POTracker: Optimizing Large Language Models for Standard-Compliant Power Outage Report Generation

The article introduces POTracker, an optimized large language model specifically designed for generating power outage reports that comply with regulatory standards in the U.S. This model fine-tunes Qwen2.5-7B-Instruct using a novel loss function, POTrackerLoss, which emphasizes both textual and structural similarity, resulting in a 51% improvement in overall accuracy and achieving 86.47% structural accuracy on a dataset of 1,000 reports. This advancement is significant for practitioners as it enhances the capability of LLMs to produce domain-specific outputs that meet strict formatting requirements, thereby facilitating better interoperability in the energy sector.