Big data analysis and algorithm optimization in tobacco equipment operation monitoring system

Traditional tobacco equipment maintenance depends on regular maintenance and empirical judgment, and it is difficult to accurately predict equipment failures. Therefore, this paper proposes a monitoring system architecture based on big data analysis and algorithm optimization. The system adopts hierarchical and modular design, including four levels of data acquisition, processing, analysis and prediction and user interaction. Real-time acquisition of equipment operation data through sensor network, and data preprocessing by using sliding window denoising, dynamic normalization and other technologies. In the analysis and prediction layer, LSTM neural network and improved One-Class SVM model are used to realize equipment health prediction and fault detection respectively. The experimental results show that the optimized algorithm is significantly superior to the traditional method in fault detection accuracy, recall and response timeliness, with the accuracy increased from 83.2% to 98.1%, recall increased from 67.5% to 96.3%, false alarm rate decreased from 9.8% to 1.2%, and F1-score increased from 0.741 to 0.972. In addition, the model still has a certain generalization ability in cross-equipment models and some cross-factory scenarios, but cross-factory scenarios need to be further optimized to narrow the data distribution differences. The research in this paper provides strong technical support for the intelligent transformation of tobacco industry, and also provides reference for equipment condition monitoring in other industries.