Data-driven modeling for wastewater treatment process (WWTP) under hybrid cloud environment, has been widely regarded as a promising solution. Existing methods managed to learn a forward mapping for WWTP, and were highly reliable on rich intermediate process parameters (IPP) such as dissolved oxygen amount. However, they cannot well handle scenes where IPP are unavailable. In fact, such situations are quite common because many wastewater treatment plants still lack relevant monitoring systems. To remedy such gap, this research collected real-world data from wastewater treatment plants to build realistic experimental scenarios. On this foundation, a probabilistic model for WWTP, named Pro-WWTP for short, is proposed in this paper. More concretely, generative processes of outlet results are expressed as conditional probability, and IPP are estimated via Gibbs sampling-based Bayesian posterior probabilistic inference. Empirically, we conduct two groups of experiments to evaluate proactivity of the proposed Pro-WWTP. Experimental results reveal that Pro-WWTP possesses proper recovery precision for IPP and is able to promote modeling efficiency. Besides, another group of experiments are further implemented to verify total robustness of Pro-WWTP.

数据驱动的混合云环境下废水处理过程（WWTP）建模已被广泛认为是一种有前途的解决方案。现有方法设法学习了WWTP的正向映射，并且在丰富的中间过程参数（IPP）（例如溶解氧量）上高度可靠。但是，它们不能很好地处理IPP不可用的场景。实际上，这种情况非常普遍，因为许多废水处理厂仍缺乏相关的监控系统。为了弥补这种差距，本研究从废水处理厂收集了现实世界的数据，以建立现实的实验方案。在此基础上，提出了污水处理厂的概率模型，简称Pro-WWTP。更具体地说，出口结果的生成过程表示为条件概率，IPP和IPP通过基于Gibbs采样的贝叶斯后验概率推断进行估算。根据经验，我们进行两组实验以评估拟议的Pro-WWTP的主动性。实验结果表明，Pro-WWTP对IPP具有适当的恢复精度，并且能够提高建模效率。此外，进一步进行了另一组实验，以验证Pro-WWTP的总体鲁棒性。