, effectively captures the long - time - series evolution law of ENSO, and finally generates the Oceanic Nino Index of the target future time period, so as to know the degree of the El Niño - Southern Oscillation phenomenon occurring in the target future time period, and improve the efficiency, accuracy, and stability of predicting the ENSO evolution trend. Therefore, through the present disclosure, the problem that the prediction accuracy of ENSO prediction in the prior art decreases with the increase of the prediction duration can be effectively solved.

[0020] Some other aspects and / or advantages of the general concept of the present disclosure will be partly elaborated in the following description, and some will be clear from the description, or can be learned through the implementation of the general concept of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Through the following description with reference to the drawings showing embodiments, the above - mentioned and other objects and features of the embodiments of the present disclosure will become clearer, where: Figure 1 is a flowchart showing the El Niño - Southern Oscillation prediction method of embodiments of the present disclosure; Figure 2 is a system flowchart showing the training method of the adaptive multi - scale spatio - temporal ENSO prediction model and the ENSO prediction method of embodiments of the present disclosure; Figure 3 is a schematic structural diagram showing the adaptive multi - scale spatio - temporal ENSO prediction model of embodiments of the present disclosure; Figure 4 is a schematic structural diagram showing the adaptive multi - scale spatial hypergraph module of embodiments of the present disclosure; Figure 5 is a schematic structural diagram showing the frequency - domain multi - scale time encoding and decoding module of embodiments of the present disclosure; Figure 6 is a schematic diagram showing the overall structure of an adaptive multi-scale spatio-temporal ENSO prediction model according to an embodiment of the present disclosure; Figure 7 is a block diagram showing an El Niño - Southern Oscillation prediction apparatus according to an embodiment of the present disclosure; Figure 8 is a schematic diagram showing the structure of an electronic device according to an embodiment of the present disclosure. Detailed Embodiments

[0022] The following detailed embodiments are provided to assist the reader in obtaining a comprehensive understanding of the methods, apparatuses, and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications, and equivalents of the methods, apparatuses, and / or systems described herein will be apparent. For example, the order of operations described herein is merely illustrative and is not limited to those set forth herein, but may be changed as will be apparent after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, descriptions of features known in the art may be omitted for greater clarity and conciseness.

[0023] The features described herein may be implemented in different forms and should not be construed as limited to the examples described herein. Instead, the examples described herein are provided only to illustrate some of the many possible ways of implementing the methods, apparatuses, and / or systems described herein, which will be apparent after understanding the disclosure of the present application.

[0024] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more of them.

[0025] Although terms such as "first", "second", and "third" may be used herein to describe various components, components, regions, layers, or parts, these components, components, regions, layers, or parts should not be limited by these terms. Instead, these terms are only used to distinguish one component, component, region, layer, or part from another component, component, region, layer, or part. Thus, a first component, first component, first region, first layer, or first part referred to in the examples described herein may also be referred to as a second component, second component, second region, second layer, or second part without departing from the teachings of the examples.

[0026] In the specification, when an element (such as a layer, region, or substrate) is described as being "on" another element, "connected to" or "coupled to" another element, the element can be directly "on" the other element, directly "connected to" or "coupled to" the other element, or there can be one or more other elements therebetween. In contrast, when an element is described as being "directly on" another element, "directly connected to" or "directly coupled to" another element, there can be no other elements therebetween.

[0027] The terms used herein are for describing various examples only and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. The terms "comprising", "including" and "having" specify the presence of the recited features, quantities, operations, components, elements and / or combinations thereof, but do not preclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0028] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains after understanding the disclosure. Unless clearly defined herein, terms (such as those defined in a general dictionary) shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and shall not be interpreted in an idealized or overly formal manner.

[0029] In addition, in the description of the examples, when a detailed description of a related structure or function that is considered to be well-known would cause an ambiguous interpretation of the disclosure, such detailed description will be omitted.

[0030] The current ENSO prediction model is based on past temporal and meteorological features (such as global sea surface temperature anomalies, heat capacity anomalies, etc.) to predict the Oceanic Nino Index (ONI) in future time series. The problem of predicting the El Niño-Southern Oscillation is essentially to predict the ONI in future time series based on past spatio-temporal correlation sequences of sea surface temperature, heat capacity, etc., which can be formally expressed as: , where t represents the current moment, T represents the number of past moments, represents the number of future moments. Traditional ENSO prediction models only consider the historical spatio-temporal correlation sequences at the original grid scale, establish a fixed spatial topology at the original grid scale, and extract time features in the time domain. Therefore, they cannot capture local and global information simultaneously, and the capture of spatio-temporal evolution laws is one-sided and single. At the same time, in the time domain, as the prediction time increases, the computational complexity and error accumulation will gradually increase, reducing the long-term prediction accuracy, resulting in low prediction efficiency, poor prediction accuracy and stability.

[0031] Starting from the difficulties in capturing spatio-temporal information, first, in the spatial dimension, an adaptive multi-scale spatial hypergraph module is introduced to dynamically construct hypergraph structures at different grid scales, realizing fine modeling and information aggregation of spatial dependence relationships, and fully extracting multi-scale spatial features in the ENSO evolution process. Second, in the time dimension, through a frequency-domain multi-scale time encoding and decoding module, the feature sequence is mapped from the time domain to the frequency-domain space, and the periodic characteristics of different frequency bands in the frequency domain are used for modeling, effectively alleviating the problem of cumulative long-term prediction errors in traditional time-domain modeling, and the time signal can be restored through the inverse Fourier transform to achieve accurate prediction of the ONI in the future time period, so as to know the degree of the El Niño-Southern Oscillation phenomenon in the target future time period, and improving the efficiency, accuracy, and stability of predicting the ENSO evolution trend.

[0032] The El Niño-Southern Oscillation prediction method and device of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0033] The present disclosure proposes an El Niño-Southern Oscillation prediction method. Figure 1 is a flowchart showing the El Niño-Southern Oscillation prediction method according to an embodiment of the present disclosure. Refer to Figure 1 , the El Niño-Southern Oscillation prediction method includes the following steps: In step S101, the meteorological feature sequence of the target sea area in the target historical time period is input into the adaptive multi-scale spatial hypergraph module to obtain an intermediate feature sequence, where the target historical time period is a time period of the first predetermined duration before the current moment, and the intermediate feature sequence includes the spatial dependence relationships between the nodes of each hypergraph structure in a plurality of hypergraph structures, and each hypergraph structure is obtained based on the feature sequence corresponding to the grid scale in a plurality of grid scales of the meteorological feature sequence.

