sigevolution spring 2006

Download SIGEVOlution Spring 2006

Post on 21-Jan-2015

3.357 views

Category:

Technology

1 download

Embed Size (px)

DESCRIPTION

Summer 2006 issue of the newsletter of the Special Interest Group on Genetic and Evolutionary Computation. The high resolution version available at www.sigevolution.org

TRANSCRIPT

  • 1. SIGEVOlutionApril 2006Volume 1 Issue 1 newsletter of the ACM Special Interest Group on Genetic and Evolutionary Computationin this issueEC@AirliquideCharles Neely HarperNew Challenges forEC Music and Art Jon McCormackOpen BeagleChristian Gagn & Marc Parizeau The Columns letters dissertation cornerforthcoming papersnew books calls & calendar

2. EDITORIALSIGEVOlutionApril 2006, Volume 1, Issue 1EditorialNewsletter of the ACM Special Interest Groupon Genetic and Evolutionary Computation.SIGEVO Ofcerselcome to the rst issue of SIGEVOlution, the newsletter of the ACM Special Inter- WErik Goodman, Chairest Group on Genetic and Evolutionary Computation (SIGEVO). One year has almost John Koza, Vice Chairpassed since this newsletter was announced during the last GECCO in Washington Erick Cantu-Paz, SecretaryD.C. and now, while many of us have already booked an airplane ticket to Seattle, Wolfgang Banzhaf, Treasurer the rst issue is ready, at last! SIGEVOlution is rst of all an opportunity. It has been conceived as a way to facilitate the sharing of information relevant to the EC community. In particular, theSIGEVOlution Board information that would not t in mainstream EC scientic journals, such as, Evolutionary Compu- tation, Genetic Programming and Evolvable Machines (GPEM), or the Transactions on EvolutionaryPier Luca Lanzi (EIC) Computation. Ideally, each issue of SIGEVOlution should include a couple of short, general interest Lawrence quot;Davidquot; Davis articles from EC related elds; a review or a tutorial of available EC software; several columns aboutMartin Pelikan various topics, such as, surveys of existing EC research groups and labs, reports of EC conferences or workshops, letters, recently discussed theses, forthcoming papers, books, and events.Contributors to this Issue This rst, inaugural issue gives a birds eye view of the many opportunities that SIGEVOlution can Charles Neely Harper offer. In the rst paper, Charles Neely Harper reports on two applications of evolutionary compu- Jon McCormack tation developed at American Air Liquide. In his position paper, Jon McCormack discusses the new Christian Gagn challenges in the eld of evolutionary art. He also provided the picture for the cover of this issue. Marc Parizeau In the software corner, Christian Gagn and Marc Parizeau introduce us to Open BEAGLE, a C++ EC framework which supports major evolutionary algorithms, including Genetic Algorithms and tree- based Genetic Programming. The subsequent columns provide various information about a newlyContents born EC lab, a recently discussed PhD thesis, the forthcoming issues of EC journals, new books, and EC@American Air Liquide 2 the calendar of EC events. by Charles Neely Harper SIGEVOlution has been possible thanks to the help of many people who supported this project New Challenges for EC Music and Art 5 in several ways. The members of the SIGEVO board, who entrusted me to be the editor of this by Jon McCormack wonderful project. Gianluca Pignalberi who gave me the rst set of LTEX classes from which this A Open BEAGLE 12 newsletter has been created and was patient enough to help me at various points. The members of by Christian Gagn & Marc Parizeau SIGEVOlution board, Dave Davis and Martin Pelikan. Last but not least, the authors, Charles Neely Letters 16 Harper, Jon McCormack, Christian Gagn, and Marc Parizeau, who provided the contents for this Dissertation Corner 17 issue without having a clue how their contributions would appear. Forthcoming Papers18 New Books 19Pier Luca Calls and Calendar20April 30th, 2006 About the Newsletter22 ISSN: 1931-8499 SIGEVOlution April 2006, Volume 1, Issue 1 3. Evolutionary Computation at American Air Liquide Charles Neely Harper Director, National Supply & Pipeline Operations Air Liquide Large Industries U.S. LPir Liquide is the world leader in industrial and medical gases supply chain. We hired BiosGroup, a complexity science company basedAand related services. The Group offers innovative solutionsin Santa Fe, NM, to help us assess the potential for cost reduction. A re-based on constantly enhanced technologies. These solu- sult of that engagement was the decision to pursue two separate streamstions, which are consistent with Air Liquides commitmentof optimization: one related to reducing the cost of producing and dis- to sustainable development, help to protect life and enable our cus-tributing liquid oxygen, liquid nitrogen and liquid argon, and one related tomers to manufacture many indispensable everyday products. Founded to reducing the cost of producing, compressing and distributing gases in in 1902, Air Liquide has nearly 36,000 employees and is present in more our pipelines. Even though they are related, these are two very different than 70 countries. Sales in 