AI-driven Optimization of Operational NOTAM Management

One of the biggest problems affecting the Notice to Air Missions (NOTAM) systems at Air Navigation Service Providers (ANSPs) around the world is the achievement of good data quality or often the lack of data quality whatsoever. Fundamentally, NOTAMs are free text data objects, without a prescribed data structure but rather conventions for syntax and semantics. This leads to the significant risk that data inconsistencies are introduced during the data lifecycle starting directly with the NOTAM origination. NOTAM solutions introduce business rule-based data quality enforcement by checking and, if necessary, correcting errors in NOTAMs. This is both time-consuming and resource-intensive, and, as such, a major source of effort and costs for Aeronautical Information Management (AIM) departments of ANSPs and other stakeholders. In addition, these corrections are always reactive by nature. This challenge is continuously growing, as the number of NOTAMs being published is increasing every year. The solution is the introduction of Digital NOTAMs. However, experience has shown that this transition is not happening overnight and takes many years. Alternate methods to reduce these efforts and still ensure the consistency and correctness of the NOTAM system are needed. This paper describes an Artificial Intelligence (AI) driven, Machine Learning (ML) based algorithm for supporting AIM specialists in identifying and correcting erroneous NOTAMs. Extensive analysis of historic (ICAO) NOTAMS has shown that most of the errors are in the encoding of the Q-code section of a NOTAM. An initial validation is performed based on static business rules, enhanced with an AI algorithm. The AI algorithm uses the concept of confidence scores to correct the Q-section of a NOTAM. It also checks for logical consistency using a trained neural net comparing it to the free-text E section of the NOTAM. The AIM specialist workload is reduced to only having to accept, modify, or decline the corrections proposed by the system. To ensure that the AI algorithm increases efficiency across the entire value chain of the NOTAM it is embedded into an optimized NOTAM processing flow. The paper will outline the NOTAM AI-enhanced correction functionality and describe the building blocks. Finally, the lessons learned from a real-world implementation will be discussed and an outlook on how this functionality can be expanded to address additional NOTAM-related pain points in AIM will be provided.