[0034] In step S102, based on the intermediate feature sequence, the Ocean Nino Index of the target historical time period is obtained.

[0035] In step S103, the Ocean Nino Index of the target historical time period and the target future time period are input into the frequency-domain multi-scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period, where the target future time period is a time period of the second predetermined duration after the current moment, and the Ocean Nino Index of the target future time period represents the degree of the El Niño-Southern Oscillation phenomenon occurring in the target future time period.

[0036] As an example, the El Niño-Southern Oscillation prediction method of this embodiment can be applied to an ENSO prediction model, that is, the ENSO prediction model includes an input adaptive multi-scale spatial hypergraph module, a transformation module, and a frequency-domain multi-scale time encoding and decoding module. Among them, the transformation module obtains the Oceanic Nino Index of the target historical period based on the intermediate feature sequence. The above transformation module can be a multi-layer perceptron (MLP), and the present disclosure does not limit this.

[0037] As an example, the ENSO prediction model can be trained based on the meteorological feature sequence of the target sea area in the historical period and the corresponding ONI. For example, the historical meteorological feature sequence of the target sea area can be used as a training sample, and the ONI of the target sea area in the future period can be used as the prediction result label of the training sample to train the ENSO prediction model. Another example is that if it is necessary to predict the ONI in the next 12 months based on the meteorological feature sequence in the past 12 months, then during the training of the ENSO prediction model, both the historical period and the future period are set to 12 months. The meteorological feature sequence of the target sea area from the 1st to the 12th month is used as a training sample, and the ONI of the target sea area from the 12th to the 24th month is used as the prediction result label.

[0038] It should be noted that the specific training process will be described in detail later and will not be elaborated here for the time being.

[0039] As an example, the meteorological feature sequence of the target sea area in the target historical period can be obtained by preprocessing the global sea surface multi-feature sample database. For example, the meteorological feature sequence of the target sea area in the target historical period can be the sea surface temperature, heat capacity, etc. of the target sea area from a certain past moment to the current moment (such as the past 12 months). It should be noted that the global sea surface multi-feature data sample library can be obtained through but not limited to the following addresses: the National Oceanic and Atmospheric Administration of the United States, the Climate Observation Center of the National Weather Service of the United States, the meteorological satellite data download website provided by the National Aeronautics and Space Administration of the United States, the climate data storage and query platform of the World Meteorological Organization, etc.

[0040] The above adaptive multi-scale spatial hypergraph module designs a construction mechanism for feature sequences with multi-grid scales from fine granularity to coarse granularity according to the characteristics and requirements of the El Niño-Southern Oscillation prediction problem in the ocean, so as to enhance the model's comprehensive extraction ability of local information and global spatial features. By introducing the hypergraph structure, multiple connections between distant and neighboring nodes are established in physical space, enabling each spatial node to aggregate information from more relevant regions. At the same time, the historical period can use months as time steps, and each time step has its own hypergraph results under multiple grid scales. Combining the dynamic evolution characteristics of the time dimension, the module can adaptively adjust the hypergraph structure under each grid scale at each time step, so as to more accurately simulate the spatial dynamic evolution process of the ENSO phenomenon in real physical time.

[0041] The above frequency-domain multi-scale time encoding and decoding module proposes a time series processing strategy based on the frequency domain according to the characteristics and requirements of the El Niño-Southern Oscillation prediction problem in the ocean. This module maps the input historical time series to the frequency domain space and decomposes it into time features in multiple different frequency bands, thereby realizing the decoupled modeling of short-term fluctuations and long-term trends. In the frequency domain, an attention mechanism is introduced for each frequency band of time features for feature enhancement and encoding processing, and then it is restored to the time domain representation through the frequency domain decoding unit and the frequency domain fusion unit.

[0042] The adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module are introduced separately below.

[0043] According to an embodiment of the present disclosure, the adaptive multi-scale spatial hypergraph module may include a multi-scale data processing unit and a plurality of parallel network blocks, and the plurality of grid scales correspond to the plurality of parallel network blocks one by one. Among them, in step S101, the meteorological feature sequence of the target sea area in the target historical time period is input into the adaptive multi-scale spatial hypergraph module to obtain an intermediate feature sequence, which can be realized in the following manner: the meteorological feature sequence of the target historical time period is input into the multi-scale data processing unit to obtain the feature sequences of the meteorological feature sequence at multiple grid scales; for the feature sequences at each grid scale among the multiple grid scales, the current feature sequence is input into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence, where the hidden feature includes the spatial dependence relationship between the nodes in the hypergraph structure of the current feature sequence; the hidden features of the feature sequences at multiple grid scales are determined as the intermediate feature sequence.

[0044] Through this embodiment, the meteorological feature sequence of the target historical time is divided into feature sequences at multiple network scales, which can accurately depict the evolution of the meteorological variables in the target sea area in the time dimension, realize the efficient aggregation of spatial information, and use multiple network blocks to process the feature sequences at each network scale in parallel, so that the hidden features of the feature sequences at each network scale can be obtained quickly, and then the intermediate feature sequence can be obtained quickly.

[0045] As an example, the above multi-scale data processing unit creates feature sequences at multi-grid scales with different granularities from the meteorological feature sequence of the target historical time period. Specifically, the multi-scale data processing unit divides the meteorological feature sequence of the target historical time period into feature sequences at multiple different grid scales from fine granularity to coarse granularity based on a preset granularity window size; for example, the original grid scale of the meteorological feature sequence of the target historical time period is latitude 、longitude a total of 24 The global real physical space range of 48 grid points, with the preset granularity window sizes of 24 and 48. Then, through two-dimensional convolution, 4 12 and 2 feature sequences at two different grid scales of 6 are obtained, representing the feature sequence representations at different grid scales from local to global.

[0046] As an example, the parameters of the above multiple parallel network blocks can be shared, and the present disclosure does not limit this. Each network block can adopt an adaptive hypergraph learning neural network, and the present disclosure also does not limit this.

[0047] Specifically, each network block is used to output the hidden features of the feature sequence at the corresponding grid scale based on the feature sequences at different grid scales. In other words, the input of each network block is respectively a feature sequence at a grid scale of the target sea area during the target historical time period, and the feature sequence at the corresponding grid scale is independently processed to obtain the feature time sequence with spatial information at the corresponding network scale of the target sea area during the target historical time period, that is, the above hidden features. And multiple network blocks can obtain the hidden features at multiple grid scales in parallel, and combine the multiple hidden features as the output of the adaptive multi-scale spatial hypergraph module.

