{"id":{"repo_id":"nus","oai_identifier":"oai:scholarbank.nus.edu.sg:10635/239059"},"canonical_url":"https://search.dev.ndltd.org/etd/nus/oai:scholarbank.nus.edu.sg:10635/239059","repository":{"repo_id":"nus","name":"National University of Singapore","base_url":"https://scholarbank.nus.edu.sg/oai/request"},"display":{"title":"DEEP REINFORCEMENT LEARNING FOR BUILDING ENERGY MANAGEMENT","abstract":"Building energy management is an increasingly complex problem, both in terms of energy production and consumption, with the integration of renew- able energy and the ever-increasing needs of building residents. Multiple studies have shown that Deep Reinforcement Learning has great potential in controlling energy allocation in buildings. This thesis aims to demonstrate the use of PPO, a recent Deep Reinforce- ment Learning algorithm with an actor-critic framework and Trust Region Policy, to control a thermal energy storage scheduling problem in a continuous state-action space with stochastic electric generation. To this end, two main steps are carried out in this thesis. First, the PPO algorithm is trained on a ten-state building environment without stochastic generation. A Rule-Based Controller is defined and serves as a benchmark to be beaten by the RL controller. Next, the algorithm is trained on a twenty-six- state building environment, with stochastic generation by solar panels. Very encouraging results have been achieved in both these stages. The trained PPO controller beat the RBC, reducing the electricity bill by 8.8% in the building fitted with solar panels.","abstract_html":"Building energy management is an increasingly complex problem, both in terms of energy production and consumption, with the integration of renew- able energy and the ever-increasing needs of building residents. Multiple studies have shown that Deep Reinforcement Learning has great potential in controlling energy allocation in buildings. This thesis aims to demonstrate the use of PPO, a recent Deep Reinforce- ment Learning algorithm with an actor-critic framework and Trust Region Policy, to control a thermal energy storage scheduling problem in a continuous state-action space with stochastic electric generation. To this end, two main steps are carried out in this thesis. First, the PPO algorithm is trained on a ten-state building environment without stochastic generation. A Rule-Based Controller is defined and serves as a benchmark to be beaten by the RL controller. Next, the algorithm is trained on a twenty-six- state building environment, with stochastic generation by solar panels. Very encouraging results have been achieved in both these stages. The trained PPO controller beat the RBC, reducing the electricity bill by 8.8% in the building fitted with solar panels.","abstract_has_math":false,"creators":["MILLE MELANIE ROLANDE COLETTE"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2023,"date_issued":"2023-01-09","date_published":"2023-01-09","updated_at":"2026-07-24T03:33:22Z","subjects":["reinforcement learning, energy management, storage scheduling"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["MILLE MELANIE ROLANDE COLETTE"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2023-01-09"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://scholarbank.nus.edu.sg/handle/10635/239059"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["reinforcement learning, energy management, storage scheduling"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholarbank.nus.edu.sg/bitstreams/488d8d7f-39bf-45f9-a2ee-5f7ad894a11c/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Building energy management is an increasingly complex problem, both in terms of energy production and consumption, with the integration of renew- able energy and the ever-increasing needs of building residents. Multiple studies have shown that Deep Reinforcement Learning has great potential in controlling energy allocation in buildings. This thesis aims to demonstrate the use of PPO, a recent Deep Reinforce- ment Learning algorithm with an actor-critic framework and Trust Region Policy, to control a thermal energy storage scheduling problem in a continuous state-action space with stochastic electric generation. To this end, two main steps are carried out in this thesis. First, the PPO algorithm is trained on a ten-state building environment without stochastic generation. A Rule-Based Controller is defined and serves as a benchmark to be beaten by the RL controller. Next, the algorithm is trained on a twenty-six- state building environment, with stochastic generation by solar panels. Very encouraging results have been achieved in both these stages. The trained PPO controller beat the RBC, reducing the electricity bill by 8.8% in the building fitted with solar panels."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["d48126c8d9c0cafd755bd827d03f0060","a0d58b87427f9d7074cfb1cff9957cef"]},{"key":"dc:title","label":"Title","values":["DEEP REINFORCEMENT LEARNING FOR BUILDING ENERGY MANAGEMENT"]}]}],"canonical_facts":{"dc:creator":["MILLE MELANIE ROLANDE COLETTE"],"dc:date.issued":["2023-01-09"],"dc:description.abstract":["Building energy management is an increasingly complex problem, both in terms of energy production and consumption, with the integration of renew- able energy and the ever-increasing needs of building residents. Multiple studies have shown that Deep Reinforcement Learning has great potential in controlling energy allocation in buildings. This thesis aims to demonstrate the use of PPO, a recent Deep Reinforce- ment Learning algorithm with an actor-critic framework and Trust Region Policy, to control a thermal energy storage scheduling problem in a continuous state-action space with stochastic electric generation. To this end, two main steps are carried out in this thesis. First, the PPO algorithm is trained on a ten-state building environment without stochastic generation. A Rule-Based Controller is defined and serves as a benchmark to be beaten by the RL controller. Next, the algorithm is trained on a twenty-six- state building environment, with stochastic generation by solar panels. Very encouraging results have been achieved in both these stages. The trained PPO controller beat the RBC, reducing the electricity bill by 8.8% in the building fitted with solar panels."],"dc:format.checksum.md5":["d48126c8d9c0cafd755bd827d03f0060","a0d58b87427f9d7074cfb1cff9957cef"],"dc:identifier.uri":["https://scholarbank.nus.edu.sg/bitstreams/488d8d7f-39bf-45f9-a2ee-5f7ad894a11c/download"],"dc:relation.isreferencedby":["https://scholarbank.nus.edu.sg/handle/10635/239059"],"dc:subject":["reinforcement learning, energy management, storage scheduling"],"dc:title":["DEEP REINFORCEMENT LEARNING FOR BUILDING ENERGY MANAGEMENT"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T03:33:22Z"}