The data stream processing framework processes the stream data based on event-time to ensure that the request can be responded to in real-time.In reality,streaming data usu-ally arrives out-of-order due to factors such as network delay.The data stream processing framework commonly adopts the watermark mechanism to address the data disorderedness.Wa-termark is a special kind of data inserted into the data stream with a timestamp,which helps the framework to decide whether the data received is late and thus be discarded.Traditional wa-termark generation strategies are periodic;they cannot dynam-ically adjust the watermark distribution to balance the respon-siveness and accuracy.This paper proposes an adaptive water-mark generation mechanism based on the time series prediction model to address the above limitation.This mechanism dynam-ically adjusts the frequency and timing of watermark distribu-tion using the disordered data ratio and other lateness proper-ties of the data stream to improve the system responsiveness while ensuring acceptable result accuracy.We implement the proposed mechanism on top of Flink and evaluate it with real-world datasets.The experiment results show that our mecha-nism is superior to the existing watermark distribution strate-gies in terms of both system responsiveness and result accuracy.