基于奇异谱分析的SPBO-ANFIS月径流组合预测模型  

SPBO-ANFIS Model of Combined Monthly Runoff Forecasting Based on Singular Spectrum Analysis

作　　者：张亚杰 崔东文 ZHANG Yajie;CUI Dongwen(Water Conservancy Bureau of Yimen County,Yuxi,Yunnan Province,Yuxi 651100,China;Water Affairs Bureau of Wenshan Prefecture,Yunnan Province,Wenshan 663000,China)

机构地区：[1]云南省玉溪市易门县水利局,云南玉溪651100 [2]云南省文山州水务局,云南文山663000

出　　处：《人民珠江》2022年第5期137-144,153,共9页Pearl River

摘　　要：针对水文时间序列月径流多尺度非平稳性等特点,提出基于奇异谱分解(SSA)的学生心理学优化(SPBO)算法-自适应神经模糊推理系统(ANFIS)月径流组合预测模型,并应用于云南省某水文站月径流预报。首先通过SSA将实例月径流时序数据分解为若干独立子序列分量,以降低时序数据的复杂性;其次介绍SPBO算法原理,通过取8个标准函数对SPBO算法进行仿真验证及比较;最后采用SPBO算法优化ANFIS条件参数和结论参数,建立SSA-SPBO-ANFIS模型对每一个子序列进行预测,叠加后作为最终月径流预测结果,并与基于集合经验模态分解(EEMD)的EEMD-SPBO-ANFIS模型和未经分解的SPBO-ANFIS模型作比较。结果表明:SPBO算法具有较好的寻优精度;SSA-SPBO-ANFIS模型对实例月径流预测的平均绝对百分比误差5.57%,平均绝对误差0.20 m^(3)/s,纳什系数0.9948,合格率96.7%,预测效果优于EEMD-SPBO-ANFIS模型,远优于SPBO-ANFIS模型。模型及方法可为相关水文时间序列预测研究提供参考。In view of the multi-scale non-stationarity and other characteristics of monthly runoff in hydrological time series,this paper proposes a singular spectrum decomposition(SSD)-based model of combined monthly runoff forecasting that integrates the student psychology based optimization(SPBO)algorithm with the adaptive network based fuzzy inference system(ANFIS),namely the SSDSPBO-ANFIS model.This model is then applied to the monthly runoff forecasting at a hydrological station in Yunnan Province.Specifically,time series data of sample monthly runoff are decomposed into various independent sub-series components through SSD to reduce the complexity of the time series data;then,the principle of the SPBO algorithm is outlined,and eight standard functions are selected for simulation verification and comparison of the SPBO algorithm;finally,the SPBO algorithm is employed to optimize the ANFIS condition and conclusion parameters.The SSD-SPBO-ANFIS model is built to forecast each sub-series,which is then superimposed to obtain the final monthly runoff forecasting result.In addition,the results of the proposed model are compared with those of the ensemble empirical mode decomposition(EEMD)-based EEMD-SPBO-ANFIS model and the SPBO-ANFIS model without decomposition.The following observations can be made from the results:The SPBO algorithm has favorable optimization accuracy;with a mean absolute percentage error of 5.57%,a mean absolute error of 0.20 m^(3)/s,a Nash coefficient of 0.9948,and a pass rate of96.7%,the SSD-SPBO-ANFIS model has an effect better than that of the EEMD-SPBO-ANFIS model and far better than that of the SPBO-ANFIS model in forecasting sample monthly runoff.The proposed model and method can provide references for related research on hydrological time series forecasting.

关 键 词：径流预测 奇异谱分析 学生心理学优化算法 自适应神经模糊推理系统 仿真测试 

分 类 号：P333[天文地球—水文科学]