Distributed, autonomous control of space habitats

Long-duration space missions require advanced life support (ALS) systems that can regenerate air, water and food. These ALS systems need complex control strategies that can maintain stable system performance and balance resources with small margins and minimal buffers. In this paper we will describe the ALS control task in detail and give some examples of previous control solutions. Then we will look at how machine learning techniques can help create a more adaptive ALS control system. We will examine reinforcement learning and genetic algorithms and their relationship to optimizing resource utilization in an ALS system. Finally, we will present an innovative multistep genetic algorithm that generates control strategies that perform much better than traditional reinforcement learning or traditional genetic algorithms.