A comprehensive inversion of the water quality information of sewage water was realized through the combination of hyperspectral technology and water quality comprehensive evaluation method. Taking the sewage sample collected by a sewage treatment plant in Shaanxi as the research object, principal component analysis (PCA) was used to comprehensively evaluate the sewage water quality to obtain a comprehensive evaluation factor for water quality. At the same time, the original wastewater spectrum was obtained by the ASD FieldSpec 3 hyperspectral instrument. After data preprocessing and different mathematical transformations, four spectral indices were obtained: spectral reflectance (SG), reciprocal logarithm (LR), standard normal variable (SNV) and continuum removed (CR). Based on partial least squares regression (PLSR), stepwise regression (SR) and extreme learning machine (ELM), a hyperspectral model of inversion water quality comprehensive evaluation factor was constructed. The results showed that the original spectral data of this group of water samples and the spectral data modeling by standard normalization transformation had good modeling results, and the prediction effect RPD of the model was above 2.5. Among the three models, the PLSR model and the ELM model had good modeling prediction effects, while stepwise regression modeling results were declined compared with PLSR model and ELM model, the R2c and R2p of the REF-SR and SNV-SR models were all above 0.8 and 0.85, and the RPD was above 3.0, which still had a very good inversion prediction effect, and it achieved the optimization of the model and the optimization of the characteristic band, and SNV-SR-ELM (R2c=0.956, R2p=0.954, RMSE=0.500, RPD=4.651) was the best model. The establishment of SNV-SR-ELM model provided a way for the optimization of hyperspectral inversion water quality model and the rapid evaluation of sewage water quality.