蛾群算法与投影寻踪耦合模型在区域水资源脆弱性评价中的应用

摘要

提出一种新型群智能算法——蛾群算法(MSA)与投影寻踪(PP)技术耦合的水资源脆弱性评价模型,以云南省16个州市水资源脆弱性评价为例进行研究.从水量、水质、洪灾和旱灾4个方面遴选28个指标构建区域水资源脆弱性评价指标体系和分级标准;通过8个典型测试函数对MSA进行仿真验证,并与差分进化(DE)算法、文化算法(CA)、教学优化(TLBO)算法和粒子群优化(PSO)算法的仿真结果进行对比.利用MSA、DE、CA、TLBO和PSO算法优化PP最佳投影方向,构建MSA-PP、DE-PP、CA-PP、TLBO-PP和PSO-PP耦合模型对实例水资源脆弱性进行评价.结果表明:MSA寻优精度优于DE等4种算法,具有较好的全局极值寻优能力和收敛稳定性能,在求解高维、复杂优化问题时有着更高的求解精度;MSA-PP模型对昆明、玉溪、曲靖、楚雄4州市的水资源脆弱性评价为”轻度脆弱“,对怒江州、迪庆州评价为“重度脆弱”,其余州市评价为“中度脆弱”.MSA-PP模型对实例评价结果与TLBO-PP模型相同,但在排序上存在差异;与CA-PP、DE-PP和PSO-PP模型的评价结果及排序上均存在差异,表明智能算法求解精度的高低决定了评价效果的优劣.本文提出的模型及验证方法具有通用性,可推广应用于其他领域.%A new group intelligence algorithm,water resources vulnerability evaluation model,which is based on the moth swarm algorithm (MSA) and the projection pursuit (PP) technique,is proposed to study the vulnerability of water resources in 16 cities of Yunnan province.The water qualitity,water quality,flood and drought were selected from 28 indicators to construct regional water resources vulnerability evaluation index system and grading standard.Eight typical test functions were used to simulate the MSA,and compared with DE algorithm,(CA),teaching optimization (TLBO) algorithm and particle swarm optimization (PSO) algorithm.The optimal PP direction was optimized by using MSA,DE,CA,TLBO and PSO algorithms.The water resources vulnerability of MSA-PP,DE-PP,CA-PP,TLBO-PP and PSO-PP model was evaluated.The results show that:The MSA has better global optimization and convergence stability than the other four algorithms,and has higher accuracy in solving high-dimensional and complex optimization problems.The vulnerability of water resources in Kunming,Yuxi,Qujing and Chuxiong was evaluated as "mildly fragile" by MSAPP model;"severe fragile" for Nujiang and Diqing,and "moderate" by other counties fragile.The evaluation results of MSA-PP model are the same as those of TLBO-PP model,but they are different in order of ranking.There are differences between CA-PP,DE-PP and PSO-PP model.The level of the evaluation of the merits of the evaluation.③The model proposed in this paper and the verification method are universal and can be applied to other fields.