Using deep reinforcement learning to optimize hydropower production and other complex systems
Machines that explore, experiment and learn are powerful tools. They build scientific models of the world. They experiment and reason about how actions can change the world to the better. Billions of alternatives are investigated. It is all about creative machines that learn how to act from experience.
One example is our project with Agder Energi on controlling hydropower production. Hydropower production is so complex that it is difficult to create mathematical models that capture the reality. One must plan according to future precipitation and electricity prices. In addition there is significant uncertainty around the physics of hydropower production systems. To solve this problem, we develop artificial intelligence that learns how to operate hydropower production by itself. It understands physics through massive scientific experimentation. It recognizes crucial patterns in precipitation and electricity prices through observation. And it discovers optimal hydropower production strategies by building models that capture how all the important factors interact.