Louis Wehenkel - Profile on Academia.edu (original) (raw)

Papers by Louis Wehenkel

We propose to post-process the results of a scenario based stochastic program by projecting its d... more We propose to post-process the results of a scenario based stochastic program by projecting its decisions on a parameterized space of policies. By doing so the risk of overfitting to the set of scenarios used in the stochastic program is reduced. A proper choice of the structure of the space of policies allows one to exploit them in the context of novel scenarios, be it for Monte-Carlo based value estimation or for use in real-life conditions. These ideas are presented in the context of planning the exploitation of electric energy resources or for evaluating the economic value of a portfolio of assets.

The paper introduces a framework for information exchange and coordination of security assessment... more The paper introduces a framework for information exchange and coordination of security assessment suitable for distributed multi-area control in large interconnections operated by a team of transmission system operators. The basic idea of the proposed framework consists of exchanging just enough information so that each operator can evaluate the impact in his control area of contingencies both internal and external to his area. The framework has been thought out with the European perspective in mind where it is presently not possible to set up a transnational security coordinator that would have authority to handle security control over the whole or part of the European interconnection. Nevertheless, it can also be considered as an approach to handle security control in North-American Mega-RTOs, where it could help to circumvent problems of scalability of algorithms and maintainability of data by distributing them over the TSOs under the authority of the RTO.

In this paper we present a new tree-based ensemble method called "Extra-Trees". This algorithm av... more In this paper we present a new tree-based ensemble method called "Extra-Trees". This algorithm averages predictions of trees obtained by partitioning the inputspace with randomly generated splits, leading to significant improvements of precision, and various algorithmic advantages, in particular reduced computational complexity and scalability. We also discuss two generic applications of this algorithm, namely for time-series classification and for the automatic inference of near-optimal sequential decision policies from experimental data.

Springer eBooks, 2003

Reinforcement learning aims to determine an (infinite time horizon) optimal control policy from i... more Reinforcement learning aims to determine an (infinite time horizon) optimal control policy from interaction with a system. It can be solved by approximating the so-called Q-function from a sample of four-tuples (xt, ut, rt, xt+1) where xt denotes the system state at time t, ut the control action taken, rt the instantaneous reward obtained and xt+1 the successor state of the system, and by determining the optimal control from the Q-function. Classical reinforcement learning algorithms use an ad hoc version of stochastic approximation which iterates over the Q-function approximations on a four-tuple by four-tuple basis. In this paper, we reformulate this problem as a sequence of batch mode supervised learning problems which in the limit converges to (an approximation of) the Q-function. Each step of this algorithm uses the full sample of fourtuples gathered from interaction with the system and extends by one step the horizon of the optimality criterion. An advantage of this approach is to allow the use of standard batch mode supervised learning algorithms, instead of the incremental versions used up to now. In addition to a theoretical justification the paper provides empirical tests in the context of the "Car on the Hill" control problem based on the use of ensembles of regression trees. The resulting algorithm is in principle able to handle efficiently large scale reinforcement learning problems.

PLOS ONE, Dec 16, 2013

Disordered regions, i.e., regions of proteins that do not adopt a stable three-dimensional struct... more Disordered regions, i.e., regions of proteins that do not adopt a stable three-dimensional structure, have been shown to play various and critical roles in many biological processes. Predicting and understanding their formation is therefore a key subproblem of protein structure and function inference. A wide range of machine learning approaches have been developed to automatically predict disordered regions of proteins. One key factor of the success of these methods is the way in which protein information is encoded into features. Recently, we have proposed a systematic methodology to study the relevance of various feature encodings in the context of disulfide connectivity pattern prediction. In the present paper, we adapt this methodology to the problem of predicting disordered regions and assess it on proteins from the 10th CASP competition, as well as on a very large subset of proteins extracted from PDB. Our results, obtained with ensembles of extremely randomized trees, highlight a novel feature function encoding the proximity of residues according to their accessibility to the solvent, which is playing the second most important role in the prediction of disordered regions, just after evolutionary information. Furthermore, even though our approach treats each residue independently, our results are very competitive in terms of accuracy with respect to the state-of-the-art. A web-application is available at x3Disorder.

