Object Trajectory Prediction Using Seq2Seq Orbital Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the trajectory of re-entering space debris are limited by uncertainties and lack generalizability, leading to imprecise predictions and inability to apply to various objects.
Innovation Solution
A system and method using a training data set constructed from past actual orbit information, employing neural networks like RNN, GRU, LSTM, and Seq2Seq to predict object trajectories by inputting sequence data before a reference point and outputting data after it, considering object shape and using TLE data for altitude calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If physical law-based dynamic modeling is used for trajectory prediction, then the method can be applied to various objects, but prediction precision is limited due to uncertainties
Solution Approach 1:
The patent replaces the mechanical/physical law-based dynamic modeling system with a data-driven neural network system. Instead of using physics equations to model trajectory, the system uses sequence-to-sequence neural networks trained on historical observation data to predict trajectories, thereby achieving higher precision while maintaining generalizability across different objects
Solution Approach 2:
The patent changes the fundamental parameters of the prediction system by transitioning from physical parameters (mass, drag coefficient, atmospheric density) to data-driven parameters (sequence data representations, neural network weights). This allows the system to capture complex patterns in the data that physical models cannot represent, improving precision while maintaining versatility
2Measurement precision
If observation data correction and control parameter optimization are performed, then trajectory prediction can be improved, but the method cannot be generalized to various objects
Solution Approach 1:
The patent creates a universal prediction framework using sequence-to-sequence neural networks that can handle multiple object types (satellites, space debris, re-entering objects) with a single model architecture. The model processes standardized sequence data inputs and produces trajectory predictions for various objects without requiring object-specific customization, achieving both precision and generalizability
3Productivity
If existing prediction methods are used, then computation can be performed with available data, but updating observation data is required for accurate predictions
Solution Approach 1:
The patent performs preliminary action by training the neural network model extensively on historical observation data before deployment. This pre-training phase captures long-term patterns and uncertainties in the data, enabling the model to provide reliable predictions without requiring continuous updates of observation data during operational use, thus maintaining both efficiency and reliability
Data Source
AI summary
A method for predicting trajectory of an object includes constructing a training data set using past actual orbital information of a target object, wherein the training data set includes a plurality of pairs of input sequence data corresponding to a trajectory in a first section before a reference point, and output sequence data corresponding to a trajectory in a second section after the reference point, training an object trajectory prediction model using the training data set, and predicting the trajectory of the prediction target object after the reference point, by inputting input sequence data corresponding to an actual trajectory of the prediction target object before the reference point into the object trajectory prediction model.