4 présentations

• 
  10h30 - 10h55

  DEA-based composite indexes: two case studies from Africa

  • Matthias P. Takouda, prés., Laurentian University
  • Mohamed Dia, Laurentian University
  • Alassane Ouattara, CESAG, Dakar, Senegal

  Data Envelopment Analysis (DEA), a mathematical programming approach to efficiency measurement based on linear programming, can be used to calculate composite indexes in various areas to assess complex multi-dimensional concepts. We present two case studies where DEA and its post-hoc models are utilized to compute composite indexes (CI). In the first one, the indexes measure the level of financial inclusion of economies from the West African Economic and Monetary Union (WAEMU). In the second case, the composite scores assess how well the policies designed to facilitate venture initiation in sub-Saharan countries perform. In both cases, the obtained CI are compared to CI based on Principal Component Analysis (PCA), and then used in further analyses.
• 
  10h55 - 11h20

  The Value of Text for Waiting Time Prediction: A Repair Shop Case Study

  • Mohammad Mosaffa, prés., University of British Columbia
  • Amir Ardestani-Jaafari, University of British Columbia
  • Javad Tavakoli, UBC Okanagan

  One of the most formidable tasks for service companies is to inform customers about the approximate waiting time. Overestimated waiting time may force customers not to proceed with a company, and underestimated waiting time may lead to an impossible responsibility for the company to accomplish. This research develops a two-phase predictive model to estimate waiting time in service systems, integrating textual and structured data. We initially utilize the Bag of Words technique to convert text into numerical form. Subsequently, our model exploits a Multi-Layer Perceptron deep learning architecture to predict waiting time, as tested on nearly 31,000 cases. Our findings reveal the superior performance of deep learning, with 87.6% accuracy, compared to traditional machine learning methods in forecasting continuous output based on textual data. Moreover, using combined structured and textual data enhanced accuracy by 5.3% over using only structured data. This model significantly assists in underestimated waiting time scenarios, improving accuracy by 50.1%. Additionally, our work underscores the importance of textual data in identifying deficiencies in service systems from a managerial viewpoint.
• 
  11h20 - 11h45

  From the Lowe’s acquisition of RONA (2016) to the “new” RONA (2022): A look at the Chains of Hardware Retail stores in Canada

  • Matthias Takouda, prés., Laurentian University
  • Mohamed Dia, Laurentian University
  • Mohamed Abdulkader, Laurentian University

  We explore the relative efficiencies of the 3 major public chains of hardware retail stores (Home Depot, Lowe’s and RONA) operating in Canada during the period 2000-2021. Data Envelopment Analysis (DEA) models are utilized to calculate these relative efficiencies. The financial performance (sales and profit) is evaluated against the human resources (number of employees), the marketing resources (number of stores), and the financial resources (capital assets and cost). We perform a benchmark analysis of the companies in our study. Moreover, we investigate the potential impact of another input which is the total sales area on these relative efficiencies. We assess also how the firms were affected by the financial situations, the dynamics of the industry, and the COVID-19 pandemic. We focus in particular on the impacts of the acquisition of RONA by Lowe’s in 2016. The observations indicated that the scale of operations was a significant factor in changing the most efficient company in the selected sample. Lowe’s was not able to gain the lead as the most efficient company in the industry after they acquired Rona.
• 
  11h45 - 12h10

  A readily implementable data-driven method for temperature management of commercial buildings

  • Marie-Christine Paré, prés., Department of Electrical Engineering, Polytechnique Montréal, GERAD, Mila, Qc, Canada
  • Antoine Lesage-Landry, Polytechnique Montréal
  • Vasken Dermardiros, BrainBox AI, Montréal, Qc, Canada

  Model predictive control (MPC) has been shown to significantly improve the energy efficiency of buildings while maintaining comfort. Recently, data-driven approaches based on neural networks have been proposed to facilitate system modelling. However, such approaches are generally nonconvex and result in computationally intractable optimization problems. In this work, we design a readily implementable optimal temperature management method for commercial buildings. We propose a data-driven and mixed-integer convex MPC which can be solved to optimality by off-the-shelves numerical solvers. We consider rooftop unit heating, ventilation, and air conditioning systems with discrete controls to accurately model the operation of most commercial buildings. Our approach uses an input convex recurrent neural network (ICRNN) to model the temperature dynamics using historical data. The ICRNN accurately fits the system behaviour and allows the optimization problem to be solved to optimality efficiently. We model real operating conditions such as temperature setpoints and equipment toggling, thus enhancing the potential for practical implementation. Performances of the controller are evaluated on a state-of-the-art building simulation. Finally, we leverage our approach in a demand response setting and discuss preliminary results.

  Keywords: Building Energy Management, Model Predictive Control, Input Convex Recurrent Neural Network, Rooftop Unit, Optimal Control, Mixed-Integer Programming.