【作者】 曹鸿兴； 谷湘潜； 封国林； 陈国范；

【Author】 Cao Hongxing, Gu Xiangqian, Feng Gaolin, Chen Guofan(Chinese Academy of Meteorological Sciences , Beijing 100081;Mathematics Physics School, Yangzhou University, Yangzhou 225009;China Meteorological Administration , Beijing 100081)

【机构】 中国气象科学研究院； 扬卅大学理学院； 中国气象局；

【摘要】 <正> 1 引言人工智能在气象中的最早应用是学习机在天气预报中的应用[1,2],运用学习机做天气预报的准确率可以与当时的业务天气预报相当。在20世纪80年代初期,专家系统在长期天气预报中被广泛使用。但这些天气预报专家系统被试用一段时间后就搁浅了。90年代则盛行人工神经网络(ANN),在美国、日本、欧洲,人工神经网络应用在数值天气预报的释用中,即根据数值天气预报产品运用人工神经网络来制作专业预报,即用户需要的预报。更多还原

【Abstract】 Based on an atmospheric memory, in introducing a memory function into the dynamic equations of a system, so-called self-memorization equations, which are difference-integral equations actually, have been formulated. The methodology of prediction based on the self-memorization equation is different from what is to solve differential equations governing motion of a system as an initial value problem, i.e. Cauchy problem. Instead, our approach is to solve difference-integral equations in which several preceding values can be contained and processed, like system forecast in modern control theory. We call a model that is constructed by help with the approach a self-memory model.The self-memorial principle is successfully combined with the traditional numerical model to formulate a specific self-memorial model, of which the scheme is designed properly and the computing time only increases not so much in comparison with the kernel model and the storage is less increasing, so it is an advantage in comparison with ensemble forecast, which consumes a great amount of computer resources. The self-memorial model can combine the self-memorial principle with the traditional numerical, it does not exclude any revision of the kernel model. Nevertheless, it can fully utilize every improvement of the kernel model; thereby the self-memorial model will be improved too.Artificial intelligence, in particular, artificial neural networks have been widely applied in meteorology, the basis of the artificial intelligence is similar to the self-memorial principle, that is, they all utilize the historical observations. Therefore, combination of the artificial intelligence with the self-memorial model will possibly develop a new forecasting approach; in particular, it can provide a new idea to formulate an ensemble forecast更多还原