| 摘 要: 通过人工智能领域中的自然语言工具将不直观的专业海上风速预报数据转化为友好易懂的分析文本，提高气象预报服务的时效性和合理性。首先提取中国海洋石油公司南海东部四个海上区块的原始预报数据，利用神经网络对数据进行回归拟合，得到一条显示未来风速上升下降的单调区间、极值点和增减幅度的光滑曲线，再利用自然语言处理基于规则和基于统计相结合的方法自动生成预报文本。该方法可将现行需要人工45—60 分钟才能完成的预报报文减少到2—4 分钟内完成，大大提高了海洋气象预报的工作效率。 |
| 关键词: 海上大风预报；自动生成；自然语言处理；神经网络拟合 |
|
中图分类号: TP39
文献标识码: A
|
| 基金项目: 国家自然科学基金国际(地区)合作交流项目“中南半岛和华南极端降水事件年代际变化的机理、影响和未来预估”(41861144014)；国家自然科学基金杰出青年科学基金项目“海气自然变率影响青藏高原热源的过程和机理”(41725018). |
|
| Study on Automatic Generation of Marine Wind Forecast Text by Artificial Intelligence |
|
JIAN Jun1 , WANG Heng1 , SUN Zheng1 , WU Guanlin1 , SU Xin1 , CHEN Sanjun2
|
( 1. Dalian Maritime University, Dalian 116026, China;
2. Shenzhen Branch, CNOOC (China ) Co., Ltd., Shenzhen 518067, China)
jianjun@dlmu.edu.cn; wheng@dlmu.edu.cn; szheng@dlmu.edu.cn; wuguanlin@dlmu.edu.cn; sux@dlmu.edu.cn; chenshj3@cnooc.com.cn
|
| Abstract: This paper proposes to apply Artificial Intelligence-Natural Language tool to transform professional and non-intuitive meteorological data into understandable and user-friendly texts, thus improving the speed and rationality of meteorological services. The raw meteorological forecast data of the CNOOC's four offshore blocks in Eastern South China Sea is retrieved firstly, and the neural network is applied to perform regression fitting on the data, so to generate a smooth curve showing the monotonic interval, extreme points, and increase/decrease range of future wind speed. Then, Natural Language tool is applied to process the forecast texts which are automatically generated based on a combination of rules and statistics. Using the proposed method, the current 45-60 minutes manual forecasting work could be completed within 2-4 minutes, which greatly improves the efficiency of sea weather forecasting. |
| Keywords: sea wind forecast; automatic generation; natural language processing; neural network fitting |
