4 présentations

• 
  10h30 - 10h55

  A computational study on imputation methods for environmental applications

  • Paul Dixneuf, prés.,
  • Fausto Errico, École de technologie supérieure
  • Mathias Glaus, École de technologie supérieure

  Recently, a considerable amount of research has been devoted to the issue of missing data. In this work we perform a computational comparison among three imputation methods based on a random forest, k-nearest neighbor and multivariate imputation by chained equations. In our study we consider nine data sets differing in terms of dimension, nature of variables, and data distribution structure. Preliminary results suggest that the Random forest based method outperforms the others. Our purpose is to give an insight into which imputation method to use, depending on the data types encountered in environmental issues.

• 
  10h55 - 11h20

  Detecting and predicting the traffic condition by applying image processing methods on CCTV’s all around the town

  • Mohsen Amoei, prés., Student

  By utilizing two fields of computer science, computer vision, and data mining we come up with a way to not only reduce the traffic congestion but also to predict it in future. By using CCTV's all around the town and applying image processing methods to evaluate the traffic status of each road.

• 
  11h20 - 11h45

  Corrosion failure prediction models for gas transmission pipelines

  • Kimiya Zakikhani, prés., Concordia University
  • Fuzhan Nasiri, Concordia University
  • Tarek Zayed, The Hong Kong Polytechnic University

  Corrosion is considered as the most frequent failure in gas transmission pipelines. A literature review reveals that some contributing parameters are ignored in the developed corrosion prediction models. This research aims to develop a corrosion failure estimation model for gas pipelines by exploiting historical data using statistical methods.

• 
  11h45 - 12h10

  Assessing Québec's hydropower value in a decarbonized future: how public data are sufficient?

  • Sébastien Debia, prés., GERAD

  Québec's hydropower is a key asset in the development of a decarbonized future in Northeastern America. However hydropower is not a perfectly flexible resource, and one must take hydro management constraints into account if he does not want to overestimate its value. Taking these constraints into account for a long-term, multi-area problem is however a difficult task: hydro modelling is dynamic--hence suffering from the curse of dimensions--, hydropower in Québec has typically supply-chain characteristics--several plants with different flexibility are placed at different point of the same river--, and only few technical data are public to calibrate such a model.
  The methodology to model and calibrate a linear model allowing for a plant-based representation of 71% of Hydro-Québec Production capacity is presented, namely the La Grande, Manicouagan and Aux Outardes valleys. Simplifying the model to an open-loop dynamic problem permits us to represent each valley with the main reservoirs and their underlying supply-chain-management.