Influent quality and quantity were important factors that caused the abnormal operation of WWTP. In this study, the prediction models of influent quality and quantity were established based on four machine learning methods of Linear Regression, Ridge Regression, ElasticNet Regression and Lasso Regression. The meteorological conditions (precipitation and air temperature) and influent indicators (influent quantity, COD, and NH₃-N) were used as training data. The influent quantity prediction of the models were evaluated using the historical data obtained from a WWTP located in western China, and the results showed that the normal rates of influent quantity were ranged from 98.9%-100%. The highest accuracy was obtained with Ridge method which was 86.19% .
For influent quality (COD) prediction, Ridge method is relatively ideal, with 82% accuracy. For influent quality (NH₃-N) prediction, because of higher data normality rates, Lasso and ElasticNet method were more ideal, both with 74% accuracy. Further, in view of the reason of low prediction accuracy, this paper puts forward the idea of model improvement from the three directions of data fluctuation, correlation and amount. It is expected that this study will provide reference for similar research and provide a reference and thought for similar research.