Electrostatic Precipitator Optimization using a model-based Evolutionary Algorithm

Abstract

A main component of the gas cleaning system in large scale power plants is an
electrostatic precipitator (ESP). In the entrance of the ESP, a gas distribution
system (GDS) is required to control and guide the gas flow through the seperator.
This GDS consists of several hundred baffles as well as blocking and perforated
plates, each of which can be set up in multiple distinct states. To find a point of
maximum efficiency for the ESP, an optimal configuration of the GDS is necessary.
However, the vast amount of possible combinations in the GDS reveals a complex
discrete optimization problem. For a single evaluation of a given configuration a
computationally expensive CFD simulation is needed, which results in hours of
computation time, even if just the momentum transport of a single phase is
simulated. The proposed solution consists of evolutionary algorithms implemented on
top of an automatically created multi-fidelity model structure. The presentation
will start off with a deeper insight into the complexity of the problem itself. The
second part covers the workflow of the evolutionary algorithm including specially
designed operators. The overall structure using the models of different fidelity
levels is explained. Lastly, the presentation concludes by showing achieved results
and remarks on the performance.