Application Of An Improved Naive Bayes Classifier In Precipitation Level Prediction Of South Region

Waterlogging is an important natural disaster that affects our country.How to improve the accuracy of precipitation forecast has always been a difficult point in weather prediction.This paper made the full use of numerical forecast product by statistical method in order to improve the accuracy of precipitation prediction.This paper adopted the model output statistical method(MOS)to synthesize the dynamic information of numerical forecast product and statistical knowledge,and used the numerical forecast product of the T511 from June to September in 2012-2014 and the fusion of China Automated Station and CMORPH precipitation grid dataset product to establish two kinds of fitness function Naive Bayes classifier precipitation level prediction models(BPSO-NB1,BPSO-NB2).On selecting predictor,an artificial intelligence method--binary particle swarm algorithm was used to combine the optimal factors.In order to test the prediction performance of the Naive Bayes classifier optimized by the binary particle swarm optimization algorithm,using the data from July to September in 2015 to forecast precipitation for the five stations in the southern region.The results showed that:1.BPSO-NB1 model improved the fitness function by 0.239 on precipitation occurrence,therefore the application effect of factor selection was better.2.Both BPSO-NB1 and BPSO-NB2 improved the accuracy rate by more than 24%compared with the T511 model on precipitation occurrence forecast.And the drizzle' TS scores of the BPSO-NB1 and BPSO-NB2 increased by more than 0.1 compared with the T511 model.3.Both BPSO-NB1 and BPSO-NB2 could significantly reduce the false alarms number of drizzle and moderate rain of the T511 model.4.From the aspect of score improvement,BPSO-NB1 was more effective than the BPSO-NB2 model,with PC score of 3%higher and TS score of 0.1 higher,so it was more suitable for precipitation forecast in the southern region.5.From the qualitative forecast of precipitation falling on June 26-27,2015,the improved model also had good application effect.