Browse ORBi by ORBi project

- Background
- Content
- Benefits and challenges
- Legal aspects
- Functions and services
- Team
- Help and tutorials

Bias-variance tradeoff of soft decision trees ; Wehenkel, Louis (2004) This paper focuses on the study of the error composition of a fuzzy decision tree induction method recently proposed by the authors, called soft decision trees. This error may be expressed as a sum of ... [more ▼] This paper focuses on the study of the error composition of a fuzzy decision tree induction method recently proposed by the authors, called soft decision trees. This error may be expressed as a sum of three types of error: residual error, bias and variance. The paper studies empirically the tradeoff between bias and variance in a soft decision tree method and compares it with the tradeoff of classical crisp regression and classification trees. The main conclusion is that the reduced prediction variance of fuzzy trees is the main reason for their improved performance with respect to crisp ones. [less ▲] Detailed reference viewed: 31 (0 ULg)Erratum to : A complete fuzzy decision tree technique (vol 138, pg 221, 2003) ; Wehenkel, Louis in Fuzzy Sets and Systems (2003), 140(3), 563-565 Detailed reference viewed: 22 (1 ULg)A complete fuzzy decision tree technique ; Wehenkel, Louis in Fuzzy Sets and Systems (2003), 138(2), 221-254 In this paper, a new method of fuzzy decision trees called soft decision trees (SDT) is presented. This method combines tree growing and pruning, to determine the structure of the soft decision tree, with ... [more ▼] In this paper, a new method of fuzzy decision trees called soft decision trees (SDT) is presented. This method combines tree growing and pruning, to determine the structure of the soft decision tree, with refitting and backfitting, to improve its generalization capabilities. The method is explained and motivated and its behavior is first analyzed empirically on 3 large databases in terms of classification error rate, model complexity and CPU time. A comparative study on 11 standard UCI Repository databases then shows that the soft decision trees produced by this method are significantly more accurate than standard decision trees. Moreover, a global model variance study shows a much lower variance for soft decision trees than for standard trees as a direct cause of the improved accuracy. (C) 2003 Elsevier B.V. All rights reserved. [less ▲] Detailed reference viewed: 84 (8 ULg)OMASES: A DYNAMIC SECURITY ASSESSMENT TOOL FOR THE NEW MARKET ENVIRONMENT ; ; et al (2003, June 26) The paper presents the efforts and results of a large consortium of European Industries, Research Centers and Universities involved in an EU research project named OMASES in the field of Power System ... [more ▼] The paper presents the efforts and results of a large consortium of European Industries, Research Centers and Universities involved in an EU research project named OMASES in the field of Power System Dynamic Security Assessment (DSA). The overall structure of an on-line DSA tool including TSA – Transient Stability Assessment, VSA – Voltage Stability Assessment, TS – Training Simulator and MS – Market Simulator is reported. Some basic assumptions and methodological aspects of the tool are presented considering its possible use within actual or future Energy Management Systems under the new electric market environment. Scenarios set up for the validation phase and results are reported with reference to the Italian test facility. [less ▲] Detailed reference viewed: 98 (1 ULg)A probabilistic approach to power system network planning under uncertainties ; ; et al in Proceedings of the IEEE Bologna Power Tech Conference (2003) This work proposes a methodology and a practical tool for the study of long-term network planning under uncertainties. In this approach the major external uncertainties during the planning horizon are ... [more ▼] This work proposes a methodology and a practical tool for the study of long-term network planning under uncertainties. In this approach the major external uncertainties during the planning horizon are modeled as macroscenarios at different future time instants. On the other hand, the random nature of actual operating conditions is taken into account by using a probabilistic model of microscenarios based on past statistics. Massive Monte-Carlo simulations are used to generate and simulate a large number of scenarios and store the detailed results in a relational database. Data mining techniques are then applied to extract information from the database so as to rank scenarios and network reinforcements according to different criteria. [less ▲] Detailed reference viewed: 28 (6 ULg)Iteratively extending time horizon reinforcement learning Ernst, Damien ; Geurts, Pierre ; Wehenkel, Louis in Machine Learning: ECML 2003, 14th European Conference on Machine Learning (2003) 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 ... [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 (x(t), u(t), r(t), x(t+1)) where x(t) denotes the system state at time t, ut