[0048] According to an embodiment of the present disclosure, the above network block may include a spatial graph construction unit and a hypergraph information aggregation unit. Among them, inputting the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden features of the current feature sequence can be achieved in the following way: inputting the current feature sequence into the spatial graph construction unit to obtain the hypergraph structure of the current feature sequence, where the hypergraph structure of the current feature sequence includes the hypergraph structures of multiple preset time lengths in the target historical time period; inputting the hypergraph structure into the hypergraph information aggregation unit to obtain the hidden features of the current feature sequence.

[0049] Through this embodiment, the spatial graph construction unit can construct the hypergraph structures of the feature sequences of each preset time length at the corresponding network scale, so that combined with the dynamic evolution characteristics of the time dimension, the hypergraph structures at each grid scale of each preset time length can be adaptively adjusted, thereby more accurately simulating the spatial dynamic evolution process of the ENSO phenomenon in real physical time. And based on this hypergraph structure, through the hypergraph information aggregation unit, the hyperedge weights are adaptively adjusted according to the changes of the feature sequences in each preset time length, realizing the effective modeling of the dynamic spatial evolution structure, and then relatively accurate hidden features can be obtained.

[0050] As an example, the above hypergraph information aggregation unit can adaptively adjust the hyperedge weights according to the changes in the network state in each preset time period, aggregate the hypergraph results, and obtain corresponding accurate hidden features; the above preset time period can be in months, and the present disclosure does not limit this. It should be noted that the preset time period can also be called a time step; the network block can include a spatial graph construction unit and a hypergraph information aggregation unit, and the number of channels of the two units is the same. It should be noted that the change in the network state is the change in the feature sequence of each preset time period (time step).

[0051] The spatial graph construction unit and the hypergraph information aggregation unit are described separately below: 1) The above spatial graph construction unit can construct a hypergraph structure based on the geographical distance between meteorological observation points and the similarity of the meteorological feature sequences of meteorological observation points. Specifically, based on the geographical coordinate information of the meteorological observation points in the current feature sequence, the Euclidean distance between each observation point can be calculated, and the observation points with the Euclidean distance less than the first set threshold can be selected and included in the same hyperedge; hyperedges can also be constructed based on the correlation between the meteorological feature sequences of the meteorological observation points in the current feature sequence. When the correlation of the feature sequences of two meteorological observation points in the historical time period exceeds the second set threshold, the two meteorological observation points are included in the same hyperedge; meteorological observation points that satisfy both of the above conditions can also be included in the same hyperedge. The present disclosure does not limit this, as long as a node set with high meteorological similarity can be formed. The hypergraph structure output by this spatial graph construction unit is used for subsequent hypergraph information aggregation operations to enhance the spatial dependence modeling ability.

[0052] It should be noted that for the s th grid scale, the hypergraph structure expressions constructed based on the geographical distance between meteorological observation points or the similarity of the meteorological feature sequences of meteorological observation points can all adopt the following formula:

[0053] Among them, when constructing the hypergraph structure based on the geographical distance between meteorological observation points, represents the first set threshold; when constructing the hypergraph structure based on the similarity of the meteorological feature sequences of meteorological observation points, represents the second set threshold; v represents the node position at the s-th grid scale, e represents the s th hyperedge at the grid scale.

[0054] 2) Based on the above hypergraph structure, the above hypergraph information aggregation unit outputs corresponding hidden features. Specifically, the hypergraph information aggregation unit can adopt hypergraph convolution operations. In each preset time period, it performs weighted summation or attention-weighted fusion on the features of multiple nodes connected by each hyperedge along the aggregation path of node-hyperedge-node to extract the high-order spatial dependence information between nodes, that is, the above hidden features, so that each node not only aggregates the information of its adjacent nodes, but also integrates the global dynamic features of multiple nodes in the same hyperedge, thereby enhancing the spatio-temporal modeling ability within each grid scale. This aggregation process can be combined with a dynamic adjustment mechanism to adaptively adjust the hyperedge weights according to the changes in the network state at each time step, achieving effective modeling of the dynamic spatial evolution structure.

[0055] The weights adopted by the above weighting method can be static weights, such as calculating fixed edge weights according to the geographical distance or meteorological feature similarity between nodes; the weights adopted by the above weighting method can also be dynamically learnable weights, such as by introducing an attention mechanism to adaptively assign different aggregation weights to different nodes according to the feature context at the current moment.

[0056] For example, in a hyperedge connecting five meteorological observation points, different weights can be dynamically assigned to each adjacent node according to the meteorological feature similarity or geographical distance between each node and the central node, and finally the spatial aggregation representation of the node under the current hyperedge is generated. In addition, the hypergraph information aggregation unit also fuses the feature information from multiple hyperedges, and further improves the spatial perception ability through multi-hyperedge aggregation strategies (such as average aggregation, max aggregation, or weighted fusion) to obtain more discriminative hidden features.

[0057] According to an embodiment of the present disclosure, the frequency-domain multi-scale time encoding and decoding module includes a frequency-domain embedding unit, a frequency-domain encoding unit, a frequency-domain decoding unit, and a frequency-domain fusion unit. Among them, inputting the Ocean Nino Index of the target historical time period and the target future time period into the frequency-domain multi-scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period can be achieved in the following way: inputting the Ocean Nino Index of the target historical time period into the frequency-domain embedding unit to obtain time features in different frequency bands; inputting the time features in different frequency bands into the frequency-domain encoding unit to obtain encoded features; inputting the encoded features into the frequency-domain decoding unit of the target future time period to obtain predicted features in the future frequency band; and inputting the predicted features in the future frequency band into the frequency-domain fusion unit to obtain the Ocean Nino Index of the target sea area in the target future time period.

[0058] Through this embodiment, the Ocean Nino Index of the input target historical time period is mapped to the frequency domain space and decomposed into time characteristics of multiple different frequency bands, so as to realize the decoupled modeling of short-term fluctuations and long-term trends; moreover, in the frequency domain, the time characteristics of each frequency band are respectively subjected to feature enhancement and encoding processing, and then restored to the time domain representation through the frequency domain decoding unit and the frequency domain fusion unit.

