CN-121503700-B - Deep learning-based ocean wave large model construction method

CN121503700BCN 121503700 BCN121503700 BCN 121503700BCN-121503700-B

Abstract

The invention relates to the technical field of deep learning, in particular to a method for constructing a large ocean wave model based on deep learning. The method comprises the steps of obtaining multi-source ocean association monitoring data, carrying out spatial interpolation mapping processing of ocean association monitoring features through the multi-source ocean association monitoring data to generate ocean association monitoring feature spatial mapping data, carrying out ocean characterization feature analysis processing of ocean association heterogeneous feature channels on the ocean association monitoring feature spatial mapping data to generate ocean characterization feature data, carrying out ocean characterization and evolution prediction mapping relation establishment on the basis of a preset deep learning algorithm and the ocean characterization feature data to generate an ocean evolution prediction model, obtaining historical ocean characterization evolution data, carrying out model training and ocean evolution constraint optimization processing on the ocean evolution prediction model by utilizing the historical ocean characterization evolution data to generate an optimized ocean evolution prediction model. The invention realizes accurate prediction analysis of ocean wave conditions.

Inventors

ZHANG CHENYU
LI DEMING
MA XIAODONG
HU XINMIAO
LIU XIUJIE
SONG MENG
SUN YUXIU

Assignees

青岛埃克曼科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260112

Claims (3)

