FiDiPro Project DeCoMo: Decision Support for Complex Multiobjective Optimization Problems 2015-2017


2nd workshop
Participants of the Workshop Artificial Intelligence Supported Decision Making in Industry dedicated to the findings and results of the project (September 5, 2017)


The project "Decision Support for Complex Multiobjective Optimization Problems (DeCoMo)" was funded by Tekes – the Finnish Funding Agency for Innovation under Finland Distinguished Professor (FiDiPro) programme. The Industrial Optimization Group was happy to host FiDiPro Professor Yaochu Jin, Chair in Computational Intelligence, Department of Computing, University of Surrey, UK. Besides Tekes, the DeCoMo project was funded by Outotec, Fingrid and the University of Jyvaskyla. This project run for 3 years in 2015-2017. It involved also DIMECC, Fortum Power and Heat, Simosol, Valmet Power, Valtra and Honda Research Institute Europe.

FiDiPro Professor Yaochu Jin is a world-leading researcher in surrogate-assisted evolutionary optimization as well as multiobjective optimization. He has rich expertise in learning systems, in particular in integrating evolution and learning. Additionally, he has long experience working with the Honda Research Institute, where he worked on various real-world aerodynamic optimization problems.

In DeCoMo, the expertise of Prof. Jin and that of the Industrial Optimization group in interactive multiobjective optimization complemented each other and novel optimization methods were developed for decision support in solving complex multiobjective optimization problems by combining modern meta-heuristics and machine learning techniques. The output was a prototype of an intelligent decision support tool that can make advanced multiobjective optimization methods available for industry, thereby significantly enhancing the innovation capability and competitiveness of the Finnish industries. The principal investigator for DeCoMo was Professor Kaisa Miettinen and Doctors Dr. Tinkle Chugh, Karthik Sindhya and Jussi Hakanen worked in it.

Abstract

Innovation, short product design cycles and resource efficiency of processes have become increasingly important for industries due to globalization and circular economy paradigm. Multiobjective optimization can be used as a powerful tool for product innovation and improving processes holistically by finding better designs and balancing between conflicting objectives efficiently and effectively. However, multiobjective optimization problems in industries are often complex and computationally expensive involving a large number of objectives, decision variables and constraints. In addition, supporting human decision makers and involving them in optimization have rarely been considered in complex multiobjective optimization problems. In this project, we develop novel optimization methods for decision support in solving complex multiobjective optimization problems by combining modern evolutionary algorithms, machine learning techniques and multiple criteria decision making methods. In this we incorporate preference information of a human decision maker. We focus on developing surrogate-assisted optimization techniques to handle computationally expensive problems having several objectives and constraints that are commonly seen in industry. The performance of the methods developed is verified with industry problems. The output of this project will be a prototype of an intelligent decision support tool that can make advanced multiobjective optimization methods available for industry, thereby significantly enhancing the innovation capability and competitiveness of the Finnish industries and wider in society.

Abstrakti

Innovoinnin, lyhyiden tuotannonsuunnitteluaikojen ja prosessien tehokkuuden merkitys teollisuudelle on kasvanut globalisaation ja resurssitehokkuuden korostuneen tarpeen myötä. Monitavoiteoptimointi tarjoaa käyttökelpoisia työkaluja uusien tuotteiden innovointiin ja prosessien tehostamiseen löytämällä parempia ratkaisuja ristiriitaisten tavoitteiden väliltä. Teollisuuden monitavoiteoptimointitehtävät ovat kuitenkin usein kompleksisia ja laskennallisesti raskaita ja niissä on paljon tavoitteita, muuttujia ja rajoitteita. Lisäksi päätöksentekijän tukemista ja osallistamista parhaan ratkaisun löytämisessä on harvoin huomioitu kompleksisia monitavoiteoptimointiongelmia ratkottaessa. Tässä hankkeessa kehitetään uusia menetelmiä ja ne teollisuuden käyttöön tuova päätöksenteon tukiohjelmiston prototyyppi kompleksisten monitavoiteoptimointitehtävien ratkaisemiseen yhdistämällä moderneja evoluutioalgoritmeja,, koneoppimisen ja monitavoitteisen päätöksenteon tuen menetelmiä. Tässä käytetään päätöksentekijän preferenssitietoa uudella tavalla. Hankkeessa kehitetään sijaismallipohjaisia optimointimenetelmiä laskennallisen vaativuuden vähentämiseen, jotta voidaan tehokkaasti käsitellä teollisuudessa usein ilmeneviä optimointiongelmia. Menetelmien toimintaa testataan teollisuusongelmilla. Kehitettävä prototyyppi tuo hankkeessa kehitetyt, edistykselliset monitavoiteoptimoinnin menetelmät teollisuuden saataville. Siten hanke parantaa merkittävästi suomalaisen teollisuuden ja laajemminkin yhteiskunnan innovointi- ja kilpailukykyä.


Final report available as pdf



Main research areas

  • Surrogate assisted multi/many-objective optimization for computationally expensive problems
  • Data-driven evolutionary optimization
  • Interactive decision support for complex problems