Modeling, Identification, and Validation of Models for Predictive Ammonia Control in a Wastewater Treatment Plant—A Case Study

Application of Gaussian processes for black-box modelling of biosystems

“…Beside commonly used performance measures such as e.g. mean relative square error (MRSE, [23]), which compares only the mean prediction of the model to the output of the process:…”

“…For identification, the same data was used as in [23]. We did not have the possibility to influence the data acquisition.…”

“…This procedure is longer and more difficult to pursue, as in principle many possible regressor combinations in the regressor pool must be tested. Based on the prior process knowledge from [23] and its authors, the following regressors were chosen for the initial GP model structure: [S 1,2,3 out S 4 in T 2 DO 3 1 DO 1 2 ], where the superscript denotes the delay, e.g. S 4 in = S in (k − 4).…”

“…The process of interest is moving bed biofilm reactor pilot plant in Domžale-Kamnik WWTP. The case study is adopted from [23], where it was identified with linear and parametric nonlinear models.…”

“…5) consists of an anoxic reactor (176 m 3 ), two aerobic reactors (130 and 117 m 3 ), a mixed reactor (115 m 3 ) and a settler (600 m 3 ) [23,25]. The main process of ammonia (NH 4 -N) removal is nitrification, which runs in aerobic reactors.…”