用于非线性时变系统噪声统计Q、R阵估计的新方法

提出了一种估计非线性时变系统过程噪声协方差阵Q和观测噪声协方差阵R的新方法.扩展卡尔曼算法结合前馈神经网络的非线性时变系统辨识过程中,噪声统计Q、R阵的估计是影响系统建模和预测精度的关键因素之一.本文所提出的估计噪声统计Q、R阵方法是基于协方差匹配技术,将M ehra估计定常系统噪声统计的方法推广到一般的非线性时变系统.仿真结果显示了本文方法的有效性.

A new method to estimate unknown noise statistics Q and R of nonlinear time-varying system

A new method is presented to estimate process noise covariances Q and observation noise covariances R of nonlinear time-varying system.It is very important to estimate noise covariances Q and R for identification and prediction of nonlinear timevarying system using extended Kalman filter based on neural network.The present method is based on covariances-matching techniques,and extends the Mehra approach of noise statistics estimation of constant system to general nonlinear time-varying system.The effectiveness of the present method is demonstrated by identification and prediction results of two nonlinear time-varying system.