针对基本粒子滤波算法存在的粒子退化问题,提出了一种基于广义回归神经网络(GRNN)的重要性样本调整的粒子滤波算法。利用广义回归神经网络优化从重要性密度函数采样的样本,将样本作为神经网络的输入,以观测值作为神经网络的目标向量,通过多次训练优化光滑因子逼近目标向量,用样本值和其周围的调整值作为训练后神经网络的输入向量,通过神经网络的输出向量指示用最优点来取代样本值。利用GRNN对样本进行调整,使得样本更接近于后验概率密度。仿真结果表明:基于广义回归神经网络的粒子滤波算法的性能在有效粒子数和均方误差参数方面优于基本粒子滤波算法,在改善滤波精度方面取得了较好的效果,验证了广义回归神经网络在粒子滤波算法中是可用的和有效的。 Aiming at the problem of particle degeneration existing in basic particle filter algorithm, a particle filter algorithm based on GRNN is proposed. Using generalized regression neural network to optimize samples sampled from important density function, the samples are taken as the input of neural network, the observed value is used as the target vector of neural network, and the target vector is approximated by training smoothing factor with multiple trainings. As the input vector of the trained neural network, the output vector of the neural network is used to indicate the optimal point to replace the sample value. The sample is adjusted using GRNN to bring the sample closer to the posteriori probability density. The simulation results show that the performance of the particle filter based on generalized regression neural network is better than that of the basic particle filter in terms of the number of effective particles and the mean square error, and has achieved good results in improving the filtering accuracy. The generalized regression neural network Available and valid in particle filter algorithms.