Runoff Prediction Using a Novel Hybrid ANFIS Model Based on Variable Screening
Zhennan Liu,Qiongfang Li,Jingnan Zhou,Weiguo Jiao,Xiaoyu Wang
DOI: https://doi.org/10.1007/s11269-021-02878-4
IF: 4.426
2021-06-29
Water Resources Management
Abstract:The accurate and reliable prediction of future runoff is important to guarantee for strengthening water resource optimization and management. The novel contribution of this article is the development of a hybrid model (FWA-ANFIS), which is based on the improvement of the adaptive neuro-fuzzy inference system (ANFIS) with the fireworks algorithm (FWA). The dominant driving factors of runoff are selected from several hydro-meteorological indices (precipitation, soil moisture content, and evaporation) as predictors by correlation coefficient (CC) analysis, mutual information (MI) analysis, correlation analysis and principal component analysis (CC-PCA), mutual information and kernel principal component analysis (MI-KPCA), MI-PCA, and CC-KPCA. The FWA-ANFIS model is applied to the Beiru River, China, with data from 1985–2016 (1985–2012 for model training and 2013–2016 for model prediction). The standard ANFIS, the GA-ANFIS, the PSO-ANFIS, the FWA-ELM, the GA-ELM, and the PSO-ELM are utilized as compared prediction models on the identical dataset. The results indicate that CC-PCA outperforms the other methods regarding the selection of predictors, and FWA-ANFIS has the best performance in terms of the root mean square error, correlation coefficient, and coefficient of determination, followed by the GA-ANFIS, PSO-ANFIS, ANFIS, FWA-ELM, GA-ELM, and PSO-ELM models. Furthermore, the degrees of uncertainty of the models increase in the following order: FWA-ANFIS, GA-ANFIS, PSO-ANFIS, ANFIS, PSO-ELM, GA-ELM, and FWA-ELM.
water resources,engineering, civil