Machine Learning ETA Prediction for Railroad Equipment Using Event Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current railroad systems provide inaccurate estimated times of arrival (ETAs) for freight, leading to delays, inefficiencies, and potential losses due to the static nature of waybill documents that do not account for real-time events and changes during the journey, especially for perishable goods.
Innovation Solution
A machine learning-based system that utilizes trained models, such as gradient boosted trees, to predict ETAs and ETIs by incorporating real-time event data, telematics data, congestion calculations, and seasonality, using inputs like waybill data, sensor data, and historical shipment routes to provide accurate arrival times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional static ETA prediction methods are used based on internal performance metrics, then the system is simple to operate, but the prediction accuracy deteriorates and ETAs vary from planned ETAs by several days
Solution Approach 1:
The system implements continuous feedback by monitoring real-time events during equipment trips and updating ETA predictions dynamically. Event data from multiple sources (train movements, switch operations, crew changes, maintenance activities) are fed back into the prediction model to adjust ETAs as conditions change, ensuring predictions remain accurate throughout the journey rather than being static initial estimates
Solution Approach 2:
The patent introduces an intermediary prediction system that acts as a mediator between raw event data and final ETA outputs. This intermediary layer processes and integrates data from multiple sources including railroad operations, weather conditions, and historical patterns, transforming disparate inputs into coherent, accurate predictions that reconcile real-time conditions with planned schedules
2Measurement precision
If real-time event data and multiple data sources are integrated into the prediction system, then prediction accuracy improves, but the quantity of data to be processed increases
Solution Approach 1:
The system extracts only the most relevant and impactful event data from the broader data stream for prediction purposes. Rather than processing all available data equally, the model identifies and extracts key features such as critical delays, significant weather events, and important operational changes, filtering out redundant information to maintain prediction accuracy while managing data volume
Solution Approach 2:
The system performs preliminary processing and filtering of event data before it enters the main prediction model. Historical data is pre-processed to identify patterns and relationships, and real-time events are pre-filtered to extract only those with potential impact on ETA, reducing the computational burden on the prediction system while preserving predictive accuracy
3Productivity
If the system continuously updates ETAs based on real-time events, then operational efficiency improves and losses are reduced, but the computational power required increases
Solution Approach 1:
The system implements periodic updates of ETA predictions rather than continuous real-time recalculations. Predictions are updated at strategically determined intervals based on the occurrence of significant events or at scheduled checkpoints along the route, maintaining operational efficiency while reducing unnecessary computational energy consumption during stable travel conditions
Data Source
AI summary
Systems and methods for predicting an estimated time of arrival for an equipment at a destination location are disclosed. The method includes inputting to a trained machine learning model a starting location, the destination location and at least one event data, generated in response to an event during a trip of the equipment via the railroad from the starting location to the destination location. The machine learning model is trained using a plurality of historical shipment route data. A particular historical shipment route data comprises at least a location and a corresponding timestamp. The method includes predicting, using the trained machine learning model, the estimated time of arrival of the equipment at the destination location.