國立高雄大學圖資館 |

Advanced Solver Development for Large-scale Dynamic Building System Simulation.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Advanced Solver Development for Large-scale Dynamic Building System Simulation.
作者:	Chen, Zhelun .
出版者:	Ann Arbor : ProQuest Dissertations & Theses, 2019
面頁冊數:	282 p.
附註:	Source: Dissertations Abstracts International, Volume: 81-04, Section: B.
附註:	Advisor: Wen, Jin.
Contained By:	Dissertations Abstracts International81-04B.
標題:	Architectural engineering.
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27540222
ISBN:	9781687980861

Advanced Solver Development for Large-scale Dynamic Building System Simulation.
Chen, Zhelun .

Advanced Solver Development for Large-scale Dynamic Building System Simulation. - Ann Arbor : ProQuest Dissertations & Theses, 2019 - 282 p.

Source: Dissertations Abstracts International, Volume: 81-04, Section: B.

Thesis (Ph.D.)--Drexel University, 2019.

This item must not be sold to any third party vendors.

Efficiently, robustly and accurately solving large and sparse nonlinear algebraic and differential equation system for dynamic building simulation is becoming more and more essential due to increasing demands to simulate large-scale problems for multiple buildings coupled with various levels of strength either through the smart grid or other means, such as district heating/cooling and shared distributed energy resources.This study is interested in advancing solving techniques that either improve the quality and efficiency of a dynamic building simulation model generically or improve the performance of the underlying equation solver. Nowadays, many commonly used tools for dynamic building system simulation still employ direct Newton methods. These methods are not only lack of convergence for stiff problems or cold starts, but also fail to meet the increased memory requirements associated with large-scale problems or more specific issues that arise in problems where the nonlinear equations resulted from the discretization of an underlying engineering differential equation. Therefore, a Newton-Krylov method that satisfies the computational need for large-scale dynamic building system simulation is investigated. An ideal preconditioner and an automatic update scheme are employed to ensure fast and robust simulation by way of the Newton-Krylov method. In addition to the comparison study focuses on the numerical solution methods, a generic function smoothing technique for the rare occasion that discontinuous functions are encountered is also investigated.Four testbeds, namely, 4Z5B, 4Z1B, 12Z5B, and 40Z5B, are developed in an HVACSIM+ environment to evaluate the advancement techniques. All testbeds simulate the airflow and thermal behaviour of building zones (from four zones, 4Z, to forty zones, 40Z) that are served by air handling unit (AHU) and variable air volume (VAV) systems. 4Z5B and 4Z1B testbeds simulate the same building system with the same number of equations but with different equation groupings while 4Z5B. 12Z5B and 40Z5B testbeds have the same equation grouping but are corresponding to very different building system sizes (four, twelve, and forty zones, respectively) and therefore different numbers of equations to be solved.The following tasks are completed and summarized in this report:(1) Develop numerical testbeds to evaluate solution methods and techniques. (2) Investigate potential numerical issues in a typical dynamic building system simulation model and seek generic techniques to improve the quality of the model. (3) Examine the performance of a Newton-Krylov method on solving dynamic building system simulation equations. (4) Improve the performance of the Newton-Krylov method by developing and employing proper preconditioning techniques. (5) Investigate potential strategies to construct physics-based preconditioners. (6) Investigate the impact of finite difference step size in Jacobian approximation on the performance of dynamic building system simulation.The major numerical issue found in the testbeds mentioned above is the discontinuity of the simple coil component model. A generic smoothing technique is employed to improve the performance of the discontinuous simple coil component model, and the smoothed model results in a more stable and more accurate solution. A Newton-Krylov method is employed to increase the computational speed of a large-scale simulation. However, the direct implementation of the Newton-Krylov method results in stability issues. Therefore, a preconditioned Newton-Krylov method that employs the ideal preconditioner and an automatic update scheme is developed in this study, referred to as INB-PSGMRES(m). This method performs as robust as the default Powell’s Hybrid (PH) method in HVACSIM+ while saving a significant amount of computational time. Its computational time saving against the PH method is at least 49.7%, 91.8%, 88.7%, and 97.1% for 4Z5B, 4Z1B, 12Z5B, and 40Z5B testbeds, respectively. It is found that because of the employment of preconditioning, two important parameters of the INB-PSGMRES(m) method, i.e., the forcing term and the restarting parameter, have little impact on the simulation performance. A few potential partitioning strategies for developing a physics-based preconditioner are investigated. Due to the strong coupling of mass flow rates and pressures between each nodal point of the airflow network system, it is difficult to construct an effective physics-based preconditioner for the airflow network of an AHU-VAV system. On the other hand, the thermal network can be effectively exploited. A preconditioner that targets the coil related equations is found effective at reducing the condition number of the Jacobian (which typically leads to fast linear convergence in a Krylov method) due to the high nonlinearity of the coil component model and its strong impact on the temperature and humidity in the HVAC system. Four finite difference step sizes for the Jacobian approximation and four finite difference step sizes for the Jacobian-vector approximation are investigated. For the Jacobian approximation, the current finite difference step size employed by HVACSIM+ is effective for the operating period. Its performance can be improved for the nonoperating period by adding a lower bound to the finite difference step size.

