Digitalisation of ozone pre-oxidation control through a fuzzy inference system: Development and open-loop validation in a full-scale drinking water treatment plant

Abstract

Drinking Water Treatment Plants (DWTPs) increasingly face operational challenges associated with the temporal variability of surface-water quality, particularly in Mediterranean regions affected by seasonal and climatic fluctuations. In ozone-based pre-oxidation systems, dosage adjustment is commonly performed through manual operator supervision, often relying on empirical criteria and limited online monitoring, which may lead to unstable operation and inconsistent control of dissolved organic matter oxidation. This study reports the development and full-scale deployment of a fuzzy inference system for ozone dosage decision-support, using online absorbance at 254 nm and water temperature as inputs and encoding operator expertise as interpretable rules. The system was integrated into the plant SCADA and operated in advisory mode for approximately seventeen months, displaying recommendations without actuating equipment, with A 254 reduction across the ozonation stage reviewed periodically as a supervisory performance indicator. Operators followed the recommended dose on 74% of operating days, but not uniformly: agreement reached 90% in cold water and fell to 35% above 17 °C in both of the summers the record spans. The departures are confined to the thermal term of the rule base, which was the only input introduced on mechanistic grounds rather than from historical practice. Over the assisted period, A 254 reduction met the 25% supervisory target on 62% of reviewed days, coagulant dosage was approximately 8% lower, and trihalomethane concentrations remained below the historical baseline; since no equipment was actuated and the two periods differ in calendar time and instrumentation, these are reported as associations rather than effects. Advisory operation therefore served as a diagnostic of the encoded knowledge, localising a disagreement with operational practice that closed-loop control would have left invisible.