摘要
利用基于深度学习的多隐层结构的时空序列预测神经网络,对风-浪实况大数据进行信息挖掘并构建智能预报模型,推理大洋-海域尺度非平稳态浪场时空演化过程,并在业务应用中与数值模型进行对比。结果表明:大数据驱动的智能预报的精度与数值预报相当;费效比比数值预报降低近700倍;业务流程与数值预报几乎一致,便于改造系统;业务应用情景比数值预报更广泛。此外,高效灵活的智能预报技术与新型计算设备相结合,可使海浪预报从业务中心进一步下沉到新兴的涉海行业实体中。
The spatiotemporal sequence prediction neural network with multiple hidden layer structure based on deep learning is used to mine the historical wind-wave big data, and to deduce the spatiotemporal evolution process of unsteady wave field from open ocean to regional sea scale, which is compared with the application of numerical model in operation. The results show that the accuracy of intelligent prediction driven by big data is nearly equivalent to that of numerical prediction and the cost-efficiency ratio is nearly 700 times lower than that of numerical prediction. Moreover, the operational process is consistent with the numerical prediction, which is convenient for system transformation. Therefore, operational application scenarios of intelligent prediction are broader than those of numerical prediction. In addition, the combination of efficient and flexible intelligent prediction technology and new computing equipment can make the wave prediction possible in marine-related industry entities beside that in the operational centers.
作者
屈远
高志一
蔡靖泽
王久珂
侯放
QU Yuan;GAO Zhiyi;CAI Jingze;WANG Jiuke;HOU Fang(National Marine Environmental Forecasting Center,Beijing 100081,China)
出处
《海洋预报》
CSCD
北大核心
2022年第5期17-26,共10页
Marine Forecasts
基金
国家重点研发计划(2018YFC1407001)。
关键词
海浪预报
智能预报
深度学习
大数据
费效比
wave forecast
intelligent prediction
deep learning
big data
cost-efficiency ratio
