基于智能计算的广西冷湿极端天气定性和定量组合预报方法研究

论文以逐日气温和降水量数据、NCEP/NCAR再分析资料以及预报场资料为基础,将表征冬季低温冷害的冷湿指数作为预报量,先利用随机森林方法进行冬季逐日冷湿极端天气定性判别预报分析,再进一步以粒子群算法为基础的模糊神经网络集成个体生成技术方法,建立一种新的非线性智能计算定量集成预报模型（PSO-FNN）,进行了广西冷湿极端天气定量预报模型的预报建模研究。结果表明,论文提出的这种以不同的智能计算方法构建的定性、定量综合预报分析方法,比较符合极端天气小概率事件的预报特点,其中随机森林算法构建的定性预报模型,对广西冷湿极端天气事件的预报TS评分（Threat Score）为0.77,空报率为0.23,漏报率为0,ETS评分（Equitable Threat Score）为0.41,TSS评分（True Skill Statistic）为0.53。而采用粒子群—模糊神经网络方法构建的极端冷湿指数定量集成预报模型比其他线性和非线性预报模型具有更好的预报精度。其中PSO-FNN集成预报模型在预报建模样本和独立预报样本个例相同的情况下,比回归方法的预报平均绝对误差下降了25%以上,比一般的普通模糊神经网络预报平均绝对误差下降了14.37%。主要原因是因为PSO-FNN集成预报模型通过改进集成个体的预报能力和增强集成个体的种群差异性,提高了集成预报模型的预报精度。因此,该智能计算集成预报模型的泛化能力显著提高,预报结果稳定可靠,为冷湿极端天气客观预报提供了新的预报工具和预报建模方法。

A Combined Qualitative and Quantitative Prediction Scheme for Cold-Wet Extreme Weather in Guangxi Based on Intelligent Computing

Taking the cold-wet index measuring chilling damage in winter as the predictand, a qualitative, discriminant prediction model has been developed for daily cold-wet extreme weather forecasting in winter. This model is based on a Random Forest (RF) algorithm, and uses daily temperature and precipitation data, NCEP/NCAR reanalysis data, and 24-h forecast field data. Further, a new, nonlinear, intelligent computing, quantitative ensemble prediction scheme has been developed for predicting cold-wet extreme weather in Guangxi by employing Particle Swarm Optimization (PSO) algorithm, and it is termed the PSO-FNN ensemble prediction model. The ensemble members of the PSO-FNN model were generated by adopting the PSO algorithm. Results showed that the qualitative and quantitative comprehensive prediction based on different intelligent computing methods proposed in this paper were in accord with the forecast characteristics of the small probability extreme weather event. Threat score of the qualitative forecast model based, on the RF algorithm for cold-wet extreme weather in Guangxi, was 0.77, false alarm rate was 0.23, missing rate was 0, equitable threat score was 0.41, and the true skill statistic was 0.53. Moreover, the forecast accuracy of the quantitative PSO-FNN ensemble prediction model was higher than those of the linear and nonlinear forecast models. Using identical modeling samples and independent samples, the PSO-FNN ensemble prediction model showed the reduction in mean absolute errors being >25% relative to the stepwise method, and 14.37% relative to the normal fuzzy neural network. Analyses of the new scheme suggested that the forecast accuracy of the ensemble prediction model was improved by enhancing the prediction ability and population diversity of the individual ensemble members. Therefore, the generalization capacity of the intelligent computing ensemble prediction model was significantly enhanced, and the forecast results were stable and reliable, providing new forecasting tools and prediction modeling methods for objective forecasts of cold-wet extreme weather.