[0059] As an example, the frequency-domain multi-scale time encoding and decoding module can, based on the self-attention mechanism architecture, transform the intermediate feature sequence of the target sea area in the target historical time period into the frequency domain and extract the ONI prediction result. It should be noted that the frequency-domain multi-scale time encoding and decoding module can include multiple frequency-domain operation units, and the following will introduce the multiple frequency-domain operation units respectively: 1) The frequency-domain embedding unit performs a Discrete Fourier Transform (DFT) on the Ocean Nino Index of the target historical time period, maps the time series of the Ocean Nino Index from the time domain to the frequency domain, and extracts the frequency band characteristics reflecting different time scale changes, that is, extracts the time characteristics of different frequency bands, for capturing the periodicity and oscillation patterns in the ENSO phenomenon. Specifically, the frequency-domain embedding unit first performs a Fourier transform on the input time series of the Ocean Nino Index to obtain its amplitude spectrum or phase spectrum representation on multiple frequency components; then, these frequency components can be further divided into several predefined frequency bands through a frequency band segmentation module, such as low frequency (long-term change trend), medium frequency (interannual fluctuation), and high frequency (seasonal oscillation), so as to respectively extract the frequency-domain characteristics representing different time scale change characteristics.

[0060] For example, when the input is the time series of the Ocean Nino Index of sea surface temperature, after the frequency-domain embedding unit performs the FFT, it can respectively obtain its characteristics in frequency intervals with periods of 1 year, 6 months, 1 month, etc. The characteristics in these frequency intervals help the subsequent frequency-domain encoding unit to judge the oscillation period related to ENSO, thereby enhancing the perception ability of the ONI trend.

[0061] It should be noted that to enhance the stability and distinguishability of the frequency representation, the frequency-domain embedding unit can also introduce processing operations such as amplitude normalization and spectrum enhancement, making the model more dependent on the key frequency bands, and thus improving the prediction accuracy.

[0062] 2) The frequency-domain encoding unit encodes the time features of multiple frequency bands extracted by the frequency-domain embedding unit. By constructing a frequency attention map to highlight the influence of key frequency bands, for example, a frequency-domain attention module based on the Transformer structure can be used to model the dependencies of multiple frequency bands. Specifically, the frequency-domain encoding unit constructs a self-attention network in the frequency dimension (Frequency-wise Transformer Encoder), captures the long-range dependencies and interaction relationships between different frequencies through the multi-head attention mechanism, and then outputs the corresponding encoded features. The frequency-domain encoding unit can stack multiple frequency-domain Transformer layers, each layer containing a frequency attention sub-layer and a feed-forward network sub-layer, and adopting residual connections and normalization operations to make the network more stably focus on the key frequency bands strongly related to ENSO.

[0063] 3) The frequency-domain decoding unit constructs a prediction structure for the future frequency distribution based on the frequency representation output by the frequency-domain encoding unit, that is, predicts the Ocean Nino Index (ONI) of the target sea area in the target future time period. For example, the frequency-domain decoding unit can adopt a frequency-time series modeling network based on the Transformer decoder structure. Specifically, this decoder structure is a frequency-aware Transformer decoder, which can fuse the historical frequency representation (i.e., the encoded features output by the frequency-domain encoding unit) and the target frequency position information (the true ONI corresponding to the encoded features) during the training phase for sequence prediction in the frequency domain; each decoding module contains a frequency attention mechanism, gradually predicts the evolution features of each future frequency band, and outputs the predicted spectrum result for subsequent fusion and restoration.

[0064] 4) The frequency-domain fusion unit performs weighted fusion on the prediction results of different frequency bands and converts the frequency-domain results back to the time domain to form the Ocean Nino Index (ONI) of the target sea area in the target future time period. Specifically, the frequency-domain fusion unit introduces a frequency-band weighting strategy, dynamically adjusts the fusion weights of each frequency band according to indicators such as the importance and confidence of the frequency-band prediction results, and then, through the inverse Fourier transform (IDFT), restores the fused spectrum signal to a time series to obtain the final prediction result of ONI.

[0065] According to an embodiment of the present disclosure, the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module are trained in the following manner: based on the meteorological feature sequence of each meteorological feature among various meteorological features of the global sea surface and the Ocean Nino Index, a training sample set is obtained, wherein each training sample includes a meteorological feature sequence of a first predetermined duration and a true Ocean Nino Index of a second predetermined duration after the first predetermined duration; for each training sample, the following processing is performed: input the meteorological feature sequence of the first predetermined duration in the current training sample into the adaptive multi-scale spatial hypergraph module to obtain a predicted intermediate feature sequence; based on the predicted intermediate feature sequence, obtain a first predicted Ocean Nino Index, wherein the first predicted Ocean Nino Index is the predicted Ocean Nino Index of the first predetermined duration; input the first predicted Ocean Nino Index and the second predetermined duration after the first predetermined duration into the frequency-domain multi-scale time encoding and decoding module to obtain a second predicted Ocean Nino Index, wherein the second predicted Ocean Nino Index is the predicted Ocean Nino Index of the second predetermined duration after the first predetermined duration; based on the second predicted Ocean Nino Index and the true Ocean Nino Index, adjust the parameters of the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module.

[0066] As an example, during the training process, the frequency-domain multi-scale time encoding and decoding module can also input the true Ocean Nino Index of a past time period, such as the true Ocean Nino Index of the first predetermined duration or the true Ocean Nino Index of the second half duration after the first predetermined duration, and the present disclosure does not limit this.

[0067] Through this embodiment, the training of the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module can be conveniently and accurately achieved.

[0068] According to an embodiment of the present disclosure, based on the meteorological feature sequence of each meteorological feature among various meteorological features of the global sea surface and the Ocean Nino Index, obtaining a training sample set includes: obtaining various meteorological features of the global sea surface; preprocessing the various meteorological features of the global sea surface to obtain the meteorological feature sequence of each meteorological feature and the Ocean Nino Index; performing sequential sliding sampling on the meteorological feature sequence and the Ocean Nino Index of each meteorological feature; based on the sampling result, obtaining a training sample set.

[0069] Through this embodiment, sampling the preprocessed meteorological feature sequence and the Ocean Nino Index can increase the sample size and alleviate the problem of insufficient effective data volume in the ocean field.

[0070] As an example, various meteorological feature data of the global sea surface can be obtained from the global sea surface multi-feature data sample library, which can be obtained from the National Oceanic and Atmospheric Administration of the United States, the Climate Observation Center of the National Weather Service of the United States, the meteorological satellite data download website provided by the National Aeronautics and Space Administration of the United States, the climate data storage and query platform of the World Meteorological Organization, etc. The present disclosure does not limit this. The prediction data samples of the El Niño / Southern Oscillation in the global sea surface multi-feature data sample library can be divided into two categories: historical observation data and model simulation data. Among them, the model simulation data is the observation data simulated based on the historical observation data. These two types of data are usually used in combination to improve the accuracy and reliability of the prediction. For example, the historical observation data can be used to calibrate the model and serve as the benchmark data for model input; the model simulation data can help predict the possibility and trend of the future El Niño / Southern Oscillation time.

