Parameter Recursive Estimation of Time-varying Systems Based on Kernel Particle Filter

摘要

In order to get a better parameter of time-varying systems, a parameter recursive estimation method based on Kernel Particle Filter (KPF) is proposed. This method is based on the feasible initial parameter which is generated by adaptive modeling in short time-varying systems. A set called particle swarm is constructed by the initial parameters with random noise, and the parameters are improved in recursive process at the same time. The KPF invokes kernels to form a continuous estimate of the posteriori density function. Particles are allocated based on the gradient information estimated from the kernel density estimate of the posterior in the process of Mean shift. The emulation results show that its prediction accuracy is better than least squares and particle filter, this method is suitable for a nonlinear and non-Gaussian noise environment, and can meet the requirement of parameter estimation of real world.