Reinforcement learning based smart home heating solution in Energy2D simulation software
Name
Erik Mukk
Abstract
During the course of this bachelor thesis, a machine learning based heating system was
created. System is based on reinforcement learning, which learns and automates heating
in simulations running on Energy2D software. To evaulate the system, solutions based on
reinforcement learning are compared to a benchmark, which is thermostat based heating.
Built reinforcement learning based solution is able to operate the heater in the room
autonomously. By configuring the system before training in a way that it depends on
electricity price, environment heating profile changes. With this reinforcement learning
based system, a 5%-15% savings in money spent on heating was achieved compared to
the benchmark, whilst being as good or even better at holding target temperature of the
room. At the end of the thesis, shortcomings of this system are indentified and tips to fix
them are given. Also, future works are propsed.
created. System is based on reinforcement learning, which learns and automates heating
in simulations running on Energy2D software. To evaulate the system, solutions based on
reinforcement learning are compared to a benchmark, which is thermostat based heating.
Built reinforcement learning based solution is able to operate the heater in the room
autonomously. By configuring the system before training in a way that it depends on
electricity price, environment heating profile changes. With this reinforcement learning
based system, a 5%-15% savings in money spent on heating was achieved compared to
the benchmark, whilst being as good or even better at holding target temperature of the
room. At the end of the thesis, shortcomings of this system are indentified and tips to fix
them are given. Also, future works are propsed.
Graduation Thesis language
Estonian
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Jakob Mass
Defence year
2020