基于Agent建模的农户土地利用行为模拟研究——以陕西省米脂县孟岔村为例

以Agent强化学习（Reinforcement Learning）模型框架为基础，选择政策、市场、Agent自身及与其它Agent相互作用、土壤肥沃程度等因素作为Agent对环境认识的基础，结合环境反馈值，初步构建了农户土地利用行为决策模型，这对于从微观主体层面揭示土地利用变化的内在机理具有重要意义。论文以陕西省米脂县孟岔村为例，采用农户Agent强化学习模型，模拟了农户土地利用行为，得出以下3个结论：①经过对Agent强化学习模型的检验和验证（按面积统计误差率为5.5%），认为该模型较适合于微观层面农户土地利用行为的研究；②该模型可定量表达政策、市场、农户自身及与其他农户相互作用等因素对农户土地利用行为决策的影响；③与离散函数的土地利用决策相比，应用连续函数表达农户的土地利用决策比较符合实际农户决策状况（农户最终决策中玉米、蔬菜和其它作物的决策值分别为0.3233、0.4035和0.2732）。同时，在文章的最后还分析了需要进一步研究和探讨的问题：①模型在实现农户决策的空间位置方面取得一定进展，但还存在决策位置失误的问题，函待进一步的研究；②对于影响农户决策的自然因素，探讨定量化表达其对农户影响作用，是下一步研究的重点。

Agent-based Modeling of Simulation on Households Land-use Behavior-A Case of Mengcha Village of Mizhi County in Shaanxi Province

Taking reinforcement learning model framework as basis, policy, market, agent and its interaction with other agents, and soil fertility as the choices of agent's awareness towards environment, the households' land-use decision-making model was constructed in combination with the value of the feedback from environment. It has great significance to reveal the inner mechanism of land use at micro-level. Taking Mengcha Village of Mizhi County in Shaanxi Province for example, the agent reinforcement learning model was used to simulate the households land-use behavior. The following results are obtained:1) The agent reinforcement learning model was suitable for study micro-level households land-use behavior after testing and verifying that model (the error rate with area statistics is only 5.5%). 2) With quantitative expression of policies, markets, households and their interactions with other households on households land-use decision-making behavior, the value of agent decision-making of corn, vegetable and other crops in 2008 are 0.3233, 0.4035 and 0.2732, respectively. 3) Compared with discrete functions of land use decision-making, using continuous function to express households land-use decision-making tallies better with the actual situation. This article also analyzed the issues needed for further study and discussion:1) The model made some progress in the aspect of space position, but it still has some errors and needs further research. 2) The next focus of study is to explore the quantitative expression of the impact of natural factors affecting household's decision-making.