ABSTRACT. Existing research on large language models (LLMs) for automated code compliance has primarily focused on performance, treating the models as black boxes and overlooking how training decisions affect their interpretive behavior. This paper addresses this gap by employing a perturbation-based attribution analysis to compare the interpretive behaviors of LLMs across different fine-tuning strategies such as full fine-tuning (FFT), low-rank adaptation (LoRA) and quantized LoRA fine-tuning, as well as the impact of model scales which include varying LLM parameter sizes. Our results show that FFT produces attribution patterns that are statistically different and more focused than those from parameter-efficient fine-tuning methods. Furthermore, we found that as model scale increases, LLMs develop specific interpretive strategies such as prioritizing numerical constraints and rule identifiers in the building text, albeit with performance gains in semantic similarity of the generated and reference computer-processable rules plateauing for models larger than 7B. This paper provides crucial insights into the explainability of these models, taking a step toward building more transparent LLMs for critical, regulation-based tasks in the Architecture, Engineering, and Construction industry.

ABSTRACT. As technological advancement accelerates, enterprises increasingly invest in smart appliance development, utilizing Internet of Things (IoT) and machine learning to achieve real-time control and energy monitoring of intelligent home devices, thereby enhancing lifestyle convenience and electricity efficiency. According to the International Energy Agency, the buildings and construction sector accounts for over one-third of global energy consumption and nearly 40% of carbon emissions, positioning residential energy conservation as a critical global priority. This research integrates building energy consumption models with household electricity data, employing Python's Flask framework to process extensive datasets with enhanced accuracy and efficiency. TensorFlow facilitates short-term energy consumption forecasting, enabling comprehension of electricity usage trends and behavioral adjustments to achieve conservation objectives. The study aims to realize the vision of Grid-interactive Efficient Buildings (GEB) and Connected Communities (CC). To strengthen prediction applicability, this research introduces neighborhood clustering, grouping households with similar electricity usage patterns to enable peer comparison, thereby enhancing conservation awareness and behavioral supervision. In the context of global transition toward net-zero emissions and intelligent energy management, this study leverages smart meter technology and deep learning methodologies as foundational elements. By reducing energy consumption in residential and commercial buildings, stabilizing electricity demand curves, and promoting renewable energy utilization, we aspire to achieve environmental sustainability and socially comfortable, secure living conditions. The anticipated outcome is a resilient, adaptive energy infrastructure that empowers communities to participate actively in the global sustainability transition while maintaining optimal quality of life and economic prosperity.