Extreme Learning Machine Based on Particle Swarm Optimization for Estimation of Reference Evapotranspiration

摘要

Reference evapotranspiration (ET_0) plays an important role in water resources scheduling of irrigation systems. This paper proposes a novel extreme learning machine (ELM) method optimized by particle swarm optimization (PSO) algorithm (PSO-SWELM) to realize more accurate evapotranspiration estimation with limited environmental and meteorological data. The weights and thresholds between input and hidden layers of ELM is optimized by PSO algorithm and a function based on the two-wave superposition is selected as the activation function of ELM, which both enhances the accuracy of estimation. The Penman-Monteith model (FAO-56 PM) is used as the standard model to estimate ET_0. The root of mean squared error (RMSE) and coefficient of determination (R~2) are set as the two evaluation criteria to compare the performances of BP, PSO-BP, SVM, ELM, PSO-ELM and PSO-SWELM in estimating ET_0. The simulation results show that the PSO-SWELM method has better performance in predicting the ET_0 than the currently prevailing methods.