2005 totalled 10,435 million euros. ways of delivering products, one by truck and one by pipeline. American Air Liquide Holdings, Inc. oversees the North American opera- tions of Air Liquide. Through its subsidiary businesses, American Air Liq-Initial optimization systems uide offers industrial gases and related services to a variety of customers including those in rening, natural gas, chemistry, metals, automotive, In late 2001 BiosGroup developed a Proof of Concept system for a small chemicals, food, pharmaceutical, electronics, specialty and healthcarearea of our business that optimized the distribution of oxygen and nitro- markets.gen in liquid form by truck from our more than 40 production plants to more than 8,000 customer sites. This system used an ant colony opti- Our products are primarily oxygen, nitrogen, and hydrogen along with mizer to determine truck routes and sourcing from our plants. The per- the services and technology involved in delivering these gases. We sepa- formance of this system was very impressive, and we realized that there rate atmospheric air into oxygen, nitrogen and argon through a cryogenic was a good deal of benet to be gained from extending the system to distillation process and we produce hydrogen by cracking natural gas. schedule the production of our liquid products. We distribute our products through several methods: in gaseous form through nearly 2,000 miles of pipelines or in compressed cylinders and In 2002 BiosGroup created a Proof of Concept system for our pipeline in liquid form, by truck transportation from our plants to our customers operations. This system used a genetic algorithm to decide how to con- tanks and facilities. More than half the cost of creating and distributing trol the pipeline, and it used a mixed integer programming approach to oxygen, nitrogen and hydrogen lies in the cost of energy, as natural gas optimize the operations of the plants. While the system did not per- or electricity. Operating air separation and hydrogen plants, cogenera- form detailed simulation of the costs and performance of our equipment, tion units and our pipeline is an energy-intensive business. its results suggested strongly that there were signicant savings to be In 1999 we began to investigate ways to substantially reduce our pro- gained if the pipeline optimization system were to be developed into a duction and distribution costs and to nd smart ways to manage ourfull-edged simulator and optimizer. 2 SIGEVOlution April 2006, Volume 1, Issue 1 4. EDITORIALThe liquid gas system the combination of the production schedule and the distribution sched- ules in order to nd out how well they work together. Each optimizer is BiosGroups consulting operations were acquired by NuTech Solutions,then given the feedback from their joint result. In this way, the ant colony Inc. in 2001 and the subsequent development of these systems was car- optimizer and the genetic algorithm adapt in conjunction with each other ried out by NuTech. Some of the BiosGroup project members continued to generate integrated schedules, even though neither system is explic- with the projects after the transition. From this point onward, the contin- itly aware of the operations of the other. ued development of both projects was performed by NuTech Solutions. A signicant insight derived from this system was the observation that, In the next major phase of the liquid supply chain production and distri- while the ant system operating alone took many thousands of iterations bution project, we wanted to nd the best solution to plan both produc- and several hours to come to a solution, it could run three or four iter- tion and distribution of our products. It did not seem to us that a goodations per solution produced by the genetic algorithm, so that the time off-the-shelf solution existed that could solve the problem of coordinating required to run the two systems linked as we have described was under production and distribution. The major supply chain software systems op-our six-hour limit. timized rst production and then distribution and the results seemed to Today we use the liquid gas system to help us schedule the production us to be substantially suboptimal. In fact, we acquired one industrial gas and distribution of our liquid products. The cost savings and operational company that had created an award-winning production and distribution efciencies are substantial. We are saving more than 1.5 million dollars optimization system based on a large commercial supply chain product, per quarter at one of our plants by utilizing optimization techniques in a and its performance seemed to be well below what could be achieved. demanding and changing environment. It was clear to us that the problem of coordinating production and dis- We are currently extending the liquid production and optimization system tribution was not one that could be adequately solved by mathematical in multiple ways, and we expect its benets to increase as these exten- techniques such as linear programming, because our plant production sions are completed. We believe that the combination o