the control action taken, rt the instantaneous reward obtained and x(t+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 four-tuples 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. [less ▲] Detailed reference viewed: 57 (7 ULg)A reinforcement learning based discrete supplementary control for power system transient stability enhancement ; Ernst, Damien ; Wehenkel, Louis in Proceedings of the 12th Intelligent Systems Application to Power Systems Conference (ISAP 2003) (2003) This paper proposes an application of a Reinforcement Learning (RL) method to the control of a dynamic brake aimed to enhance power system transient stability. The control law of the resistive brake is in ... [more ▼] This paper proposes an application of a Reinforcement Learning (RL) method to the control of a dynamic brake aimed to enhance power system transient stability. The control law of the resistive brake is in the form of switching strategies. In particular, the paper focuses on the application of a model based RL method, known as prioritized sweeping, a method proven to be suitable in applications in which computation is considered to be cheap. The curse of dimensionality problem is resolved by the system state dimensionality reduction based on the One Machine Infinite Bus (OMIB) transformation. Results obtained by using a synthetic four-machine power system are given to illustrate the performances of the proposed methodology. [less ▲] Detailed reference viewed: 33 (2 ULg)An empirical comparison of machine learning algorithms for generic image classification Marée, Raphaël ; Geurts, Pierre ; et al in Proceedings of the 23rd SGAI international conference on innovative techniques and applications of artificial intelligence, Research and development in intelligent systems XX, (2003) Detailed reference viewed: 48 (3 ULg)Une méthode générique pour la classification automatique d'images à partir des pixels Marée, Raphaël ; Geurts, Pierre ; Wehenkel, Louis in Revue des Nouvelles Technologies de l'Information (2003), 1 Dans cet article, nous évaluons une approche générique de classification automatique d'images. Elle repose sur une méthode d'apprentissage récente qui construit des ensembles d'arbres de décision par ... [more ▼] Dans cet article, nous évaluons une approche générique de classification automatique d'images. Elle repose sur une méthode d'apprentissage récente qui construit des ensembles d'arbres de décision par sélection aléatoire des tests directement sur les valeurs basiques des pixels. Nous proposons une variante, également générique, qui réalise une augmentation fictive de la taille des échantillons par extraction et classification de sous-fenêtres des images. Ces deux approches sont évaluées et comparées sur quatre bases de données publiques de problèmes courants: la reconnaissance de chiffres manuscrits, de visages, d'objets 3D et de textures. [less ▲] Detailed reference viewed: 147 (19 ULg)An implementation of on-line transient stability screening and control using distributed processing ; Pavella, Mania ; Wehenkel, Louis in Proc. of Intelligent Systems Application to Power Systems (2003) This paper describes the implementation of an online transient stability assessment software, composed of algorithms for contingency screening and for the design of preventive control actions. The ... [more ▼] This paper describes the implementation of an online transient stability assessment software, composed of algorithms for contingency screening and for the design of preventive control actions. The implementation of the two parts rely on a hybrid method called SIME, coupled with a time domain simulation engine and power flow program. The speed up of the contingency screening module is obtained by distributing contingencies on a cluster of computers to comply with extended real-time speed requirements. A compensation scheme is used to determine active power rescheduling alternatives in order to stabilize the dangerous contingencies identified at the screening step. The software has been coupled with an industrial EMS platform, and tested in the simulation environment. [less ▲] Detailed reference viewed: 25 (3 ULg)Using artificial neural networks to estimate rotor angles and speeds from phasor measurements ; ; Wehenkel, Louis (2003) This paper deals with an improved use of phasor measurements. In particular, the paper focuses on the development of a technique for estimation of generator rotor angle and speed, based on phasor ... [more ▼] This paper deals with an improved use of phasor measurements. In particular, the paper focuses on the development of a technique for estimation of generator rotor angle and speed, based on phasor measurement units, for transient stability assessment and control in real-time. Two multilayered feed-forward artificial neural networks are used for this purpose. One for the estimation of rotor angle and another for the estimation of rotor speed. The validation has been made by simulation in a power system because techniques for the direct measurement were not available. Results obtained with the