CONVOLUTIONAL NEURAL NETWORKS IN FAULT DIAGNOSIS OF INDUSTRIAL INTERNET OF THINGS

Abstract

Benefiting from the rapid development of Internet technology and communication technology, the Internet of Things industry has risen rapidly. With the rapid development of Internet technology, network security has become increasingly prominent. Moreover, intrusion attacks can cause system failures or reduce system performance, so intrusion detection is an important aspect of ensuring system reliability. Aiming at the great security risks faced by industrial Internet of Things during operation, this study proposes an industrial Internet of Things fault detection model based on a convolutional neural network, which initially screens the intrusion attacks by convolutional neural network, and introduces a particle swarm optimization algorithm to identify the screened intrusion attacks. The experimental results demonstrated that when the training set size was 1600, the accuracy rates of random forest, K-mean clustering algorithm, convolutional neural network and improved convolutional neural network algorithms were 93.2%, 94.9%, 96.3%, and 98.6%, respectively, and the false alarm rates were 6.9%, 5.0%, 3.8%, and 2.1%, respectively. The random forest, K-mean clustering, convolutional neural network, and improved convolutional neural network algorithms had root mean square error values of 0.32, 0.22, 0.18, and 0.11, respectively. The corresponding F1 values were 0.81, 0.84, 0.87, and 0.98 when the training set size was 800. The results of the study demonstrate that the improved algorithmic model outperforms the other strategies, offering a solid foundation for application in the industrial Internet of things.