Cost-Based Vehicle Routing Using Machine Learning

The patent segments the routing cost into multiple components: direct costs (fuel, tolls) and indirect costs (maintenance, third-party expenses). By dividing the cost structure into segments and evaluating each separately using machine learning models, the system achieves comprehensive cost optimization without excessive complexity. Each cost segment is modeled independently and then aggregated to determine the total route cost.

The patent applies preliminary action by training machine learning models in advance with historical cost data from multiple vehicles. These pre-trained models enable the routing system to quickly estimate indirect costs for different route segments without performing complex calculations in real-time. The preliminary training phase stores learned patterns that are then reused during route planning, reducing computational complexity while improving energy cost optimization.

The patent performs preliminary action by training machine learning models offline using historical cost data from multiple vehicles. The trained models are then deployed to predict indirect costs for route segments during navigation. This separation of training and inference phases allows complex learning to occur beforehand, while real-time route planning uses simpler predictions, reducing computational complexity during actual use while still achieving total cost optimization.

The patent applies partial action by using machine learning models only for predicting indirect costs (maintenance, third-party expenses) rather than attempting to model all route attributes. Direct costs are calculated using traditional methods, while ML models supplement only the uncertain indirect portions. This partial application of complex technology achieves total cost optimization without unnecessarily increasing computational complexity across the entire system.

The patent applies preliminary action by collecting and processing actual cost data from multiple vehicles in advance to train machine learning models offline. Once trained, the models can quickly predict indirect costs for new routes without requiring real-time access to raw data or complex processing. This preliminary data processing phase separates bulk data handling from real-time prediction, improving accuracy while minimizing data processing time during actual route planning.

The patent uses copying by creating trained machine learning model representations that capture patterns from historical multi-vehicle cost data. Instead of processing raw data from multiple vehicles during each routing calculation, the system copies the learned patterns into pre-trained models that can be rapidly applied to predict costs for individual routes. This copying approach preserves the accuracy benefits of multi-vehicle data while eliminating the time cost of processing that data in real-time.