SBIR-STTR Award

Predictive Analytics for NOrmalcy Reasoning and AnoMaly Analysis (PANORAMA) is a machine learning (ML) tool for automatic identification system (AIS) data that learns maritime patterns of life and detects anomalous vessel behavior. PANORAMA learns what is normal for a given ship, taking into account: (1) that ship’s past behavior, (2) the past behavior of similar ships, (3) normalcy patterns in the ship’s current location, and (4) normalcy pattern in the ship’s current environmental conditions (e.g., weather). PANORAMA then assesses the likelihood that each subsequent movement is consistent with these patterns, generating alerts for the most significant anomalies. By subsuming context and local normalcy patterns into the ML model, we learn from data more efficiently and reduce the false alarm rate. The PANORAMA team has already completed a successful study that demonstrates this concept of “predictive analytics” using AIS data. Phase I will expand this work to (1) improve resolution on shipping lanes and harbors; (2) transfer learned behavior across similar ships; and (3) improve sensitivity to drift or abrupt changes in normalcy patterns. In Phase I will also identify Government and commercial stakeholders in the areas of maritime domain awareness, space situational awareness, and vehicle surveillance.

Predictive Analytics for NOrmalcy Reasoning and AnoMaly Analysis (PANORAMA) is a machine learning (ML) tool for automatic identification system (AIS) data that learns maritime patterns of life and detects anomalous vessel behavior. PANORAMA learns normalcy patterns, taking into account: (1) that ship’s past behavior, (2) the past behavior of similar ships, (3) normalcy patterns in the ship’s current location, and (4) normalcy pattern in the ship’s current environmental conditions (e.g., weather). PANORAMA then assesses the likelihood that each subsequent movement is consistent with these patterns, generating alerts for the most significant anomalies. By subsuming context and local normalcy patterns into the ML model, we learn from data more efficiently and reduce the false alarm rate. In Phase I, we showed that this framework can also address related Navy problems, such as learning patterns of life, monitoring shipping lanes and harbors, and responding rapidly to changes in patterns. In Phase II, we will build PANORAMA into a full-scale Navy system that handles a huge data volume/velocity, ingests data from other Navy sources, and makes additional important functionality — such as event recognition and behavior-based ID — available to the Navy. Phase II’s ultimate goal is transition to NIWC and to commercial systems; to facilitate this, we have added Raytheon/IDS to the team.