Research On UAV Anomaly Data Detection Method Based On One-class Classifier

As unmanned aerial vehicle(UAV)technology becomes increasingly ubiquitous in various domains of social life,the demands for the safety and reliability of UAV systems have become a crucial issue.It has been widely accepted in the industry to enhance the safety performance of UAV systems by detecting potential hidden dangers and eliminating possible risks through data inspection of the UAV system.Among them,the effective detection of abnormal flight data,especially anomalous flight data,is a significant challenge in the current research field.Usually,UAV flight data contains both ordinary and abnormal data.There is an obvious embarrassment of imbalanced distribution between the two,which significantly reduces the accuracy of effective detection of abnormal data.To address the above issues,this article applies the one-class classification method to the abnormal data detection process,proposes and implements an abnormal data detection method based on the One Class Kernel Extreme Learning Machine(OCKELM)algorithm,and further ameliorates its robustness on this method.1.To enhance the detection efficiency of abnormal data in UAV data,the Fast ICA algorithm is employed to reconstruct the overall data so that the reconstructed abnormal data can deviate significantly from the normal data distribution,thus optimizing the basic data model for abnormal data detection.By introducing the Triangular Global Alignment Kernel(TGAK)function and utilizing an elastic similarity measure,the distance between normal and abnormal data is increased to increase the accuracy of abnormal data detection.Experimental results show that this method improves the accuracy of abnormal data detection by about 5% or more compared to the original method.2.The selection of hyperparameters usually directly affects the accuracy of detecting abnormal flight data for UAVs.An adaptive hyperparameter optimization method based on the Fruit Fly Optimization Algorithm(FOA)is proposed to solve the problem of hyperparameter selection.By introducing the Levy Flight strategy(LF),Sticky Molecule Algorithm(SMA),and Simulated Annealing Algorithm(SA),the global search and local optimization ability of the FOA algorithm are further optimized.The implementation results show that the optimized method achieves better results in the hyperparameter selection optimization task,with higher accuracy and more substantial stability in detecting abnormal data than other methods.