Sequencing batch-reactor control using Gaussian-process models

• A novel framework for the optimisation of phases duration in the SBR operation.
• Signal smoothing and the transitions recognition are done within the same framework.
• Treatment efficiency is increased and energy consumption is decreased.
• Validation on on-line signals confirmed usefulness of Gaussian-process-model framework.

This paper presents a Gaussian-process (GP) model for the design of sequencing batch-reactor (SBR) control for wastewater treatment. The GP model is a probabilistic, nonparametric model with uncertainty predictions. In the case of SBR control, it is used for the on-line optimisation of the batch-phases duration. The control algorithm follows the course of the indirect process variables (pH, redox potential and dissolved oxygen concentration) and recognises the characteristic patterns in their time profile. The control algorithm uses GP-based regression to smooth the signals and GP-based classification for the pattern recognition. When tested on the signals from an SBR laboratory pilot plant, the control algorithm provided a satisfactory agreement between the proposed completion times and the actual termination times of the biodegradation processes. In a set of tested batches the final ammonia and nitrate concentrations were below 1 and 0.5 mg L−1, respectively, while the aeration time was shortened considerably.