Advancing Water and Wastewater Treatment through the Fusion of Dynamical Systems Theory and Data Science: Perspectives from Bifurcation Structures and Model Reduction
| Reference No. | 2026a027 |
|---|---|
| Type/Category | Grant for General Research-Short-term Joint Research |
| Title of Research Project | Advancing Water and Wastewater Treatment through the Fusion of Dynamical Systems Theory and Data Science: Perspectives from Bifurcation Structures and Model Reduction |
| Principal Investigator | Tomoaki Itayama（Nagasaki University, Graduate School of Integrated Science and Technology・Professor） |
| Research Period |
May 14,2026. -
May 15,2026. August 24,2026. - August 26,2026. |
| Keyword(s) of Research Fields | Dynamical Systems, Neural Networks, Data-driven Prediction, Water and Wastewater Treatment |
| Abstract for Research Report |
Water and wastewater treatment are critical infrastructures with high energy consumption. Consequently, technical innovations for energy efficiency and carbon neutrality are in high demand. In biological wastewater treatment using the activated sludge method, some studies have successfully reduced aeration energy through predictive control based on nonlinear differential equations (Activated Sludge Models). However, typical models involve more than 13 variables and nearly 30 parameters, making site-specific parameter estimation extremely difficult. As a result, the field is shifting toward data-driven methods, such as neural networks, to learn and predict water quality time series. While applying these to predictive control is being tested, the primary challenge for practical implementation is achieving sufficient predictive performance within a short training period. Recent research has explored treating time-series learning as the learning of the underlying latent dynamical systems to predict complex phenomena with minimal training; however, such applications remain rare in water and wastewater treatment. This research aims to clarify the bifurcation structures within the primary parameter space of the differential equations (dynamical systems) describing these systems. By focusing the pre-training of Neural-ODEs on time series generated near bifurcation points, we aim to accelerate learning and enhance predictive capability. Furthermore, focusing on multi-time scale properties, we will utilize singular perturbation methods to construct reduced equation systems, thereby alleviating the "stiffness" of the original systems and facilitating the training process. Research incorporating dynamical systems perspectives into this field is scarce globally and virtually non-existent in Japan. Through this collaboration between Professor Matsue (a specialist in dynamical systems at IMI), water engineering and data science researchers, and industrial engineers, we aim to realize advanced control systems by integrating a dynamical systems perspective into water treatment predictions. This work will contribute significantly to energy reduction in the water sector and the realization of a decarbonized society. |
| Organizing Committee Members (Workshop) Participants (Short-term Joint Usage) |
Tomoakii Itayama（Graduate School of Integrated Science and Technology, Nagasaki University・Professor） Kaname Matsue（Institute of Mathematics for Industry , Kyushu University・Professor） Tomoya Sakai（Graduate School of Integrated Science and Technology, Nagasaki University・Associate Professor） Tesuo Imai（Graduate School of Integrated Science and Technology, Nagasaki University・Associate Professor） Keita Takeda（Graduate School of Integrated Science and Technology, Nagasaki University・Assistant Professor） Tetsuro Ueyama（Kyowakiden Industry, Co., Ltd.・Director） Guyen Binh Minh（Graduate School of Engineering, Nagasaki Uinversity・3rd-year, Ph.D Student） Quoc Binh Diep（Graduate School of Integrated Science and Technology, Nagasaki University・2nd-year, Ph.D Student） |