论文部分内容阅读
为避免多假设检验及概率不等式的复杂性,采用数据驱动的检测法,无需无人机的先验结构信息,仅利用无人机模型的输入-输出观测数据序列来实现多无人机编队的异常检测。对于各架无人机的非线性未知关系式,利用可无限逼近的基函数簇将原非线性未知关系式展开,将其表示成回归矢量与参数矢量的线性回归形式。采用最小二乘法求解参数估计矢量,再通过残差来设计异常检测器。当非线性关系式中仅包含有输入量时,通过最小二乘法得到的残差异常检测器可达到较好的性能。当非线性关系式中同时包含有输入和输出量时,由最小二乘法得到参数估计矢量是有偏估计,此有偏估计势必会影响最终的残差异常检测器。因此在有偏参数估计矢量中添加偏差补偿项,使之成为无偏估计矢量;并推导此偏差补偿项的表示形式,证明添加此偏差补偿项后的无偏性,提出替换偏差补偿项中某矩阵的构造方法。最后用仿真算例验证所提方法的有效性。
In order to avoid the complexity of multi-hypothesis testing and probability inequality, the data-driven detection method is used to realize the UAV formation only by using UAV input-output observation data sequence without the UAV prior structure information abnormal detection. For each unknown UAV unknown relation, the original nonlinear unknown relation can be expanded by means of an infinitely approachable basis function cluster, which is represented as a linear regression form of the regression vector and the parameter vector. The least squares method is used to solve the parameter estimation vector, then the residual error is used to design the anomaly detector. When the non-linear relationship contains only the input, the residual anomaly detector by the least squares method can achieve better performance. When the non-linear relationship contains both input and output, the least square method to obtain the parameter estimation vector is biased estimation, which will inevitably affect the final residual error detector. Therefore, a bias compensation term is added to the biased parameter estimation vector to make it an unbiased estimation vector. The representation form of this bias compensation term is deduced, and the unbiasedness after adding this bias compensation term is proved. Method of constructing matrix Finally, a simulation example is used to verify the effectiveness of the proposed method.