Multi-objective optimization design of process parameters for hydrodynamic regulation of dual circulation wells based on deep learning
The optimal design of groundwater circulation wells(GCWs)is crucial for the implementation of this technique to site remediation.This work proposed a deep-learning-based optimal technology by coupling circulation wells simulation with optimization.The key parameters affecting the operation of a dual-circulation wells,a typical unit of well cluster,were treated as input variables;The modules MODFLOW,MODPATH,and MT3DMS were invoked by FloPy to construct the groundwater flowing model,particle tracking model,and solute transporting model.The simulated performance indicators(circulation efficiency(Pr),and pollutant removal efficiency(η))are treated as output variables to build a training dataset.Convolutional neural network model was employed to establish an alternative model.The non-dominated sorting genetic algorithm(NSGA-Ⅱ)was applied in building a multi-objective optimization model with the operation effect characterization indicators of the dual-circulation wells.The optimal design of dual-circulation wells at the demonstration site in Xi'an City,Shaanxi Province was carried out.Considering both Pr and η,the ideal design for a dual-circulation well,with Pr as the primary target,maintains the same pumping mode with the distraction screen in the upward and injection screen in the downward,with a pumping rate of 185 m3·d-1 and a well spacing of 20 m.This scheme results in a 99％Pr rate and a 27％η over a set duration;Conversely,when η is the primary target,the ideal design scheme involves the reverse pumping mode,a 163 m3·d-1 pumping rate,and a 13 m well spacing.This scheme attains a 95％Pr and a 34％η over a set period.The parameters for the well structure of the both schemes are also provided in the research.The findings showed that the alternative model based on the convolutional neural network method has high computational precision,evidenced by an R2 of 0.972 and a MAPE of 0.008 in forecasting Pr in the dual-circulation wells,alongside an R2 of 0.986 and a MAPE of 0.009 in predicting η.This approach significantly reduces computational time relative to conventional numerical computation.Based on the input hydrogeological parameters of the site,the proposed simulation-optimization technique can produce an ideal dual-circulation well design before engineering,which is crucial for advancing groundwater circulation wells technology.
国家科技期刊平台
NSTL
万方数据
