使用 GenAI 解码 NOTAM.pdf

1、Decoding NOTAMs with GenAIEnhancing aviation efficiencyAgata Migalska,PhDAI/ML EngineerRyanair LabsFLY AI Forum 202522-23 April 2025BrusselsAIRPORTSAIRPORTS235235FLIGHTSFLIGHTS36003600#(+9COUNTRIESCOUNTRIES3737MILLION PAXMILLION PAX183.7183.7P.A.FY24MILLION PAXMILLION PAX300300/z z%9Z“Driving innova

2、tion,efficiency,and exceptional value across Europes favourite airline”B2515/22 NOTAMN Q)ESAA/QMRLC/IV/NBO/A/000/999/5935N01638E005A)ESOW B)2208040600 C)2208040900 E)RWY 01/19 CLSD DUE MAINT.NOTAM:Notice to AirmenNOTAM:Notice to AirmenB2515/22 NOTAMN Q)ESAA/QMRLC/IV/NBO/A/000/999/5935N01638E005A)ESO

3、W B)2208040600 C)2208040900 E)RWY 01/19 CLSD DUE MAINT.(20%NOTAMS w/o defined Q-code)Q Q-CODESCODES13K+13K+$+(-$(free text)Q Q-CODECODE(structured)NOTAM:Notice to AirmenNOTAM:Notice to Airmen Hard to read Inconsistent Uses heavy abbreviationsThe NOTAM Overload ProblemThe NOTAM Overload ProblemSafety

4、-critical decisions depend on fast and accurate NOTAM processingVVNOTAMS DAILY(Each issuing own NOTAMs)MEMBERSMEMBERS9393ICAO Part 2:Trustworthy Part 2:Trustworthy ClassificationsClassifications Focused on 3 critical cases:Airport Closure,Runway Closure,Firefighting Downgrade Trained a cascade of tr

5、ansformer models to assign alternative Q-codes Step 1:Identify topic(airport,runway,fire)Step 2:Assign correct Q-code within topicA tool for Ops,A tool for Ops,Built for SafetyBuilt for Safety Service now in production,used daily by Ryanair OPS NOTAMs enriched with:Tag Color-coded categorization No

6、filtering,just smart augmentation for faster human comprehensionKey TakeawaysKey Takeaways Problem:Information overload from legacy NOTAM system Solution:Tagging with Gen AI+reclassification with classic ML Benefit:Enhanced human comprehens