Using a hybrid modelling approach for high time-resolution prediction of influent orthophosphate load in a water resource recovery facility

Abstract

Water resource recovery facilities face challenges with increasingly stringent effluent demands, complexity and demand for capacity increasing investments. Emerging technologies such as digital twins could alleviate these problems but require high frequency influent data. This work presents a method for utilising measurements in the primary clarifier effluent with a model of the processes between the influent and primary clarifier effluent to predict influent orthophosphate load for a plant with considerable internal load. Five functions for describing daily load variations were tested and compared for accuracy and computational time. All functions were shown to reproduce the measured primary effluent orthophosphate concentration with high accuracy, although the function based on four normal distributions was deemed the most suitable due to its short computational time, realistic influent concentration variations and accurate estimated primary effluent orthophosphate concentration. Validation of the optimised influent concentrations shows that it follows similar patterns but might overpredict the afternoon load, which could be due to deviating daily patterns by inhabitants during the COVID-19 pandemic (although this requires further investigation). The presented methodology can be extended also to estimate influent COD-fractions, automate plant calibration and optimise plant performance.