Parallel Methods for Dynamic Optimization
Researchers: Johan Åkesson, Carl Laird (Texas A&M University, TX, USA)
Optimization is used extensively in many contexts in control engineering. Applications include design optimization to develop optimal processes, set-point optimization to minimize raw material and energy consumption, and on-line optimal control strategies such as Model Predictive Control (MPC). As systems are becoming increasingly complex, the need for efficient computational methods is put into focus.
The proposed research project is motivated by Moore's law, which states that the maximum number of transistors that be fit into an Integrated Circuit to a reasonable cost is doubled every other year. For decades, Moore's law has been closely related to important performance measures, for example the computational power of processors found in desktop computers. During the last 3-4 years this situation has changed, however. While the number of transistors on an Integrated Circuit continues to increase rapidly, many software applications does not run at correspondingly higher execution speeds. The explanation is that modern processors are equipped with multiple cores. Also, the clock frequency, which directly affect execution speed, is increasing only moderately. Many applications are capable of utilizing only one core, and cannot benefit from the availability of multi-core architectures.
In order to utilize more than one core, new methods and/or application of known methods in new contexts are needed. Such methods are typically specific for different application areas. In the field of dynamic optimization, development of parallel and distributed methods is essential in order to efficiently meet the challenges outlined above. In principle, there are two different scenarios that require attention. In the first scenario, the main challenge is the complexity of the problem. In this case, decomposition and parallelization is important in order to obtain manageable subproblems to distribute amongst the available cores. In the second scenario, the complexity of the problem may be moderate, but the computation time is critical. For example, MPC falls into this category. In this case, parallel algorithms are needed in order to fully explore the computational power of multi-core architectures and thereby reduce computation times.
Carl Laird, Angelica Wong, Johan Åkesson: "Parallel Solution of Large-Scale Dynamic Optimization Problems". In 21st European Symposium on Computer-Aided Process Engineering, May 2011.