Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach
District heating for the capital area of Iceland heavily relies on geothermal water, with one of the key components being the deep well pump network located in Reykjahlíð, Mosfellsdalur, near Reykjavík. However, the network’s operation is yet to be fully optimized for overall efficiency. The combina...
Main Author: | |
---|---|
Other Authors: | |
Format: | Master Thesis |
Language: | English |
Published: |
2024
|
Subjects: | |
Online Access: | https://hdl.handle.net/1946/48707 |
_version_ | 1821557899719081984 |
---|---|
author | Friðrik Tryggvi Róbertsson 1997- |
author2 | Háskólinn í Reykjavík |
author_facet | Friðrik Tryggvi Róbertsson 1997- |
author_sort | Friðrik Tryggvi Róbertsson 1997- |
collection | Skemman (Iceland) |
description | District heating for the capital area of Iceland heavily relies on geothermal water, with one of the key components being the deep well pump network located in Reykjahlíð, Mosfellsdalur, near Reykjavík. However, the network’s operation is yet to be fully optimized for overall efficiency. The combination of electric submersible pumps (ESPs) and vertical shaft pumps (VSPs) presents a computationally intensive optimization challenge. This work addresses the challenge by integrating a hydraulic simulation model, developed using Epanet, with the Dueling Deep Q-network (DQN) architecture, in which the neural network functions as a core component, serving as a function approximator to optimize the complex, non-linear relationship between fluid flow and power consumption, achieving a multi-objective goal. Two distinct approaches are explored, with the more successful yielding a 6.5% reduction in the network’s power consumption compared to the current operation, while accurately meeting demand. This optimization is performed in near real-time, making it highly suitable for the fluctuating demand conditions typical of district heating systems. Hitaveitan á höfuðborgarsvæðinu byggir mikið á nýtingu jarðhita, einn af lykilþáttum hita veitunnar er djúpdælukerfi sem staðsett er í Reykjahlíð í Mosfellsdal. Enn á þó eftir að fullkomna virkni kerfisins til að ná fram hámarks heildarnýtni. Dælukerfið samanstendur af háspenntum djúpdælum og öxuldælum. Það að uppsetning dælukerfisins innihaldi mis munandi gerðir dæla með mismunandi stýringum gerir það að verkum að bestun kerfisins verður þung í útreikningi. Í þessari ritgerð er leitast við að ná fram hagræðingu með því að samþætta vökvahermilíkan, þróað með Epanet, við Dueling Deep Q-Network (DQN) að ferð, þar sem taugakerfið er notað sem nálgunarfall á flókna, ólínulega og ókúpta sambandi vatnsflæðis og orkunotkunar. Tvær útfærslur eru skoðaðar, sú árangursríkasta skilar 6,5% lækkun á orkunotkun kerfisins miðað við núverandi virkni, en uppfyllir jafnframt eftirspurn á nákvæman ... |
format | Master Thesis |
genre | Iceland Reykjavík Reykjavík |
genre_facet | Iceland Reykjavík Reykjavík |
geographic | Reykjahlíð Reykjavík |
geographic_facet | Reykjahlíð Reykjavík |
id | ftskemman:oai:skemman.is:1946/48707 |
institution | Open Polar |
language | English |
long_lat | ENVELOPE(-16.912,-16.912,65.642,65.642) |
op_collection_id | ftskemman |
op_relation | https://hdl.handle.net/1946/48707 |
publishDate | 2024 |
record_format | openpolar |
spelling | ftskemman:oai:skemman.is:1946/48707 2025-01-16T22:41:05+00:00 Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach Bestun á stýringu djúpdælna í lághita vatnsöflun: Nálgun með Dueling Deep Q-Network Friðrik Tryggvi Róbertsson 1997- Háskólinn í Reykjavík 2024-10 application/pdf https://hdl.handle.net/1946/48707 en eng https://hdl.handle.net/1946/48707 Vélaverkfræði Meistaraprófsritgerðir Jarðhitanýting Bestun Dælur Orkunotkun Mechanical engineering Geothermal space heating Mathematical optimization Pumping machinery Reinforcement Learning Heating from central stations Thesis Master's 2024 ftskemman 2024-10-23T00:02:50Z District heating for the capital area of Iceland heavily relies on geothermal water, with one of the key components being the deep well pump network located in Reykjahlíð, Mosfellsdalur, near Reykjavík. However, the network’s operation is yet to be fully optimized for overall efficiency. The combination of electric submersible pumps (ESPs) and vertical shaft pumps (VSPs) presents a computationally intensive optimization challenge. This work addresses the challenge by integrating a hydraulic simulation model, developed using Epanet, with the Dueling Deep Q-network (DQN) architecture, in which the neural network functions as a core component, serving as a function approximator to optimize the complex, non-linear relationship between fluid flow and power consumption, achieving a multi-objective goal. Two distinct approaches are explored, with the more successful yielding a 6.5% reduction in the network’s power consumption compared to the current operation, while accurately meeting demand. This optimization is performed in near real-time, making it highly suitable for the fluctuating demand conditions typical of district heating systems. Hitaveitan á höfuðborgarsvæðinu byggir mikið á nýtingu jarðhita, einn af lykilþáttum hita veitunnar er djúpdælukerfi sem staðsett er í Reykjahlíð í Mosfellsdal. Enn á þó eftir að fullkomna virkni kerfisins til að ná fram hámarks heildarnýtni. Dælukerfið samanstendur af háspenntum djúpdælum og öxuldælum. Það að uppsetning dælukerfisins innihaldi mis munandi gerðir dæla með mismunandi stýringum gerir það að verkum að bestun kerfisins verður þung í útreikningi. Í þessari ritgerð er leitast við að ná fram hagræðingu með því að samþætta vökvahermilíkan, þróað með Epanet, við Dueling Deep Q-Network (DQN) að ferð, þar sem taugakerfið er notað sem nálgunarfall á flókna, ólínulega og ókúpta sambandi vatnsflæðis og orkunotkunar. Tvær útfærslur eru skoðaðar, sú árangursríkasta skilar 6,5% lækkun á orkunotkun kerfisins miðað við núverandi virkni, en uppfyllir jafnframt eftirspurn á nákvæman ... Master Thesis Iceland Reykjavík Reykjavík Skemman (Iceland) Reykjahlíð ENVELOPE(-16.912,-16.912,65.642,65.642) Reykjavík |
spellingShingle | Vélaverkfræði Meistaraprófsritgerðir Jarðhitanýting Bestun Dælur Orkunotkun Mechanical engineering Geothermal space heating Mathematical optimization Pumping machinery Reinforcement Learning Heating from central stations Friðrik Tryggvi Róbertsson 1997- Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach |
title | Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach |
title_full | Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach |
title_fullStr | Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach |
title_full_unstemmed | Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach |
title_short | Optimal operation of a low-temperature geothermal well pump network : a Dueling Deep Q-network approach |
title_sort | optimal operation of a low-temperature geothermal well pump network : a dueling deep q-network approach |
topic | Vélaverkfræði Meistaraprófsritgerðir Jarðhitanýting Bestun Dælur Orkunotkun Mechanical engineering Geothermal space heating Mathematical optimization Pumping machinery Reinforcement Learning Heating from central stations |
topic_facet | Vélaverkfræði Meistaraprófsritgerðir Jarðhitanýting Bestun Dælur Orkunotkun Mechanical engineering Geothermal space heating Mathematical optimization Pumping machinery Reinforcement Learning Heating from central stations |
url | https://hdl.handle.net/1946/48707 |