Multi-Objective Optimal Design Of Seawater Intrusion Monitoring Network In Coastal Areas

With the development of economy and society in coastal areas,the demand for fresh water resources increases gradually.Excessive exploitation of groundwater often breaks the original hydrodynamic balance between seawater and fresh water,thus causing the problem of seawater intrusion.Seawater intrusion will make groundwater salty,which will cause a series of ecological damage and economic development problems.Therefore,it is very important to grasp the current situation and trend of seawater intrusion in time,so as to formulate reasonable prevention and control measures.In order to realize the extent and scope of seawater intrusion timely and accurately,it is necessary to monitor the situation of seawater intrusion.A good monitoring scheme can obtain more effective information of seawater intrusion status with less monitoring cost.At present,there are few research results on the optimization design of seawater intrusion monitoring network,and only single-objective optimization to maximize the effective information of monitoring is considered,and there is a lack of multi-objective optimization studies that balance the contradictory relationship between the effective information of monitoring and the total number of monitoring Wells.In addition,due to the relatively irregular shape of the chloride ion concentration plume of seawater intrusion,it is difficult to monitor,and it is urgent to improve the mathematical expression to describe and measure the effective information amount of seawater intrusion monitoring.Aiming at the above problems,this study takes Longkou,Shandong Province as a case study area,and simulates and predicts the future seawater intrusion in the study area through the comprehensive application of three-dimensional variable density groundwater numerical simulation model,alternative model,integrated learning,image quality evaluation method,simulation-optimization method,and multi-objective optimization algorithm.The optimization design of seawater intrusion monitoring network is studied deeply.Firstly,a three-dimensional variable density groundwater numerical simulation model is established according to the actual situation of the study area,and the operation simulation model is used to simulate and predict the future seawater intrusion.In order to analyze the influence of the uncertainty of the simulation model on the output results of the model,sensitivity analysis method was used to screen out the sensitive factors,and Monte Carlo method was used to analyze the influence of random changes of the above sensitive factors on the uncertainty of the model output results.Then,in the process of uncertainty analysis and solving optimization model of seawater intrusion monitoring network,variable density groundwater numerical simulation model needs to be invoked repeatedly,which will generate huge computational load.Therefore,this study established an alternative model of simulation model to solve the above problems.A good alternative model can fit the input-output relation of the simulation model with high precision and reduce the calculation time greatly.However,in existing studies,the establishment of alternative models is mostly based on shallow learning methods.In the face of the complex nonlinear mapping relationship of the numerical simulation model of variable density groundwater,the fitting accuracy of the alternative model established by shallow learning methods needs to be improved.In recent years,there have also been researches on the application of deep neural network method to establish alternative models,which can well solve the problem of low fitting accuracy of shallow learning method.However,its model structure is relatively complex,with too many super parameters,and it is time-consuming and laborious to construct.In this study,we established the alternative model of variable density groundwater numerical simulation model by using the more convenient e Xtreme Gradient Boosting(XGB)method in ensemble learning,and compared them with the two shallow learning methods(Kriging method and support vector regression method).The applicability of this method is analyzed.Finally,the optimization design of seawater intrusion monitoring network is studied.First of all,with the total number of monitoring Wells as the constraint condition,the location distribution of monitoring Wells as the decision variable,and the maximum effective information of seawater intrusion monitoring as the objective function,this study established a single objective optimization model of seawater intrusion monitoring network.Under the constraint condition of a limited number of monitoring Wells,the seawater intrusion monitoring precision could be as high as possible by optimizing the location of monitoring Wells.To obtain as much effective information of seawater intrusion as possible.Aiming at the construction of mathematical expression of effective information content of seawater intrusion monitoring,Feature Similarity(FSIM)method in image quality evaluation is adopted to measure the similarity between seawater intrusion state obtained by monitoring data interpolation and the real seawater intrusion state.Then a mathematical expression is established