1. The ocean wave large model construction method based on deep learning is characterized by comprising the following steps of: Step S1, acquiring multi-source ocean correlation monitoring data, performing spatial interpolation mapping processing of ocean correlation monitoring characteristics through the multi-source ocean correlation monitoring data, and generating ocean correlation monitoring characteristic spatial mapping data; wherein, step S1 comprises the following steps: Step S11, multi-source ocean associated monitoring data are obtained, wherein the multi-source ocean associated monitoring data comprise ocean typhoon path intensity data, sea surface wind field data, sea surface air pressure data, sea wave monitoring data and ocean geographic information data, and the ocean geographic information data comprise ocean astronomical tide data, ocean water depth data, ocean submarine topography data and ocean offshore line data; Step S12, performing preliminary three-dimensional space modeling of ocean topography based on ocean geographic information data to obtain a preliminary ocean topography three-dimensional space model, performing ocean geographic boundary feature analysis according to the ocean geographic information data to obtain ocean geographic boundary feature data, performing self-adaptive mapping grid design of ocean geographic spatial features according to the preliminary ocean topography three-dimensional space model and the ocean geographic boundary feature data to obtain ocean geographic space mapping grid data, and performing mapping grid establishment optimization treatment of the preliminary ocean topography three-dimensional space model to obtain an ocean topography three-dimensional space model; Step S13, carrying out multi-source data time sequence synchronous reference preprocessing on multi-source marine associated monitoring data to generate synchronous multi-source marine associated monitoring data; step S14, transmitting the synchronous multi-source ocean correlation monitoring data to an ocean topography three-dimensional space model to perform space interpolation mapping processing of ocean correlation monitoring features to generate ocean correlation monitoring feature space mapping data, and step S2, performing ocean characterization feature analysis processing of ocean correlation heterogeneous feature channels on the ocean correlation monitoring feature space mapping data to generate ocean characterization feature data; Step S3, establishing a prediction mapping relation for wave characterization and evolution based on a preset deep learning algorithm and wave characterization characteristic data, and generating a wave evolution prediction model; wherein, step S3 includes the following steps: Step S31, establishing a deep learning large model framework for wave characterization and evolution based on a preset deep learning algorithm, and generating a wave evolution prediction model framework, wherein the preset deep learning algorithm adopts a hybrid framework combining a transducer encoder and a convolutional neural network; Step S32, performing sea wave characterization spatial multi-scale feature analysis according to sea wave characterization feature data to generate sea wave characterization spatial multi-scale feature data, performing time sequence feature analysis on the sea wave characterization spatial multi-scale feature data to generate sea wave characterization space multi-scale feature data, performing sea wave characterization clustering attribute analysis according to the sea wave characterization feature data to generate sea wave characterization clustering attribute data, designing a sea wave characterization spatial multi-scale fusion strategy of clustering attribute differences through the sea wave characterization clustering attribute data, performing time sequence dynamic sea wave characterization spatial multi-scale fusion feature analysis on the sea wave characterization spatial multi-scale feature data based on the sea wave characterization spatial multi-scale fusion strategy to generate sea wave characterization space-time cross-scale fusion feature data; s33, carrying out wave characterization space-time cross-scale fusion layer embedding processing on a wave evolution prediction model framework through wave characterization space-time cross-scale fusion characteristic data to generate an embedded wave evolution prediction model framework; Step S34, carrying out evolution influence analysis of the wave characterization singlestate according to the wave characterization characteristic data to generate wave characterization singlestate evolution influence data, carrying out space specificity analysis of the wave characterization superposition state on the wave characterization space-time cross-scale fusion characteristic data to generate wave characterization superposition state space specificity data, carrying out evolution influence analysis of the wave characterization superposition state according to the wave characterization superposition state space specificity data to generate wave characterization superposition state evolution influence data, and carrying out multi-mode wave evolution self-adaptive attention strategy design according to the wave characterization singlestate evolution influence data and the wave characterization superposition state evolution influence data to generate a wave evolution multi-mode self-adaptive attention strategy; Step S35, setting and processing an overlay state attention mechanism of the wave characterization space scene by using a wave evolution multi-mode self-adaptive attention strategy to an embedded wave evolution prediction model framework to obtain a wave evolution prediction model; step S4, historical wave characterization evolution data are obtained, model training and wave evolution constraint optimization processing are carried out on the wave evolution prediction model by utilizing the historical wave characterization evolution data, and an optimized wave evolution prediction model is generated; Wherein, step S4 includes the following steps: Step S41, historical sea wave characterization evolution data are obtained; S42, analyzing the wave evolution physical constraint characteristic of the historical wave characterization evolution data to generate wave evolution physical constraint characteristic data; Step S43, historical wave characterization evolution data are transmitted to a wave evolution prediction model for training treatment, and a training wave evolution prediction model is generated; Step S44, carrying out wave evolution constraint training target decision analysis according to the wave evolution physical constraint characteristic data, generating a wave evolution constraint training target decision, carrying out wave evolution constraint auxiliary loss function design based on the wave evolution constraint training target decision, generating a wave evolution constraint auxiliary loss function, and carrying out wave evolution constraint model training parameter optimization adjustment processing on the training wave evolution prediction model by utilizing the wave evolution constraint training target decision and the wave evolution constraint auxiliary loss function, so as to generate an optimized wave evolution prediction model.
2. The deep learning-based ocean wave large model construction method according to claim 1, wherein the step S2 comprises the steps of: s21, analyzing the influence relation between the ocean correlation heterogeneous characteristic channels and the ocean characterization according to the ocean correlation monitoring characteristic space mapping data, and generating ocean correlation heterogeneous characteristic-characterization influence relation data; S22, carrying out sea wave characterization influence relation integrated analysis according to sea wave association heterogeneous characteristic-characterization influence relation data to generate sea wave characterization influence relation integrated data; and S23, carrying out wave characterization feature analysis on the wave characterization influence relation integrated data to generate wave characterization feature data.
3. The deep learning-based ocean wave large model construction method according to claim 2, wherein the step S21 comprises the steps of: Step S211, extracting characteristic space mapping data of ocean typhoon path intensity data, sea surface wind field data, sea surface air pressure data, ocean water depth data and ocean bottom topography data according to ocean correlation monitoring characteristic space mapping data, and carrying out ocean storm tide representation analysis through the characteristic space mapping data of the ocean typhoon path intensity data, the sea surface wind field data, the sea surface air pressure data, the ocean water depth data and the ocean bottom topography data to generate ocean storm tide representation data; Step S212, extracting characteristic space mapping data of ocean astronomical tide data according to ocean association monitoring characteristic space mapping data, and carrying out ocean wave tide level characterization analysis by combining the characteristic space mapping data of the ocean astronomical tide data with the ocean storm tide characterization data to generate ocean wave tide level characterization data; Step S213, extracting characteristic space mapping data of sea wave monitoring data and sea coastline data according to sea association monitoring characteristic space mapping data, and carrying out sea wave flood characterizing analysis by combining the characteristic space mapping data of the sea wave monitoring data and the sea coastline data through sea wave storm tide characterizing data and sea wave tide level characterizing data to generate sea wave flood characterizing data; And step S214, analyzing the influence relation between the wave association heterogeneous characteristic channel and the wave characterization based on the wave storm tide characterization data, the wave tide level characterization data and the wave flood characterization data, and generating wave association heterogeneous characteristic-characterization influence relation data.

