Chemical Engineering and Danish Hydraulic Institute (DHI) are looking for a Master student interested in data analysis and mathematical modelling in
wastewater treatment engineering.
Biological processes in wastewater
often exhibit time-varying and highly dynamic characteristics, which can be
influenced by various known and unknown factors. Water quality can vary depending on
concentration of different substrates and pollutants, the oxygen level, the solids
retention time, pH, temperature and many more. Various mathematical models have
been developed to evaluate and diagnose the performance of wastewater treatment
plants (WWTPs). The dynamic models are primarily based on first principles modelling approaches where physico-chemical and biological processes are often
described through sets of differential or algebraic equations to simulate the
variation of model components (e.g. concentration of different pollutants) with
respect to time. These models are usetful for process simulation, optimization
and control. Alternatively, Machine learning (ML) techniques can yield models that have shown great
potential in predicting the behavior of complicated biological systems with
high accuracy. The main advantage of such models is that they can predict the
output values based on input values after training and evaluation steps, without
the requirement of having a complex physical calibrated model. Hence, these
models can demonstrate the cause-and-effect relationship between input and
output values that can ultimately be helpful for process dynamics understanding
and as a decision-making tool. One of the main obstacles in achieving satisfactory
predictions with ML models is the scarcity of available data and lthe ack of accurate
measurements. This is a frequent problem in making decisions relating to
operational costs on WWTPs. To keep the monitoring cost low and yet have
satisfactory ML model predictions, the type, frequency and the location of
measurements should be well adjusted.
- Perform dynamic simulations (in WEST by DHI) to predict
targeted effluent concentrations (e.g. total nitrogen and COD) in a benchmark
simulation, considering default settings of the model.
Develop machine learning (ML) models, in Python, R, or Matlab
(e.g. artificial neural networks and support vector machines) to predict the targeted effluent concentrations based on
a set of already simulated components (e.g. TSS, COD, total nitrogen, total
phosphorous, temperature and pH) at different locations (e.g. influent,
aeration tank and sedimentation).
Assess ML model performance against WEST predictions based on well-defined criteria
Sensitivity analysis using ML model to identify the input
parameters that may have potential influence on the prediction of targeted
Test the ML prediction accuracy under different scenarios by
changing: i - frequency of data; ii - precision of data e.g. considering an
uncertainty with normal distribution; iii - location of data (i.e. change location where data are collected in teh WWTP).
I samarbejde medPedram Ramin (DHI) (firstname.lastname@example.org)