A few-shot learning based method for industrial internet intrusion detection

Abstract

In response to the issue of insufficient model detection capability caused by the lack of labeled samples and the existence of new types of attacks in the industrial internet, a few-shot learning-based intrusion detection method is proposed.The method constructs the encoder of the prototypical network using a one-dimensional convolutional neural network (1D-CNN) and an attention mechanism, and employs the squared Euclidean distance function as the metric function to improve the prototypical network. This approach aims to enhance the accuracy of intrusion detection in scenarios with scarce labeled samples and the presence of new types of attacks.inally, simulation experiments are conducted on the few-shot learning-based intrusion detection system. The results demonstrate that the method achieves accuracy rates of 86.35% and 91.25% on the CIC-IDS 2017 and GasPipline datasets, respectively, while also exhibiting significant advantages in detecting new types of attacks.