Description

Deep learning-based ocean wave large model construction method Technical Field The invention relates to the technical field of deep learning, in particular to a method for constructing a large ocean wave model based on deep learning. Background Ocean sea waves serve as a core component of an ocean power environment, and the evolution process of the ocean sea waves directly affects a plurality of important fields such as ocean engineering construction, shipping safety guarantee, ocean disaster prevention and reduction. The method can accurately predict the generation, development and evolution rules of sea waves, and can provide key technical support for disaster early warning of coastal area infrastructure planning, offshore operation scheduling, typhoon storm surge and the like. With the aggravation of global climate change, extreme ocean weather events are frequent, the wave evolution process presents stronger nonlinearity, randomness and space-time correlation, and the prediction requirements of high precision and high timeliness are required to be met. However, the existing sea wave prediction method has a short plate in the aspect of processing multisource sea monitoring data, is difficult to convert scattered and heterogeneous sea association monitoring data into unified characteristic space data which can be used for model training, in sea wave characterization characteristic analysis, cannot systematically establish influence relations between multisource heterogeneous characteristic channels and sea wave characterization, so that the effectiveness and pertinence of model input characteristics are insufficient, moreover, a single data source is adopted for time sequence or space characteristic modeling, the result that sea wave evolution is the comprehensive effect of multisource heterogeneous factors is ignored, complex association characteristics of sea wave evolution are difficult to comprehensively describe, the space-time cross-scale characteristic fusion capability of sea wave characterization is lacked, sea wave evolution rules of different spatial scales (such as offshore and open sea) cannot be effectively captured, and meanwhile, the physical constraint characteristics of sea wave evolution are not fully combined, so that the model prediction result is easy to deviate from the actual physical rules, and the prediction stability and reliability under the sea condition are insufficient. Disclosure of Invention Based on the above, the invention provides a deep learning-based ocean wave large model construction method to solve at least one of the above technical problems. In order to achieve the above purpose, the ocean wave large model construction method based on deep learning comprises the following steps: Step S1, acquiring multi-source ocean correlation monitoring data, performing spatial interpolation mapping processing of ocean correlation monitoring characteristics through the multi-source ocean correlation monitoring data, and generating ocean correlation monitoring characteristic spatial mapping data; s2, carrying out sea wave characterization feature analysis processing on sea wave correlation monitoring feature space mapping data to generate sea wave characterization feature data; Step S3, establishing a prediction mapping relation for wave characterization and evolution based on a preset deep learning algorithm and wave characterization characteristic data, and generating a wave evolution prediction model; and S4, acquiring historical wave characterization evolution data, and carrying out model training and wave evolution constraint optimization processing on the wave evolution prediction model by utilizing the historical wave characterization evolution data to generate an optimized wave evolution prediction model. The method has the advantages that the method ensures the comprehensiveness of data coverage by acquiring multi-dimensional multi-source ocean associated monitoring data covering ocean typhoon path intensity, sea surface wind field, sea surface air pressure, sea wave monitoring, ocean astronomical tide, water depth, submarine topography, near shore line and the like, lays a foundation for subsequent accurate modeling, simultaneously eliminates the time sequence deviation of multi-source heterogeneous data by time sequence synchronous reference preprocessing, combines an ocean topography three-dimensional space model constructed and optimized based on ocean geographic information data to develop spatial interpolation mapping processing, solves the problem that the distributed heterogeneous data is difficult to integrate, realizes accurate matching of monitoring characteristics and real ocean geographic space, and generates ocean associated monitoring characteristic space mapping data with time sequence consistency and space accuracy, thereby providing high quality for subsequent characteristic analysis, And the adaptive mapping grid design and optimization process of t