Seminars run by PhD students and postdocs with the Control and Power Group

Every week, an invited speaker gives a talk on a relevant topic (preferably in control and power) consisting of a 20-minute presentation followed by a 10-minute discussion. The content of talks can be

  • conference contributions,
  • initial research proposals,
  • preparation for Early Stage Assessment/Late Stage Review, or
  • tutorials on topics of the speaker's expertise.

After the talk, the participants can enjoy coffee and snacks (funded by the group) together in the Faraday Cage on EEE Level 6 and continue the discussion.

The CAP Research Roulette is open to everyone in the Control and Power Group. MSc students from Control Systems and Future Power Network Courses, who are willing to continue pursuing an academic career, are also strongly encouraged to participate.

Contacts: Kaiwen Chen (Control talks) [kaiwen.chen16@imperial.ac.uk], Olayinka Ayo (Power talks) [o.ayo17@imperial.ac.uk]

Talk of the week

TitleA hybrid SDDP Machine Learning approach to represent non-convexities in the hydrothermal operation problem

Speaker: Joaquim Garcia

Venue: EEE 1109B

Date and Time: Friday 26/07/2019 15:00-16:00

AbstractThe Stochastic Dual Dynamic Programming (SDDP) algorithm allows to uncouple in time the multistage stochastic operation of hydrothermal systems. In this framework, any stage’s future cost of the operation, conditioned the decision of the current stage, can be approximated by means of hyperplane constructed iteratively in the backward phase of the algorithm. If the future cost (cost to-go) function is convex in relation to the state variables, the approximation determined by this hyperplane is guaranteed a lower bound for the real future cost function and the algorithm it is guaranteed to converge to the optimal solution, under mild conditions. However, if the future cost function is not convex, the optimality of the algorithm cannot be guaranteed.

The treatment of non-convexities of the future cost function in the SDDP algorithm is a recurrent theme in the literature. In this work, a new alternative based on machine learning is proposed: we apply support vector machine (SVM) as a classification approach to recognize and delimit the validity regions of the hyperplanes generated by the SDDP algorithm. This allows the algorithm for stochastic dynamic operation to evaluate solutions that were not feasible, considering only the convex envelope constructed by the original cuts.

The SVM method is well suited for the classification of cuts employed, because of its robustness to solve large problems, which is something desired in the scope of this proposal, given the real dimensions of the problem of operation of hydrothermal systems. Moreover, the ability of the method to do classification even when classes are not linearly separable and the fact that the classification function is differentiable are key requirement for the proposed algorithm. Finally, the model selection step (optimization of SVM model hyperparameters) is fully parallelizable, just like many steps in the SDDP algorithm.

The classification of cuts employed and the proposed algorithm also do not guarantee the optimality of the solution, since they do not necessarily construct lower bounds for the nonconvex future cost functions. However, the results of the application of the method demonstrate that this approach is able to better approximate of these functions, making possible to obtain better solutions to the hydrothermal system operation problem.

Past talks

2018-2019

  • 14/06/2019 Zhongda Chu - "Towards Optimal System Scheduling with Synthetic Inertia Provision from Wind Turbines"
  • 24/04/2019 Magnus Jamieson - "Quantification and visualisation of the impact of extreme wind events on power systems"
  • 10/04/2019 Lingxin Lan - "Wireless Charging for Autonomous Drones"
  • 03/04/2019 Thiago Mendonca - "Distributed Active Network Management for Cost-Effective Integration of Energy Resources"
  • 27/03/2019 Jochen Cremer - "Full-body scanner of the electrical power system: A Probabilistic Perspective on Dynamic Security Assessment"
  • 06/03/2019 Xi Zhang - "Objective-Based Approach for Representative Day Selection for Power System Investment Planning"
  • 12/12/2018 Caspar Collins - "Comparative Analysis of an MV Neutral Point Clamped AC-CHB Converter with DC Fault Ride-through Capability"
  • 05/12/2018 Kaiwen Chen - "I&I adaptive control for systems with varying parameters"
  • 28/11/2018 Xiaohong Dong - "Planning of EV fast-charging stations on a round freeway"
  • 21/11/2018 Yue Yu - "Global transportation electrification and relative charging infrastructure"
  • 07/11/2018 Anastasis Georgiou - "Robust fault tolerant Model Predictive Control and for high order system"

