Research And FPGA-based Implementation On Data Anomaly Detection Algorithm Of Vehicular CAN Bus

The Controller Area Network(CAN)bus protocol is widely used in the vehicular system and is an efficient standard bus enabling communication between all Electronic Control Units(ECUs).However,the CAN bus is easy to be attacked because of a lack of security defense features.As the functions of Intelligent Connected Vehicles(ICVs)become more and more complex,the interfaces open to the outside world are gradually increasing.These functions may require access to the CAN bus inside the vehicle,which will cause unpredictable security threats to the internal communication system.By cracking the rule of CAN messages,the attacker can launch a precise replay attack toward the CAN bus,which has small-batch characteristics.For the design of vehicular intrusion detection systems,this paper proposes an algorithm model that merges Grid Long Short-Term Memory(Grid LSTM)and SelfAttention Mechanism(SAM),shortly named GLSAM.The Self-Attention Mechanism can enhance the characteristics of attack behaviors,and the Grid LSTM can effectively extract the depth features of the time series data.After analyzing the characteristics of various attack types toward the CAN bus,this paper designed a simulated system to generate five attack datasets by extracting the normal CAN dataset from the actual car,including Do S,fuzzy,spoofing,replay,and delete attacks.On this basis,this paper trained and tested the GLSAM and other comparative models.The experimental results demonstrate that the proposed GLSAM model can efficiently identify small-batch attacks with an overall detection accuracy of 98.98% in detecting the data anomaly on the CAN bus.Based on the trained model parameters and FPGA-embedded devices,this paper uses parallel optimization and quantization techniques to design and implement three LSTM-related hardware-accelerated models,including Stacked LSTM,Grid LSTM,and the proposed GLSAM.The experimental results show that the energy efficiency of the three acceleration models is better than that of the high-performance platform of the Intel i9 CPU.GLSAM still displays a high detection accuracy of98.81% and a low latency of 1.88 ms,even with a certain degree of quantification.The investigation provides a new idea and the hardware acceleration scheme for designing the high-precision and real-time vehicular intrusion detection systems.