摘要
Abstract
To accurately predict runoff volumes for water resources management,this study employs monthly runoff data from the Luan River Station in Tangshan from 1980 to 2020,along with concurrent climate data.The Empirical Mode Decomposition(EMD)method is used to decompose the runoff sequence into modal components.Bayesian Feature Selection(BFS)is then applied to select the optimal input variables from these components,followed by the design of a runoff prediction model using the Extreme Learning Machine(ELM)regression method.The results indicate that EMD can effectively identify multiple component characteristics within the runoff time series;the decomposition of the Luan River Station's runoff volume yielded 11 characteristic components and a residual trend term.BFS technology extracted six effective variables,significantly reducing model complexity by eliminating redundant features.The ELM model achieved a validation accuracy of R2=0.96 for runoff prediction,with Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)of 0.29 billion m3 and 0.40 billion m3,respectively.The proposed EMD-BFS-ELM strategy provides valuable support for runoff prediction.关键词
EMD分解/BFS变量选择/ELM回归/径流量Key words
EMD decomposition/BFS variable selection/ELM regression/runoff分类
建筑与水利
