智慧农业控制系统半监督异常检测模型研究  

Research on Semi Supervised Anomaly Detection Models for Smart Agriculture Control Systems

作　　者：王义元 陶志清 李科 孟浩[1] 朱军[1] WANG Yiyuan;TAO Zhiqing;LI Ke;MENG Hao;ZHU Jun(School of Information and Artificial Intelligence,Anhui Agricultural University,230036,Hefei,Anhui,China)

机构地区：[1]安徽农业大学信息与人工智能学院,安徽合肥230036

出　　处：《淮北师范大学学报(自然科学版)》2025年第1期45-50,共6页Journal of Huaibei Normal University:Natural Sciences

基　　金：贵州省科技计划项目(黔科合成果2021-119);颍上县农业农村局委托项目(KJ2022063)。

摘　　要：异常检测对于维持智慧农业控制系统稳定性和提高农业生产效率至关重要。针对异常发生频率高且标签数据稀少问题,构建基于半监督学习一维卷积变分自编码异常检测模型(1DCNN-VAE)。模型结合一维卷积神经网络与变分自编码器,以处理样本特征间时序关系并提取有效特征,通过重构误差进行异常判断。在智慧滴灌异常检测数据集中,与次优模型相比,1DCNN-VAE准确率、召回率、F1值分别提高8.2%、14.8%和6.3%。1DCNN-VAE模型在智慧农业控制系统异常检测任务中具有良好适用性。Anomaly detection is crucial for maintaining the stability of smart agriculture control systems and improving agricultural productivity.To address the high frequency of anomalies and the scarcity of labeled data,a semi-supervised learning based 1D convolutional variational autoencoder anomaly detection model(1DCNN-VAE)was developed.The model integrates 1D convolutional neural networks and variational autoencoders to handle temporal relationships among sample features and extract effective features,with anomaly detection based on reconstruction error.In the smart irrigation anomaly detection dataset,the improved accuracyrate,recall rate,and F1value of IDCNN-VAE increased by 8.2%,14.8%,and 6.3%respectively,compared with sub-optimal model.The 1DCNN-VAE model demonstrates good applicability in the anomaly detection tasks of smart agriculture control systems.

关 键 词：智慧农业 半监督学习 1DCNN-VAE 异常检测 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]