Optimal Design Of Coastal Groundwater Management Schemes Under Uncertainty

Seawater intrusion refers to the phenomenon that the balance between seawater and freshwater is destroyed in coastal areas due to overexploitation of groundwater,resulting in direct seawater intrusion into terrestrial aquifers.Seawater intrusion can cause the shortage of freshwater resources in coastal land,thereby affecting the normal domestic water use of residents,reducing the original irrigation capacity of agriculture,and restricting industrial production.Therefore,it is very important to study the reasonable exploitation of groundwater resources and the optimal design of seawater intrusion management schemes in coastal areas.Simulation-optimization method is an effective means to obtain the optimal groundwater management schemes in seawater intrusion area.In the simulationoptimization method,the numerical simulation model of seawater intrusion is used to analyze and evaluate the impact of various management schemes on the coastal groundwater system,and the optimization algorithm is used to search the optimal management scheme under certain constraints according to the output results of the simulation model.Global climate change can cause the changes of precipitation and sea level height,and then affect the process of seawater intrusion in coastal aquifers.Therefore,it is necessary to consider the influence of sea level rise and precipitation variation under climate change in the numerical simulation of seawater intrusion.At present,in the simulation-optimization research of coastal groundwater management,the core task is to design reasonable groundwater exploitation schemes.As an effective measure to prevent and control seawater intrusion,artificial recharge of groundwater is rarely considered in the optimization problem.Under the premise of controlling the degree of seawater intrusion,maximizing the pumping water and minimizing the artificial recharge are a pair of contradictory objectives,so it is necessary to carry out multi-objective optimization research.In the process of optimization,complex simulation models need to be called repeatedly for iterative calculation,so directly coupling the numerical model and optimization model will cause huge computational load.Currently,in order to solve this problem,some artificial intelligence methods are usually used to construct surrogate models with high computational efficiency to replace the simulation model to describe the excitation-response relationship of coastal aquifer system.The reliability of the optimal management scheme obtained by solving the optimization model mainly depends on the accuracy of the description of the aquifer system in the optimization process.However,there are some differences between the prediction results of the surrogate model and the calculation results of the simulation model,and the simulation model itself has inevitable uncertainties(such as the uncertainty of parameter values).Therefore,it is necessary to consider the influence of the uncertainty of the simulation model and the surrogate model on the reliability of the optimal management schemes in the simulation-optimization method.In this paper,taking Longkou,Shandong Province as an example,combined with previous research results and the latest data,the optimization design of coastal aquifer seawater intrusion management schemes is carried out by comprehensively applying three-dimensional variable density groundwater numerical simulation model,statistical downscaling model,multi-gene genetic programming,surrogate model,ensemble learning theory,simulation-optimization method and stochastic optimization model theory.Firstly,a three-dimensional variable density groundwater numerical model of Longkou area is established and solved by using the SEAWAT code,and then the simulation model is used to predict the future seawater intrusion in the aquifer of the study area.The effects of sea-level rise and precipitation variation on seawater intrusion under climate change are considered in the forecast.The statistical downscaling model is used to predict the local response of precipitation in Longkou to global climate change,while the sea-level rise is determined based on the authoritative prediction data.The sensitivity of the parameters of the established numerical simulation model is analyzed,and the most sensitive parameter is selected.Then,the multi-gene genetic programming method is used to establish the surrogate model of the variable density groundwater numerical simulation model in Longkou area.In order to further improve the accuracy of the surrogate model,the Bagging method of ensemble learning is applied to establish the ensemble surrogate model based on stand-alone multi-gene genetic programming surrogate models.In the subsequent optimization model solving process,the ensemble multi-gene genetic programming surrogate model will be directly used for calculation instead of the seawater intrusion numerical simulation model.Through statistical analysis method,the distribution of prediction residuals on the testing samples of the ensemble multigene genetic programming surrogate model is determined to approximately represent the uncertainty of the surrogate model.Finally,a stochastic multi-objective optimization model for groundwater management in Longkou is established.In the optimization problem,the pumping or injection rates of main pumping wells and fixed injection wells near the coast in the study area are taken as decision variables,and the parameter of the simulation model with the highest sensitivity and the prediction residuals of the surrogate model are taken as random variables.The optimization objectives include maximizing the total pumping rate of pumping wells and minimizing the total water injection rate of injection wells.The constraints mainly include the restriction of seawater intrusion degree in Longkou area.For the optimization algorithm,the traditional second generation non-dominated sorting genetic algorithm NSGA-II is improved based on the information entropy mechanism.Then,the stochastic simulation technique and the improved multiobjective evolutionary algorithm are combined together to solve the optimization model.Based on the above researches,the following main conclusions are drawn:(1)Through model calibration and verification,the three-dimensional variable density groundwater numerical simulation model of Longkou area is established according to the actual data.The simulated values fit well with the actual observation values,and thus the numerical model can be used to simulate and predict the complex seawater intrusion process in the aquifer of the study area in the future.(2)The prediction scenarios with and without considering climate change are input to the simulation model for calculation respectively,and the results indicate that during the forecast period,the degree of seawater intrusion in the study area considering climate change is lighter than that without considering climate change.(3)Compared with the Kriging surrogate model and kernel extreme learning machine surrogate model,the multi-gene genetic programming surrogate model has higher approximation accuracy for the variable density seawater intrusion numerical simulation model,which indicates that multi-gene genetic programming is a potential surrogate modeling method.In view of the over fitting problem of the multi-gene genetic programming surrogate model,the Bagging method of ensemble learning is applied to construct the ensemble multi-gene genetic programming surrogate model.The results show that the accuracy of the ensemble surrogate model is significantly improved compared with the stand-alone surrogate model.(4)Based on the information entropy mechanism,the traditional NSGA-II multiobjective optimization algorithm is improved.The experimental results show that compared with NSGA-II algorithm,the improved algorithm can obtain Pareto optimal solution set with more uniform population distribution.Using the improved algorithm to solve the optimization model can provide more diverse optimal management schemes for decision makers.(5)By establishing the stochastic multi-objective optimization model,the Pareto optimal solution sets under different confidence levels can be obtained.Decision makers can select the non-dominated solution with different reliability levels as the management scheme according to the actual conditions of the study area.