粒子群优化极限学习机模型在河南省干旱预测中的应用

Application of Extreme Learning Machine Model Based on Particle Swarm Optimization in Drought Prediction of Henan Province

干旱预测是提高防旱抗旱能力的重要非工程措施。在评价不同时间尺度标准化降水蒸散指数(SSPEI)对河南省干旱识别能力的基础上,以能有效表征河南省干旱状况的SSPEI为模型输出,以基于信息变化率和条件互信息的特征变量选择方法(ICR-CMIFS)筛选得到的河南省主要致旱气候系统指数为模型输入,构建了基于粒子群算法优化极限学习机(PSO-ELM)的干旱预测模型,通过对比该模型与标准极限学习机(ELM)、差分进化算法优化极限学习机(DE-ELM)模型的干旱预测结果,验证PSO-ELM模型在河南省干旱预测中的适用性。结果表明,SSPEI-3能有效识别河南省典型干旱事件,从时间和空间上可较准确地反映河南省干旱状况;ICR-CMIFS筛选出的河南省主要致旱气候系统指数为西太平洋副高面积指数和NINO指数;PSO-ELM模型能较准确地预测河南省干旱,且预测精度优于DE-ELM模型和标准ELM模型,在河南省干旱预测中具有较好的适用性。

Drought prediction is an important non-engineering measure to improve drought prevention and resistance. This paper firstly evaluated the ability of multi-scalar standardized precipitation evapotranspiration index(SSPEI) to identify drought events in Henan Province. Then a drought prediction model based on particle swarm algorithm optimized extreme learning machine(PSO-ELM) was constructed, which used SSPEIas model outputs and major drought-causing climate system indices selected by Information Changing Rate and Conditional Mutual Information-Based Input Feature Selection Method(ICR-CMIFS) as model inputs. The applicability of the PSO-ELM model in drought prediction in Henan Province was verified by comparing the drought prediction results of this model with standard extreme learning machine(ELM) and differential evolutionary algorithm optimized extreme learning machine(DE-ELM) models. The results show that the SSPEI-3 can effectively identify specific drought events in Henan Province and reflect the drought situation in Henan Province accurately in terms of time and space;The main drought-causing climate system indices in Henan Province screened by ICR-CMIFS are the western Pacific paratlantic area index and the NINO index;The PSO-ELM model can predict drought in Henan Province accurately, and the prediction accuracy is better than that of the DE-ELM model and standard ELM model, which has better applicability in drought prediction of Henan Province.