基于LSTM-OCSVM的无人机传感器数据异常检测

Anomaly Detection Method for UAV Sensor Data Based on LSTM-OCSVM

无人机是一种典型的依靠通信和控制系统实现自主飞行的信息物理系统,在安全性和可靠性方面引起了广泛的关注.本文考虑无人机传感器易受网络攻击问题,充分利用数据的时间相关性,提出了针对无人机传感器数据的异常检测模型.首先采用LSTM神经网络对传感器数据进行预测,再将预测值与实际值做差,并将差值输入LSTM分类器进行训练得到包含正样本的超平面,最后计算测试数据到超平面的距离函数值,根据其正负判定异常与否.并且,选择了合适的滑动窗口,在保证异常检测准确率的同时,缩短LSTM神经网络的训练时长.通过仿真实验,验证了该异常检测模型的可行性和有效性.

Unmanned aerial vehicle(UAV) is a typical cyber physical system that relies on communication and control system to realize autonomous flight,and its security and reliability have attracted extensive attention.Considering the vulnerability of UAV sensors to data attacks,this paper made full use of the time correlation of data and proposed an anomaly detection model for UAV sensor data.Firstly,LSTM neural network was used for sensor data prediction,and then the difference between train and prediction was put into OCSVM for training to obtain the hyperplane containing positive samples.Finally,the distance function value of the test data to the hyperplane was calculated,and the anomaly value was determined according to the distance function value.In addition,a suitable sliding window is selected which can ensure the accuracy of anomaly detection and reduce the training time of LSTM neural network.The feasibility and effectiveness of the anomaly detection model are verified by simulation experiments and tests.