In terms of water quality monitoring based on the ultraviolet-visible spectroscopy, different optical path lengths of spectrometer probe need to be set to maintain higher signal-to-noise ratio of the spectra when the water body measured is complex and changeable. However, large numbers of experiments always have to be undertaken to get the appropriate optical path length, which is difficult to meet the demand of real-time, accurate, sensitive and stable online monitoring system. In this paper, an optimized spectra fusion algorithm was developed to improve the signal-to-noise ratio of fused spectra from two independent spectra that were acquired using two different optical path lengths. In the fusion algorithm, the sliding-pane method was applied to obtain the distribution of noise variance of the spectra, so the region of strong noise in the spectra could be determined. Due to different signal intensity of spectra with long and short optical path length, genetic algorithm was used to calculate the optimal gain matching rate of fusion. Finally, according to the distribution of noise variance, piecewise weighted method is applied to achieve a fusion spectrum with higher signal-to-noise ratio. The experimental results showed that the fusion algorithm could effectively enhance the signal-to-noise ratio of the fused spectra for each sample without altering the optical path length; the noise within 200~250 nm was suppressed and the low-noise and high-sensitivity spectra in the visible band were preserved; Zero interference was moved to the left of 220 nm of the spectrum. This means the fusion algorithm not only shows improvements in both signal-to-noise ratio and the detailed characteristics of the spectrum, but also reduces the excessive number of experiments in order to optimize optical path length and minimize noise in spectra. It has important practical significance to broaden the application range of the ultraviolet-visible spectroscopy based water quality monitoring system.