Application of Robust Model Identification Techniques in the Activated Sludge Process

Research Engineer: Anthony Wu

Supervisors: Tao Chen, Matt McEwan, Dave Lovett

E-mail: awu@perceptiveapc.com

This Engineering Doctorate Research Project is a collaboration between the University of Surrey and Perceptive Engineering Limited, and is funded by the Engineering and Physical Sciences Research Council

ExperimentationGrey-Box Model

Identification

Sequential Design of Experiments

MPC Model

Process Knowledge

MPC models describe how the output responds to a change in an input signal

Automated controller uses the MPC model to predict the output andadjust the input to bring the output to the set point

3

1

2

Step tests on the input signals to observe how the output responds

Capturing more information-rich data from experiments

• Design of step tests to obtain better (more information-rich) data

• Better data → better MPC models → better process optimisation

• The Fisher Information Matrix (FIM) is used to assess the information

content of a dataset (the inputs and outputs)

• The step-tests for the next experiment is designed to maximise the

information content (the output is estimated from MPC model)

• Constraints are added so the designed input trajectory should not

violate output consents or damage process equipment

Enhancing black-box modelling with process knowledge

• Black-box models: built purely by statistical analysis

Without sufficient information-rich data, black-box

models overfit and perform poorly

• Grey-box models : process knowledge is incorporated

as constraints to black-box models

• Constraints: gain magnitude, minimum/non-minimum

phase, dead-time

[1] M. O’ Brien, J. Mack, B. Lennox, D. Lovett and A. Wall, “Model Predictive Control of an ActivatedSludge Process: A Case Study,” Control Engineering Practices, vol. 19, no. 1, pp. 54-61, 2011.

FRAMEWORK

Output

Input(s)

Noise(disturbance)

Timer

ASP simulation Slow process, disturbance dominated,Stringent output consents

MPCSet point changesDisturbance response

Target/Set Point

Output

InputPredicted Input

Output Prediction

What is my research? I am investigating modelling techniques to build more robust MPC models in wastewater treatment processes.

What is MPC? MPC stands for Model Predictive Control. It is used in automated controllers and improving process efficiency.

What are MPC models? They describe how the output signal responds to a change in input. MPC use these models to make output predictions.

How does MPC benefit the activated sludge process? The activated sludge process requires aeration and this consumes a lot of energy. MPC has

demonstrate energy savings of over 25% [1] in a typical plant without compromising the quality of treated water.

What are the challenges when building MPC models? While MPC is useful, building the MPC model is difficult due to a number of reasons,

including: high non-linearity, long time-delays, resource limitations, stringent output consent limits and process disturbances.

What solutions am I looking at? Two avenues are being explored: Grey-Box Model Identification and the Sequential Design of Experiments.

First principles knowledge and engineering experience

Dead time

Gain magnitude

Gain direction specified to be

negative

Grey-box

Black-box

Example: black-box vs grey-box model

Model overfit

Model Predictive Control

SUMMARY

OUTPUT