Operations and maintenance (O&M) expenses can vary greatly from one energy solution to another. While a solar farm or geothermal system may need minimal ongoing maintenance, wind turbines require a skilled crew to keep them operating efficiently.
In this research, the authors use a scaled-down wind farm case study
- to demonstrate the potential of one of the maintenance optimization techniques which uses Reinforcement Learning (RL) algorithms to identify an optimal O&M policy;
- to show the ease of use of AnyLogic, a multimethod simulation software, and Pathmind, a tool that enables effectively exploiting the RL capabilities without deep knowledge of machine learning.
RL is a machine learning framework in which a learning agent optimizes its behavior by consecutive trial and error interactions with a white-box model of the system being optimized. Using agent-based modeling technique supported by AnyLogic, the authors build a wind farm simulation model which encodes two types of agents: Wind Turbine and Maintenance Crew.
When the maintenance optimization model is ready, they export it and upload it into the Pathmind machine learning web application to perform the training in the cloud. On the cloud platform, the researchers run multiple experiments with different reward functions. Once the RL training is concluded, they import the operations and maintenance policy back into AnyLogic to test its performance.
Mainenance optimization using the agent-based modeling technique - Maintenance Crew statechart