Anomaly Detection Algorithm of Industrial Internet of Things Data Platform Based on Deep Learning

Abstract

The development of the Internet of Things (IoT) causes most industrial applications to utilize IoT devices to improve their productivity. Applications such as smart cities, energy management, smart homes, smart cars, and supply chain management widely utilize the IoT to manage the industries’ efficiency. Industrial IoT devices are frequently affected by cybercriminals and damage information and productivity. Criminal activities can be overcome by applying various machine-learning techniques. Existing methods can process intermediate attacks; however, traditional machine learning techniques have difficulties predicting adversarial and catastrophic attacks. In addition, most of the AI-based industrial applications have heterogeneous and mixed data, requiring robust intruder detection systems. The research issues are addressed by introducing the Meta-Heuristic Optimized Deep Random Neural Networks (MH-DRNN). The system uses the optimization process in feature selection and classification, reducing the heterogeneous data analysis issues. The optimization method selects the features from the feature set according to the sunflower movement, which minimizes the difficulties in computation. In addition, three MLP and three recurrent layers are incorporated into this system to maximize the prediction rate up to 99.2% accuracy.