摘要
为了提高降水场预测模型的学习效率与预测性能,在预测模型的训练阶段提出一个改善的训练策略,使其可以充分学习物体运动轨迹以及物体运动时的外观变化。通过在一个雷达回波数据集和一个公开数据集上进行对应实验,可以显示出该方法在两项指标的性能表现上具有明显提高,证明了该方法的有效性。
In order to enhance the learning efficiency and predictive performance of precipitation field forecasting models,an improved training strategy during the training phase of the prediction model was proposed.This strategy enabled the model to fully learn the trajectories of object movements as well as the appearance changes of objects during movement.Through corresponding experiments conducted on a radar echo dataset and a publicly available dataset,it was demonstrated that this method could significantly improved the performance on two metrics,thus validating its effectiveness.
作者
于霞
朱智睿
段勇
李冰洁
杨海波
YU Xia;ZHU Zhirui;DUAN Yong;LI Bingjie;YANG Haibo(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《沈阳大学学报（自然科学版）》
CAS
2024年第2期121-131,共11页
Journal of Shenyang University：Natural Science
基金
辽宁省教育厅服务地方项目(LJKFZ20220184)。
关键词
机器学习
深度学习
降水预测
循环神经网络
帧预测
machine learning
deep learning
precipitation prediction
recurrent neural network
frame prediction
