top of page

Knowledge-Driven Multi-Objective Evolutionary Algorithms and Decision Support

Updated: Nov 15, 2019

-Welcome to a research proposal seminar


Held by PhD Student Henrik Smedberg

Date, Time and Venue: Tuesday, 26 November 2019, 13:15 at E112, University of Skövde

Discussion Leader/External Reviewer: Dr. Kjell Orsborn, Uppsala University


Real-world optimization problems often involve the simultaneous optimization of several conflicting objectives. This means that instead of a single optimal solution, a tradeoff between the different objectives exists, which leads to multiple optimal solutions. Multi-Objective Evolutionary Algorithms (MOEAs) are often applied to such problems in order to provide the decision maker with a diverse set of trade-off solutions.


For better decisions in real-world scenarios

This thesis work will have three parts, in the first part, we will apply knowledge discovery techniques on the solutions generated by such algorithms. The extracted knowledge may reveal interesting and previously unknown problem characteristics that can help the decision maker take better decisions in real-world scenarios. Moreover, when such knowledge is extracted from high-performing solutions, it has the potential to improve the overall optimization process.


Knowledge-driven optimization

In this thesis, the application of knowledge discovery in multi-objective optimization is referred to as knowledge-driven optimization (KDO), and two approaches to realize KDO have been identified: offline KDO and online KDO, which together form the second part of the thesis. Offline KDO refers to knowledge discovery from completed optimizations, where the obtained knowledge is stored in a knowledge base to be used by an expert system to influence future optimization runs and problem formulations. On the other hand, online KDO refers to the application of knowledge discovery to a currently running optimization algorithm, and using the obtained knowledge for faster convergence towards the entire Pareto-optimal front or a preferred region thereof. As previously mentioned, knowledge extracted, whether online or offline, can provide better decision support to a decision maker.


Knowledge visualization techniques

The third part of this thesis will be dedicated to knowledge visualization. New techniques to visualize knowledge generated from optimization data can greatly enhance the decision making process. The developments made in this thesis will be validated through benchmark problems and subsequently applied on real-world multi-objective optimization problems from the manufacturing industry.


Read more about subject area OPT-KNOW


Henrik Smedberg PhD Student School of Engineering Science University of Skövde Email: henrik.smedberg@his.se Work: +46 (0)500-448622

27 views0 comments

Comments


bottom of page