help of a simple one machine to infinite bus system are presented and compared against those obtained using analytical formulas derived from the generator classical model. [less ▲] Detailed reference viewed: 33 (1 ULg)On the construction of the inclusion boundary neighbourhood for markov equivalence classes of bayesian network structures ; Wehenkel, Louis in Proceedings of Uncertainty in Artificial Intelligence (2002) The problem of learning Markov equivalence classes of Bayesian network structures may be solved by searching for the maximum of a scoring metric in a space of these classes. This paper deals with the ... [more ▼] The problem of learning Markov equivalence classes of Bayesian network structures may be solved by searching for the maximum of a scoring metric in a space of these classes. This paper deals with the definition and analysis of one such search space. We use a theoretically motivated neighbourhood, the inclusion boundary, and represent equivalence classes by essential graphs. We show that this search space is connected and that the score of the neighbours can be evaluated incrementally. We devise a practical way of building this neighbourhood for an essential graph that is purely graphical and does not explicitely refer to the underlying independences. We find that its size can be intractable, depending on the complexity of the essential graph of the equivalence class. The emphasis is put on the potential use of this space with greedy hillclimbing search. [less ▲] Detailed reference viewed: 6 (1 ULg)FACTS devices controlled by means of reinforcement learning algorithms Ernst, Damien ; Wehenkel, Louis in Proceedings of the 14th Power Systems Computation Conference (PSCC 2002) (2002) Reinforcement learning consists of a collection of methods for approximating solutions to deterministic and stochastic optimal control problems of unknown dynamics. These methods learn by experience how ... [more ▼] Reinforcement learning consists of a collection of methods for approximating solutions to deterministic and stochastic optimal control problems of unknown dynamics. These methods learn by experience how to adjust a closed-loop control rule which is a mapping from the system states to control actions. This paper proposes an application of reinforcement learning methods to the control of a FACTS device aimed to damp power system oscillations. A detailed case study is carried out on a synthetic four-machine power system. [less ▲] Detailed reference viewed: 37 (1 ULg)Improving the bias/variance tradeoff of decision trees - towards soft tree induction Geurts, Pierre ; ; Wehenkel, Louis in Engineering intelligent systems (2001), 9 One of the main difﬁculties with standard top down induction of decision trees comes from the high variance of these methods. High variance means that, for a given problem and sample size, the resulting ... [more ▼] One of the main difﬁculties with standard top down induction of decision trees comes from the high variance of these methods. High variance means that, for a given problem and sample size, the resulting tree is strongly dependent on the random nature of the particular sample used for training. Consequently, these algorithms tend to be suboptimal in terms of accuracy and interpretability. This paper analyses this problem in depth and proposes a new method, relying on threshold softening, able to signiﬁcantly improve the bias/variance tradeoff of decision trees. The algorithm is validated on a number of benchmark problems and its relationship with fuzzy decision tree induction is discussed. This sheds some light on the success of fuzzy decision tree induction and improves our understanding of machine learning, in general. [less ▲] Detailed reference viewed: 70 (4 ULg)Application of reinforcement learning to electrical power system closed-loop emergency control ; Ernst, Damien ; Wehenkel, Louis in Principles of Data Mining and Knowledge Discovery, 4th European Conference, PKDD 2000 (2000) This paper investigates the use of reinforcement learning in electric power system emergency control. The approach consists of using numerical simulations together with on-policy Monte Carlo control to ... [more ▼] This paper investigates the use of reinforcement learning in electric power system emergency control. The approach consists of using numerical simulations together with on-policy Monte Carlo control to determine a discrete switching control law to trip generators so as to avoid loss of synchronism. The proposed approach is tested on a model of a real large scale power system and results are compared with a quasi-optimal control law designed by a brute force approach for this system. [less ▲] Detailed reference viewed: 45 (3 ULg)Temporal machine learning for switching control Geurts, Pierre ; Wehenkel, Louis in Proceedings of PKDD 2000, 4th European Conference on Principles of Data Mining and Knowledge Discovery (2000) In this paper, a temporal machine learning method is presented which is able to automatically construct rules allowing to detect as soon as possible an event using