Railcar ETA Prediction Using Engineered Features for Multi-Carrier Trips
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing railcar tracking systems and APIs provide inaccurate ETA predictions, especially for multi-carrier journeys, leading to uncertainties in supply chain management and resource inefficiencies.
Innovation Solution
A machine learning (ML) model is trained using railcar tracking data and location coordinates to predict ETA and generate fleet readiness alerts, employing feature engineering and gradient boosting techniques for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional rail tracking systems use statistical average transit time to predict ETA, then the system is simple to operate, but the measurement precision of ETA becomes increasingly inaccurate for longer trips
Solution Approach 1:
The patent replaces traditional statistical averaging methods with a machine learning model that processes multiple data features (trip information, location coordinates, railroad characteristics) to predict ETA. This substitution of the prediction mechanism achieves higher accuracy for long-distance trips while maintaining operational simplicity through automated model inference.
Solution Approach 2:
The patent transforms the ETA prediction approach by changing from a single average transit time parameter to multiple engineered features including trip duration, distance, railroad identifiers, and location coordinates. This parameter expansion enables the model to capture complex patterns in long-distance rail transport while managing complexity through structured feature engineering.
2Measurement precision
If proprietary APIs are used for ETA prediction, then measurement precision improves for single-railroad trips, but the system becomes less adaptable to multi-carrier journeys
Solution Approach 1:
The patent creates a universal ETA prediction model that works across multiple railroads and carriers by engineering features that capture railroad-specific characteristics and trip patterns. The model ingests data from various sources including trip information and location coordinates from different carriers, enabling it to predict ETA for multi-carrier journeys while maintaining accuracy through learned patterns rather than carrier-specific rules.
Solution Approach 2:
The patent segments the ETA prediction problem by creating separate engineered features for different railroads and trip characteristics. By breaking down the prediction into learnable components from historical data across multiple carriers, the model achieves both precision and adaptability to various routing scenarios including multi-carrier journeys.
3Ease of manufacture
If speed/time approach is used for ETA calculation, then the method is simple to implement, but reliability decreases for long-duration trips
Solution Approach 1:
The patent replaces the simple speed/time calculation mechanism with a machine learning model that learns actual travel patterns from historical data. This substitution maintains ease of implementation through automated model deployment while significantly improving reliability for long-duration trips by capturing real-world variations in rail transport performance that simple calculations cannot predict.
4Device complexity
If traditional rail tracking systems are used, then device complexity remains low, but loss of information occurs for trips involving multiple railroads
Solution Approach 1:
The patent enriches the tracking system by transforming basic trip data into multiple engineered features including railroad identifiers, location coordinates, trip duration, and distance metrics. This parameter expansion captures information about multi-carrier journeys without significantly increasing system complexity, as the feature engineering is performed automatically from existing data sources.
Data Source
AI summary
A method for using a trained machine learning (ML) model to determine an ETA of a railcar to a destination and to generate a fleet readiness alert, including: collecting railcar trip information; collecting coordinates of a rail station associated with a rail carrier; creating a joined dataset by combining the trip information and coordinates; performing feature engineering (FE) on the joined dataset to create a feature engineered output including a plurality of engineered features; combining the joined dataset with the feature engineered output to create ML model feature inputs for the trained ML model; and activating the ML model, which accepts the model feature inputs as an input, determines the ETA of the railcar, and generates the fleet readiness alert based on the determined ETA of the railcar, indicating at least a usability status of a fleet of railcars at the depot yard within a fixed time.