Evaluating a building's performance and comfort after it has been occupied is crucial for identifying issues such as high energy consumption, particularly in older structures built before 1970s energy crises, like those found on university campuses. This study is part of a broader Post-Occupancy Evaluation (POE) project assessing four major academic buildings across Texas Tech University. It intends to enhance their performance, energy efficiency, and occupants' comfort by using retrofit strategies and building energy modeling. Data on actual energy usage, construction details, operating schedules, and equipment was collected using spot and long-term measurement techniques to evaluate building performance and user comfort. Building energy models were developed using the Integrated Environment Solution- Virtual Environment (IESVE). Simulations explored potential retrofits, including passive design strategies and renewable energy options such as set-back for HVAC systems, double-glazing windows, LED lighting, and cool roofs. The results showed that specific retrofits, such as double-glazing windows, yielded 15% and 18% energy reduction in two buildings, achieving Energy Use Intensity (EUI) values of 133 and 125, respectively, in two buildings. Furthermore, combining different retrofits could lead to significant annual reductions in energy consumption of 15.32%, 19.53%, 17.52%, and 9.09% for each of the four buildings in this study. These findings highlight the value of POE in identifying opportunities for energy savings and performance improvements in academic settings, providing a framework for broader application in other educational environments.

Energy benchmarking is essential for policymakers, building owners, and architects to monitor energy consumption and implement conservation strategies. However, challenges persist, including limited data availability for annual and monthly energy usage, the complexity of machine learning model interpretation, and difficulties in classifying buildings based on energy use patterns. This study proposes a data-driven energy benchmarking framework utilizing white-box and grey-box AI models at both annual and monthly scales, focusing on building energy data from Washington, DC. The first stage applies an unsupervised K-Means clustering algorithm to categorize buildings into four distinct groups based on monthly energy usage patterns, providing insights into their energy consumption characteristics. Subsequently, two predictive models are employed: a white-box multi-linear regression (MLR) model and a grey-box LightGBM (LGBM) model, to estimate building energy consumption at both annual and monthly levels. Sensitivity analysis is conducted to assess the influence of various building attributes, such as building type and weather conditions, on energy usage and carbon emissions. Results indicate that the LGBM model outperforms MLR in predictive accuracy, though both models exhibit similar sensitivity to key attributes. Notably, Energy Star^® ratings, building type, weather conditions, and building area emerge as the most significant factors influencing energy consumption. This research establishes a replicable classification framework and demonstrates the value of combining annual and monthly benchmarking for enhanced AI-driven energy analysis. The proposed methodology can be generalized to cities without existing benchmarking programs, supporting broader efforts toward decarbonization and climate resilience. By advancing AI-based energy benchmarking techniques, this study contributes to achieving net-zero carbon emissions by 2050.

This research examines the feasibility of Artificial Intelligence (AI) in achieving contextual coherence in architectural design, identifying limitations and proposing a future framework. While AI, particularly through Generative Adversarial Networks (GANs), theoretically promises contextually coherent designs, this goal remains largely unachieved. Conventional computational methods, while effective at process optimization using linear evolutionary and deterministic algorithms, lack the capacity to replicate human design intelligence. They often fail to integrate critical environmental and regulatory factors, limiting their ability to produce meaningful architectural outcomes.

The study investigates whether current AI capabilities can achieve contextual coherence comparable to human designers. Contextual coherence is defined as a design’s ability to harmonize with its surroundings (3D physical objects), environmental factors (e.g., solar radiation), and local building codes. To address these challenges, the research identifies gaps in existing synthetic intelligence frameworks and proposes new developments leveraging contextual data and advanced learning algorithms. This approach positions AI as a collaborative design partner rather than merely a tool for process optimization.

Methodologically, the study deconstructs synthetic intelligence frameworks through an extensive literature review and conceptualizes a metric-based pathway to qualitatively measure contextual coherence. Computational experiments using evolutionary algorithms and existing AI models evaluate their ability to integrate contextual factors such as the 3D objectivity of design elements, solar radiation, and building code compliance. The proposed framework combines quantitative (e.g., climate, site conditions) and qualitative (e.g., spatial relationships) data through iterative feedback loops to refine outputs for contextual alignment.

Preliminary findings highlight AI’s strengths in novel object generation and its ability to handle individual or multiple contextual factors. However, challenges persist in holistically integrating all factors. This study positions AI as a design collaborator, enhancing human creativity while ensuring contextual relevance and unlocking new possibilities for architectural innovation.