past and present measurements made on a ... [more ▼] In this paper, a temporal machine learning method is presented which is able to automatically construct rules allowing to detect as soon as possible an event using past and present measurements made on a complex system. This method can take as inputs dynamic scenarios directly described by temporal variables and provides easily readable results in the form of detection trees. The application of this method is discussed in the context of switching control. Switching (or discrete event) control of continuous systems consists in changing the structure of a system in such a way as to contreol its behavior. Given a particular discrete control switch, detection trees are applied to the induction of rules which decide based on the available measurements whether or not to operate a switch. Two practical applications are discussed in the context of electrical power systems emergency control. [less ▲] Detailed reference viewed: 22 (1 ULg)Investigation and reduction of discretization Variance in decision tree induction Geurts, Pierre ; Wehenkel, Louis in Proceedings of ECML 2000, European Conference on Machine Learning (2000) This paper focuses on the variance introduced by the discretization techniques used to handle continuous attributes in decision tree induction. Different discretization procedures are first studied ... [more ▼] This paper focuses on the variance introduced by the discretization techniques used to handle continuous attributes in decision tree induction. Different discretization procedures are first studied empirically, then means to reduce the discretization variance are proposed. The experiments shows that discretization variance is large and that it is possible to reduce it significantly without notable computational costs. The resulting variance reduction mainly improves interpretability and stability of decision trees, and marginally their accuracy. [less ▲] Detailed reference viewed: 11 (3 ULg)Probabilistic design of power-system special stability controls Wehenkel, Louis ; ; et al in Control Engineering Practice (1999), 7(2), 183-194 A probabilistic approach to the design of power-system special stability controls is presented here. Using Monte-Carlo simulations, it takes into account all the potential causes of blackouts, slow and ... [more ▼] A probabilistic approach to the design of power-system special stability controls is presented here. Using Monte-Carlo simulations, it takes into account all the potential causes of blackouts, slow and fast dynamics, and modeling uncertainties. A large number of scenarios are simulated in parallel by time-domain numerical integration, and the relevant parameters of the resulting system trajectories are stored in a database. Data-mining tools are used to identify the most important system weaknesses and possible improvements. The approach is tested on a large-scale study on the SouthÐEastern part of the extra-high-voltage system of Electricité de France. [less ▲] Detailed reference viewed: 26 (4 ULg)Transient stability-constrained optimal power flow ; ; Ernst, Damien et al in Proceedings of the IEEE Power Tech'99 (1999) This paper proposes a new approach able to maximize the interface flow limits in power systems and to find a new operating state that is secure with respect to both, dynamic (transient stability) and ... [more ▼] This paper proposes a new approach able to maximize the interface flow limits in power systems and to find a new operating state that is secure with respect to both, dynamic (transient stability) and static security constraints. It combines the Maximum Allowable Transfer (MAT) method, recently developed for the simultaneous control of a set of contingencies, and an Optimal Power Flow (OPF) method for maximizing the interface power flow. The approach and its performances are illustrated by means of simulations carried out on a real world power system. [less ▲] Detailed reference viewed: 95 (5 ULg)Data mining tools and application in power system engineering ; Geurts, Pierre ; Wehenkel, Louis in Proceedings of the 13th Power System Computation Conference, PSCC99 (1999) The power system field is presently facing an explosive growth of data. The data mining (DM) approach provides tools for making explicit some implicit subtle structure in data. Applying data mining to ... [more ▼] The power system field is presently facing an explosive growth of data. The data mining (DM) approach provides tools for making explicit some implicit subtle structure in data. Applying data mining to power system engineering is an iterative and interactive process, requiring an acquainted user with the application specifics. The paper describes data mining tools like statistical methos, visualization, machine learning and neural networks, exemplifying by results obtained with a DM software developed for dynamic security assessment studies. Power system engineering applications where data mining would be useful are reviewed in the second part of the paper. [less ▲] Detailed reference viewed: 200 (0 ULg) |
||