Nowadays water treatment process is significant and of great interest due to the difficult environmental situation caused by industrial and household activities of people. It is important to maintain the water treatment quality at the proper level, as it directly affects human health. Eventually the state of the water source may be changed, so it is necessary to adapt the water treatment process to these changes. It is necessary to constantly monitor not only the state of an incoming water source, but also the results of each water treatment. An early warning system was developed to response to a possible abnormal situation, in which drinking water quality parameters are out of rage, in time. Physical and chemical parameters of water source and drinking water obtained after purification as well as operated factors, which affect the water treatment quality, are controlled regularly and form time series system. The procedure of possible abnormal situation detection consists of two stages. The first stage includes modeling and forecasting the water source state factors based on vector autoregression approach. At the second stage piecewise linear regression dependences of quality factors (with a response break) from physical and chemical parameters of water source and operated factors, which characterize the work of water treatment system, are used to forecast drinking water quality parameters. In addition, for each indicator of drinking water quality there is its own model of optimal order based on the modeling sample of the optimal volume. The system quality is estimated using a test sample.