Due to its miniaturization, integration, and fast detection speed, the miniature optical fiber spectrometer was suitable for on-line detection of multi-component substances in industrial sites. In view of the problem that the micro spectrometer was affected by the stray light, photometric noise and light source fluctuation, the measurement accuracy of the spectral signal was poor, and an optical system error correction method was proposed for on-line detection of micro spectrometer. Firstly, a micro spectrometer was used to measure the spectral intensity online, and the spectral signal was obtained from the reference solution and the solution to be measured. Secondly, the adaptive wavelet threshold denoising method was used to eliminate the photometric noise of the spectral signal and improve the sensitivity of the spectral signal. Then, a light source stabilization method based on the invariance of the dual wavelength light intensity ratio was proposed to dynamically eliminate the light source error. Finally, the spectral signals of the copper and cobalt were dynamically corrected, and the performance analysis of the correction curve was performed. The results show that the method proposed in this paper is simple and fast, and it was suitable for dynamic correction of spectral signal to meet the needs of online detection of multi metal impurity ions.