Data to intelligence: The role of data-driven models in wastewater treatment

Increasing energy efficiency in wastewater treatment plants (WWTPs) is becoming more important. An emerging approach to addressing this issue is to exploit development in data science and modelling. Deployment of sensors to measure various parameters in WWTPs opens greater opportunities for exploiting the wealth of data. Artificial intelligence (AI) is emerging as a solution for automation and digitalization in the wastewater sector. This review aims to comprehensively investigate, summarize and analyze recent developments in AI methods applied to the modelling of WWTPs. The review shows that among the standalone models, Artificial Neural Networks (ANN) was the most popular model followed by, in descending order: Decision Trees (DT), Fuzzy Logic (FL), Genetic algorithm (GA) and Support Vector Machine (SVM). In the case of incomplete data, FL was the most frequently used method as it uses linguistic expert rules to find an approximation for the missing data. Regarding accuracy and precision, hybrid models demonstrated relatively better performance than the standalone ones. Among these models, the Machine Learning (ML)-metaheuristic, which integrates an AI model with a bioinspired optimization method, was the most preferred type as it was used in more than 45% of the hybrid models. Correlation coefficient (R), Correlation of Determination (R2) and Root Mean Square Error (RMSE) were the frequently used metrics for model performance evaluation. Finally, the review shows that despite recent developments, industrial deployment is still lacking. The industrial application requires close interaction of interested parties, among which research institutes, private sector and public sector play an inevitable role. The future research should focus on mitigating the barriers for more in-depth collaboration of interested parties and finding new paths for more cooperative and harmonized activity of them.