一种耦合进化算法与FLUS模型的土地利用变化模拟模型

A Land Use Change Simulation Model:Coupling of Evolutionary Algorithm and FLUS Model

研究如何在土地利用变化模拟过程中更为科学、客观地设置模型参数,对于避免复杂模型由于参数设置不当导致模拟效果不佳的问题具有重要意义。本文通过耦合进化算法(Evolutionary Algorithm,EA)与FLUS模型,构建了具有参数寻优功能的EA-FLUS模型。该模型首先通过进化策略对FLUS模型中人工神经网络模型的参数进行寻优,以提升对各土地利用类型出现概率分布的预测准确率,在此基础上结合地理分区,利用带精英策略的遗传算法与进化策略的组合对FLUS模型中元胞自动机模型的参数进行调整,以提升模拟精度。在实证研究阶段,本文以桂林市为实验区,通过对土地利用变化进行分区模拟来分析EA-FLUS模型的改进效果,此外还设置了自然发展、耕地保护、生态优先3种情景来模拟桂林市2020—2030年的土地利用变化。结果表明:(1)进化算法对参数的寻优结果相比于基于经验以及土地利用变化历史特征的参数设置,更加贴近实验区的政策导向,更能体现各土地利用类型在不同区域间多样的发展趋势;(2)EA-FLUS模型相较于FLUS模型,在加入地理分区的土地利用变化模拟中更具优势,其模拟结果总体精度、Kappa系数、FoM系数分别提升0.56%、0.011、0.009,更贴近真实的土地利用格局;(3)桂林市的建设用地与耕地具有较强的扩张趋势,但林地却呈现收缩趋势,进一步加强对生态空间的保护有助于建设用地与耕地扩张速度的减缓。研究结果在丰富现有土地利用变化模拟技术与方法的同时,也可为城市规划、可持续性研究等提供一定的理论基础与科学依据。

It is of great significance to study how to set parameters of land use change simulation models more scientifically and objectively,in order to avoid the problem of poor simulation caused by improper parameters setting in a complex model.In this paper,the EA-FLUS model with parameter optimization function was constructed by coupling Evolutionary Algorithm(EA)and FLUS model.This model first optimized the parameters of the artificial neural network model in the FLUS model through evolutionary strategy to improve the prediction accuracy of the probability distribution of each land use type.On this basis,combined with geospatial partition,the parameters of the cellular automaton model in the FLUS model were adjusted by using the combination of elitist genetic algorithm and evolutionary strategy to improve the simulation accuracy.In the empirical study phase,taking Guilin as the study area,this paper analyzed the improvement of EA-FLUS model by partition simulation of land use change.In addition,the natural development scenario,cultivated land protection scenario,and ecological priority scenario were set up to simulate the land use change in Guilin from2020 to 2030.The results show that:(1)Compared with the parameters setting based on experience and historical characteristics of land use change,the parameters optimization result using evolutionary algorithms was closer to the policy orientation in the study area,and better reflected the diversified development trends of various land use types in different geospatial partition;(2)Compared with the FLUS model,the EA-FLUS model had more advantages in land use change simulation with geospatial partition.The overall accuracy,Kappa coefficient,and FoM coefficient of the simulation result were increased by 0.56%,0.011,and 0.009,respectively;(3)The construction land and cultivated land in Guilin showed a strong expansion trend,but the forested land showed a shrinking trend.Further strengthening the protection of ecological space would help to slow down the expansion of construction land and cultivated land.The research results not only enrich the existing land use change simulation techniques and methods,but also provide a certain theoretical basis and scientific basis for urban planning and sustainability research.