Simulation Of Farm Households'land-use Model Based On ABM And Machine Learning

In recent years,more and more scholars have paid attention to the evolution process of complex man-land system with land use and cover change as the core.Modeling based on Micro-land use subject is the main way to study land use evolution.ABM land use model can accurately simulate agent decision-making behavior and is easy to model and analyze.It has become the mainstream model of current land use change research.The establishment of land use conversion rules is the core of ABM land use model.Due to the complexity of land use transformation and the huge amount of geographic data,how to effectively set the rules of ABM model transformation has become one of the main ways to improve its explanatory power.At present,many geographers at home and abroad gradually turn their attention to machine learning and deep learning methods.Machine learning and deep learning have obvious advantages in data mining,feature extraction and modeling.How to utilize the advantages of machine learning and deep learning algorithms to mine the rules of land use transformation of ABM model and build a new model coupling it with ABM model has become one of the current research hotspots.Gaoqu Township,located on the Loess Plateau,is a typical ecologically fragile area in China.In this paper,PRA and traditional household surveys are used to obtain the required data,and a new simulation platform for simulating and predicting land use change is constructed by combining machine learning algorithm,deep learning algorithm and traditional land use change model.The platform mainly includes the following two aspects:(1)Based on the experience-based BDI behavior decision-making model,the RF-BDI model is constructed by using random forest to extract variables affecting farmers' decision-making;(2)Under the guidance of BDI framework,the stochastic forest algorithm is used to select the optimal eigenvector,and the deep neural network algorithm is used to independently mine farmers' decision-making rules and construct land use change.Simulate the model.By analyzing the simulation results of the platform,the following conclusions can be drawn:(1)The combination of random forest algorithm and BDI model can effectively represent the limited rational decision-making behavior of farmers.In order to explore the effectiveness of random forest algorithm for land use change simulation,this paper compares the RF-BDI model with the traditional BDI model in Gaoqu Township,Mizhi County,Shaanxi Province.The results show that the RF-BDI model is a traditional BDI model.After the decision behavior was revised,the simulation accuracy was improved by 14% without significantly increasing the amount of calculation.(2)Whether the setting of the conversion rule is appropriate directly determines the simulation result.The traditional BDI model is a human-based decision-making rule based on experience.This rule is simpler and more rigid.The random forest algorithm can mine farmers' land use behavior rules through machine learning based on big data.Farmers have random planting behaviors when planting,random forest algorithm design is relatively simple,mining rules are too rational,and random forest mining rules and experience-based rules are mutually confirmed,which can effectively eliminate the errors caused by random planting behaviors of farmers and improve model simulation Precision.(3)Although RF-BDI model can better simulate farmer's planting behavior,it still has about 20% error.Deep learning has better mining and simulation ability than machine learning.Based on Keras deep learning module,this paper designs and adjusts network structure.Taking Gaoqu Township of Mizhi County in Shaanxi Province as data source,constructs a deep neural network model,and compares the deep neural network model with RF.-BDI model simulation.On the premise of no empirical rules,the depth neural network model independently extracts farmers' bounded rational decision rules,simulates and predicts land use change in Gaoqu Township and compares it with the actual land use change.The accuracy of the model is 85.2%.Compared with the RF-BDI model,the simulation accuracy of the depth neural network model is improved by about 5%.The model can effectively simulate the random planting behavior of farmers and predict their future planting tendency.It provides a new way to describe the future land use change and its driving mechanism in Gaoqu Township.(4)The deep neural network model has a strong ability to describe the land use behavior of farmers,and has a strong ability to fit the random planting behavior of farmers based on their own wishes.However,the deep neural network model can only show the final planting results of farmers,among which the intermediate process parameters are too complex to analyze the driving mechanism of human-land system evolution at the micro level.The RF-BDI model can directly reflect the relative importance of various factors in the process of human-land system evolution,and synthesize the simulation ability of the deep neural network model and the interpretation of the RF-BDI simulation in machine.The advantage is that the next step needs to be done.