ISBN: 9781687980861Subjects--Topical Terms:

857338
Architectural engineering.

Advanced Solver Development for Large-scale Dynamic Building System Simulation.
LDR:06360nmm a2200313 4500 001 570813
005 20200514112006.5
008 200901s2019 ||||||||||||||||| ||eng d
020 $a 9781687980861
035 $a (MiAaPQ)AAI27540222
035 $a AAI27540222
040 $a MiAaPQ $c MiAaPQ
100 1 $a Chen, Zhelun . $3 857533
245 1 0 $a Advanced Solver Development for Large-scale Dynamic Building System Simulation.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2019
300 $a 282 p.
500 $a Source: Dissertations Abstracts International, Volume: 81-04, Section: B.
500 $a Advisor: Wen, Jin.
502 $a Thesis (Ph.D.)--Drexel University, 2019.
506 $a This item must not be sold to any third party vendors.
520 $a Efficiently, robustly and accurately solving large and sparse nonlinear algebraic and differential equation system for dynamic building simulation is becoming more and more essential due to increasing demands to simulate large-scale problems for multiple buildings coupled with various levels of strength either through the smart grid or other means, such as district heating/cooling and shared distributed energy resources.This study is interested in advancing solving techniques that either improve the quality and efficiency of a dynamic building simulation model generically or improve the performance of the underlying equation solver. Nowadays, many commonly used tools for dynamic building system simulation still employ direct Newton methods. These methods are not only lack of convergence for stiff problems or cold starts, but also fail to meet the increased memory requirements associated with large-scale problems or more specific issues that arise in problems where the nonlinear equations resulted from the discretization of an underlying engineering differential equation. Therefore, a Newton-Krylov method that satisfies the computational need for large-scale dynamic building system simulation is investigated. An ideal preconditioner and an automatic update scheme are employed to ensure fast and robust simulation by way of the Newton-Krylov method. In addition to the comparison study focuses on the numerical solution methods, a generic function smoothing technique for the rare occasion that discontinuous functions are encountered is also investigated.Four testbeds, namely, 4Z5B, 4Z1B, 12Z5B, and 40Z5B, are developed in an HVACSIM+ environment to evaluate the advancement techniques. All testbeds simulate the airflow and thermal behaviour of building zones (from four zones, 4Z, to forty zones, 40Z) that are served by air handling unit (AHU) and variable air volume (VAV) systems. 4Z5B and 4Z1B testbeds simulate the same building system with the same number of equations but with different equation groupings while 4Z5B. 12Z5B and 40Z5B testbeds have the same equation grouping but are corresponding to very different building system sizes (four, twelve, and forty zones, respectively) and therefore different numbers of equations to be solved.The following tasks are completed and summarized in this report:(1) Develop numerical testbeds to evaluate solution methods and techniques. (2) Investigate potential numerical issues in a typical dynamic building system simulation model and seek generic techniques to improve the quality of the model. (3) Examine the performance of a Newton-Krylov method on solving dynamic building system simulation equations. (4) Improve the performance of the Newton-Krylov method by developing and employing proper preconditioning techniques. (5) Investigate potential strategies to construct physics-based preconditioners. (6) Investigate the impact of finite difference step size in Jacobian approximation on the performance of dynamic building system simulation.The major numerical issue found in the testbeds mentioned above is the discontinuity of the simple coil component model. A generic smoothing technique is employed to improve the performance of the discontinuous simple coil component model, and the smoothed model results in a more stable and more accurate solution. A Newton-Krylov method is employed to increase the computational speed of a large-scale simulation. However, the direct implementation of the Newton-Krylov method results in stability issues. Therefore, a preconditioned Newton-Krylov method that employs the ideal preconditioner and an automatic update scheme is developed in this study, referred to as INB-PSGMRES(m). This method performs as robust as the default Powell’s Hybrid (PH) method in HVACSIM+ while saving a significant amount of computational time. Its computational time saving against the PH method is at least 49.7%, 91.8%, 88.7%, and 97.1% for 4Z5B, 4Z1B, 12Z5B, and 40Z5B testbeds, respectively. It is found that because of the employment of preconditioning, two important parameters of the INB-PSGMRES(m) method, i.e., the forcing term and the restarting parameter, have little impact on the simulation performance. A few potential partitioning strategies for developing a physics-based preconditioner are investigated. Due to the strong coupling of mass flow rates and pressures between each nodal point of the airflow network system, it is difficult to construct an effective physics-based preconditioner for the airflow network of an AHU-VAV system. On the other hand, the thermal network can be effectively exploited. A preconditioner that targets the coil related equations is found effective at reducing the condition number of the Jacobian (which typically leads to fast linear convergence in a Krylov method) due to the high nonlinearity of the coil component model and its strong impact on the temperature and humidity in the HVAC system. Four finite difference step sizes for the Jacobian approximation and four finite difference step sizes for the Jacobian-vector approximation are investigated. For the Jacobian approximation, the current finite difference step size employed by HVACSIM+ is effective for the operating period. Its performance can be improved for the nonoperating period by adding a lower bound to the finite difference step size.
590 $a School code: 0065.
650 4 $a Architectural engineering. $3 857338
650 4 $a Applied mathematics. $3 377601
650 4 $a Mechanical engineering. $3 190348
690 $a 0462
690 $a 0364
690 $a 0548
710 2 $a Drexel University. $b Architectural Engineering (College of Engineering). $3 857534
773 0 $t Dissertations Abstracts International $g 81-04B.
790 $a 0065
791 $a Ph.D.
792 $a 2019
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27540222