Yuhong Liu

Bidirectional electric vehicles (EVs) are increasingly proposed as distributed energy storage to mitigate grid stress, yet their performance is typically evaluated using metrics such as peak demand reduction and consumer cost savings that are weakly coupled to system-level conditions. As a result, strategies that appear beneficial at the household level may have negligible impacts on the grid. This study introduces a grid-aware evaluation framework centered on an Aggregate Grid Impact (AGI) metric, which quantifies the alignment between behind-the-meter load shifting and real-world grid demand. Unlike conventional metrics, AGI captures whether shifted loads coincide with periods of high system stress, enabling consistent comparison of electricity rate designs. The framework is implemented using a vehicle-to-home (V2H) simulation platform with stochastic, regionally transferable models of driving behavior and building energy use. A case study of 100 households in a high-demand residential region compares unmanaged charging with optimized bidirectional charging under multiple electricity rates. Results show that an existing EV rate yields minimal improvement in AGI (~5%), long consumer payback periods (~9 years), and reduced utility profits (~21%), despite reducing peak demand. In contrast, a real-time pricing structure aligned with wholesale signals reduces AGI by up to 40% and significantly improves economic outcomes. These findings demonstrate that commonly used evaluation metrics can misrepresent the system-level effectiveness of load shifting strategies and highlight the need for grid-aligned performance measures in electricity tariff design.

The performance of modern machine learning systems depends on access to large, high-quality datasets, often sourced from user-generated content or proprietary, domain-specific corpora. However, these rich datasets inherently contain sensitive personal information, raising significant concerns about privacy, data security, and compliance with regulatory frameworks. While conventional anonymization techniques can remove explicit identifiers, such removal may result in performance drop in downstream machine learning tasks. More importantly, simple anonymization may not be effective against inference attacks that exploit implicit signals such as writing style, topical focus, or demographic cues, highlighting the need for more robust privacy safeguards during model training. To address the challenging issue of balancing user privacy and data utility, we propose a reinforcement learning framework that fine-tunes a large language model (LLM) using a composite reward function that jointly optimizes for explicit and implicit privacy, semantic fidelity, and output diversity. To effectively capture population level regularities, the privacy reward combines semantic cues with structural patterns derived from a minimum spanning tree (MST) over latent representations. By modeling these privacy-sensitive signals in their distributional context, the proposed approach guides the model to generate synthetic rewrites that preserve utility while mitigating privacy risks. Empirical results show that the proposed method significantly enhances author obfuscation and privacy metrics without degrading semantic quality, providing a scalable and model-agnostic solution for privacy preserving data generation in the era of large language models.