湖南大学学报（自然科学版）2025，Vol.52Issue(10)：99-107,9.DOI:10.16339/j.cnki.hdxbzkb.2025210
基于卷积自编码器的综合传动异常检测研究
Research on Integrated Transmission Anomaly Detection Based on Convolutional Autoencoder
摘要
Abstract
To address the challenges in the integrated transmission devices of special vehicle working process,including complex data,the imbalance between normal and abnormal data and the high false alarm and underreporting rates of traditional statistical methods in detecting sensor anomalies,a new method,called ACA-SVM anomaly detection method,combining an attention mechanism convolutional autoencoder(ACA)with a support vector machine(SVM)is proposed.Using operation data from armored tracked vehicles,the monitoring data from integrated transmission sensors is preprocessed.Key features in the transmission data are identified and focused by the attention mechanism.The convolutional autoencoder(CAE)then reduces the dimensionality of the original data and extracts features to facilitate data detection,resulting in the calculation of reconstruction errors and eigenvalues.The support vector machine classifies and calculates the abnormal scores of the training set data samples,and the detection performance is compared with the traditional anomaly detection models.Experimental results show that the proposed ACA-SVM method outperforms CAE,gated recurrent unit(GRU),and other models in detecting anomalies in the integrated transmission data of special vehicles,achieving a detection accuracy of 97.2%and an F1 of 0.976.关键词
异常检测/传动装置/漏油故障/卷积自编码器/注意力机制Key words
anomaly detection/transmission device/oil leakage/convolutional autoencoder/attention mechanism分类
计算机与自动化引用本文复制引用
贾然,吴傲,陈涛,郝乃芃,王立勇,赵津..基于卷积自编码器的综合传动异常检测研究[J].湖南大学学报（自然科学版）,2025,52(10):99-107,9.基金项目
北京市教委科研计划科技一般项目(KM202311232006),Beijing Municipal Commission of Education Science Research Program General Science and Technology Project(KM202311232006) （KM202311232006）
国家自然科学基金资助项目(52175074),National Natural Science Founda-tion of China(52175074) （52175074）
