BPSO优化朴素贝叶斯分类器的降水分级预报试验

Experiment of precipitation level prediction with binary particle swarm optimized Nave Bayes classifier

为进一步研究朴素贝叶斯分类器在单站降水预报方面的应用效果,利用2008年至2011年6~9月份的T511数值预报产品和单站观测资料,采用2种不同适应度函数的二进制粒子群算法（简称BPSO）优化朴素贝叶斯分类器算法（ BPSO-NB）,对石家庄、太原、林西3站13~24 h时段的晴雨和降水等级进行了预报试验。试报结果表明：BPSO-NB、BPSO-NB2模型3站平均晴雨预报准确率明显高于T511,均在85%以上,且BP-SO-NB2（87.1%）最优；2种模型小雨、中雨TS评分也高于T511,BPSO-NB1（0.403、0.167）最优。 BPSO-NB模型能有效降低T511空报次数。

For the further study of the application effect of Na？ve Bayes classifier （ N-Bayes） on single station pre-cipitation forecasting,two different fitness functions of binary particle swarm optimization（BPSO）,were used opti-mize N-Bayes models （ BPSO-NBs） . The BPSO-NBs forecast 13 through 24 hour period precipitation occurrence and precipitation levels of Shijiazhuang, Taiyuan and Linxi stations with T511 numerical prediction products and corresponding station observations of summer months （June through September） from 2008 to 2011. Results show that the BPSO-NB1, BPSO-NB2 average precipitation occurrence forecast accuracies of the three stations are obvi-ously higher than those of T511 with the BPSO-NB2 being the optimal （87. 1%）, and that the accuracies are more than 85%. As to flurry and median rain, the two models’ TS are significantly higher than those of T511, with the highest performance of BPSO-NB1 （0. 403 and 0. 167 respectively）. BPSO-NB models can effectively reduce the false alarm number compared to T511 .