[0071] As an example, the meteorological feature sequence and the Oceanic Nino Index of each meteorological feature can be obtained by preprocessing the global sea surface multi-feature sample database. For example, the meteorological feature sequence of the target sea area within the historical time period can be the sea surface temperature, heat capacity, etc. of the target sea area from a certain past moment to the current moment (such as the past 12 months), and the ONI within the historical time period can be the true ONI of the target sea area from a certain past moment to the current moment (such as the past 12 months). It should be noted that the true ONI within the historical time period can be used to characterize the ENSO phenomenon degree from a certain past moment to the current moment (such as the past 12 months).

[0072] The above preprocessing can be carried out by using data preprocessing software, python data processing packages, etc. The present disclosure does not limit this. The purpose of preprocessing the obtained global sea surface multi-feature sample database is to obtain a sample set in a specific format containing features related to the prediction of the El Niño / Southern Oscillation. The preprocessed features are prediction factors closely related to the El Niño / Southern Oscillation time, such as sea surface temperature, heat capacity, etc. For the sake of understanding, the following will elaborate on the sea surface temperature and heat capacity. The data preprocessing can be carried out by using data preprocessing software, python data processing packages, etc., and the final data formats of the sea surface temperature and heat capacity are respectively the meteorological feature sequence and the ONI time series data. Among them, the meteorological feature sequence can be expressed as , and the ONI time series data can be expressed as , where n represents the maximum moment of the meteorological feature sequence, R H×W represents X i has a spatial dimension, that is X iIn addition to having a time dimension, it also has longitude and latitude. H Indicates longitude. W Indicates latitude. R Indicates Y i Does not have a spatial dimension, that is Y i Only has a time dimension.

[0073] As an example, after obtaining the meteorological feature sequence and ONI in the target sea area during the historical time period, the meteorological feature sequence and ONI can be sampled by sliding according to the sampling interval The sampling interval Can be set for years, or can be set for the time resolution of sequence data (usually months). The advantage of the latter is that it can greatly increase the sample size and alleviate the problem of insufficient effective data volume in the ocean field. The disadvantage is that model overfitting may occur due to the repeated use of data. When the sampling interval for months Is set to 12, it is equivalent to the sampling interval for years Set to 1. Therefore, considering the insufficient data volume and the risk of model overfitting, the value range of the sampling interval for months Can be .

[0074] As an example, based on the second predicted Oceanic Niño Index and the true Oceanic Niño Index, the loss can be determined, and the parameters of the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module can be adjusted through this loss. The loss function used to calculate the above loss can be the mean square error loss between the second predicted Oceanic Niño Index and the true Oceanic Niño Index, but the present disclosure does not limit this. It should be noted that in terms of training, relevant training parameters, methods, and functions can be selected, and the meteorological feature sequence samples are input into the adaptive multi-scale spatio-temporal ENSO prediction model to train the model to obtain a trained adaptive multi-scale spatio-temporal ENSO prediction model.

[0075] To better understand the present disclosure, the following is a systematic description in combination with Figure 2 For illustration.

[0076] Figure 2 Shows the training method of the adaptive multi-scale spatio-temporal ENSO prediction model and the system process of the ENSO prediction method. As Figure 2 Shown, it can include the following steps: 1) Obtain the global sea surface multi-feature data sample library; 2) Preprocess the data in the global sea surface multi-feature data sample library, and extract the meteorological feature sequence data and ONI time series data containing features related to the prediction of the El Niño-Southern Oscillation.

[0077] 3) Based on the spatio-temporal characteristics and evolution trends of the El Niño-Southern Oscillation in the ocean, an adaptive multi-scale spatio-temporal ENSO prediction model is constructed and trained for meteorological characteristic sequence data such as sea surface temperature and heat capacity, and ONI time series data.

[0078] 4) Based on the trained adaptive multi-scale spatio-temporal ENSO prediction model, predict the ONI time series data for future time periods.

[0079] The typical problems addressed by the above-mentioned adaptive multi-scale spatio-temporal ENSO prediction model are spatio-temporal sequence feature extraction and time series prediction problems, that is, extracting an intermediate feature sequence containing spatial feature information from meteorological characteristic sequence data based on past time, then transforming the intermediate feature sequence into ONI time series data for past time, and predicting ONI time series data for future time based on ONI time series data for past time. The key here lies in capturing the spatial feature information and time evolution characteristics of meteorological characteristic sequence data.

[0080] Figure 3 shows the structure of the adaptive multi-scale spatio-temporal ENSO prediction model, as Figure 3 shown, the input data of the adaptive multi-scale spatio-temporal ENSO prediction model is meteorological characteristic sequence data at the target sea area for the past T time moments , and at the same time, the data for the next τ time moments is also input. Figure 3 (not shown in ), after the meteorological characteristic sequence data is extracted by the adaptive multi-scale spatial hypergraph module, an intermediate feature sequence containing spatial feature information is obtained, and then the intermediate feature sequence is transformed into ONI time series data for the past T time moments . Then, the time series data passes through the frequency-domain multi-scale time encoding and decoding module to obtain the ONI prediction result for the target sea area at the next τ time moments

[0081] Figure 4 shows the structure of the adaptive multi-scale spatial hypergraph module, as Figure 4 shown, the adaptive multi-scale spatial hypergraph module includes a multi-scale data processing unit and multiple parallel network blocks, and each network block includes a spatial graph construction unit and a hypergraph information aggregation unit.

[0082] The adaptive multi-scale spatial hypergraph module is aimed at meteorological characteristic sequence data for past time, and its purpose is to extract the spatial feature information in the meteorological characteristic sequence data for past time periods, obtain an intermediate feature sequence containing spatial feature information, and then transform it into the ONI time series data for the corresponding target sea area in past time periods. Figure 4The case where the feature dimension (i.e., the number of channels) at each moment is 1 is shown. In fact, the feature dimension at each moment is greater than 1, but the principle is the same. The functions of each unit in the adaptive multi-scale spatial hypergraph module are described as follows: The above multi-scale data processing unit (Multi-scale Data Processing Unit) can perform scale processing on the meteorological feature sequence data in the past time period. For example, a coarse-grained spatial representation can be formed based on two-dimensional convolution. Figure 4 What is shown is the processing of t - T + 1,..., t moments. Exemplarily, two-dimensional convolution can be used to perform convolution processing on the spatial grid to obtain a spatio-temporal correlation grid sequence representation (i.e., the above sequence features) of spatial grid points at multiple different grid scales, such as Figure 4 where S shown represents the number of different grid scales. Each spatio-temporal correlation grid sequence at each grid scale in the spatio-temporal correlation grid sequences at multiple grid scales output by the multi-scale data processing unit is input into a combined unit (i.e., a network block) composed of its corresponding spatial graph construction unit (Spatial Graph Construction Unit) and hypergraph information aggregation unit (Hypergraph AggregatorModule) for subsequent spatial structure modeling and aggregation.

