Undergraduate, School of Optoelectronics
Published research results in EAAI, a top journal in the field of artificial intelligence
Recently, Huang Xinyu, a 2019 undergraduate student of the School of Optoelectronic Science and Engineering and a member of the "Light Leads the Future" scientific research and education class, wrote a paper "LSTM-NV: A Combined Scheme Against Selective Forwarding Attacks in Event-Driven Wireless Sensor Networks under Harsh Environments" Accepted by Engineering Applications of Artificial Intelligence (EAAI), the top journal in the field of artificial intelligence. Huang Xinyu, a 2019 undergraduate student of the School of Optoelectronics, is the first author of the paper, Professor Wu Yuanming of the School of Optoelectronics is the corresponding author of the paper, and the School of Optoelectronics is the first unit of the paper. This is also the second Top journal paper published by Huang Xinyu as the first author at the undergraduate level after IEEE Sensors Journal.
Engineering Applications of Artificial Intelligence (EAAI) is internationally recognized as the top academic journal in the field of computer science and artificial intelligence, including the application of artificial intelligence and machine learning methods in solving engineering problems; The journal is the top of JCR Region I, Chinese Academy of Sciences Engineering, and Multidisciplinary, with a latest impact factor of 7.802 in 2022.
With the continuous research and development of the Internet of Things, wireless sensor networks, as one of the basic networks of the Internet of Things, have gradually become a research hotspot. Due to the broadcast communication characteristics and unattended characteristics of wireless sensor networks, it is highly vulnerable to external network attacks, especially selective forwarding attacks with more advanced attack methods. Selective forwarding attacks are often manifested as malicious nodes randomly dropping received packets in the time and data dimensions, causing random anomalies in node forwarding behavior. When the network is trapped in a harsh communication environment and the channel quality is poor, the forwarding rate of normal nodes is often mixed with the forwarding rate of malicious nodes that launch selective forwarding attacks, making it difficult to distinguish. However, the previously studied computer network security system requires huge energy and time costs to achieve attack detection due to its high algorithm complexity. Therefore, how to establish a lightweight and effective detection system has become the biggest difficulty for researchers at home and abroad.
The forwarding rate of each round of malicious nodes that launch selective forwarding attacks tends to fluctuate and fall on the time scale, which is very different from the stable normal forwarding rate. In view of this, this research work uses the LSTM network, which is commonly used to process time series data, to predict the forwarding rate data of each node. The predicted forwarding rate data of each round is compared with the real forwarding rate data of each round, and then combined with the node neighbor voting method designed in this work, the node is judged, so as to realize the detection of selective forwarding attacks.
Figure 1. Detection mechanism of LSTM-NV scheme
In this study, LSTM-NV is composed of the time series data processing model VMD-LSTM and Neighbor Voting, and is divided into four steps: the first step is to use the generated normal node data into the VMD-LSTM model for training, master the forwarding characteristics of normal nodes and save the trained model parameters, and provide the predicted value of the current round forwarding rate of each node, and compare the predicted value with the real value to obtain the error series; In the second step, a dynamic threshold method based on data statistical law is established to process the error series, and the local abnormal points corresponding to each node are obtained. In the third step, a dynamic adaptive node neighbor voting mechanism is constructed to further determine the local anomalies of each node and determine the abnormal characteristics of nodes. Finally, according to the pre-selected number of constraint judgments, the nodes that have been voted abnormal by the system are judged to be malicious nodes and isolated from the network.
After the construction of the attack detection scheme is completed, the most commonly used event-driven network in the real world is used as the simulated network mode, and the intrusion detection process is simulated by MATLAB 2022b. When the harsh environment in the setting network changes with time, the proportion of malicious nodes in the network is 5%, 10%, 15%, and 20%, respectively, and the missed detection rate (MDR), false detection rate (FDR), and overall detection accuracy (DAR) corresponding to the detection scheme are obtained.
Figure 2.Deployment diagram of an event-driven wireless sensor network
Figure 3: System Average Missed Detection Rate (MDR) and Average False Detection Rate (FDR)
Figure 4: System Average Detection Accuracy (DAR)
From the experimental results, the overall average missed detection rate (MDR) of the LSTM-NV scheme proposed in this work is less than 3.3%, the average false detection rate (FDR) is less than 0.6%, and the average detection accuracy is higher than 96%, which proves the effectiveness of the scheme. At the same time, after the complexity analysis of the detection scheme, the detection complexity of the LSTM-NV scheme is the order of magnitude of parameter pairs, which is much lower than that of other schemes with the complexity of parameter square.
Huang Xinyu, a 2019 undergraduate student of the School of Optoelectronic Science and Engineering, a member of the Dongliang Engineering Liren Class of the University of Electronic Science and Technology of China, an active member of the third party branch of the undergraduate, was selected into the "Elite Program" of the School of Optoelectronics, "Light Introduction Future Scientific Research and Education Program", etc., and has been elected as an IEEE Student Member since 2021, and serves as a reviewer of IEEE Sensor Journal, IEEE Access, Sensors and other international authoritative journals. During the school, he won more than 20 university-level cultural and sports science innovation awards, including the first prize of the National College Students Mathematics Competition, the first prize of the Mathematical Modeling Competition of the Higher Education Society Cup, and the M Award of the American Mathematical Modeling Competition. He joined Professor Wu Yuanming's research group in the School of Optoelectronic Science and Engineering since his sophomore year, dedicated to the scientific research project of Internet of Things security technology, and published papers in the IEEE Sensor Journal, the top journal of the Chinese Academy of Sciences, as the first author. At present, Huang Xinyu has gone to the Chinese University of Hong Kong to carry out research on 6G communication perception integration technology under the tutelage of Professor Yin Feng and Professor Zhang Zonghui (IEEE Fellow) of CUHK.
Scan the code to view
Related papers
This article is transferred from: University of Electronic Science and Technology Undergraduate Admissions