基于贝叶斯分类器的气象预测研究

将机器学习的理论和方法应用于气象预报领域,基于贝叶斯推理学习的理论,使用朴素贝叶斯分类器(Na(i)ve Bayes classifier)对降雨量预测问题进行了分类预测研究,提出了预测降雨量的朴素贝叶斯算法learn-and-classify--rainfall,将各预测因子及预测目标按照气象学分级标准进行分级,以历年气象数据为训练集,在训练集上学习各预测目标的先验概率及各预测因子的条件概率,用NBC计算出极大后验假设作为预测目标值,该算法具有鲁棒性强、易实现等优点,表现出较强的实用性和有效性,经实验表明,预测精度明显高于目前短期气候预测中采用的回归分析、聚类分析等其它预测方法.同时它还对困扰气象工作者的如何选择预测因子的问题具有指导作用.

Research on meteorological prediction with Bayesian classifier

Applying the theory and methods of machine leaning in meteorological prediction,research of rainfall amount classification is conducted based on Bayesian deducing and learning theory,an algorithm named learn-and-classify--rainfall is presented for predicting rainfall amount with Na?ve Bayesian classifier to improve prediction accuracy.Firstly,various predictors and predicting objectives are categorized according to meteorological standards,secondly,prior probabilities of prediction objectives and conditional probabilities of predictors on the history meteorological data which is taken as training set are learned,lastly,it calculates maximum a posteriori probability hypotheses with NBC as the rainfall classification objective.The experiments shown that it is practicable and robust and effective and easy to be realized,and the results also suggested it acquired better accuracy rate than many other existing prediction methods such as re-gression analysis and cluster analysis which are adopted in short term meteorological prediction at present.At the same time,it can offer an instruction for selecting predictors,which often puzzles the researchers in the field of meteorology.