2017-2018

  • 12 Sep 2018 Willem Leterme - "Modeling for AC/DC Converters with Controlled DC Fault Blocking Capability"
  • 25 Jul 2018 Joan Marc Rodriguez - "Implementation of Model Predictive Control Techniques in Power Electronic Voltage Source Converters"
  • 18 Jul 2018 Evelyn Heylen - "Reliability-based electricity consumption: Individualized reliability levels in probabilistic short-term reliability management of power systems"
  • 11 Jul 2018 Dawei Qiu - "Investigating the impact of demand flexibility on electricity retailers"
  • 20 Jun 2018 Kaiwen Chen - "Adaptive Control of Linear Systems with Time-Varying Parameters"
  • 13 Jun 2018 Jochen Cremer - "Game over, no more Jenga! Machine Learning based Security Assessment & Secure Operation of the Power System"
  • 06 Jun 2018 Tiago Rodrigues - "Computationally Efficient Risk-averse Energy Storage Investment for Renewables Coordination"
  • 31 May 2018 Xi Zhang - “Economic Assessment of Alternative Heat Decarbonisation Strategies through Coordinated Operation with Electricity System”
  • 21 May 2018 Georgios Anagnostou - "Identifying Unknown Dynamics in Power Systems"
  • 15 May 2018 Michael Evans - "Capability assessment and optimal dispatch of distributed stores"
  • 09 May 2018 Anastasios Oulis Rousis - "Optimal Design and Operation of a Hybrid AC/DC Microgrid"
  • 02 May 2018 Efstratios Batzelis - "Mathematical models of explicit form for photovoltaic systems"
  • 25 Apr 2018 James Wylie - "Modular Multi-level Converter with Increased Number of Effective Levels by using Asymmetric Sub-module Voltages for Medium Voltage Applications"
  • 18 Apr 2018 Arghavan Nazemi - "Energy Management of Parallel Hybrid Electric Vehicles"
  • 04 Apr 2018 Firdous Ul Nazir – "Scenario-based chance constrained Volt/VAR control in distribution systems"
  • 21 Mar 2018 Luis Badesa Bernardo – "Optimal Provision of Frequency Services in Decarbonised Grids"
  • 28 Feb 2018 Nicolas Faedo - "Optimal control of wave energy converters: fundamental principles and a moment-based approach"
  • 21 Feb 2018 David Burth Kurka - "Git and Github for research management"
  • 14 Feb 2018 Julio Perez Olvera - "Using Multi-terminal DC Networks to Improve the Hosting Capacity of Distribution Networks"
  • 31 Jan 2018 Hussain Akhtar - "Optimal Energy Management of Microgrids"
  • 17 Jan 2018 Francesca Boem - "Distributed Fault Detection for Interconnected Large-Scale Systems: a Scalable Plug & Play Approach"
  • 10 Jan 2018 Temitayo Oderinwale - "Strategic Investment Planning incorporating demand flexibility"
  • 13 Dec 2017 Nicolas Cifuentes - "Network Allocation of Battery Energy Storage Systems (BESS) considering stability criteria"
  • 6 Dec 2017 Alexander Gallo - "A distributed attack detection method for multi-agent systems governed by consensus-based control"
  • 29 Nov 2017 Omar Faqir - "Joint optimization of transmission and propulsion in aerial communication networks"
  • 22 Nov 2017 Thiago Mendonca - "Distributed Coordination of Energy Resources for Improving Network Hosting Capacity"
  • 15 Nov 2017 Alexandre Moreira - "An Adjustable Robust Optimization Approach for Contingency-Constrained Transmission Expansion Planning"