: ..
:
: 119
:
: 2026
: .. // . - 2026. - . 119. - .79-90.
: , ,
(.): persistent homology, topological analysis, automated control system
: , . , , , , , . 89% , . , , , . . . 89% 6% , . .
(.): The paper describes the development, mathematical justification and implementation of a security element model at industrial facilities based on convolutional neural networks and persistent homology. The following works were performed: analysis of requirements for monitoring systems at industrial facilities, formalization of the feature space, development of a hybrid data analysis model, optimization of the model in embedded systems, experimental study of the model's effectiveness, methods for integrating the hybrid model into the automated control system structure at energy facilities. An increase in the accuracy of anomaly detection to 8-9% has been achieved on a dataset simulating thermal and radiation maps. The key factor in achieving improved recognition accuracy is the persistent image method, which allows the use of gradient methods to optimize the persistent homology method. This makes it possible to integrate the topological data analysis module into the architecture of neural networks. The experimental part was carried out on three datasets of different architectures. The method shows an increase in accuracy of up to 8-9% according to various metrics on the dataset of heat maps and up to 6% on the dataset used to monitor the integrity of structures. The conclusion is made about the universality of the method for use in working with data sets with a pronounced topological structure.
PDF
: 37, : 12, : 11.