Dynamic predictive intelligence

Entopy’s software was used to deliver predictive intelligence across the strategic road network of a major UK port to inform operators of future freight flows to support data-driven operational intervention to help mitigate and control congestion.  

Delivering effective predictive intelligence across the strategic network is challenging. The network is dynamic, comprising many dynamics and has frequent outlier occurrences such as traffic accidents, scheduled road networks and one-time events that alter traffic activity.

Entopy’s software was used to create a semantic model of the road network, depicting key aspects. This creates a dynamic model able to capture and organise data in a way that can deliver the actionable intelligence that the port needs. This included understanding historical dynamics and using real-time data to deliver operational intelligence. But it also required a novel approach to Artificial Intelligence, able to combine computational and stochastic models to be able to deliver predictive intelligence, considering ‘black swan’ events that are more difficult to predict.

Entopy’s novel micromodels technology offers a more dynamic approach to predictive intelligence. The technology focuses on breaking complex predictive problems into smaller, specific ‘chunks’. Machine learning models can then be focused on more ‘atomic’ pieces of the problem. Entopy’s foundational software orchestrates the outputs of multiple ‘micromodels’ and networks them together semantically with real-time, event-based data, creating a network capable of delivering dynamic predictive intelligence.

Entopy deployed multiple micro-machine-learning models at junctions across the strategic road network. These models predicted traffic by type using multiple inputs including weather, day, time, and seasonality. The models are then linked semantically, combining spatial and temporal logic to capture relevant relationships. For example, there is a directional relationship between Junction 10 and Junction 11, each with independent micro-machine-learning models predicting traffic flows at those specific parts of the road at given time intervals.

Real-time event-based data is captured from multiple source systems including RNS, highways and social media. These events, including traffic accidents, scheduled road networks and one-time events were captured, located, categorised, and formed new nodes on the network, again, semantically mapped to other events and models based on key parameters.  

The result is a dynamic predictive network capable of delivering accurate predictive intelligence, considering key inputs and outlier events. Current performance shows predictions derived from Entopy’s software >90% accuracy, using the port’s commercial data as a validation dataset. The software is live in the ports control tower, supporting operators to making more informed and data-driven decisions.