【摘要】 本文以多输入多输出(MIMO)非线性传感器系统为背景,在Ferguson-Srikantan检验法和RBF神经网络拟合法的基础上提出了一种训练样本集中粗差定位与修复方法。传统粗差检验方法以残差作为诊断统计量,容易对高杠杆点和粗差点产生误判。而建立在学生氏残差和外学生氏残差基础上的F-S检验法能高效地区分两者,并定位粗差点,然后利用RBF神经网络拟合法估计并替换粗差点,从而完成训练样本集的修复。实验表明,该方法具有很强的鲁棒性,在精确定位和准确修复粗差数据的同时提高了传感器信号重构的效率。更多还原

【Abstract】 Based on multi-input multi-output (MIMO) nonlinear sensor, a novel method combined Ferguson-Srikantan test and RBF neural network regressing is proposed in this paper to detect and correct gross error data of sample set. Conventional gross error detecting method regards residual error as diagnostic statistics and is liable to misjudge potential case and gross error. Whereas, F-S test introduced in the research is founded upon studentized residual and externally studentized to be capable of distinguishing these two cases efficiently. Thereafter, gross errors will be located and replaced with the estimations calculated by RBF neural network regressing method. Since all gross errors are corrected, the sample set is recovered to be taken as training data for the following signal reconstruction. Emulation experiments and corresponding analysis indicate that the proposed method is provided with higher robustness. Additionally, the precise detection and accurate recovery of gross error remarkably enhance the signal reconstruction of MIMO nonlinear sensor.更多还原