2016 INFORMS Annual Meeting Program

POSTER SESSION

INFORMS Nashville – 2016

Non-convex Problems For Long-term Hydropower Scheduling Martin N. Hjelmeland, PhD Candidate, Norwegian University of Science and Technology, O.S. Bragstads plass 2E, Trondheim, 7491, Norway, martin.hjelmeland@ntnu.no With its large operational flexibility, hydropower may provide spinning reserves at a low cost. With the introduction of a market for providing capacity in the scheduling, the unit commitment problem in the stochastic multistage hydropower scheduling problem becomes essential. This is mainly due to the minimum generation restriction for a hydropower station that should be enforced to ensure a practicable dispatch. This work will focus on methods to solve these kinds of problems, and evaluate its importance for the hydropower scheduling problem. Methods that will be applied are Stochastic Dynamic Programming (SDP) and Stochastic Dual Dynamic Programming (SDDP) with possible extensions. A Two-echelon Decomposition Method On Fresh Product Distribution Problem hongtao HU, Associate Professor, Shanghai Maritime University, Room101, No 96, 555 Guzong Road, Shanghai, 201306, China, hu.hongtao@foxmail.com Refrigerator cars are widely used for fresh product distribution. The energy consumption of these vehicles is sensitive to the environment temperature. To reduce operation costs of third-party transportation providers (TPTP), the refrigerator car scheduling problem is addressed in this research. A time- dependent mixed-integer programming model is established to reduce total operation costs. An adaptive heuristic method is proposed by combining the variable neighborhood search and particle swarm optimization.Numerical experiments are conducted to demonstrate the effectiveness of the proposed time- dependent decision model. Modeling Capacity Planning Projects In The Automotive Industry Using A Markov Decision Process Paul Jana, Technische Universität München (TUM), Arcisstr 21, Munich, 80333, Germany, paul.jana@tum.de, Martin Grunow To support automotive OEMs in time-phased decision making during capacity planning projects connected to new vehicle introductions, we present a dynamic programming approach based on a Markov Decision Process. We employ Bayesian updating to anticipate forecast updates and adapt standard risk measures. Our methodology is superior over alternative stochastic approaches. Assessment Of Clustering Algorithms Based On A Data Mining Technique Youngseon Jeong, Assistant Professor, Chonnam National University, 77 Yongbong-ro, Buk-gu, Gwangju, 61186, Korea, Republic of, youngseonjeong@gmail.com This research presents a novel performance assessment of clustering compactness based on one-class classification algorithms. The proposed method evaluates the compactness of each cluster by using support vector data description (SVDD) and Bayesian support vector data description (BSVDD), which is robust to arbitrary shapes of a clustering. In addition, the proposed approach can accurately evaluate a clustering compactness in kernel space. The experimental results show that the proposed method can evaluate the accurate compactness for arbitrary shapes of a clustering. The Status, Development Potential Prediction And Policy Recommendations Of New Energy In China HongDian Jiang, Master Degree Candidate, China University of Petroleum, No.18, Fuxue Road, Changping District,, Beijing, 102249, China, cupjhd@163.com, KangYin Dong, RenJin Sun By combining the GM (1, 1) grey model with BP neural network model and establishing a combined Grey-BP modelling tool, future the development potential of China’s new energy is forecasted. In addition, the ranking of the development potential for the different new energy fuel types is performed, from both the development scale and growth rate perspective. According to our estimation, China’s total new energy consumption will increase to 690.5 Mtoe in 2020, accounting for 19.7% of the domestic energy need. Besides, according to the rank results, nuclear and solar energy will be considered as future oriented composition of the new energy, as well as hydropower considered as the key element in China. Predictive Maintenance From Analysis Of Airplane Sensor Data Ruiwei Jiang, Data Scientist, Boeing Vancouver, 1146 Homer Street, Vancouver, BC, V6B 2X6, Canada, ruiwei.jiang@aeroinfo.com, Phillip Mah, Dawen Nozdryn-Plotnicki, Benji Shieh, Hubert Duan Unscheduled maintenance drives 10% of the annual operational cost to airlines worldwide. Predictive Maintenance could reduce those costs, particularly when synchronized with airline’s operations. By using engineering expertise, statistics and machine learning on aircraft sensor and fault data, as well as analysis of an airline’s flight and maintenance schedule, we detect impending issues on the aircraft and suggest maintenance tasks in accordance with the prediction and an airline’s working rhythm. These predictive maintenance tasks will increase reliability and reduce unscheduled maintenance.

