MétaCan
Menu
Back to cohort
Record W7067312275

Learning-Accelerated Exact Mixed-Integer Second-Order Cone Programming for Unit Commitment

2023· other· fr· W7067312275 on OpenAlexfundno aff

Bibliographic record

VenuePolyPublie (École Polytechnique de Montréal) · 2023
Typeother
Languagefr
FieldEngineering
TopicGeotechnical and Geomechanical Engineering
Canadian institutionsnot available
FundersNatural Sciences and Engineering Research Council of CanadaInstitut de Valorisation des Données
KeywordsGraphIndependent setSkill mix
DOInot available

Abstract

fetched live from OpenAlex

RÉSUMÉ: Dans le domaine des réseaux électriques, la gestion de l’engagement des unités – mieux connu sous son appellation anglaise "unit commitment" (UC) – est une problématique d’optimisation majeure. Ces enjeux sont fondamentaux puisqu’ils engagent des décisions cruciales en matière de planification des unités de production d’électricité pour répondre efficacement à la demande pour une période spécifique. Leur complexité, due à leur nature binaire et à la non-convexité des contraintes d’écoulement de puissance, exige leur modélisation en tant que programmes coniques mixtes entiers du second ordre (MISOCPs) et programmes quadratiques convexes mixtes entiers (MICQPs). Les méthodes d’apprentissage machines offre un socle robuste pour développer des heuristiques sur mesure, destinées à résoudre ces problématiques d’optimisation du UC. En puisant dans des motifs et des données, elles offrent des solutions spécifiquement personalisé à chaque cas d’utilisation. Dans ce mémoire, nous proposons une méthodologie visant à augmenter l’efficacité des solveurs d’optimisation pour UC à l’aide d’un graph Covonlutional neural network (GCNN). Notre approche tire avantage de la structure inhérente de ces problèmes mixtes entiers en les représentant sous forme de graphes k-partites variables-contraintes. Cette représentation nous donne l’opportunité de saisir les interactions subtiles entre les variables et les contraintes du problème. Grâce à l’utilisation du GCNN, nous sommes en mesure d’extraire des informations précieuses de ces graphes. Le processus d’apprentissage, réalisé par imitation, s’appuie sur la règle d’expert du "strong branching" comme référence. Cette approche permet à notre modèle d’élaborer des stratégies de sélection de variables efficaces pour l’algorithme de séparation et évaluation, accélérant de manière significative l’ensemble du processus d’optimisation. Il est notable que notre méthode assure le maintien de l’optimalité de la solution, un aspect souvent négligé dans les approches reposant systématique sur l’apprentissage de bout en bout. La préservation de l’optimalité est essentielle dans le contexte de l’UC, où la planification optimale des unités de production d’électricité peut impacter fortement l’efficacité opérationnelle et la fiabilité des systèmes électriques. Un élément clé de notre approche est l’intégration d’informations spécifiques au problème dans le processus de sélection des variables, au cœur de l’optimisation de séparation et évaluation. ABSTRACT: The complexity of Unit Commitment (UC) problems in power systems is underscored by their binary attributes and the non-convex nature of power flow constraints. This results in these problems being best represented as mixed-integer second order cone programs (MISOCPs) and mixed-integer convex quadratic programs (MICQPs). The considerable optimization challenge posed by UC problems is deeply rooted in their significant role in decision-making. Specifically, they necessitate critical scheduling of power generation units to efficiently fulfil demand within a predetermined time period. Machine learning method provide a robust framework for creating customized heuristics designed to solve optimization problems like UC, by learning from patterns and data, hence offering solutions that are tailored specifically to individual use-cases. In this Master thesis, we present a methodology that aims to improve the efficiency of UC optimization solvers using a GCNN. Our approach harnesses the structure of these mixed-integer problems, representing them as variable-constraint k-partite graphs. This represen-tation allows us to capture the intricate relationships between variables and constraints in the problem. Utilizing GCNN, we extract valuable insights from these graphs. The training process is conducted via imitation learning and uses the strong branching as expert rule. This approach enables our model to learn effective variable selection policies for the branch and bound algorithm, thus accelerating the overall optimization process. Notably, our method ensures the preservation of solution optimality, a feature that is often compromised in end-to-end learning approaches. This preservation of optimality is crucial in the context of UC problems, where the optimal scheduling of power generation units can have significant implications for the operational efficiency and reliability of power systems. A key aspect of our approach is the integration of problem-specific information into the variable selection process within the branch and bound optimization. Our work demonstrates the effectiveness and robustness of our machine learning-based ap-proach in addressing two different formulation of the UC problem across various power grid models. Our method consistently outperformed the SCIP solver it is based on, even un-der varied scenarios. These results, obtained from numerical analysis performed on seven standard grid models, illustrate the adaptability and speed of our approach, highlighting its potential for significant advancements in power system optimization.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.001
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesMeta-epidemiology (narrow), Research integrity, Insufficient payload (model declined to judge)
Consensus categoriesMeta-epidemiology (narrow), Research integrity
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: none
GenreCandidate signal: Methods · Consensus signal: none
Teacher disagreement score0.691
Threshold uncertainty score1.000

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0010.001
Meta-epidemiology (narrow)0.0020.002
Meta-epidemiology (broad)0.0020.001
Bibliometrics0.0010.002
Science and technology studies0.0000.000
Scholarly communication0.0000.000
Open science0.0010.000
Research integrity0.0020.003
Insufficient payload (model declined to judge)0.0030.001

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.015
GPT teacher head0.226
Teacher spread0.211 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; both teacher heads agree on what is shown here.

Study designSimulation or modeling
Domainnot available
GenreMethods

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2023
Admission routes1
Has abstractyes

Explore more

Same venuePolyPublie (École Polytechnique de Montréal)Same topicGeotechnical and Geomechanical EngineeringFrench-language works237,207