Mingyang Sun

Electric vehicles (EVs) are key to transportation decarbonization, yet their battery performance and longevity are vulnerable to temperature extremes, which will be affected by climate change. Battery technology advancements moderate this vulnerability, a dynamic rarely captured in technology assessments under future climates. Here we show that technological advancements have largely mitigated battery lifetime reductions driven by climate change. We combine bottom-up EV simulation and battery degradation models with high-resolution downscaled climate data, capturing warming and variability, for 300 global cities. Under 2 °C warming, old (2010–2018) batteries would experience lifetime declines of 8% (average) and 30% (maximum), whereas new batteries (2019–2023) would experience lifetime declines of 3% (average) and 10% (maximum). New batteries also mitigate regional inequities in battery lifetime reductions driven by climate change. Increasing cell temperature primarily drives lifetime declines. Our findings emphasize climate adaptation co-benefits from technological advancement and increasing thermal resiliency of emerging battery technologies.

Probabilistic wind power forecasting (PWPF) is a critical tool to manage the inherent uncertainty of wind power generation, providing probability predictions essential for economic dispatch and stability assessment. However, the increasing threat of cyberattacks necessitates a comprehensive investigation of the vulnerabilities of PWPFs. This letter introduces a novel adversarial attack framework, termed PDF-Reshape Attack, which targets to manipulate the statistical measures of the probability density functions (PDFs) while maintaining stealthiness by minimizing changes to the expectations. We evaluate the effectiveness of the proposed attack on mainstream PDF distributions (e.g., Gaussian, Beta, Gamma, Weibull) and assess its economic impact on downstream economic dispatch.

Deep reinforcement learning (DRL) has become a promising approach for sequential decision-making in large-scale, complex, and dynamic environments, offering an effective solution for demand-side resource (DSR) control. However, the decision-making process of DRL is highly dependent on the accuracy of its input data, making it vulnerable to data manipulation and perturbations. This vulnerability is exacerbated in DSR systems, where large-scale distributed deployment and reliance on communication networks expose them to widespread cybersecurity threats. In this paper, we propose a robust smoothing adversarial reinforcement learning algorithm to enhance the robustness of DSR control systems under malicious attacks. First, we model the interaction process between attackers and the DSR control center as a partially observable Markov game. Then, we develop a robust adversarial training framework, incorporating randomized smoothing to enhance defense capability against unknown and dynamic attacks. Furthermore, we provide a theoretical robustness guarantee for the proposed algorithm across a wide range of attack scenarios. Finally, comparison studies illustrate and verify that the proposed scheme enhances the robustness of DRL-based DSR control systems under state-altering attacks.

Deep reinforcement learning (DRL) has been implemented across various critical applications, including smart grids, trac management systems, and autonomous vehicles. To safeguard intellectual property and mitigate security vulnerabilities, access to DRL models is typically restricted to a black-box format. is means specic details like the structure of the policy network and optimization processes are not openly available to users. It is crucial to determine if the hyper-parameters can be inferred from observable states and actions within these models, presenting two primary challenges: 1) limited data available from the black-box model and 2) the intertwined eects of hyperparameters on the model's behavior. Since DRL models exhibit varying behaviors in identical tasks depending on their hyper-parameter congurations, we introduce a novel hyper-parameter inference attack against DRL, named HyperInfer, which allows adversaries to deduce the settings of a black-box DRL model. In order to fully assess the risk of model hyper-parameter leakage, we design two novel state generation methods that provoke divergent responses from DRL models. We also develop an inference framework to elucidate the relationship between model behavior and hyper-parameter settings. rough comprehensive experiments involving multiple DRL models and environments, we demonstrate that model behaviors can indeed reveal hyper-parameter settings, with inference accuracy surpassing 90% in scenarios such as PPO with CartPole. We also discuss keyndings relevant to practical applications and explore how knowledge of hyperparameters can facilitate more sophisticated attacks. Lastly, we propose potential defensive strategies to minimize the risk of hyper-parameter leakage in DRL models.