Fiber-optic plasma sensors continuously collect substantial real-time data during detection processes in biomedical applications, aquatic environment monitoring, and food safety assessments. However, the relatively small variations in output spectra, combined with anomalies and noise from environmental factors or human interference, make it challenging to rapidly extract valid data, thus affecting detection accuracy. To address this, we propose an efficient hybrid intelligent processing method using the Robust iterative reweighted penalized least squares (RirPLS) algorithm to tackle issues including slow data processing, multiple outliers, significant noise, and baseline drift. This algorithm effectively eliminates outliers by continuously updating weights and fitting valid data. Additionally, we implement a two-level improved complete ensemble empirical mode decomposition with adaptive noise (2L-ICEEMDAN) for denoising, preventing information loss in high-frequency intrinsic mode functions. Simulation experiments demonstrate that this method improves the signal-to-noise ratio to 18.0539 dB. In a biological antigen-antibody detection case study, after processing extensive complex data collected from fiber-optic plasma sensors, the noise intensity decreased by 0.5018, correlation dimension reduced by 0.1797, and amplitude-aware permutation entropy decreased by 0.6932. The results indicate that the proposed method demonstrates excellent performance and broad applicability in processing complex data.