Combining machine learning and modelling has shown promise for supporting efficient tetracycline degradation and minimising electrical energy consumption in wastewater treatment.
Applying both machine learning and the modelling technique response surface methodology (RSM) could advance pharmaceutical wastewater treatment design, research suggests.
The study by Ezati et al. demonstrated this approach can precisely and efficiently analyse, predict and optimise Electro-Fenton system performance in pharmaceutical wastewater treatment.
The Electro-Fenton process has shown “remarkable potential” to degrade persistent organic pollutants. However, its performance is dependent on several interacting operational parameters, of which traditional approaches cannot accurately analyse or identify optimal operational conditions.
When applied via central composite design (CCD), RSM models accurately assesed input and output variables, showing “high predictive accuracy and good agreement with experimental data”. Ultimately, when combined with analysis of variance (ANOVA), optimal process conditions could be identified.
Ezati et al. utilised RSM and machine learning models to perform multi-objective optimisation of the Electro-Fenton process. This was used to maximise the removal efficiency of the antibiotic tetracycline and minimise electrical energy consumption using the heterogeneous catalyst MIL-100(Fe) and persulfate.
Of all the parameters assessed, pH was the most influential on process performance. “Under optimal conditions, the removal efficiency reached 75.13 percent and the specific energy consumption 165.20 kWh/kg, with deviations from the experimental results of 79.25 percent for removal efficiency and 132.13 kWh/kg for energy consumption, which were considered minimal”.
Overall, integrating RSM, machine learning models and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) algorithm could be applicable to other pharmaceutical contaminants, “by redefining response variables and experimental domains”.
Ezati et al. concluded that while real wastewater matrices may introduce additional complexity, “the data-driven nature of the framework allows recalibration based on site-specific conditions, supporting its potential applicability to practical treatment scenarios”.
This paper was published in Results in Engineering.
No comments yet