Trip History Destination Prediction Using Bayesian Road Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current destination prediction methods in navigation systems are limited by their reliance on absolute statistical properties, failure to consider individual conditions, and lack of flexibility in adding or removing features, leading to reduced predictability and inefficiency.
Innovation Solution
A method utilizing a trip history system that determines road segments and represents trips as connected links in a road graph, employing probabilistic Bayesian models to analyze trip data and update probabilities based on observed characteristics, allowing for flexible adaptation and handling of missing data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If absolute statistical properties are used for prediction, then the prediction method is simple, but the prediction accuracy is reduced when current observation does not totally match the history
Solution Approach 1:
The patent transforms the prediction approach from using absolute statistical properties to using relative statistical properties. Instead of comparing absolute values of current observations with historical data, the system calculates relative statistics (e.g., ratios, differences normalized by historical variance) that are invariant to changes in absolute values. This allows the prediction model to maintain accuracy even when absolute conditions change, while keeping the computational complexity manageable.
2Device complexity
If conditions for each person are weighted equally, then the model is simple, but the predictability is reduced when regularity patterns do not apply to all destinations
Solution Approach 1:
The patent implements differential weighting of conditions based on their relevance to specific destinations and individuals. Instead of applying uniform weights to all conditions for all users, the system learns and applies customized weightings that reflect the importance of different conditions (e.g., time-of-day, day-of-week, weather) for each user-destination pair. This local customization improves predictability for destinations with regular patterns while maintaining flexibility for those without.
3Adaptability or versatility
If the algorithm is re-trained for all recorded data when adding new features, then the model adapts to new features, but the computational time and resources increase significantly
Solution Approach 1:
The patent pre-processes and stores historical data in a structured format with pre-computed statistical properties and feature representations. When new features are added to the model, the system can leverage the pre-processed data structure and only needs to compute the additional feature statistics rather than re-training the entire model from scratch. This preliminary preparation significantly reduces the computational burden of adaptation.
Solution Approach 2:
The patent implements an incremental learning approach where the model continuously updates its parameters as new data becomes available or when new features are introduced, rather than performing discrete full re-training cycles. This allows the system to maintain adaptability while minimizing interruptions and computational overhead, as the useful action of learning continues smoothly without complete resets.
Data Source
AI summary
A method for utilizing a trip history of a vehicle during a trip from an original position to a destination includes: (a) determining the original position; (b) comparing the original position to a mapping database covering the trip; (c) determining a road segment of the mapping database associated to the original position; (d) determining a current position during the trip; (e) comparing the current position to the mapping database; (f) determining a road segment of the mapping database associated to the current position; (g) setting the road segment as a link of the trip; (h) repeating (e)-(g) until the destination is reached; (i) determining the destination; (j) comparing the destination to the mapping database; (k) determining a road segment of the mapping database associated to the destination; and (l) representing the trip as connected links between the original position and destination, each link corresponding to a road segment.