In the context of a deterministic Lipschitz continuous environment over continuous state spaces, ... more In the context of a deterministic Lipschitz continuous environment over continuous state spaces, finite action spaces, and a finite optimization horizon, we propose an algorithm of polynomial complexity which exploits weak prior knowledge about its environment for computing from a given sample of trajectories and for a given initial state a sequence of actions. The proposed Viterbi-like algorithm maximizes a recently proposed lower bound on the return depending on the initial state, and uses to this end prior knowledge about the environment provided in the form of upper bounds on its Lipschitz constants. It thereby avoids, in way depending on the initial state and on the prior knowledge, those regions of the state space where the sample is too sparse to make safe generalizations. Our experiments show that it can lead to more cautious policies than algorithms combining dynamic programming with function approximators. We give also a condition on the sample sparsity ensuring that, for a given initial state, the proposed algorithm produces an optimal sequence of actions in open-loop.

IEEE Transactions on Power Systems, Nov 1, 2007

He is a full Professor of electrical engineering and computer science with the University of Lièg... more He is a full Professor of electrical engineering and computer science with the University of Liège. His research interests lie in the fields of stochastic methods for systems and modeling, machine learning, and data mining, with applications in power systems planning, operation and control, and bioinformatics.

ROC curves on PDB30 dataset

Accuracy evaluation on the CASP10 dataset

<p>Top: the scores obtained when evaluating Casp10 on models learnt on Disorder723 through ... more <p>Top: the scores obtained when evaluating Casp10 on models learnt on Disorder723 through the relevant feature functions found on Disorder723. Bottom: comparison of a number of predictors, which participated in or evaluated their model to the 10th CASP experiment. These results were reported by <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0082252#pone.0082252-Eickholt1" target="_blank">[33]</a>. In parenthesis: the group number of the methods that participated in the CASP10 experiment. The standard deviations were calculated by a bootstrapping procedure in which 80% of the dataset was sampled 1000 times, as it was done by <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0082252#pone.0082252-Eickholt1" target="_blank">[33]</a>.</p

Raphaël Marée2Benelearn is the annual machine learning conference of Belgium and The Netherlands.... more Raphaël Marée2Benelearn is the annual machine learning conference of Belgium and The Netherlands. It serves as a forum for researchers to exchange ideas, present recent work, and foster collaboration in the broad field of Machine Learning and its applications. Benelearn 2008 is organised by the Systems and Modeling and Bioinformatics and Modeling research units of the Department of Electrical Engineering and Computer Science and GIGA-Research of the University of Liège. The conference takes place in the Solcress seminar center, at walking distance from the center of the city of Spa located in the Belgian Ardennes. Conference Chair

IEEE Transactions on Power Systems, 2019

Shie Mannor received the B.Sc. degree in electrical engineering, the B.A. degree in mathematics, ... more Shie Mannor received the B.Sc. degree in electrical engineering, the B.A. degree in mathematics, and the Ph.D. degree in electrical engineering from the Technion

IFAC Proceedings Volumes, 1997

We develop a general hybrid k Nearest Neighbors (kNN) approach, where kNNs take advantage of prob... more We develop a general hybrid k Nearest Neighbors (kNN) approach, where kNNs take advantage of problem-specific information provided by decision trees and of generalpurpose optimization provided by genetic algorithms. This general methodology is then adapted to two concerns of power system dynamic security that kNNs are conceptually well appropriate to handle. One such question of paramount importance is how to detect outliers; these are cases "too far away" from the preanalyzed cases of the data base used to train kNNs. The other question is how to avoid dangerous diagnostics which could arise from an erroneous identification of the relevant majority class of neighbors. In this paper, these two questions are tackled in the context of transient stability and illustrated on the Hydro-Quebec power system.

