摘要
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。
One data driven method is used to detect the anomalies in multi UAV formation from only input and output data sequences about each UAV. Because this special method needs not any prior structure information of UAV and avoids the complexity coming from multi hypothesis test and some probability inequalities. The established nonlinear unknown form corresponding to each UAV can be well approximated by basis functions and converted to be a linear regressor form which includes one regressor and parameter vector. The parameter vector is identified by least square method,then one anomaly detector is constructed based the derived residuals. When only the inputs are included in this nonlinear form,the anomaly detector based on the consistent parameter estimations can satisfy the better performance. But when the inputs and outputs are all included simultaneously,the parameter estimations are biased using the common least squares method. These biased estimations will affect the final anomaly detector seriously. So in order to avoid this bias,the bias compensated terms are added into the bias estimations. Furthermore the specified expressions about the bias compensated terms are derived and the consistent property is also proven. The way that how to replace some matrices in the bias compensated terms is constructed. Finally the simulation example results confirm the identification theoretical results.
出处
《科学技术与工程》
北大核心
2015年第29期87-94,共8页
Science Technology and Engineering
基金
国家自然科学基金(61402426)资助
关键词
多无人机编队
异常检测
非线性辨识
偏差补偿估计
multi UAV formation anomaly detection nonlinear identification bias compensated estimation
