Accurate Parameter Estimation Based on the Two-Sided KZ Filter in Low Signal-to-Noise Ratio Conditions

System identification is the process of developing models based on data collected from dynamic systems. A significant challenge in this field is accurately estimating model parameters using input-output data. This data, typically gathered through sensors, often contains noise due to inaccuracies in the sensors and errors during communication. The presence of such noise can lead to biased estimates of parameters, which undermines the model’s effectiveness. While traditional methods strive to minimize this bias during parameter estimation, they frequently encounter difficulties when dealing with high levels of noise. An alternative strategy involves employing filtering techniques to clean the data before estimation takes place. Low-pass (LP) filters are designed to reduce frequencies above a specific threshold and are widely used for mitigating noise issues. Among these filters, the Kolmogorov–Zurbenko (KZ) filter operates as multiple iterations of a moving average (MA) filter and has shown strong capabilities in reducing noise; it is commonly applied in high-resolution signal analysis and image processing tasks. In this study, we introduce two KZ filtering approaches aimed at denoising data prior to parameter estimation. Theoretical evaluations and experimental findings indicate that the proposed KZ filtering method surpasses LP filters and wavelet denoising techniques, particularly in environments characterized by low Signal-to-Noise Ratios (SNR).