基于野马优化算法改进随机森林模型的区域农业水质综合评价方法

Comprehensive Evaluation of Regional Agricultural Water Quality Based on Improved Random Forest Model Using Wild Horse Optimization Algorithm

摘要:

针对农业实践造成的农业水质污染问题,亟需构建系统的农业水质评价并揭示其时空分异特征及驱动机制。本文利用野马优化算法随机森林(Wild horse optimizer-Random forest,WHO-RF)模型评价了黑龙江省北大荒农垦集团有限公司建三江分公司下辖15个农场地表水及地下水水质,分析其时空演化规律,探究水质变化成因。结果表明:建三江分公司水质符合农业用水要求。2018—2020年地表水质量先降后升,地下水质量则先升后降,2020年两者水质差异最大,2021—2022年地表水及地下水质量均上升,农业水质呈显著改善趋势。与中部农场相比,临江农场地表水及地下水水质变化均较大,同时,区域水质空间异质性由强变弱。通过OOB指数确定了地表水中 TP、NH3-N 含量以及地下水中Cl - 、CODMn、NH3-N含量5个指标是影响区域农业水质的关键指标,顷均氮肥施用量对区域农业水质影响最为显著。为检验WHO-RF模型综合性能,选用传统随机森林(Random forest,RF)模型、蜻蜓算法随机森林(Dragonfly algorithm-Random forest,DA-RF)模型与其进行对比分析,通过比较平均绝对百分比误差、决定系数、均方根误差和均方误差等4类评价指标,验证了该模型在区域水质评价中的优越性与可靠性。研究成果为水质评价提供一种新途径,同时也为区域水质风险应对提供指导。

Abstract:

Water quality is a core element related to human health and food security. To address the issue of agricultural water pollution caused by agricultural practices,it is urgent to establish a systematic evaluation of agricultural water quality and reveal its spatiotemporal differentiation characteristics and driving mechanisms. Using the wild horse optimizer-random forest (WHO-RF) model to evaluate the surface water and groundwater quality of 15 farms under the jurisdiction of Jiansanjiang Branch of Heilongjiang Beidahuang Agricultural Reclamation Group Co. ,Ltd. , their spatiotemporal evolution laws were analyzed,and the causes of water quality changes were explored. The results indicated that the water quality of Jiansanjiang Branch met the requirements for agricultural water use. From 2018 to 2020,the surface water quality firstly decreased and then increased,while the groundwater quality firstly increased and then decreased. In 2020,the difference in water quality between the two was the largest. From 2021 to 2022, the surface water and groundwater quality both increased, and the agricultural water quality showed a significant improvement trend. Compared with the central farms, the surface water and groundwater quality of farms near the riverbank underwent significant changes while the spatial heterogeneity of regional water quality was weakened. The OOB index was used to determine the content of TP and NH3-N in surface water,as well as the content of Cl - ,CODMn,and NH3-N in groundwater, which were five key indicators affecting regional agricultural water quality. The average nitrogen fertilizer application rate had the most significant impact on regional agricultural water quality. To test the comprehensive performance of the WHO-RF model,traditional random forest (RF) model and dragonfly algorith-random forest (DA-RF) model were selected for comparative analysis. By comparing four evaluation indicators based on mean absolute percentage error, coefficient of determination, root mean square error,and mean square error,the superiority and reliability of the model in regional water quality evaluation were verified. The results can provide an approach for water quality evaluation, and also provide guidance for regional water quality risk response.