摘要
青霉素的生产发酵过程呈现时变性强和非线性强的特征,针对药厂青霉素发酵工艺中残糖时序数据突然出现的异常行为,可以及时发现并以此来避免重大经济损失节约更多资源,提出将LSTM算法的特征嵌入到AE算法的结构中,并以发酵培养基中残糖浓度作为研究对象,可以很好的捕捉到残糖数据中的时变性和非线性特征,从而更好地进行异常检测功能。AE负责捕获变量的潜在空间,进一步提升LSTM的检测能力。结果表明,该算法对比一般的重构异常检测算法其准确率大于90%,具有良好的准确性和适应性。该算法能有效识别数据中的时变性和非线性特征,从而为药厂数据的异常检测部分提供理论参考与方法依据。
The production and fermentation process of penicillin exhibits pronounced time-varying and non-linear characteristics.This is exemplified by the sudden and anomalous behaviour of residual sugar time series data in the penicillin fermentation process in pharmaceutical factories.Such behaviour can be identified in a timely manner and employed to avert significant economic losses and conserve resources.The features of the LSTM algorithm are embedded into the structure of the AE algorithm,with the concentration of residual sugar in the fermentation medium taken as the research object.This allows the time-varying and non-linear features in the residual sugar data to be captured,thereby further improving the detection capability of LSTM.The time-varying and non-linear features in the data can be effectively captured,thereby enhancing the anomaly detection function.The AE is responsible for capturing the latent space of the variables,which further enhances the detection ability of the LSTM.The results demonstrate that the algorithm exhibits excellent accuracy and adaptability when compared to a general reconstruction anomaly detection algorithm,with a detection accuracy exceeding 90%.The algorithm effectively identifies time-varying and nonlinear features in the data,providing valuable theoretical insights and a methodological foundation for anomaly detection in pharmacy data.
作者
张秀清
王杰
王晓君
赵春丽
ZHANG Xiuqing;WANG Jie;WANG Xiaojun;ZHAO Chunli(School of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050091,China;Sinopharm Group Weiqida Pharmaceutical Co,Datong 037000,China)
出处
《通信与信息技术》
2025年第6期22-25,共4页
Communication & Information Technology
关键词
异常检测
时间序列
长短期记忆网络
自编码网络
重构误差
Anomaly detection
Time series
Long and short-term memory network
Self-coding network
Reconstruction error
