Method for predicting dynamic response of offshore engineering structure based on recursive deep learning

By employing recursive deep learning methods, combined with convolutional layers and bidirectional long short-term memory networks, the problem of model decoupling under multiple excitations of earthquakes and waves was solved, enabling efficient long-term prediction of the dynamic response of marine engineering structures and enhancing the model's generalization ability and prediction stability.

CN122154484APending Publication Date: 2026-06-05OCEAN UNIV OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
OCEAN UNIV OF CHINA
Filing Date
2026-04-09
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

When dealing with multi-electromagnetic excitations of earthquakes and waves, existing technologies struggle to effectively decouple the contribution of these excitations to the structural response, resulting in weak generalization ability, inability to handle long sequences, and problems of error accumulation and phase drift during long-term extrapolation.

Method used

A recursive deep learning-based approach is adopted to extract local spatial features of multi-source environmental stimuli through convolutional layers. Combined with a bidirectional long short-term memory network and a feature cascade linear modulation mechanism, the deep fusion of static physical parameters and dynamic temporal features is achieved. Furthermore, the recursive architecture is used to pass historical features for long-term time-series prediction.

Benefits of technology

It enhances the model's generalization ability, enabling it to handle dynamic response sequences of arbitrary length, ensuring the physical consistency and stability of the predictions, and effectively solving the error accumulation problem in long-term extrapolation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122154484A_ABST
    Figure CN122154484A_ABST
Patent Text Reader

Abstract

The application discloses a kind of marine engineering structure dynamic response prediction method based on recursive deep learning, belong to engineering structure safety and disaster prevention technical field.This method first divides long-time excitation sequence into overlapping time segments;For first segment, dynamic characteristics are extracted by convolution layer and bidirectional long short memory network, and static physical parameters are used to perform dimension-by-dimension fusion on time series characteristics by FiLM mechanism;For subsequent segments, the hidden state and cell state of the previous segment are recursively passed as initial memory, and its latent space feature abstract is introduced to maintain long-term consistency;Finally, the complete response sequence is output through the linear fading splicing strategy of the overlapping area.The application effectively decouples the contribution of multiple sources of excitation, significantly improves the model generalization ability and the stability of long sequence extrapolation, and is suitable for real-time safety evaluation of key infrastructure such as offshore platforms.
Need to check novelty before this filing date? Find Prior Art