to describe and measure the effective information amount of seawater intrusion monitoring.Then,the advantages of FSIM method are analyzed by comparing it with the existing spatial moment method.On this basis,FSIM method is used as the mathematical expression to measure the effective information of monitoring,and a multi-objective optimization model is established,which takes the location distribution of monitoring Wells as the decision variable,maximizes the effective information of monitoring seawater intrusion and minimizes the total number of monitoring Wells as the objective function.A series of optimal monitoring network layout schemes are obtained under the condition of mutual fluctuation of the total number of monitoring Wells and effective monitoring information.Due to the large number of decision variables in the multi-objective optimization model,the mapping from decision space to objective space is nonlinear and difficult to solve.In this study,Differential Evolution strategy(DES)was used to improve the traditional Non-dominated Sorting Genetic Algorithm-II(NSGA-Ⅱ).The population diversity of NSGA-Ⅱalgorithm was improved and the optimization ability of the algorithm was enhanced.On this basis,the improved hybrid difference-non-dominated sorting genetic algorithm(DE-NSGA-Ⅱ)was applied to solve the above multi-objective optimization model,and finally a series of seawater intrusion monitoring network alternatives were obtained which fully reflected the tradeoff between the total number of monitoring Wells and the effective monitoring information.Through the above research,the following conclusions can be drawn:(1)The 3D variable density groundwater numerical simulation model established by this research in Longkou area of Shandong Province has been calibrated and verified,and the simulated value fits well with the actual observed value,meeting the conditions for future groundwater level and groundwater quality prediction.In the uncertainty analysis of the simulation model,it is proved that groundwater extraction,precipitation and porosity are the three sensitive parameters that have the greatest influence on the output results of the model.The above three sensitive parameters are taken as random variables to carry out Monte Carlo simulation.The results show that the seawater intrusion area is expected to be between 63.99 and 76.12 km~2 in 30 years.The uncertainty degree of seawater intrusion state is relatively high.(2)This study explores an alternative model modeling method with better performance--the limit gradient Lifting Tree method(XGB).Compared with the Kriging method and the nuclear extreme learning machine method,the XGB alternative model has higher fitting accuracy for the numerical simulation model of seawater intrusion with variable density,and the parameters are easy to adjust.It is a potential lightweight alternative model modeling method.(3)Combining image quality evaluation with well pattern design of seawater intrusion monitoring,this study proposed a more effective method to describe the amount of effective monitoring information--feature similarity method(FSIM).In this study,the feature similarity method and the conventional spatial moment method were respectively applied to optimize the well pattern design of seawater intrusion monitoring.The results show that the monitoring accuracy of the optimization results of the FSIM method is higher,and the seawater intrusion state obtained by the monitoring data interpolation is closer to the real seawater intrusion state,indicating that the monitoring information obtained by the FSIM method is more.Therefore,it is more effective and practical to use the feature similarity method to establish a mathematical expression describing the amount of effective information of seawater intrusion monitoring.(4)In this study,by improving the traditional NSGA-Ⅱalgorithm,an effective solution to the multi-objective optimization model of seawater intrusion monitoring network is explored,which is the DE-NSGA-Ⅱalgorithm.The optimization results reflect the tradeoff relationship between the total number of Wells monitored and the effective information of monitoring more completely.Compared with NSGA-Ⅱalgorithm,DE-NSGA-Ⅱalgorithm sacrifices a part of local search ability,and brings a significant improvement in the distribution universality of Pareto frontier,which can bring broader tradeoff space for decision makers.The optimization results show that the newly added monitoring Wells have obvious diminishing marginal effect on the improvement of effective monitoring information.When the number of newly added monitoring Wells ranges from 0 to 20,the addition of monitoring Wells can significantly improve the monitoring accuracy,and the newly added monitoring Wells are cost-effective and suitable for the situation where the monitoring budget is relatively limited.When the number of newly added monitoring Wells ranges from 20 to 40,the income of newly added monitoring Wells decreases,but the monitoring accuracy is still improved steadily,which is suitable for the situation with high monitoring accuracy requirements.When the number of newly added monitoring Wells is more than 40,the monitoring accuracy cannot be improved basically,so it is not recommended to add more new Wells.