动态选择专家模型，用更少参数和数据实现高精度天气预报
VA-MoE: Variables-Adaptive Mixture of Experts for Incremental Weather Forecasting
- 根据输入上下文动态选专家，自适应融合不同气象变量模式
- 仅用25%参数、50%训练数据，达到顶尖模型的1天与5天预报水平
- 适合需要实时更新、计算资源有限的持续天气预测场景
本文提出变量自适应混合专家（VA-MoE）框架，用于增量式天气预报。传统模型在实时数据更新时面临计算开销大、需频繁重训练的问题。VA-MoE采用专家混合架构，各专家专注捕捉温度、湿度、风速等大气变量的特定子模式，并通过变量自适应门控机制，依据输入上下文动态选择并组合相关专家，实现高效知识迁移与参数共享。该设计显著降低计算负担，同时保持高预测精度。在真实世界ERA5数据集上的实验表明，VA-MoE在短时（1天）与长时（5天）预报任务中表现接近当前最优模型，仅需约25%的可训练参数和50%的初始训练数据。
原文摘要 · Abstract (English)
This paper presents Variables Adaptive Mixture of Experts (VAMoE), a novel framework for incremental weather forecasting that dynamically adapts to evolving spatiotemporal patterns in real time data. Traditional weather prediction models often struggle with exorbitant computational expenditure and the need to continuously update forecasts as new observations arrive. VAMoE addresses these challenges by leveraging a hybrid architecture of experts, where each expert specializes in capturing distinct subpatterns of atmospheric variables (temperature, humidity, wind speed). Moreover, the proposed method employs a variable adaptive gating mechanism to dynamically select and combine relevant experts based on the input context, enabling efficient knowledge distillation and parameter sharing. This design significantly reduces computational overhead while maintaining high forecast accuracy. Experiments on real world ERA5 dataset demonstrate that VAMoE performs comparable against SoTA models in both short term (1 days) and long term (5 days) forecasting tasks, with only about 25% of trainable parameters and 50% of the initial training data.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。