Water Resources Research Act Program

Details for Project ID 2019DC048B

Application of deep reinforcement learning in optimizing the operation of biofilters in carbon and nitrogen removal from stormwater

Institute: District of Columbia
Year Established: 2019 Start Date: 2019-03-01 End Date: 2020-02-28
Total Federal Funds: $9,999 Total Non-Federal Funds: $22,050

Principal Investigators: Arash Massoudieh

Abstract: A pilot system will be constructed aiming at facilitating the application of deep reinforcement learning on optimizing the operation of biofilters for treating stormwater. Various water quality parameters including DO, NO3, ORP, pH and temperature will be monitored continuously, and the data will be saved on a cloud MySQL database. The inflow will be obtained from a bioretention system currently under construction at CUA right in front of the Engineering building. Operation of the pilot including the aeration rate, return flow and water storage in the system will be automatically controlled via a number of solenoid valves. A Deep Reinforcement Learning (DRP)-based program will be written to learn from the collected data on-line and to optimize the operation of the pilot in terms of achieving carbon and nitrogen thresholds by dynamically adjusting the storing or release of water, aeration and also by recommending backwash time. Also a process-based model of the biofiltration will be constructed using the Green Infrastructure Flexible model (GIFMod) and will be calibrated using the collected data aiming at providing a process-level understanding of factors affecting biofiltration performance.