Harald Siegfried Waschl,
"Self Tuning Strategies for Model Predictive Control of Integral Gas Engines"
, 1-2010
Original Titel:
Self Tuning Strategies for Model Predictive Control of Integral Gas Engines
Sprache des Titels:
Englisch
Englische Kurzfassung:
Mainly, this work is focused on possible methods and strategies for self tuning of model predictive controllers and state observers. These two structures were combined in a control application for integral natural gas engines at gas compressor stations. Such engines are used to maintain the gas flow in the pipeline network and to compensate for load deviations, whereas many of them are internal combustion engines with standard SISO PI control for engine speed and fuel to air ratio. Usually the compressors operate in a narrow operating range of engine speed and load. However, during compressor load changes high disturbances arise on the crankshaft torque and consequently high NOx emission levels. Due to the strengthened emission legislation these engines have to meet lower emission ratings, which is not possible with the current engine setup. To this end a possible approach consists into renewing the fuel and air mixture control with an optimized control strategy.
The diploma thesis is based on a former thesis and an ongoing research project, where different MIMO control strategies for these compressor stations were looked for. The earlier work showed that a predictive MIMO control provides advantages over the standard SISO control. One drawback of that MPC and state observer scheme is the need of a trained control engineer to setup the control at the station. A key intention of this work is to determine a self tuning method that requires a minimum user input and is able to work on different plant types. This is necessary because most of the compressor stations contain legacy engines with different fuel injection types, different numbers of cylinders and different air systems and therefore require an individual setup. That main objective can be divided into two essential parts, the design of a self tuning state observer and a self tuning model predictive control structure. An additional requirement is that the solution has to run online on the control hardware, th