[0083] The above spatial graph construction unit constructs a spatial graph connection structure (i.e., a hypergraph structure) at its respective grid scale according to the geographical distance between ocean physical variables or the similarity of meteorological features. For example, at each time step t in the past time (taking a month as the time step, that is, the preset duration in the above embodiment), all grid points within a grid scale at the current time step are regarded as graph nodes, and the Euclidean distance and the similarity of meteorological features between each graph node are calculated to form a weighted adjacency matrix, thereby establishing a spatial graph with physical significance. Further, in order to capture the potential dynamic correlation between distant regions, a hypergraph structure is introduced, and multiple nodes with similar change patterns are aggregated in each hyperedge, thereby constructing a hypergraph connection relationship reflecting the spatial coupling pattern.

[0084] The above hypergraph information aggregation unit uses a hypergraph convolution or attention aggregation mechanism to perform information propagation on the constructed hypergraph structure. In each time step, the module performs feature fusion in the aggregation path of node - hyperedge - node, so that each node not only aggregates the information of its adjacent nodes, but also integrates the global dynamic features of multiple nodes in the same hyperedge, thereby enhancing the spatio-temporal modeling ability within each grid scale. This aggregation process can be combined with a dynamic adjustment mechanism to adaptively adjust the hyperedge weights according to the change of the network state in each time step, and achieve effective modeling of the dynamic spatial evolution structure.

[0085] It should be noted that in a hypergraph, the node-hyperedge-node aggregation path is a mechanism for information transmission and feature update. By adopting this aggregation path, nodes and hyperedges can enhance each other, thereby better representing the complex relationships in the graph structure, which helps to model and analyze hypergraph data and has extensive applications in multiple fields such as image segmentation, gene medicine, and natural language processing. For example, in image segmentation, nodes can be pixel points, and hyperedges represent certain relationships between pixel points. Through this aggregation path, the context information in the image can be better utilized for pixel classification; in gene medicine, nodes represent genes, and hyperedges correspond to gene sets or genetic pathways, which helps to analyze the interaction and functional relationships between genes.

[0086] Figure 5 shows the structure of the frequency-domain multi-scale time encoding and decoding module, as Figure 5 shown, the frequency-domain multi-scale time encoding and decoding module includes a frequency-domain embedding unit, a frequency-domain encoding unit, a frequency-domain decoding unit, and a frequency-domain fusion unit.

[0087] The frequency-domain multi-scale time encoding and decoding module targets the ONI time series data in the past time period of the target sea area, aiming to capture the time evolution law in the ONI time series data in the past time period and predict the ONI time series data in the future time period. The functions of each unit of the frequency-domain multi-scale time encoding and decoding module are described as follows: The above-mentioned frequency-domain embedding unit (Frequency-domain Embedding Unit) transforms the input ONI time series data in the past time period from the time domain to the frequency domain to more effectively extract periodic and trend features, that is, extract time features in different frequency bands. Specifically, the discrete Fourier transform (DFT) can be used to perform window segmentation and spectrum extraction on the ONI time series data to obtain frequency component information at different local times, decoupling the long sequence signal into multiple sub-features in different frequency ranges.

[0088] The above-mentioned frequency-domain encoding unit (Frequency-domain Encoding Unit) deeply encodes the time features in different frequency bands, that is, a frequency attention map can be constructed through the frequency attention mechanism to identify the key frequency bands that have a significant impact on the ENSO evolution.

[0089] The above-mentioned frequency-domain decoding unit (Frequency-domain Decoding Unit) performs feature transformation on the time features in multiple frequency bands output by the frequency-domain encoding unit to generate prediction results for each frequency band. Here, it is equivalent to transforming the time features in multiple frequency bands output by the frequency-domain encoding unit into a unified feature space representation convenient for fusion, that is, the prediction features in the future frequency band, which are used for the fusion of subsequent prediction outputs.

[0090] The above frequency-domain fusion unit (Frequency-domain Fusion Unit) integrates the prediction results of each frequency band to predict the final ONI. That is, by applying the inverse Fourier transform (IDFT), the fused frequency-domain features are restored to a time-domain signal to predict the ONI time series data in the future time period of the target area.

[0091] Figure 6 shows the overall structure of the adaptive multi-scale spatio-temporal ENSO prediction model. As Figure 6 shown, the result of preprocessing the meteorological feature sequence of the target sea area in the target historical time period is input into the adaptive multi-scale spatial hypergraph module to obtain a multi-scale spatially enhanced meteorological feature time series, that is, the intermediate feature sequence, and the information of multiple scales of the intermediate feature sequence is aggregated through a multi-layer perceptron (MLP) to obtain the ONI time series data of the target sea area in the target historical time period , and then, this ONI time series data is input into the frequency-domain multi-scale time encoding and decoding module to predict the final result of ONI in the target future time period of the target sea area .

[0092] In summary, the present disclosure introduces an adaptive multi-scale spatial hypergraph module to dynamically model the spatio-temporal correlation structure in the meteorological feature sequence, accurately depict the time-varying spatial evolution pattern of meteorological variables in the target sea area in the time dimension, achieve efficient aggregation of spatial information, and construct an ONI time feature sequence for the historical time period with spatial dependence characteristics; subsequently, use the frequency-domain multi-scale time encoding and decoding module to perform frequency-domain modeling on the above ONI time series data of the historical time period, so as to effectively capture the long-time evolution law of ENSO-related variables, and finally predict the ONI time series data in the future time period, significantly improving the prediction efficiency, accuracy and stability of the evolution trend of ENSO events.

[0093] Figure 7 is a block diagram showing an El Niño-Southern Oscillation prediction device according to an embodiment of the present disclosure. As Figure 7 shown, the device includes an extraction unit 70, an acquisition unit 72 and a prediction unit 74.

