Few-shot learning based intrusion detection method of industrial control system
US11218502B1 · kind B1 · utility
Assignee
Inventors
Key dates
| Filing date | Jun 27, 2021 |
| Grant date | Jan 4, 2022 |
| Priority date | — |
| Expiry date | Jun 27, 2041 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06N20/00
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A few-shot learning based intrusion detection method of an industrial control system, including: dividing an original data set extracted from a data flow of the industrial control system into a detection model training set and a basic model training set; using principal component analysis method to reduce dimension of a continuous data matrix M in the two training sets; using one-hot encoding method to process a discrete data matrix V in the two training sets; using processed basic model training set to construct few-shot training tasks required for basic model training; training a basic model based on convolutional neural networks with help of constructed few-shot training tasks; based on trained basic model, using processed detection model training set for further training to obtain the detection model; effectively detecting attacks in real-time data streams with help of center vectors of three different types of samples in the detection model.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.