RTU Improves Algorithm for Water Quality Detection

«In the PROTOTECH project, in cooperation with Waterson Technologies, we developed an algorithm for detecting water quality anomalies,» says Sergejs Paršutins, Associate Professor, Doctor of Engineering, Institute of Information Technology, Faculty of Computer Science, Information Technology and Energy, Riga Technical University (RTU). «The algorithm we have developed tracks the readings coming from the plant's physicochemical sensors and predicts the probability of an event occurring in the network to which the plant is connected - atypical changes in water composition. This event may be due to regular fluctuations, such as pressure changes, or it may be a so-called baseline change, which indicates a serious change in the composition of the water,» explains the expert. «It can be due to severe turbidity, but it can also indicate that there is some kind of intrusion, for example from groundwater, from sewage, from rainwater. The algorithm we developed has improved accuracy: we made it less sensitive to normal fluctuations in the system and more focused on pollution-related anomalies,» says S.Paršutins.

To train the models, the experts used real data from the RTU Water Research Laboratory, conducting experiments with different types of pollution - in wastewater, groundwater, surface water - and turbidity, simulating intrusions into the network. In turn, a partially labelled Waterson Technologies dataset was to be used to evaluate the performance of the algorithm.

The model uses artificial intelligence technologies. «As a result, the model is less affected by noise and can detect pollution changes more accurately,» adds project co-developer Arnis Kiršners, PhD in Engineering, Assistant Professor at the RTU Institute of Information Technology. In addition, the algorithms developed can be further developed so that they continuously and automatically learn and improve depending on the data they receive. The biggest challenge in this project, according to S.Paršutins, has been to understand the fluctuations that can occur in the water network so that it is possible to distinguish between important and unimportant anomalies: this is the basis for the data labelling that is needed to develop the algorithms. It has also been interesting to see how time series anomaly detection methods work on a real dataset. «There are different noises, which is not the standard by which students are taught. It was a good experience,» says the associate professor. According to him, applying theoretical knowledge to a real, practical task is extremely important. «There are theoretical projects that end with a scientific publication, but here it is the product that is going to be used. So it's also personal innovation and experience,» he explains.

A. Kiršners also agrees with the colleague's comments. «Theory is one thing, but if you have a real task that needs to be framed in a certain way, improving an existing product: this requires the relevant knowledge, for example to pass on information to data markers correctly, because everything has to be accurate, otherwise it will go wrong,» he explains.

PROTOTECH is a prototyping programme created by the Technology Business Centre of the Investment and Development Agency of Latvia, which provides support to natural and legal persons in the design and development of prototypes of science-intensive technology products or services in cooperation with leading Latvian universities - RTU, University of Latvia and Riga Stradins University. The project is implemented under the Norwegian Financial Mechanism 2014-2021 Programme «Entrepreneurship Development, Innovation and Small and Medium-sized Enterprises». The development of this prototype has received support of EUR 15 000. 

Working together for a competitive Europe!