信息扩散模型及其海温序列插值应用

摘要

针对海洋观测资料中实际存在的数据缺损、零散等小样本问题，引入信息扩散原理，并通过遗传算法对其扩散系数的选取进行规范和改进，建立了遗传优化的新信息扩散模型，该模型通过对非规则、非对称的贫乏样本数据进行模糊化处理，可实现对其中蕴含信息的概率扩散。为验证新优化算法的可行性和有效性，选取美国国家环境预报中心和国家大气研究中心1982年1月至2011年10月间海温场月平均再分析资料作为样本数据，运用本文模型进行了多组不同样本容量的海温资料时间序列插值试验，并与传统的正态信息扩散模型和最优信息扩散模型进行对比分析。结果表明，新扩散模型在一定比例缺测的情况下能够较好地插补填充原海温时间序列，可为海洋稀疏数据的客观分析及应用提供参考。%To tackle sparse observed data in asymmetrical and abnormal distribution and the problem of small sample,a new interpolation technique based on the information diffusion and optimal window width theory was improved by genetic algorithm in this paper.With fuzzy mapping route,the scattered samples can be diffused and mapped into corresponding fuzzy sets in the form of probability.By interpolating monthly sea surface temperature time series from National Center for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR)reanalysis,the rationality and validity of the improved information diffusion model was validated using different size of samples.The conventional information diffusion models were also introduced for comparison.All results show the new idea and technique may be a-vailable for interpolating such incomplete time series in ocean science or imperfect information condition.