An Agricultural Decision System Based On Surrogate-assisted Multi-objective Optimization

In order to achieve sustainable agricultural intensification,there is a need to increase the yield of available agricultural land and to reduce the environmental impact.Therefore,irrigation and fertilization strategies that consider crop yield and environmental impacts are important to achieve this goal.In this paper,we propose to combine the Decision Support System for Agrotechnology Transfer(DSSAT)with multi-objective optimization techniques to provide decision makers with the required irrigation and fertilization strategies for agriculture.To address the problem that the DSSAT model consumes a long time for each run,this paper proposes a new surrogate-assisted multi-objective optimization algorithm based on a fast non-dominated ranking genetic algorithm(NSGA-Ⅱ)and Radial Basis Function(RBF)to accelerate the optimization convergence process.Due to the complexity of agricultural problems,this paper proposes a two-level screening and twolevel optimization framework to achieve a tight integration of the surrogate-assisted multiobjective algorithm and DSSAT.Specifically,the main work of this paper is as follows:(1)A new surrogate-assisted multi-objective evolutionary algorithm is proposed.The main idea of applying the surrogate model to the multi-objective algorithm is to generate a large number of trial offspring using traditional simulated binary crossover(SBX)and polynomial mutation(PM).It is worth mentioning that we employ the ideal point and MED methods to maintain the convergence and diversity of the algorithm,and finally,select a small number of children from a large number of children as individuals for the true function evaluation by the surrogate model.Finally,by comparing with K-RVEA,par EGO,and SMSEGO algorithms,the results show that the proposed method not only significantly outperforms other algorithms in benchmarking problems but also takes less time in surrogate running time.(2)A two-level screening and two-level optimization framework is proposed,allowing the proposed algorithm to be well integrated with DSSAT.Firstly,in the upper level screening,the date ranges for irrigation and fertilization are pre-screened by precipitation and solar radiation information.Second,in the upper-level optimization,a series of irrigation and fertilization date combinations are generated by Latin Hypercube Sampling(LHS),and again,in the lower-level optimization,a large number of irrigation and fertilization strategies are generated by combining the surrogate-assisted evolutionary algorithm and DSSAT.and fertilization strategies that meet the needs of local decisionmakers.(3)Applying the proposed surrogate-assisted multi-objective evolutionary algorithm to DSSAT not only resulted in a reduction of the final code runtime to less than 20%,but the experiment yielded better irrigation and fertilization strategies with 48% less water use and 36% less nitrogen application.The economic benefits increased by 9% to 10%.(4)The optimization problem of existing decision support systems for agricultural technology transfer(DSSAT)is studied and the need to optimize agricultural irrigation and fertilizer application strategies is presented.The advantages and shortcomings of the surrogate-assisted multi-objective optimization algorithm are analyzed,and the application prospects and development directions of the algorithm are discussed.