Eliminating Preventable Motor Vehicle Accidents Through Simulation Scenarios Of Vehicle Modifications

Sahar Khamsehi, Prospective Phd Student/Practical trainer, Binghamton University, 4400 Vestal Pkwy E,, Binghamton, NY, 13902, United States, skhamse1@binghamton.edu, Constance Dwyer Facilitator for express work out initiative to determine ways to improve driving performance in enterprise divisions. Developed SharePoint team site. Supported effort with Jack 7.1 simulation tool model to evaluate statistical significance of vehicle improvements with extended graduate student team resulted in optimized scenarios where the certain vehicle modification resulted in an improved visibility measured by simulation software. Consequently, the study led to proved solutions in order to eliminate preventable accidents. Building A Luxury Products Evaluation Model Youn Sung Kim, Professor, Inha University, 253 Yonghyun-Dong, Nam-Gu, Incheon, 402-751, Korea, Republic of, keziah@inha.ac.kr Considering the market size and history of luxury product, it is very surprising that there is no holistic model to evaluate luxury product market. In this study we try to propose comprehensive perspectives and policy considerations on existing research about the evaluation of luxury products in terms of both product characteristics and value. It has been focused on the brand value. Even though it needs the verification process of proposed evaluation model, this research indicates that it considers the new viewpoint of the evaluation of luxury products. Therefore, the following study will be substantiated based on case study and focused expert group interview. Strengthening Keystroke Dynamics Based User Authentication Based On User Adaptive Feature Construction For One Class Classification Junhong Kim, Korea University, Seoul, Korea, Republic of, junhongkim@korea.ac.kr, Haedong Kim, Boseop Kim This study present a new KDA method based on one-class classification(OCC) with user-adaptive feature construction scheme. Since users have their own typing patterns, the average typing speeds of digraphs are also different between users. Hence, we construct eight features by considering the rank of the typing speed of digraph for four typing speed measures for each user. A total of five OCC is then trained for a valid user is applied to classify a new keystroke data. We collected more than 10,000 keystrokes from 150 participant. Based on the experiment with 25 combinations of training and test keystroke size, the proposed model yielded lower EER than the conventional feature construction method. Evaluating Information Quality For News Articles Based On Topic Modeling Hyungseok Kim, Korea University, Anam-dong Seongbuk-gu, Korea University, School of Industrial Management Engineering, Seoul, 163-713, Korea, Republic of, hskim0263@korea.ac.kr, Kim Boseop We propose two topic model-based information quality evaluation measures for news articles: Relevance and Uniqueness. Relevance of an article is assessed by the weighted sum of the average per-topic posterior probabilities of user-provided keywords and the per-document topic distributions. Uniqueness of an article is assessed by the score of a novelty detection algorithm based on per-document topic distributions. The proposed model is applied to Korean economic news articles and qualitatively verified by domain experts. Why Firms Disappear: Bankruptcy In The Thoroughbred Horse Industry’s Social Network Angela King, Chapman University CMB: 1406, 1 University Drive, Orange, CA, 92866, United States, amxk96@gmail.com Darcy Fudge Kamal, Cristina Nistor I look at how social influence in the Thoroughbred Horse Industry network can influence the value for goods at auction. Firms going through a bankruptcy are forced to sell off their goods while the network of social connections is affected by the news of their impending bankruptcy. I analyze whether the network takes into account that these firms will not exist in the future. I use clustering analysis to find network patterns in which account for the differences in network structures related to the node deletion from 2010-2014. Nondestructive Quality Inspection Using Piezoelectric Transducers Affixed To A Fixture Tomilayo Komolafe, PhD Candidate, Virginia Polytechnic Institute and State University, 1145 Perry Street, Blacksburg, VA, 24061, United States, tomilayo@vt.edu Product change detection is of utmost importance in any manufacturing environment and it is one of the major goals in quality control practices. Some changes could be due to a malicious cyber-physical attack which is inherently very difficult to detect by traditional inspection schemes. This study proposes to use piezoelectric transducers (PZT) affixed to a fixture to perform nondestructive quality inspection. Mechanical impedance information is obtained through exciting the PZT bonded to the fixture-part combination and signal processing is used to identify presence of an alteration.

178