You are on page 1of 30

Decomposition Approaches for Building Design Optimization

PhD Candidate: Yin Li


Supervisor: Dr. Bruno Lee

Decomposition approaches for building design optimization 02/25/2022 1


1. Introduction
2. Hierarchical Decomposition
3. Parallel Decomposition
4. Sensitivity analysis for dimension reduction
5. Conclusion
1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

1. INTRODUCTION
1.1 Background
• Building design optimization combines simulation with optimization to achieve high performance.

• The dimension increases drastically when more variables are considered for optimization.
• Solving the high-dimensional problem is inefficient and sometimes not technically feasible.

Decomposition approaches for building design optimization 02/25/2022 3


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

1. INTRODUCTION
1.1 Background
• Decomposition divides the original problem into smaller subproblems to reduce the dimensionality.

Best burger in Best burger in


Quebec ……
Ontario

Best burger in Canada

Best burger in
Canada

Decomposition approaches for building design optimization 02/25/2022 4


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

1. INTRODUCTION
1.2 Problem decomposition
• Three types of problem decomposition.

• Decomposition might miss the global optimum.

Local
optimum

Global
optimum

Decomposition approaches for building design optimization 02/25/2022 5


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

1. INTRODUCTION
1.3 This work
Propose two decomposition approaches that can save computation time and achieve global optimum.

Chapter 2 proposes hierarchical decomposition. Appy to the problem minimizing operating costs and emissions.
Chapter 3 proposes parallel decomposition. Apply to the problem minimizing the energy consumptions.
Chapter 4 discusses the feasibility of combining decomposition and variable screening using sensitivity analysis.

Decomposition approaches for building design optimization 02/25/2022 6


2. Hierarchical Decomposition

2.1 Introduction
2.2 Methodology
2.3 Case study
2.4 Results
2.5 Conclusion
1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

2. HIERARCHICAL DECOMPOSITION
2.1 Introduction
• The hierarchical decomposition transforms the single-level optimization into multi-levels.
• To achieve the global optimum, the first-level optimization is required to keep the solutions of original problem for the second-level optimization.

Decomposition approaches for building design optimization 02/25/2022 8


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

2. HIERARCHICAL DECOMPOSITION
2.1 Introduction
2.1.2 Proposed work
• This study finds that this goal can be achieved if the objective functions of first-level optimization are linear-sum of the ones of the
original problem.
• It is proved using Proof By Contradiction

Original problem:
• Minimize i=1, 2, …n
Where

After decomposition:
• First-level: Minimize
• Second-level: Minimize

Decomposition approaches for building design optimization 02/25/2022 9


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

2. HIERARCHICAL DECOMPOSITION
2.2 Methodology
This study applies the hierarchical decomposition to the problem:
• Minimize the Operating Cost and Carbon Emission under different scenarios.
Where and

Applying the hierarchical decomposition:


• First-level: Minimize Energy Consumptions.
• Second-level: Minimize Operating Cost and Carbon Emission under different scenarios.

The benefits:
The proposed approach avoids repeating simulation-based optimization, save computation time, achieve global optimum.

Decomposition approaches for building design optimization 02/25/2022 10


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

2. HIERARCHICAL DECOMPOSITION
2.3 Case study: Four-stories residential building
• Design variables

Variable Value

Building Orientation (0°, 45°, 90º)

RSI Value of Roof [m2K/W] (3, 4, 5)

RSI Value of Wall [m2K/W] (2.5, 3, 3.5, 4)

Window to wall ratio (South wall) (10%, 13%, 16%, 19%, 22%)

U-Value Window [W/m2k] (1, 2, 3, 4)

•   Energy priceCost
andofcarbon
Electricityemission scenarios
Cost of Natural Gas Emission Factor of Electricity Emission Factor of Natural Gas
Units [$/kWh] [$/kWh] [kgCO2eq/kWh] [kgCO2eq/kWh]
Scenario 1 (S1) 0.05 0.05 0.60 0.30

Scenario 2 (S2) 0.07 0.03 0.79 0.18

Scenario 3 (S3) 0.10 0.01 1.00 0.10

Decomposition approaches for building design optimization 02/25/2022 11


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

2. HIERARCHICAL DECOMPOSITION
2.4 Results
Results without decomposition VS with decomposition: Same optimal solutions, save 37% computation time.
Without decomposition With decomposition

Decomposition approaches for building design optimization 02/25/2022 12


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

2. HIERARCHICAL DECOMPOSITION
2.5 Conclusions
This study defines the hierarchical decomposition criterion to achieve global optimum: the first-level optimization should keep the
solution of original problems.

Using proof by contradiction, identify more condition when the criterion is valid and apply to different problems: linear cubic sum
Original problem:
• Minimize
Where , i=1, 2, …n

After decomposition: Will achieve the same optimal solutions


• First-level: Minimize
• Second-level: Minimize

Decomposition approaches for building design optimization 02/25/2022 13


3. Parallel Decomposition

3.1 Introduction
3.2 Development of the parallel decomposition approach
3.3 Case study (1)
3.4 Case study (2)
3.5 Discussion
3.6 Conclusion
1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.1 Introduction
3.1.1 Background
Parallel decomposition
1) Variable grouping
The variable grouping criterion will determine the condition
when the variables should be separated.

