Searcharxiv⌕ Search

arXiv subjects

Prasad Enjeti

Publications and source records attributed to Prasad Enjeti.

6 recordsLinked to original sources

Low-Voltage Ride-Through in AI Data Centers via Coordinated Battery and Computing Load Control

The ability of large AI data centers to ride through grid voltage sags is critical to power system stability and reliability. This paper frames data center low-voltage ride-through (LVRT) as a DC-link power balance problem. Once the rectifier reaches its current limit, the power drawn from the grid decreases in proportion to the grid voltage. The DC-link capacitor can supply the resulting power deficit only briefly before the DC-link voltage falls low enough to trigger undervoltage protection tripping. To address this issue, this paper proposes effective control strategies for coordinating uninterruptible power supply (UPS) battery and computing load curtailment to offset the power deficit and enable LVRT at minimum control cost. Three control scenarios are studied: UPS battery support alone, computing load curtailment alone, and the coordinated use of both resources. Electromagnetic transient (EMT) simulations using the TAMU-ERCOT AI data center model demonstrate that battery support and the coordinated strategy achieve LVRT for all tested sags, including a near-zero-voltage fault, while load curtailment alone rides through moderate voltage sags.

eess.SY↗

How AI Load Swings Reduce Datacenter Power Conversion Efficiency

AI training and inference workloads make GPU power swing repeatedly between compute and communication phases. As a result, the cascaded converters in a datacenter rack keep moving between operating points with different efficiencies. This paper uses an energy based, cycle averaged method to quantify that effect along a 240 V AC to 400 V DC to 51 V DC rack power path, using measured efficiency curves of commercially available GaN based PFC and LLC converters. For a 1 Hz, 50% duty-cycle load swinging between 5 kW and 510 W, conversion efficiency falls to 95.97%, compared with 96.82% at the equivalent constant average load, a loss of 0.88%. When 51/48V to 1V point of load conversion to the chip rail is included, it is shown that the loss grows to 1.66% (89.45% vs. 90.96%). These results show that efficiency rated at average load overstates real world performance, and that the penalty grows with each load dependent conversion stage between the grid and the GPU.

cs.AR↗

Predicting DC-Link Capacitor Current Ripple in AC-DC Rectifier Circuits Using Fine-Tuned Large Language Models

Foundational Large Language Models (LLMs) such as GPT-3.5-turbo allow users to refine the model based on newer information, known as ``fine-tuning''. This paper leverages this ability to analyze AC-DC converter behaviors, focusing on the ripple current in DC-link capacitors. Capacitors degrade faster under high ripple currents, complicating life monitoring and necessitating preemptive replacements. Using minimal invasive noisy hardware measurements from a full bridge rectifier and 90W Power Factor Correction (PFC) boost converter, an LLM-based models to predict ripple content in DC-link currents was developed which demonstrated the LLMs' ability for near-accurate predictions. This study also highlights data requirements for precise nonlinear power electronic circuit parameter predictions to predict component degradation without any additional sensors. Furthermore, the proposed framework could be extended to any non-linear function mapping problem as well as estimating the capacitor Equivalent Series Resistance (ESR).

eess.SY↗

Electromagnetic Transient Model of Cryptocurrency Mining Loads for Low-Voltage Ride Through Assessment in Transmission Grids

In this paper, we developed an Electromagnetic Transient (EMT) model tailored for large cryptocurrency mining loads to understand the cross-interaction of these loads with the electric grid. The load model has been built using Electromagnetic Transients Program (EMTP) software. We have cross-validated the performance of the EMT model of the load with commercial application-specific integrated circuit miners, typically used by large-scale mining facilities, by comparing their low-voltage ride-through (LVRT) capabilities. Subsequently, LVRT capabilities of the large-scale miners have been tested against various fault scenarios both within the miner's remote facility as well as at one of the distant buses of the interconnected grid. The significance of this model lies in its scalability to accommodate larger blocks of mining loads and its seamless integration into a larger electric grid.

eess.SY↗

Detection of Cyber Attacks in Renewable-rich Microgrids Using Dynamic Watermarking

This paper presents the first demonstration of using an active mechanism to defend renewable-rich microgrids against cyber attacks. Cyber vulnerability of the renewable-rich microgrids is identified. The defense mechanism based on dynamic watermarking is proposed for detecting cyber anomalies in microgrids. The proposed mechanism is easily implementable and it has theoretically provable performance in term of detecting cyber attacks. The effectiveness of the proposed mechanism is tested and validated in a renewable-rich microgrid.

eess.SY↗

Information theoretically secure, enhanced Johnson noise based key distribution over the smart grid with switched filters

We introduce a protocol with a reconfigurable filter system to create non-overlapping single loops in the smart power grid for the realization of the Kirchhoff-Law-Johnson-(like)-Noise secure key distribution system. The protocol is valid for one-dimensional radial networks (chain-like power line) which are typical of the electricity distribution network between the utility and the customer. The speed of the protocol (the number of steps needed) versus grid size is analyzed. When properly generalized, such a system has the potential to achieve unconditionally secure key distribution over the smart power grid of arbitrary geometrical dimensions.

cs.CR↗