Complex components of lakes and water difference between different sampling points contribute to the matrix effect influence on precision and accuracy of quantitative analysis with laser induced breakdown spectroscopy (LIBS). To reduce the fluctuation of system parameters and the matrix effect caused by the sample enrichment and water quality difference, characteristic spectral lines of the element under test are corrected with the background and the internal standard element. The influence of different input vectors on support vector regression (SVR) model is studied, including the peak intensity, integral intensity and signal to background ratio. Comparison results show that the effect of SVR model with binary input vector of signal to background ratio and intensity corrected by internal standard element is the best, and the root mean square error and the correlation coefficient of training set are 0.367 and 0.981, respectively, and the relative standard deviation and the average relative error of test set are 4.5% and 12.1% respectively. Multiple input vector carrying more characteristic spectral lines can effectively reduce the influence of parameter fluctuations and substrate difference. The experimental conclusion provides reference for data input in LIBS quantitative analysis of heavy metals in natural water.