[0094] The extraction unit 70 is configured to input the meteorological feature sequence of the target sea area in the target historical time period into the adaptive multi-scale spatial hypergraph module to obtain an intermediate feature sequence, where the target historical time period is a time period of a first preset duration before the current moment, and the intermediate feature sequence includes the spatial dependencies between the nodes of each hypergraph structure in a plurality of hypergraph structures, and each hypergraph structure is obtained based on the feature sequence at the corresponding grid scale among a plurality of grid scales of the meteorological feature sequence; the acquisition unit 72 is configured to obtain the Ocean Nino Index of the target historical time period based on the intermediate feature sequence; the prediction unit 74 is configured to input the Ocean Nino Index of the target historical time period and the target future time period into the frequency-domain multi-scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period, where the target future time period is a time period of a second preset duration after the current moment, and the Ocean Nino Index of the target future time period characterizes the degree of the El Nino-Southern Oscillation phenomenon occurring in the target future time period.

[0095] According to an embodiment of the present disclosure, the adaptive multi-scale spatial hypergraph module includes a multi-scale data processing unit and a plurality of parallel network blocks, and the plurality of grid scales correspond to the plurality of parallel network blocks one by one. Among them, the extraction unit 70 is further configured to input the meteorological feature sequence of the target historical time period into the multi-scale data processing unit to obtain the feature sequences of the meteorological feature sequence at a plurality of grid scales; for the feature sequence at each grid scale among the plurality of grid scales, input the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence, where the hidden feature includes the spatial dependencies between the nodes in the hypergraph structure of the current feature sequence; determine the hidden features of the feature sequences at the plurality of grid scales as the intermediate feature sequence.

[0096] According to an embodiment of the present disclosure, the network block includes a spatial graph construction unit and a hypergraph information aggregation unit. Among them, the extraction unit 70 is further configured to input the current feature sequence into the spatial graph construction unit to obtain the hypergraph structure of the current feature sequence, where the hypergraph structure of the current feature sequence includes the hypergraph structures of a plurality of preset durations in the target historical time period; input the hypergraph structure into the hypergraph information aggregation unit to obtain the hidden feature of the current feature sequence.

[0097] According to an embodiment of the present disclosure, the frequency-domain multi-scale time encoding and decoding module includes a frequency-domain embedding unit, a frequency-domain encoding unit, a frequency-domain decoding unit, and a frequency-domain fusion unit. Among them, the prediction unit 74 is further configured to input the Ocean Nino Index of the target historical time period into the frequency-domain embedding unit to obtain time features in different frequency bands; input the time features in different frequency bands into the frequency-domain encoding unit to obtain encoded features; input the encoded features and the target future time period into the frequency-domain decoding unit to obtain predicted features in the future frequency band; input the predicted features in the future frequency band into the frequency-domain fusion unit to obtain the Ocean Nino Index of the target sea area in the target future time period.

[0098] According to an embodiment of the present disclosure, the training unit is configured to train the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module in the following manner: Based on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature among various global meteorological features, obtain a training sample set, where each training sample includes a meteorological feature sequence with a first predetermined duration and the true Ocean Nino Index with a second predetermined duration after the first predetermined duration; for each training sample, perform the following processing: input the meteorological feature sequence with the first predetermined duration in the current training sample into the adaptive multi-scale spatial hypergraph module to obtain an estimated intermediate feature sequence; based on the estimated intermediate feature sequence, obtain a first estimated Ocean Nino Index, where the first estimated Ocean Nino Index is the estimated Ocean Nino Index with the first predetermined duration; input the first estimated Ocean Nino Index and the second predetermined duration after the first predetermined duration into the frequency-domain multi-scale time encoding and decoding module to obtain a second estimated Ocean Nino Index, where the second estimated Ocean Nino Index is the estimated Ocean Nino Index with the second predetermined duration after the first predetermined duration; based on the second estimated Ocean Nino Index and the true Ocean Nino Index, adjust the parameters of the adaptive multi-scale spatial hypergraph module and the frequency-domain multi-scale time encoding and decoding module.

[0099] According to an embodiment of the present disclosure, the training unit is further configured to obtain various meteorological features of the global sea surface; preprocess the various meteorological features of the global sea surface to obtain the meteorological feature sequences and the Ocean Nino Index of each meteorological feature; perform sequential sliding sampling on the meteorological feature sequences and the Ocean Nino Index of each meteorological feature; based on the sampling results, obtain a training sample set. Figure 8 Shows a schematic structural diagram of an electronic device provided by the present disclosure, such as Figure 8As shown in the figure, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 may invoke the logic instructions in the memory 830 to execute any of the ENSO prediction methods provided in the foregoing embodiments.

[0100] In addition, when the logic instructions in the foregoing memory 830 are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0101] It should be noted here that the electronic device of the present disclosure can implement all the method steps implemented in the foregoing method embodiments and can achieve the same technical effects. Therefore, the same parts and beneficial effects as those in the method embodiments will not be specifically described in this embodiment.

[0102] According to an embodiment of the present disclosure, there is provided a computer-readable storage medium storing instructions, wherein when the instructions are run by at least one computing device, at least one computing device is caused to execute the El Niño - Southern Oscillation prediction method as described in any of the foregoing embodiments.

[0103] It should be noted here that the non-transitory computer-readable storage medium provided by the present disclosure can implement all the method steps implemented in the foregoing method embodiments and can achieve the same technical effects. Therefore, the same parts and beneficial effects as those in the method embodiments will not be specifically described in this embodiment.

[0104] According to an embodiment of the present disclosure, there is provided a system including at least one computing device and at least one storage device storing instructions, wherein when the instructions are run by at least one computing device, at least one computing device is caused to execute the El Niño - Southern Oscillation prediction method as described in any of the foregoing embodiments.

[0105] According to an embodiment of the present disclosure, a computer program product is provided, including computer instructions which, when executed by a processor, implement the El Niño - Southern Oscillation prediction method as described above.

[0106] It should be noted here that the computer program product provided by the present disclosure can implement all the method steps implemented by the above - mentioned method embodiment and can achieve the same technical effects. The same parts and beneficial effects as those in the method embodiment will not be specifically described in this embodiment.

[0107] Although some embodiments of the present disclosure have been shown and described, those skilled in the art should understand that these embodiments can be modified without departing from the principles and spirit of the present disclosure as defined by the claims and their equivalents.

