4 présentations

• 
  10h30 - 10h55

  Mixed-Integer Convex Optimization to Infer Electric Vehicle Charging Patterns for Impact Studies

  • Feng Li, prés., Polytechnique Montreal
  • Elodie Campeau, Polytechnique Montreal
  • Ilhan Kocar, Polytechnique Montreal
  • Antoine Lesage-Landry, Polytechnique Montréal

  As the penetration level of electrical vehicles (EVs) is increasing on the power distribution networks, utility planners need to evaluate the impacts of EV charging to maintain system reliability and power quality. EV chargers are often installed behind-the-meter (BTM); hence the charging events are invisible to utilities, making it challenging to model customers' charging behaviours. In this work, we propose a non-intrusive and training free method to detect EV charging events from the data measured by advanced metering infrastructure (AMI) such as smart meters. By leveraging the contextual information of EV charging, we formulate a mixed-integer convex quadratic program (MICQP) to detect EV charging events from customers' daily meter data. No labelled training data or hyperparameter tuning are required, and the MICQP can be efficiently solved. Based on the detection results, we infer customers' charging patterns in terms of probabilities of charging profiles through a data-driven approach. In the case study, we demonstrate that our proposed approach can achieve similar detection accuracy as that of other learning-based approaches which use high-solution meter data. Finally, impacts of EV charging on the large-scale IEEE-8500 test feeder are presented by using the inferred charging patterns.

• 
  10h55 - 11h20

  Physics-informed neural networks for optimal transmission switching

  • Abdelrahman Ayad, prés., McGill University
  • François Bouffard, McGill University

  Optimal transmission switching (OTS) is an effective control mechanism that optimizes power systems operation by changing the topology of the transmission networks to alter the power flows. The reconfiguration of the lines can relieve transmission congestion and minimize the cost of generation dispatch. The OTS is modelled as a mixed-integer programming (MIP) problem, which makes it a NP-hard, results in high computation time, and potentially renders the problem intractable for real-world large systems. This work presents a physics-informed neural network (PINNs) approach that improves the OTS solving time and improves its accuracy. PINNs embeds the underlying physical laws of the electric grids power flow dynamics and network configuration into the machine learning model which shifts the computational burden to an offline training stage, using relatively low number of training data. The performance of the proposed PINNs model is evaluated against several power system test cases to test its performance in both solution optimality and computational efficiency.

• 
  11h20 - 11h45

  Nature-Inspired Techniques for Reverse Energy Auctions

  • Sifat E Jahan, prés., University of Regina
  • Malek Mouhoub, University of Regina
  • Mehdi Sadeghilalimi, University of Regina

  With the growing electricity demand, the likelihood of experiencing power outages is also rising. Utility companies have started buying electricity through e-auctions to address this issue. To meet the increasing electricity demand, we propose a solution to procure energy from various sources, trading off multiple objectives while solving a complex winner(s) determination problem for resource procurement optimization. Winner determination is an NP-hard problem, and applying exact methods is impractical. Instead, we rely on nature-inspired techniques as they are appropriate for trading the quality of the returned solution for the required running time. In particular, Genetic Algorithms (GAs), Whale Optimization Algorithms (WOAs), Ant Colony Optimization (ACOs), Particle Swarm Optimizations (PSOs) and Firefly Algorithms (FAs) are explored and evaluated in terms of effectiveness in producing high-quality solutions for several instances of the Combinatorial Reverse Auction (CRA) problem.

• 
  11h45 - 12h10

  A Heuristic Algorithm to solve the One-Warehouse Multi-Retailer Problem with an Emission Constraint

  • Matthieu Gruson, prés., UQÀM - CIRRELT
  • Qihua Zhong, HEC Montréal
  • Ola Jabali, Politecnico di Milano
  • Jans Raf, HEC Montréal

  We consider the one-warehouse multi-retailer problem with a global carbon emission cap constraint (OWMR-EC). This constraint aims at limiting the carbon emissions related to the production, setup and inventory holding operations. We develop a penalized relaxation method to heuristically solve the problem, with and without the possibility of having initial inventory. This heuristic uses in itself another heuristic that we propose to solve the standard one-warehouse multi-retailer problem (OWMR). Our penalized relaxation method is tested on instances adapted from the literature. Our results indicate that the penalized method finds between 87.4 and 89.8% of feasible solutions, with an average optimality gap of 2.1 and 2.2%. Our method is highly effective in terms of run-time and solution quality, when a feasible solution is found. We further perform a sensitivity analysis on the optimal solutions of the OWMR-EC to better understand the implications of the carbon emission cap constraint. The analysis indicates that the marginal cost of reducing carbon emissions increases as the emission cap decreases. The analysis also shows that the correlation between the cost and emission parameters has an important impact on the potential to further lower the emissions.