Event Date: 13 Aug 2015 01:30 AM - 13 Aug 2015 03:00 AM

Event Venue: Tower Building, Level 2, CREATE Theatrette

CAM.CREATE.C4T Seminar Series

Dr. Bhagyesh Patil, C4T Research Fellow, NTU

Title: Global optimization algorithms: design, implementation and application in the process system engineering

 
Abstract:

The optimization of a mixed-integer nonlinear programming (MINLP) problems constitutes an active area of research in the field of engineering and applied science. The present work focuses on development of novel algorithms for unconstrained and constrained global optimization of multivariate polynomial MINLP problems using the Bernstein form. The proposed algorithms in this work are based on the classical interval techniques. The proposed algorithms are of the branch-and-bound, as well as of the branch-and-prune type. In the former (branch-and-bound) approach, branching is done by subdividing the domain of the variables, and bounding operations are done using the Bernstein coefficients. Similarly, in the later (branch-and-prune) approach, branching is done by subdividing the domain of the variables, and pruning is done with the consistency techniques, called Bernstein hull consistency and Bernstein box consistency. To enhance the efficiency of the proposed algorithms, several new devices are developed in this work, such as a modified subdivision procedure, a new branching rule for the variables, and a speed up device, called vectorized Bernstein cut-off test. The performance of all the proposed algorithms is tested and compared on a variety of polynomial MINLP problems from the literature. Similarly, the performance of the proposed algorithms is also compared with state-of-the-art MINLP solvers, such as AlphaECP, BARON, Bonmin, BNB20, and DICOPT. The obtained test results show the superiority of the proposed algorithms over these existing solvers. Lastly, the proposed algorithms are applied to solve a key problem in engineering: optimal feedback control of nonlinear hybrid system using a multiple model approach. Through this successful application, the power and capability of the proposed algorithms in tackling real world challenging problems is demonstrated.

Prof. Jan Maciejowski, C4T IRP4 PI, Uni. of Cambridge

Title: Smart coordination of power generation and consumption

Abstract:

Energy could be saved in power generation and distribution systems if the amount generated and the amount consumed were better coordinated with each other. In traditional power systems the limits to how well this can be done are set by different time-scales, and by the absence of effective means of storing electrical energy. The increased penetration of renewable generation (wind, tidal, solar) and the real possibility of being able to store large amounts of electrical energy in the near future (batteries, super-capacitors) shifts those limits and perhaps allows us to do better. Networks with large smart (industrial) consumers, who may be able to generate thermal and electrical energy as well as to consume it, offers further possibilities for smart coordination. But how could such coordination be achieved? This talk addresses one technical aspect of this (there are also socio-economic aspects). ‘Model Predictive Control’ (‘MPC’) is a very successful control technology which exploits modern computing and sensor technology, and optimisation algorithms, to make complex decisions ‘on the fly’ in dynamically-changing situations. The talk will outline this technology, and summarise MPC-related research being done within the CARES C4T project.

Please indicate your attendance through this link: http://goo.gl/forms/DcXCGhzQJz