基于One-Class SVM的青鱼异常行为识别方法
来源:用户上传
作者:罗毅 王伟 刘勇 姜杰 刘翠棉 赵乐 李歆琰 李治国 廖日红 王艳 王新春 饶凯锋
摘 要：为了更准确地解析青鱼在突发污染环境中的行为变化趋势，提出了一种基于One-Class SVM模型的青鱼异常行为识别方法。以青鱼的生理及行为特征作为观测指标，将采集到的暴露在不同类型和舛忍卣魑廴疚锵碌那圜鱼行为强度信号作为经验数据，利用直方图统计和主成分分析（PCA）对行为强度数据进行降维，实现行为特征提取，基于One-Class SVM构建模型，并以五水合硫酸铜和三氯酚作为特征污染物进行暴露实验对算法进行验证。结果表明，One-Class SVM模型可以准确地识别正常行为和污染物暴露时发生的异常行为;对于有机污染物最快可在10 min内完成预警，重金属污染物可在1 h内完成预警，并且污染物浓度越高，模型的识别效果越好。识别方法可对水源突发性水质污染进行更有效的监测和预警，也可为水污染应急决策提供技术支撑。
关键词：环境质量监测与评价;模式识别;青鱼;异常行为;One-Class SVM
中图分类号：X832 文献标识码：A
DOI： 10.7535/hbgykj.2022yx03008
Abnormal behavior recognition method of medaka based on One-Class SVM
LUO Yi1，WANG Wei2，LIU Yong2，JIANG Jie2，LIU Cuimian1，ZHAO Le3，LI Xinyan3，LI Zhiguo3，LIAO Rihong4，WANG Yan4，WANG Xinchun4，RAO Kaifeng5，6，7
（1.Shijiazhuang Environmental Monitoring Center，Shijiazhuang，Hebei 050022 ，China;2.CASA Environmental Technology （Wuxi） Company Limited，Wuxi，Jiangsu 214024，China;3.Hebei Province Ecology Environmental Monitoring Center，Shijiazhuang，Hebei 050037，China;4.Beijing South-to-North Water Diversion Loop Management Division，Beijing 100176，China;5.Research Center for Eco-Environmental Sciences，Chinese Academy of Sciences，Beijing 100085，China;6.State Key Joint Laboratory of Environment Simulation and Pollution Control，Beijing 100085，China;7.Key Laboratory of Drinking Water Science and Technology，Research Center for Eco-Environmental Sciences，Chinese Academy of Sciences，Beijing 100085，China）
Abstract：In order to analyze the behavior change trend of medaka in sudden polluted environment more accurately，an abnormal behavior recognition method of medaka based on One-Class SVM model was proposed.Taking the physiological and behavioral characteristics of medaka as observation indexes，the behavioral intensity signals of medaka exposed to different types and concentrations of characteristic pollutants were taken as empirical data.The dimension of behavioral intensity data was reduced by histogram statistics and principal component analysis （PCA），so as to realize the extraction of behavioral features.The model was constructed based on One-Class SVM，and the algorithm was verified by exposure experiments with copper sulfate pentahydrate and trichlorophenol as characteristic pollutants.The experimental results show that the One-Class SVM model can accurately identify normal behavior and abnormal behavior during pollutant exposure.For organic pollutants，the early warning can be completed within 10 minutes at the fastest speed，and for heavy metal pollutants，the early warning can be completed within 1 hour.In addition，the higher the pollutant concentration is，the better the recognition effect of the model is.The identification method can carry out more effective monitoring and all-round early warning of sudden water source pollution，and provide technical support for water pollution emergency decision-making.
nlc202206201608
转载注明来源:https://www.xzbu.com/1/view-15434052.htm