IEEJ Transactions on Power and Energy, 1998

The Single Machine Equivalent (SIME) is a hybrid method resulting from the coupling of a time-dom... more The Single Machine Equivalent (SIME) is a hybrid method resulting from the coupling of a time-domain transient stability program with the equal-area criterion. Its aim is to set up a software as general as the time-domain program with which it is coupled but much more powerful and faster than it. The paper shows how the method reaches the threefold objective : to properly and readily assess transient stability limits (such as critical clearing times and power limits) under any power system modelling and stability scenario, to identify the relevant system machines and to appraise stability margins. In turn, these latter pieces of information open possibilities towards handy sensitivity analysis and transient stability monitoring. In short, they pave the way for real-time transient stability preventive monitoring and control. A variety of sample simulations highlight the method and its specifics.

to Electrical Power System Control

Copyright c○2005 by the authors. All rights reserved. No part of this publication may be reproduc... more Copyright c○2005 by the authors. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the publisher, bepress, which has been given certain exclusive rights by the

Extended equal area criterion revisited

This paper reports on a case study conducted on the EHV French power system in order to revisit t... more This paper reports on a case study conducted on the EHV French power system in order to revisit the extended equal area criterion and test its suitability as a fast transient stability indicator. The assumptions underlying the method are reexamined, causes liable to invalidate them are identified, and indices are devised to automatically circumvent them. The selection of candidate critical machines is also reconsidered and an augmented criterion is proposed. The various improvements are developed and tested on about 1000 stability scenarios, covering the entire 400-kV system; the severity of the scenarios, resulting from the combination of weakened both pre- and post-fault configurations, subjects the method to particularly stringent conditions. The obtained results show that the devised tools contribute to significantly reinforce its robustness and reliability.

Multi-period power loss optimization with limited number of switching actions for enhanced continuous power supply

2014 16th International Conference on Harmonics and Quality of Power (ICHQP), 2014

ABSTRACT This paper deals with multi-period active power loss minimization. We formulate this pro... more ABSTRACT This paper deals with multi-period active power loss minimization. We formulate this problem as a mixed-integer nonlinear programming (MINLP) problem, including constraints that specifically limit the number of switching actions between two successive anticipated system states. We solve this problem using the Mixed Integer Hybrid Differential Evolution (MIHDE), a modified Differential Evolution algorithm, which we adapt for our application. The effectiveness of the proposed approach is proven using a 60-bus system.

Power system security assessment: A position paper

2009 IEEE Bucharest PowerTech, 2009

This paper focuses on optimal power flow (OPF) computations in which no more than a pre-specified... more This paper focuses on optimal power flow (OPF) computations in which no more than a pre-specified number of controls are allowed to move. The benchmark formulation of this OPF problem constitutes a mixed integer nonlinear programming (MINLP) problem. To avoid the prohibitive computational time required by classical MINLP approaches to provide a (potentially sub-optimal) solution, we propose instead two alternative approaches. The first one consists in reformulating the MINLP problem as a mathematical program with equilibrium constraints (MPEC). The second approach includes in the classical OPF problem a nonlinear constraint which approximates the integral constraint limiting the number of control variables movement. Both approaches are solved by an interior point algorithm (IPA), slightly adapted to the particular characteristics of each approach. We provide numerical results with the proposed approaches on two test systems and for two practical problems: minimum cost to remove thermal congestion, and minimum cost of load curtailment to restore a feasible equilibrium point.

Lecture Notes in Computer Science, 2012

We consider in this paper look-ahead tree techniques for the discrete-time control of a determini... more We consider in this paper look-ahead tree techniques for the discrete-time control of a deterministic dynamical system so as to maximize a sum of discounted rewards over an infinite time horizon. Given the current system state xt at time t, these techniques explore the lookahead tree representing possible evolutions of the system states and rewards conditioned on subsequent actions ut, ut+1,. . .. When the computing budget is exhausted, they output the action ut that led to the best found sequence of discounted rewards. In this context, we are interested in computing good strategies for exploring the look-ahead tree. We propose a generic approach that looks for such strategies by solving an optimization problem whose objective is to compute a (budget compliant) tree-exploration strategy yielding a control policy maximizing the average return over a postulated set of initial states. This generic approach is fully specified to the case where the space of candidate tree-exploration strategies are "best-first" strategies parameterized by a linear combination of look-ahead path features-some of them having been advocated in the literature before-and where the optimization problem is solved by using an EDA-algorithm based on Gaussian distributions. Numerical experiments carried out on a model of the treatment of the HIV infection show that the optimized tree-exploration strategy is orders of magnitudes better than the previously advocated ones.