Correlations of performance for activated sludge using multiple regression with autocorrelation

A method is developed to correlate activated sludge response variables, such as SVI, effluent volatile suspended solids or effluent BOD, with state variables such as F/M ratio, influent BOD, dynamic sludge age, etc. The method is based on multiple linear regression with autocorrelated errors. The method was applied to three data sets from two full-scale activated sludge plants; one a regional municipal utility, the other an industrial facility. Highly statistically significant models were found which could explain 65-82% of the variability in effluent total oxygen demand, 22-60% in effluent volatile suspended solids and 48-88% in sludge volume index. It was only necessary to model up to second order lags. The models were tested for bias using different data sets and produced correlation coefficients between predicted and observed values as high as 0.96. These results show that useful predictive relationships can be developed for full-scale activated sludge processes. The methods could be used to develop a range of automatic process control schemes.