Research

Power system outage prediction models remain limited by data fragmentation, interpretability challenges, and operational deployment difficulties. With the rise of large language models (LLMs), we explore their potential to assist utilities in outage analysis and prediction—specifically through retrieval-augmented generation (RAG), which augments the model’s context with retrieved historical records. OutageGPT, a multi-agent RAG framework, is introduced. It integrates a mixture-of-experts architecture and advanced prompting strategies to address diverse outage-related queries. In retrospective 2021 severe-weather test cases, OutageGPT outperformed a state-of-the-art open-source LLM queried directly without retrieval, with ground-truth values more frequently within its predicted ranges due to contextual grounding from historical data. While it posses some limitations, like underestimating extreme events and producing broad prediction intervals, it demonstrates promise while highlighting future needs, including multimodal integration and domain-specific foundation models for energy systems.

National Laboratory of the Rockies, Colorado

Effective contrastive learning relies on choosing useful, descriptive "hard" negative samples. Current hard negative sampling methods designed for traditional artificial neural networks (ANNs) are, in general, computationally costly and are ill-suited to create the temporal patterns relevant for neuromorphic data. To tackle this issue, in this work, we formulate hard negative sampling (HNS) for the energy-efficient paradigm of spiking neural networks (SNNs). We propose a hybrid convolutional-recurrent SNN model that is trained utilizing a supervised contrastive (SupCon) loss. Furthermore, we incorporate a spike regularization term that promotes computational sparsity. This approach learns a highly discriminative, computationally efficient embedding space from temporal data, as evidenced by our experiments with the Neuromorphic-MNIST (N-MNIST) dataset. Our model significantly outperforms baselines with respect to both classification accuracy and computation efficiency, without requiring complex memory banks or implicit mining protocols.

Rochester Institute of Technology, New York

The project presents integrating deep learning methods along with Kalman filter preprocessing yielding better accuracy in fault detection in industrial pumps systems. By utilizing ITT Gould Pumps sensor data the execution runs a simulation for normal operation and range of frequent failure modes like cavitation, misalignment, soft footing, and dry running. Real-time collection of vibration and flux signals is shown in the time-waveform and FFT plots with a split-screen observing that contrasts the raw sensor inputs and the Kalman-filtered data. The visual representations reflect the filter's capability to stabilize during fault conditions and measure noise. A model of deep learning checks both of the data streams and the side-by-side prediction results exhibit the performance enhancements attained through noise reduction. In summary, the project illustrates that when classical state estimation is combined with modern AI approaches their application in predictive maintenance is stronger because it leads to higher sensor reliability and diagnostic accuracy.

Rochester Institute of Technology, New York

Electrical energy is the backbone of modern industry and an essential tool for modern life. Because of the growing demand for energy and the stress on conventional energy resources with undesirable impact on environment, the industry has been urged to accelerate the researches on alternative energy resources. Among the available alternative energies, Solar or Photovoltaic (PV) energy is one of the most promising renewable energies, which are freely available and environment friendly. PV equipment generates electricity without any gas emissions and its operation is virtually silent. Its construction as a stand-alone system can provide a large power supply for remote areas, whereas grid-connected PV systems are still quite expensive. PV power systems are now widely being installed around the world and the demand of such power is increasing every year.

National Aerospace Laboratories, Bengaluru, India