Data-Driven Control Optimization withMachine Learning - Wind Farm Power Maximization as an Exemplar

9月14日上午10:00~11:30

清华大学土木工程系（何善蘅楼）多功能厅

KinchoH. Law

Professorof Civil and Environmental Engineering

EngineeringInformatics Group

Stanford University, Stanford, CA 94305-4020,USA

Abstract

This presentation discusses a data-driven optimization approach tomaximize the wind farm power production.Conventionally, under a given windcondition, each individual wind turbine is controlled to maximize its ownenergy production without taking into consideration the power production ofother wind turbines. For this greedystrategy, because of the reduced wind speed and increased turbulence intensityinside the wake, the downstream wind turbines would produce only a fraction ofthe power produced by the upstream wind turbines. In this work, we formulate the wind farm power maximization problem as acooperative control problem that minimizes the wake interference among windturbines and, thus, maximizes the overall wind farm power production. The optimal, coordinated actionsby controlling the yaw offset and blade angles of wind turbine are determinedusing a probabilistic model-free Bayesian optimization (BO) approach. In BO framework, the wind farm power functionis modeled as Gaussian Process (GP) using the historical input (controlactions) and the output (total wind farm power production) data. Theconstructed wind farm power function is then used to determine the trialcontrol actions with the objective to improve the wind farm power as well asexploring (learning) the wind farm power function. Numerical simulations and wind tunnel experiments are conducted toillustrate the feasibility of the data driven optimization approach.

Biographical Information:

KinchoH. Law is currently Professor of Civil andEnvironmental Engineering at Stanford University. He obtained his BS in Civil Engineering andBA in Mathematics from University of Hawaii in 1976, and his MS and PhD in CivilEngineering in 1979 and 1981, respectively, from Carnegie Mellon University. Professor Law’s research has been focused onthe application of advanced computing principles and techniques to civil andstructural engineering. His work hasdealt with various aspects of structural dynamics, structural health monitoring, wirelesssensing and control, high performance computing, computational science andengineering, engineering and legal information management, engineeringenterprise integration, and Internet computing,. He has authored and co-authored over 350articles in journals and conference proceedings.