Akul Bansal, Simge Küçükyavuz · Feb 05, 2026
Normalization of ReLU dual for cut generation in stochastic mixed-integer programs involves reformulating non-anticipativity constraints using ReLU functions to generate tight, non-convex, and mixed-integer representable cuts. Normalizing the dual in the extended space identifies solutions that yield stronger cuts, which are proven to be tight and Pareto-optimal in the original state space. The proposed approach outperforms existing methods by consistently yielding stronger cuts and reducing solution times on harder instances.
Why This Matters
This paper's proposal to normalize the ReLU dual for cut generation can improve resilience in stochastic mixed-integer programs, which is crucial for power system operators and grid planners, especially when dealing with uncertain renewable energy sources or load forecasts. By yielding stronger cuts, normalization can enhance the accuracy of optimization models used in ISO operations, FERC filings, and NERC standards.
Hongyi Li, Jun Xu, Jinfeng Liu · Feb 05, 2026
The paper proposes a Piecewise Affine (PWA)-based distributed scheme for Model Predictive Control (MPC) that optimizes energy and comfort in large buildings, utilizing Alternating Direction Method of Multipliers (ADMM) for efficient decomposition. The PWA technique is used to handle nonlinear components, and a convex ADMM algorithm transforms the nonconvex problem into smaller convex problems, significantly enhancing computational efficiency. A case study shows that the proposed method reduces execution time by 86% compared to the centralized version.
Why This Matters
The proposed Distributed Model Predictive Control for Energy and Comfort Optimization in Large Buildings can be applied to improve the resilience and efficiency of power grid operations, particularly in managing large-scale buildings with non-linear components, which is relevant to grid operators and utility planners. This method can also be integrated into capacity markets and utility planning processes to optimize energy consumption and comfort levels.
Zhirui Liang, Jae-Won Chung, Mosharaf Chowdhury et al. · Feb 04, 2026
Researchers have developed a framework called "GPU-to-Grid" that couples device-level GPU control with power system objectives to optimize power consumption and regulation. By using batch size as a control knob, they can trade off voltage impacts against inference latency and token throughput, finding that reducing GPU power consumption alleviates lower voltage limit violations while increasing GPU power mitigates upper voltage limit violations. This challenges the common belief that minimizing GPU power consumption is always beneficial to power grids.
Why This Matters
This paper matters for power industry professionals as it provides a novel approach to managing the flexibility of data center loads in distribution systems, enabling more efficient voltage regulation and potentially improving grid resilience during peak periods or when renewable energy sources are intermittent. By optimizing GPU utilization control, this framework can help grid operators better manage the impacts of emerging data centers on their infrastructure and improve overall system stability.