2) Solve subproblems
Each subproblem is solved while other variables set to
default values.

3) Combine solutions

Decomposition approaches for building design optimization 02/25/2022 15


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.1 Introduction
3.1.2 Existing work
The existing variable grouping criterion: If there is no interaction between variables.
• This criterion ensures the subproblems generate the global optimal solutions.
• It is not applicable in building design since most of the variables interact.

Decomposition approaches for building design optimization 02/25/2022 16


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.1 Introduction
3.1.3 Proposed work
Propose new criterion as a complement to the existing criterion, which results in a dual-criteria variable grouping
method:
• Global optimum
• Can decompose variables with strong interactions.

Decomposition approaches for building design optimization 02/25/2022 17


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.2 Development of the parallel decomposition approach

Subproblems are solved while variables in other subproblems are


set as defaults.
Decomposition approaches for building design optimization 02/25/2022 18
1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.4 Case study-Low-rise office building

Subproblem 4, only one variable, it is solved while other variables are set as
defaults

Decomposition approaches for building design optimization 02/25/2022 19


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.4 Case study-Low-rise office building
• The optimal solutions using proposed approach are the same as the results of (1) Optimization without decomposition;
(2) Full factorial design ;
• The proposed approach saves 53% computation time.

Decomposition approaches for building design optimization 02/25/2022 20


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.5 Discussion

The subproblems are solved while the other


subproblems are set as defaults.

Will the solutions change while the defaults of


other subproblems change?

Solve subproblems for all scenarios to evaluate.

Decomposition approaches for building design optimization 02/25/2022 21


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.5 Discussion
Five out of seven subproblems stay the same whole the
defaults change.
If the subproblems are solved by main effects from strong to
weak.

Subproblem The optimal solutions are the same


1 Yes
2 Yes
3 Yes
4 Yes
5 Yes
6 Yes
7 Yes

Decomposition approaches for building design optimization 02/25/2022 22


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

3. PARALLEL DECOMPOSITION
3.6 Conclusion
• Introduce the concept of parallel processing in integrated building design.
• Reduce the computation time while achieving global optimum.
Future work:
• Extend the methodology to multi-objective optimization problems.

Decomposition approaches for building design optimization 02/25/2022 23


4. Sensitivity analysis for dimension reduction

4.1 Introduction
4.2 Methodology
4.3 Conclusion
1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

4. SENSITIVITY ANALYSIS FOR DIMENSION REDUCTION


4.1 Introduction
Sensitivity analysis is used for variable grouping in the parallel decomposition. Another common use of it is
variable screening:
• Variable screening (main effects): filter out variables that are not important
• Variable grouping (two-variable interactions): group variables with interactions.

Decomposition approaches for building design optimization 02/25/2022 25


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

4. SENSITIVITY ANALYSIS FOR DIMENSION REDUCTION


4.1 Introduction
Objective: Identify a method can satisfy the needs of both:

Decomposition approaches for building design optimization 02/25/2022 26


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

4. SENSITIVITY ANALYSIS FOR DIMENSION REDUCTION


4.1 Introduction
Method selection criteria:
1. The method can control the number of experiment to achieve different levels of sensitivity.
2. The method does not need to be 100% accurate, but accurate enough.

Decomposition approaches for building design optimization 02/25/2022 27


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

4. SENSITIVITY ANALYSIS FOR DIMENSION REDUCTION


4.2 Methodology
Propose regression-based sensitivity analysis using fractional factorial design as DOE.
Case studies to evaluate the performance of this method for the criteria:
1. The method can control the number of experiment to achieve different levels of sensitivity.
2. The method does not need to be 100% accurate, but accurate enough.
Models Effects Evaluation for

Case 1 Benchmark function Main effects Criterion 1

Case 2 Mid-rise residential building Two-variable interactions Criterion 1

Case 3 Low-rise office building Main effects, two-variable interactions Criterion 2

4.3 Conclusions
The proposed method could be used for both variable screening and grouping.
A dimension reduction methodology for building design optimization problems using variable screening and grouping simultaneously could
be developed

Decomposition approaches for building design optimization 02/25/2022 28


1. Introduction 2. Hierarchical 3. Parallel 4. Sensitivity Analysis 5. Conclusion

5. CONCLUSION
Decomposition saves the computation cost but might miss the global optimum. The proposed decomposition
approaches ensure global optimum while save the computation cost.
Future work: Hybrid decomposition .
Example: the minimization of the cost and CO2

Decomposition approaches for building design optimization 02/25/2022 29


THANK YOU
Yin Li| yin.li.0314@gmail.com

02/25/2022 30

You might also like