Research on Lightweight Abnormal Traffic Filtering Algorithm of Internet of Things Based on Neural Network

With the rapid development of the Internet of Things technology, the scale of the Internet of Things continues to expand, the application of the Internet of Things increases rapidly, and the behavior of the Internet of Things becomes increasingly complex, which poses new challenges to the management and maintenance of the Internet of Things. Accurately predicting IoT data in the future can help operators allocate IoT resources more reasonably, provide better service quality, and help evaluate the carrying capacity of IoT and analyze the health status of IoT. Real-time anomaly detection can help operators discover abnormal devices in the Internet of Things in a timely manner, quickly find the root cause of the problem, and avoid unnecessary losses. Neural network is a flexible and powerful artificial intelligence model, which is applied to IoT data to analyze and process massive data generated by IoT devices, and to identify abnormal patterns in IoT data, such as burst traffic, faulty devices, malicious attack. Thereby improving the security and stability of the IoT system. Therefore, in the face of the increasingly complex IoT environment, how to establish an effective data prediction model and how to detect IoT anomalies in real time is crucial and practical for cloud service providers. Based on this, this paper proposes a BSCN traffic anomaly detection method, which is obtained through the backpropagation neural network (Cheng in Pattern Recogn. 121, 2022) and visual saliency detection technology (Rydell in Linguistic and Philosophical Investigations 21:25–40, 2022), which can effectively extract the hidden abnormal traffic characteristics in the Internet of Things data, and can realize real-time or near real-time abnormal traffic detection. It can also adapt to different types and scales of IoT data, because it can automatically adjust the structure and parameters of the neural network according to the size and number of data blocks, and can adapt to different degrees of abnormal traffic through threshold adjustment.