基于RF-MLP集成模型的耕地生态安全预警系统设计与应用  

Design and Implementation of Cultivated Land Ecological Security Early Warning System Based on Multi-layer Perceptron and Random Forest

作　　者：谭术魁[1] 邹尚君 曾忠平[1] 张路 吴静怡 TAN Shu-kui;ZOU Shang-jun;ZENG Zhong-ping;ZHANG Lu;WU Jing-yi(College of Public Administration,Huazhong University of Science and Technology,Wuhan 430074,China;College of Public Administration,Central China Normal University,Wuhan 430079,China)

机构地区：[1]华中科技大学公共管理学院,湖北武汉430074 [2]华中师范大学公共管理学院,湖北武汉430079

出　　处：《长江流域资源与环境》2022年第2期436-446,共11页Resources and Environment in the Yangtze Basin

基　　金：国家自然科学青年科学基金项目(41801205);教育部人文社会科学研究青年基金项目(18YJC790219);华中科技大学自主创新研究基金(人文社科)(2019WKZDJC001)。

摘　　要：开发耕地生态安全预警系统,针对耕地生态安全情况做出预先的分析和警报对指导区域耕地资源合理利用和制定科学、合理的农业发展政策具有重要作用。从PSR模型视角构建指标体系,采用熵权法、多因素综合加权法对耕地生态安全状况进行评价,提出随机森林与多层神经网络集成模型(RF-MLP集成模型)对耕地生态安全水平进行预测,构建耕地生态安全预警系统,进而以中国粮食主产区为实例代入该系统,系统运行流畅,精度检验合格,可信度较高,为实现耕地生态安全监测及预警的业务化运行做了有益尝试,具有较强应用价值。结果表明:(1)以RF-MLP集成模型为关键技术构建的预警系统,利用随机森林模型优化多层神经网络的输入属性,对相关特征进行优化选择的提高模型的预测准确度,又避免特征过多导致过拟合,为实现耕地生态安全监测及预警的业务化运行做了有益尝试。(2)应用结果显示2006~2017年的耕地生态安全值总体呈上升趋势但增幅不高,预警等级由中警转为轻警,警情空间格局整体上呈现出北部省份警情级别较南部省份高的分布格局;2018~2027年,整个粮食主产区耕地生态安全值有所提升,但预警等级仍维持在轻警。预测期间警情空间格局已基本趋于稳定,部分地区例如东北地区、黄淮海地区耕地生态朝着良性方向发展的潜力较大,长江中下游地区耕地生态安全状况虽整体较好但安全值有所下降,预警等级由无警转为轻警。Designing an early warning system for cultivated land ecological security can help us make advance analysis and warnings for the ecological security of cultivated land, which plays an important role in guiding the rational use of regional cultivated land resources and formulating scientific and reasonable agricultural development policies. From the perspective of the PSR model to construct an index system, this paper uses the entropy method and multi-factor comprehensive weighting method to evaluate the ecological security of cultivated land, and then proposes an integrated model of random forest and multi-layer neural network(RF-MLP integrated model) to predict the level of ecological security of cultivated land, and finally the early warning system for cultivated land ecological security was constructed. We used the main grain producing areas in China as an example to verify the system. The system runs smoothly, the accuracy is qualified, and the reliability is high. It has made a useful attempt to realize the commercial operation of cultivated land ecological security monitoring and early warning, and it has strong application value. The results show that:(1) The early warning system built with the RF-MLP integrated model as the key technology uses the random forest model to optimize the input attributes of the multi-layer neural network, which not only optimizes the selection of relevant features to improve the prediction accuracy of the model, but also avoids excessive features leading to over-simulation. It has made a useful attempt to realize the commercial operation of cultivated land ecological security monitoring and early warning.(2) The application results show that the ecological security value of cultivated land from 2006 to 2017 showed an overall upward trend, but the growth rate was not high. the warning level is changed from medium to light, and the spatial pattern of the warning situation shows the distribution pattern of the northern provinces with higher warning level than the southern provinc

关 键 词：耕地生态安全 随机森林 多层神经网络 预警系统 粮食主产区 

分 类 号：X171[环境科学与工程—环境科学]