Abstract: The accuracy of cyberspace coordinate systems depends on high-quality delay data, yet random noise in real-world network environments severely limits the effectiveness of traditional filtering methods. To address the insufficient balance among noise suppression, statistical feature preservation, and step-change detection in existing approaches, this paper proposes a segmented delay filtering (SDF) method based on the Savitzky-Golay (SG) filter. The method dynamically segments delay data using a change rate detection strategy. For long segments, the SG filter is applied to fit the trend, while median filtering is used for short segments to suppress noise. This hybrid approach effectively suppresses outliers and smooths delay curves while preserving step-like abrupt changes. Validation experiments on real network datasets demonstrate that the SDF method reduces the 95% percentile relative error (NFPRE) by approximately 18.59% and significantly decreases delay variation rates, thereby enhancing the positioning accuracy and stability of cyberspace coordinate systems.