[0108] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0109] Through the description of the above - mentioned implementation manners, those skilled in the art can clearly understand that each implementation manner can be realized by means of software plus a necessary general - purpose hardware platform, and of course, it can also be realized by hardware. Based on such an understanding, the above - mentioned technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer - readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

Claims

1. An El Niño - Southern Oscillation prediction method, characterized in that, Including: Input the meteorological feature sequence of the target sea area in the target historical time period into the adaptive multi-scale spatial hypergraph module to obtain an intermediate feature sequence, where the target historical time period is a time period of a first predetermined duration before the current moment, and the intermediate feature sequence includes the spatial dependence relationships between the nodes of each hypergraph structure in a plurality of hypergraph structures, and each hypergraph structure is obtained based on the feature sequence of the corresponding grid scale among a plurality of grid scales of the meteorological feature sequence; Based on the intermediate feature sequence, obtain the Ocean Nino Index of the target historical time period; Input the Ocean Nino Index of the target historical time period and the target future time period into the frequency-domain multi-scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period, where the target future time period is a time period of a second predetermined duration after the current moment, and the Ocean Nino Index of the target future time period characterizes the degree of the El Nino-Southern Oscillation phenomenon occurring in the target future time period.

2. The method according to claim 1, wherein The adaptive multi-scale spatial hypergraph module includes a multi-scale data processing unit and a plurality of parallel network blocks, and the plurality of grid scales correspond one-to-one to the plurality of parallel network blocks. Among them, inputting the meteorological feature sequence of the target sea area in the target historical time period into the adaptive multi-scale spatial hypergraph module to obtain an intermediate feature sequence includes: Input the meteorological feature sequence of the target historical time period into the multi-scale data processing unit to obtain the feature sequences of the meteorological feature sequence at a plurality of grid scales; For the feature sequence at each grid scale among the plurality of grid scales, input the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence, where the hidden feature includes the spatial dependence relationships between the nodes in the hypergraph structure of the current feature sequence; Determine the hidden features of the feature sequences at the plurality of grid scales as the intermediate feature sequence.

3. The method according to claim 2, wherein Among them, The network block includes a spatial graph construction unit and a hypergraph information aggregation unit. Among them, inputting the current feature sequence into the network block corresponding to the current feature sequence to obtain the hidden feature of the current feature sequence includes: Input the current feature sequence into the spatial graph construction unit to obtain the hypergraph structure of the current feature sequence, where the hypergraph structure of the current feature sequence includes the hypergraph structures of a plurality of preset durations in the target historical time period; Input the hypergraph structure into the hypergraph information aggregation unit to obtain the hidden feature of the current feature sequence.

4. The method according to claim 1, wherein, The frequency-domain multi-scale time encoding and decoding module includes a frequency-domain embedding unit, a frequency-domain encoding unit, a frequency-domain decoding unit, and a frequency-domain fusion unit. Among them, inputting the Ocean Nino Index of the target historical time period and the target future time period into the frequency-domain multi-scale time encoding and decoding module to obtain the Ocean Nino Index of the target sea area in the target future time period includes: Input the Ocean Nino Index of the target historical time period into the frequency-domain embedding unit to obtain time features of different frequency bands; Input the time features of different frequency bands into the frequency-domain encoding unit to obtain encoded features; Input the encoded feature and the target future time period into the frequency domain decoding unit to obtain the predicted feature of the future frequency band; Input the predicted feature of the future frequency band into the frequency domain fusion unit to obtain the Ocean Nino Index of the target sea area in the target future time period.

5. The method according to claim 1, characterized in that The adaptive multi-scale spatial hypergraph module and the frequency domain multi-scale time encoding and decoding module are trained in the following manner: Based on the meteorological feature sequence of each meteorological feature among various meteorological features of the global sea surface and the Ocean Nino Index, obtain a training sample set, where each training sample includes the meteorological feature sequence of the first predetermined duration and the true Ocean Nino Index of the second predetermined duration after the first predetermined duration; For each training sample, perform the following processing: Input the meteorological feature sequence of the first predetermined duration in the current training sample into the adaptive multi-scale spatial hypergraph module to obtain a predicted intermediate feature sequence; Based on the predicted intermediate feature sequence, obtain a first predicted Ocean Nino Index, where the first predicted Ocean Nino Index is the predicted Ocean Nino Index of the first predetermined duration; Input the first predicted Ocean Nino Index and the second predetermined duration after the first predetermined duration into the frequency domain multi-scale time encoding and decoding module to obtain a second predicted Ocean Nino Index, where the second predicted Ocean Nino Index is the predicted Ocean Nino Index of the second predetermined duration after the first predetermined duration; Based on the second predicted Ocean Nino Index and the true Ocean Nino Index, adjust the parameters of the adaptive multi-scale spatial hypergraph module and the frequency domain multi-scale time encoding and decoding module.

6. The method according to claim 5, wherein The obtaining of the training sample set based on the meteorological feature sequence of each meteorological feature among various meteorological features of the global sea surface and the Ocean Nino Index includes: Obtain various meteorological features of the global sea surface; Preprocess various meteorological features of the global sea surface to obtain the meteorological feature sequence and the Ocean Nino Index of each meteorological feature; Perform sequential sliding sampling on the meteorological feature sequence and the Ocean Nino Index of each meteorological feature; Based on the sampling result, obtain the training sample set.

7. An El Niño - Southern Oscillation prediction device, characterized in that, Includes: An extraction unit configured to input the meteorological feature sequence of the target sea area in the target historical time period into the adaptive multi-scale spatial hypergraph module to obtain an intermediate feature sequence, where the target historical time period is a time period of the first predetermined duration before the current moment, and the intermediate feature sequence includes the spatial dependence relationship between the nodes of each hypergraph structure among multiple hypergraph structures, and each hypergraph structure is obtained based on the feature sequence at the corresponding grid scale among multiple grid scales of the meteorological feature sequence; An obtaining unit configured to obtain the Ocean Nino Index of the target historical time period based on the intermediate feature sequence; A prediction unit, configured to input the Ocean Nino Index of the target historical time period and the target future time period into a frequency-domain multi-scale time encoding and decoding module, to obtain the Ocean Nino Index of the target sea area in the target future time period, where the target future time period is a time period of a second predetermined duration after the current moment, and the Ocean Nino Index of the target future time period characterizes the degree of occurrence of the El Nino-Southern Oscillation phenomenon in the target future time period.

8. A computer-readable storage medium storing instructions, characterized in that, When the instruction is run by at least one computing device, it causes the at least one computing device to execute the El Nino-Southern Oscillation prediction method according to any one of claims 1 to 6.

9. A system comprising at least one computing device and at least one storage device storing instructions, characterized in that, When the instruction is run by the at least one computing device, it causes the at least one computing device to execute the El Nino-Southern Oscillation prediction method according to any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that, When the computer instruction is executed by a processor, it implements the El Nino-Southern Oscillation prediction method according to any one of claims 1 to 6.

Images