The MLMs development has four main phases, which is defined for each model separately, i.e., load normalized data, MLMs development, training, testing, and validation, and finally deploying the models. Two different MLMs have been developed in this study, i.e., ANN and VAE. The most important phases in MLM development are selecting the network architecture, training, and validating the model’s performance.
3.5.1 Surrogate ANN Models (Module 2)
The ANNs have been developed using the results of the previous parts. The surrogate ANNs have been developed for selecting near-optimal building energy renovation methods considering TEC, LCC and LCA pairwise. The proposed model will be used to predict TEC, LCC and LCA of the potential renovation scenarios of existing institutional buildings as shown in Figure 3-3. The ANNs
will be used as surrogate models for emulating computationally expensive, real building optimization models. The effectiveness of the proposed method will be examined using Mean Squared Error (MSE).
3.5.2 Generative VAE Models (Module 3)
The second MLM in this study develops generative VAEs to generate different renovation scenarios for building envelope, HVAC, and lighting system considering TEC and LCC as shown in Figure 2-11. The developed semi-supervised VAEs learn the inner data structure by discovering unlabeled data and utilize labeled data for fine-tuning, better discrimination and accurate classification. Therefore, the use of unlabeled and labeled data for semi-supervised training can be considered as an advantage of this method over traditional ANN. The performance of the developed model is demonstrated using a simulated renovation dataset to prove its potential. The effectiveness of the proposed method is examined using two validation methods, i.e., MSE (internal validation) and validation of results using DesignBuilder as BEM (external validation).
Figure 3-3. Artificial Neural Network architecture.
R: Roof
EW: External Wall W: Window FT: Façade Type
WWR: Window to Wall Ratio HVAC: HVAC system
COS: Cooling Operation Schedule HOS: Heating Operation Schedule Li: Lighting system
EWO: External Window Open
Feedforward Multi-Layer Perceptron (MLP) ANN
TEC vs. LCC ANN1
ANN2
TEC vs. LCA
3.6 Summary
This chapter presented an overview of the proposed framework. This research consists of four main components that are necessary to realize the proposed methodology: (1) developing a framework for data collection and preparation to define the renovation strategies; (2) proposing SBMO model to define near-optimal renovation scenarios based on the available methods; (3) applying data processing methods to reduce the effects of magnitude and range of variations of the input variables throughout the MLMs training process and remove the inconsistencies of different attributes; and (4) developing two surrogate MLMs by learning from the generated SBMO datasets and reducing the computing time while achieving acceptable accuracy.
Initially SBMO model has been developed for renovation of existing buildings envelope, HVAC, and lighting considering TEC, LCC, and LCA. SBMO model uses NSGA-II optimization and simulation tools simultaneously to create feasible renovation scenarios including Pareto Front results (as explained in Chapter 4). Consequently, two MLMs have been developed using the results of SBMO. ANNs have been used to predict TEC vs. LCC (ANN1) and TEC vs. LCA (ANN2) for different building energy renovation scenarios (as explained in Chapter 5). VAEs have been used to generate feasible renovation scenarios considering TEC and LCC (VAE1), TEC (VAE2), and LCC (VAE3) (as explained in Chapter 6). The proposed MLMs will be used to: (1) predict the energy performance, LCC and LCA of the potential renovation scenarios for existing institutional buildings using surrogate ANNs, and (2) develop a DNN to generate different renovation scenarios for building envelope, HVAC, and lighting system considering TEC and LCC. The main advantage of these models is to improve the computing time while achieving acceptable accuracy.
SIMULATION-BASED MULTI-OBJECTIVE BUILDING RENOVATION OPTIMIZATION CONSIDERING TEC, LCC, AND LCA
4.1 Introduction
As mentioned in Section 2.1, it is necessary to reduce the energy consumption of buildings by improving the design of new buildings or renovating existing buildings. Heat losses or gains through building envelopes affect the energy used and the indoor conditions. Renovating building envelopes and energy consuming systems to lessen energy losses is usually expensive and has a long payback period. Despite the significant contribution of research on optimizing energy consumption, there is limited research focusing on the renovation of existing buildings to minimize their LCC and environmental impact using LCA. This chapter aims to find the optimal scenario for the renovation of buildings considering TEC and LCA while providing an efficient method to deal with the limited renovation budget considering LCC. Different scenarios can be compared in a building renovation strategy to improve energy efficiency. Each scenario considers several methods including the improvement of the building envelopes, HVAC and lighting systems. However, some of these scenarios could be inconsistent and should be eliminated. Another consideration in this research is the appropriate coupling of renovation scenarios. For example, the HVAC system must be redesigned when renovating the building envelope to account for the reduced energy demand and to avoid undesirable side effects. An efficient GA method, coupled with a simulation tool, is used for simultaneously minimizing the energy consumption, LCC, and environmental impact of a building. A case study is developed to demonstrate the feasibility of the proposed method.
Chapter 4 is organized into sections that include the proposed methodology (Section 4.2), implementation and case study (Section 4.3), and finally, summary and conclusions (Section 4.4).