Network Fault Lightweight Prediction Algorithm Based on Continuous Knowledge Distillation

Network fault prediction is one of the important means to ensure network security and stable operation. Efficient fault prediction can improve the ability of operation and maintenance personnel to deal with faults and reduce losses caused by faults. In edge scenarios, device resources may be limited by hardware resources, such as storage space, memory, processing power. They may also be limited by unstable network connections, such as limited bandwidth, high packet loss rate, and large delay. This paper investigates the research status of network fault prediction at home and abroad. Currently, commonly used network fault prediction methods include methods based on statistics, based on machine learning, and based on deep learning. These network fault prediction methods can learn the characteristics of network faults and have achieved good results in network fault prediction tasks. However, the methods based on neural networks have a large computational resource overhead and are easily limited by device performance in edge scenarios. The methods based on statistics and machine learning have low cost but low accuracy. In this paper, an edge side network fault prediction model based on improved BiLSTM is designed, and improve the continuous distillation technology to design Stage Continuous Knowledge Distillation (SCKD). The simulation experiments prove that the student model performs similarly to the teacher model in terms of accuracy and F1-Score, and has lower memory usage and parameter volume.