Train Movement Model Simplification for Real-Time Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing train modeling systems face challenges in efficiently integrating numerous variables and providing real-time control due to high computational demands, particularly in managing braking and regulator operations, without accounting for complex interactions between train components.
Innovation Solution
A train controller with a computation engine uses an optimized empirical model to predict train behavior based on node data, reducing the number of inputs and processing requirements, enabling real-time control through a train model generator and engine controller.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics engines are used to perform simulations based on models with many variables, then the accuracy of train behavior calculation is improved, but the computational power required increases significantly
Solution Approach 1:
The patent transforms the complex physics-based train model into a neural network model by changing the mathematical representation parameters. The neural network learns the underlying patterns from training data and provides accurate predictions with significantly reduced computational requirements during inference, resolving the contradiction between accuracy and computational power.
Solution Approach 2:
The patent creates a simplified copy of the complex physics model in the form of a neural network. The neural network is trained to replicate the behavior of the detailed physics model but requires minimal computational resources when deployed for real-time control decisions, enabling accurate predictions without heavy computational power.
2Reliability
If comprehensive models integrating many variables are used for real-time control, then the operational safety is improved, but the real-time processing capability deteriorates
Solution Approach 1:
The patent changes the computational parameters from complex physics calculations to a trained neural network model that provides rapid predictions. The neural network maintains safety-critical accuracy while executing in real-time on train-mounted computers, resolving the contradiction between safety and processing speed.
Solution Approach 2:
The patent performs preliminary training of the neural network model offline using comprehensive physics simulations and historical data. This preliminary action creates a optimized model that can be deployed for rapid real-time inference, separating the computationally intensive learning phase from the fast execution phase to achieve both safety and speed.
3Adaptability or versatility
If the number of variables in the train model is increased, then the comprehensiveness of train behavior prediction is improved, but the device complexity increases
Solution Approach 1:
The patent creates a universal neural network model that handles multiple train behavior predictions (tractive effort, braking force, coupling forces) through a single integrated system. The model accepts various input variables and provides comprehensive predictions without requiring separate complex models for each function, reducing overall device complexity while maintaining comprehensiveness.
Data Source
AI summary
Systems and methods for operating a train using a train model are described herein. A The train model is a concatenated version of an empirical model. The empirical model is corrected and optimized using data generated by an empirical physics engine. The optimization of the empirical model reduces the number of inputs into the model, allowing for a simpler version of the empirical model to be deployed on a train as the train model. The train model can allow a computation engine of the train to receive data and calculate one or more predicted behaviors of the train. The calculations are used by an engine controller to control various operational aspects of the train in real-time while the train is operating.