Explore wind farm optimization techniques using reinforcement learning and large-scale simulation in this 28-minute conference talk from Anyscale. Discover how a collaboration between Microsoft and Vestas led to the development of wind farm controllers capable of increasing annual energy production by 1-2% through yaw adjustments of upstream turbines. Learn about the challenges of training deep reinforcement learning-based controllers using extensive computational fluid dynamics simulations, and how the DeepSim platform leveraged Ray for distributed computing to efficiently manage up to 15,000 CPU cores in parallel. Gain insights into the project's workflow, deployment strategies, and the application of reinforcement learning in addressing wake effects and steering in wind farms. Witness a demonstration of the controller in action and examine the impressive results achieved in this innovative approach to renewable energy optimization.
Overview
Syllabus
Introduction
About Mindside
Use Cases
Deep Sim
Product Stack
DeepSim
DeepSimpler
Workflow
Deployment
Urban farm
Why RL
Objectives
Wake effect
Wake steering
Making a controller
Demo
Results
Taught by
Anyscale