遥测数据驱动的无人机飞行状态识别方法

无人机飞行状态的识别是无人机飞行状态分析必要的基础,可为无人机任务调度、智能维护维修和设计优化提供参考信息。无人机的遥测数据是对其飞行状态识别的重要依据,针对无人机遥测数据量大、各飞行状态持续时间不同、数据混有噪声、无法直接提供飞行意图信息等问题,提出一种基于切比雪夫特征提取和随机森林分类(Chebyshev-random forest,C-RF)算法的无人机状态识别方法。采用Chebyshev拟合法对遥测数据进行特征提取和降维,利用随机森林算法实现飞行状态的自适应分类。所提出方法将Chebyshev拟合系数计算简单、接近最佳拟合的优点与随机森林算法的训练速度快、分类准确率高和抗噪能力强等优点相结合,可覆盖无人机的各类样本且避免过拟合问题,实现了无人机飞行状态的有效识别。采用真实无人机遥测数据进行验证,总体识别准确率高于90%,少类样本亦可被准确识别,证明了所提出方法的有效性和实用性。

Flight mode recognition method of the unmanned aerial vehicle based on telemetric data

Flight mode recognition of the unmanned aerial vehicle (UAV) is necessary for the analysis of its flight modes. It can provide a reference for task scheduling, intelligent maintenance and design optimization for the UAV. The telemetric data of the UAV are the most significant resource of flight mode recognition with the characters of large amount, different duration of each flight mode, and associated with noise. The data can not directly reflect the flying intention of the UAV. A flight recognition method for the UAV based on Chebyshev random forest (C RF) algorithm is proposed. The method applies Chebyshev fitting algorithm for the feature extraction and dimension reduction, and utilizes the random forest algorithm is used for the adaptive classification of fight modes. The proposed method combines the advantages of Chebyshev fitting algorithm, whose coefficient calculation is simple with good fitting effect, and the random forest algorithm is with the fast training speed, high accuracy and strong anti noise capability. It can achieve effective recognition for flight modes of the UAV, avoiding over fitting in multi classificationwith its randomness. With the verification of real telemetric data of the UAV, the accuracy of the proposed method is over 90% and the category with minority samples is also recognized correctly. The experimental results prove the effectiveness and practicability of the proposed method.