Small language models improve power outage report accuracy with risk training
Enhancing Small Language Models for Power Outage Report Generation via Minimum Risk Training
Artificial Intelligence
Summary
Power outages can generate messy reports that are hard to use. The authors improved a smaller language model so it better converts these reports into a neat, standardized format. They used a method called Minimum Risk Training, which helps the model focus on producing better overall reports, not just guessing one word at a time. This made the model much more accurate when creating these structured reports.
What this means in practice
- •For utility data teams: Automatically convert diverse power outage reports into standardized structured formats for easier analysis and sharing.
- •For energy management software developers: Integrate advanced small language models to improve accuracy in generating compliant outage documentation, reducing manual correction.
Authors
Hung Phan, Waqwoya Abebe, Youssef Hussein, Supriya Chinthavali, Dalton Lunga, Ali Jannesari
Abstract
Minimum Risk Training (MRT) enables neural machine translation models to directly optimize sequence-level evaluation metrics instead of relying only on token- level maximum-likelihood objectives Shen et al. [2016]. Although introduced a decade ago, recent work shows renewed potential for risk-based optimization in modern language models Yang et al. [2024], Jinnai et al. [2025]. We apply MRT to power outage report generation for the Outage Data Initiative Nationwide (ODIN), transforming heterogeneous reports into standardized XML compliant with CIM IEC 61968-3. Our MRT approach improves Qwen2.5-7B-Instruct overall accuracy from 16.20% to 68.95%, demonstrating the effectiveness of sequence- level optimization for domain-specific structured generation