Influence of Uncertainty in User Behaviors on the Simulation-Based Building Energy Optimization Process and Robust Decision-Making

Abstract

Computer-based simulations have been widely used to predict building performances. Building energy simulation tools are generally used to perform parametric studies. However, the building is a complex system with a great number of variables. This leads to a very high computational cost. Therefore, using a building optimization algorithm coupled with an energy simulation tool is a more promising solution. In this study, EnergyPlus is connected to a genetic algorithm that uses a probabilistic search technique based on evolutionary principles. Various sources of uncertainty exist in simulation-based building optimization problems. This study aims to investigate the influence of occupant behavior-related input variables on the optimization process. To integrate the uncertainty into the optimization process, a stochastic approach using the Latin hypercube sampling (LHS) method is employed. The varying input variables are defined by the LHS method, and each sampling run generates 14 samples. Five optimization parameters are used, and the recommendations for parameter settings of each parameter are generated as the optimization result. It is important to provide a decision maker with a decision-making framework to support robust decision-making from the generated recommendations. A clear or relatively clear tendency of recommendations toward a particular parameter setting is observed for three parameters. For these three parameters, the frequency of recommendation is identified to be a good indicator for the robustness of the most recommended setting. The test of proportion is performed to investigate the statistical significance between parameter settings. For the other two parameters, recommendations are comparatively evenly distributed among parameter settings, and the statistical significance is not shown. In this case, the Hurwicz decision rule is utilized to select an optimal solution. This dissertation contributes to the field of building optimization as it proposes a method to integrate uncertainty in input variables and shows the method generates reliable results. Computational time is reduced by using the LHS method compared to the case of using a random sampling method. While this study does not include all potential input variables with uncertainties, it provides significant insight into the role of input variables with uncertainty in the building optimization process.