混合滤波去噪与微粒群算法优化的辨识方法

Denoising and Disturbance-rejecting Identification Algorithm Based on Hybrid Filter and Swarm Optimization

针对实际系统信号中不可避免会存在噪声和瞬时扰动,提出了多项式预测与中值滤波相结合的混合实时滤波算法,消除噪声污染.对于去噪后的数据,由于包含瞬时扰动,故用最小二乘辨识算法仍然不能获得满意的结果.为此,在混合滤波去噪的基础上,采用了一种用微粒群算法优化的最小绝对误差辨识算法.仿真实验表明,所提出的方法能够同时克服噪声和瞬时扰动的不利影响,并能获得较好的辨识结果.

Additive noise and instantaneous disturbance may produce adverse influences to system identification. However, an output of a real system is often affected by both noise and disturbance. To reduce these disadvantageous influences, a hybrid filter that combines polynomial prediction filter and median filter for denoising was employed. After the output signal is denoised by the hybrid filter, the identification results obtained by employing the least squares identification method are still unsatisfactory due to instantaneous disturbances. Hence, an identification algorithm based on the hybrid filter and least absolute errors optimized by swarm optimization was proposed. The simulation study shows that the presented approach can overcome the influences of noise and disturbance simultaneously.