Monitoring and fault detection methods are increasingly important to achieve a robust and resource efficient operation of wastewater treatment plants (WWTPs). The purpose of this paper was to evaluate a promising machine learning method, Gaussian process regression (GPR), at WWTP monitoring applications. We evaluated GPR at two WWTP monitoring problems: estimate missing data in a flow rate signal (simulated data), and detect a drift in an ammonium sensor (real data). We showed that the GPR with the standard estimation method, maximum likelihood estimation (GPR-MLE), suffered from local optima during estimation of kernel parameters, and was not robust enough for WWTP monitoring applications. However, GPR with a state-of-the-art estimation method based on sequential Monte Carlo estimation (GPR-SMC) gave good predictions and did not suffer from local optima. Comparisons with simple standard methods revealed that GPR-SMC performed better than linear interpolation in estimating missing data in a noisy flow rate signal. We conclude that GPR-SMC is both a general and powerful method for monitoring full-scale WWTPs. However, this paper also shows that it does not always pay off to use more sophisticated methods. New methods should be critically compared against simpler methods, which might be good enough for some scenarios.
Changes in dilution of wastewater to a treatment plant due to infiltration or surface runoff can have a great impact on treatment process performance. This paper presents a model-based approach in which realistic influent scenarios are generated and used as inputs to a dynamic plant-wide process model of the wastewater treatment plant. The simulated operation is subsequently evaluated using life-cycle assessment (LCA) to quantify the environmental impacts of the future influent scenarios. The results show that increased infiltration led to higher environmental impact per kg nitrogen removed. The increase in surface runoff had a minor impact.
Anaerobic digestion is today internationally acknowledged as an environmentally sound process for energy and nutrient recovery from organic wastes, and it is the dominant sludge treatment technology in most countries’ wastewater treatment plants. Laboratory- or pilot-scale experiments are commonly used as a first step to investigate the potential of new ideas or to confirm research hypothesis before confirmation in full-scale. The objectives of this study were to investigate transferability of methane yield assessments between laboratory- and full-scale, and to compare the influence of experimental uncertainties on experimental power in parallel continuous digester experiments for the two scales. Both batch experiment data (used in a simple equation), as well as continuous laboratory experiments, could be used to predict full-scale methane yield with a high accuracy (<5% difference). Full-scale digesters significantly outperformed hand-fed laboratory digesters in terms of experimental power regarding relative differences in methane yield between two digesters operated in parallel. However, to justify costly long-term continuous laboratory-scale experiments with sufficient experimental power and potentially high transferability, resources also have to be allocated to measures that ensure a high data quality from full-scale reference facilities.
I artikeln beskrivs försök med att automatisk upptäcka mätfel i syregivare på reningsverk.
Uncertainty analysis is important for wastewater treatment plant (WWTP) model applications. An important aspect of uncertainty analysis is the identification and proper quantification of sources of uncertainty. In this contribution, a methodology to identify an ensemble of behavioural model representations (combinations of input data, model structure and parameter values) is presented and evaluated. The outcome is a multivariate conditional distribution of input data that is used for generating samples of likely inputs (such as Monte Carlo input samples) to perform WWTP model uncertainty analysis. This article presents an approach to verify uncertainty distributions of input data (otherwise often assumed) by using historical observations and actual plant data.