摘要
成山头处于山东半岛最东端,由于其特殊的地理位置,使得成山头站的风对黄海海面风有很好的指示作用。本文使用2005年10月—2010年9月NCEP再分析资料(1o×1o)和实况观测资料,采用动态学习率BP网络(前馈反向传播Back Propagation,简称BP网络)算法的人工神经网络建立模型,在对T639数值预报产品解释释用基础上,针对成山头站进行了24 h和48 h模拟预测。模型预测结果显示,BP网络模型对成山头站的风力预报相对T639模式的平均绝对误差降低了28.2%(24 h)和19.7%(48 h)。对容易致灾的6级以上大风准确率提高显著,尤其是在T639模式对8级以上大风完全漏报情况下,BP模型在24 h仍有25%的预报准确率,48 h能达到50%的准确率。
Due to the special position of ChengShanTou, located at the Eastern most Shandong Peninsula, the winds at ChengShanTou can represent the sea surface winds over the entire Yellow Sea. Based on the NCEP re- analysis data （1°×1°） and the observed data from October 2005 to September 2010, and the interpretation of nu- merical forecasting products of the T639 model, an artificial neural network model is constructed with the dynam- ically learning rate back propagation （BP） algorithm to predict the winds at ChengShanTou for 24 and 48 hours. Compared to the winds in ChengShanTou forecasted by the T639 model, the BP model decreases the mean abso- lute errors of 28.2% for the 24-hour forecast and of 19.7% for the 48-hour forecast. The prediction accuracy is al- so improved obviously for strong winds above 6 Beaufort scale. Especially, when the gale is completely missed by the T639 forecast, the accuracy rates of the BP prediction are still 25% within 24 hours, and 50% within 48 hours.
出处
《海洋预报》
北大核心
2013年第1期20-24,共5页
Marine Forecasts
基金
行业专项(GYHY201106006)
关键词
人工神经网络
BP算法
成山头
风
数值预报产品释用
artificial neural network
BP algorithm
ChengShanTou
wind
interpretation of numerical forecast-iag products
