基于神经网络的无迹滤波改进算法

The Improved Unscented Kalman Filter Based on Neural Network

介绍了采用无迹变换(UT)描述随机变量通过非线性系统后的均值及方差的方法,提出可以将神经模糊推理系统(ANFIS)用于确定无迹变换中的参数,使其对随机变量均值的描述达到二次以上精度,并给出了改进的无迹滤波器(UKF)结构和神经网络训练方法;仿真结果表明,该算法适用于系统含有未知输入或系统噪声为非高斯的情况,并可解决一些典型的非线性估计问题,改进算法的性能优于传统无迹滤波器。

The paper describes a method called unscented transformation （UT） for predicting mean and covariance in nonlinear system. A new approach to selecting sigma points by using adaptive neuro-fuzzy inference systems （ANFIS） is proposed and the sample points capture the posterior mean accurately to the 2rd or higher order for any nonlinearity. The superior performance of the improved unscented kalman filter （UKF） is clearly shown in a numerical example compared with the standard one. The new method has the great advantage of being able to handle unknown input and non-Gaussian noise.