The condition of current building stock in the United States raises the question of whether the energy performance of existing buildings can ever be environmentally sustainable. In the United States, buildings accounted for 39% of total energy consumption and 72% of total electricity consumption (USEPA 2009). In addition, current building energy use is projected to increase by 1.7% annually until 2025 (J.D. Ryan 2004). The great potential for energy reduction in existing buildings has created opportunities in building energy retrofit projects (Noris et al. 2013). A building renovation project must not only be affordable, taking into account factors such as investor budgets, payback period, economic risks and uncertainties, but also create a thermally comfortable indoor environment and is sustainable through its lifetime. The research objective of this dissertation is to develop a novel method to optimize the performance of buildings during their post-retrofit period in the future climate. The dissertation is organized in three sections: a) Develop a data-driven method for the hourly projection of energy use in the coming years, taking into account global climate change (GCC). Using machine learning algorithms, a validated data-driven model is used to predict the building’s future hourly energy use based on simulation results generated by future extreme year weather data and it is demonstrated that GCC will change the optimal solution of future energy conservation measure (ECM) combination. b) Develop a simplified building performance simulation tool based on a dynamic hourly simulation algorithm taking into account the thermal flux among zones. The tool named SimBldPy is tested on EnergyPlus models with DOE reference buildings. Its performance and fidelity in simulating hourly energy use with different heating and cooling set points in each zone, under various climate conditions, and with multiple ECMs being applied to the building, has been validated. This tool and modeling method could be used for rapid modeling and assessment of building energy for a variety of ECM options. c) Use a non-dominated sorting technique to complete the multi-objective optimization task and design a schema to visualize optimization results and support the decision-making process after obtaining the multi-objective optimization results. By introducing the simplified hourly simulation model and the random forest (RF) models as a substitute for traditional energy simulation tools in objective function assessment, certain deep retrofit problem can be quickly optimized. Generated non-dominated solutions are rendered and displayed by a layered schema using agglomerative hierarchical clustering technique. The optimization method is then implemented on a Penn campus building for case study, and twenty out of a thousand retrofit plans can be recommended using the proposed decision-making method. The proposed decision making support framework is demonstrated by its robustness to the problem of deep retrofit optimization and is able to provide support for brainstorming and enumerate various possibilities during the process of making the decision.