Zhaoxuan Li

Ahmad F. Taha, Nikolaos Gatsis, Bing Dong et al.

This paper puts forth a mathematical framework for Buildings-to-Grid (BtG) integration in smart cities. The framework explicitly couples power grid and building's control actions and operational decisions, and can be utilized by buildings and power grids operators to simultaneously optimize their performance. Simplified dynamics of building clusters and building-integrated power networks with algebraic equations are presented---both operating at different time-scales. A model predictive control (MPC)-based algorithm that formulates the BtG integration and accounts for the time-scale discrepancy is developed. The formulation captures dynamic and algebraic power flow constraints of power networks and is shown to be numerically advantageous. The paper analytically establishes that the BtG integration yields a reduced total system cost in comparison with decoupled designs where grid and building operators determine their controls separately. The developed framework is tested on standard power networks that include thousands of buildings modeled using industrial data. Case studies demonstrate building energy savings and significant frequency regulation, while these findings carry over in network simulations with nonlinear power flows and mismatch in building model parameters. Finally, simulations indicate that the performance does not significantly worsen when there is uncertainty in the forecasted weather and base load conditions.

Keke Lian, Bin Wang, Lei Zhang et al.

The increasing adoption of large language models (LLMs) in software engineering necessitates rigorous security evaluation of their generated code. However, existing benchmarks often lack relevance to real-world AI-assisted programming scenarios, making them inadequate for assessing the practical security risks associated with AI-generated code in production environments. To address this gap, we introduce A.S.E (AI Code Generation Security Evaluation), a repository-level evaluation benchmark designed to closely mirror real-world AI programming tasks, offering a comprehensive and reliable framework for assessing the security of AI-generated code. Our evaluation of leading LLMs on A.S.E reveals several key findings. In particular, current LLMs still struggle with secure coding. The complexity in repository-level scenarios presents challenges for LLMs that typically perform well on snippet-level tasks. Moreover, a larger reasoning budget does not necessarily lead to better code generation. These observations offer valuable insights into the current state of AI code generation and help developers identify the most suitable models for practical tasks. They also lay the groundwork for refining LLMs to generate secure and efficient code in real-world applications.