Incremental Learning for El Nino Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network-based methods for predicting El Nino extreme weather events, particularly extreme rainfall, lack expandability and temporal/spatial inheritance when faced with new data, leading to inadequate prediction accuracy and limited practicality in real-world applications.
Innovation Solution
An El Nino extreme weather warning method and device utilizing incremental learning, which involves down-sampling marine data, extracting multi-scale features through parallel convolutional neural networks, selectively constraining low-frequency component drift using multi-scale feature frequency domain distillation, and adaptively fusing features for enhanced learning, thereby improving prediction accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional convolutional neural networks are used for El Nino prediction, then basic prediction functionality is achieved, but expandability and temporal-spatial inheritance are insufficient when new data emerges
Solution Approach 1:
The patent implements incremental learning that enables the neural network to dynamically update its parameters as new El Nino data becomes available. The system transitions from a static model to a dynamic one that continuously adapts to new patterns while preserving learned knowledge through parameter freezing and selective updating mechanisms.
Solution Approach 2:
The patent divides the neural network into multiple parallel convolutional networks, each specializing in different aspects of El Nino prediction. This segmentation allows independent training and updating of specific network components, improving expandability while maintaining overall prediction accuracy through modular architecture.
2Adaptability or versatility
If neural network parameters are updated with new data, then adaptability to new patterns improves, but knowledge from old tasks is forgotten
Solution Approach 1:
The patent freezes parameters of existing parallel convolutional networks before training new ones. This preliminary action preserves learned knowledge from old tasks while creating a stable foundation for incorporating new patterns, preventing catastrophic forgetting through parameter isolation.
Solution Approach 2:
The patent introduces a parameter freezing mechanism as an intermediary between old and new knowledge. By selectively freezing parameters during incremental training, the system acts as a mediator that allows new patterns to be learned without disrupting established knowledge representations in the neural network.
3Reliability
If multi-scale features are fully fused, then comprehensive information is captured, but ability to adapt to different time spans is reduced
Solution Approach 1:
The patent applies different fusion strategies to different multi-scale features based on their temporal characteristics. Rather than uniform fusion, the system adapts the fusion approach for each scale and time span, allowing optimal adaptation to varying prediction horizons while maintaining comprehensive information capture through localized fusion quality adjustments.
Data Source
AI summary
The present invention discloses an El Nino extreme weather warning method based on incremental learning, comprising: through supervised representation learning, selectively constraining, by a multi-scale feature frequency domain distillation technology, drift of low-frequency components of the multi-scale features based on incremental training, and memorizing knowledge learned by the parallel convolutional neural networks in old tasks; adaptively learning different fusion parameters according to different time spans of the input multi-scale data by using a multi-scale feature adaptive fusion technology, so as to enhance the ability to learn new tasks; and outputting a Nino3.4 index reflecting a change rule of El Nino through fully connected layers according to the adaptively fused features, establishing a mapping function of an extreme rainfall probability r based on the Nino3.4 index, and in response to predicting that the value r goes beyond a threshold value k.