Arrival Time Estimation via Multi-Model Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current arrival time estimation methods for vehicle dispatch platforms are inaccurate due to the lack of consideration for varying travel speeds across different road segments, leading to unreliable predictions.
Innovation Solution
A machine learning model comprising a wide network, deep neural network, and recurrent neural network coupled with a multilayer perceptron network is trained using historical transportation data to estimate arrival times, incorporating features such as driver identification, weather, and real-time traffic conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a simple arrival time estimation method is used, then the device complexity is reduced, but the measurement precision of arrival time is deteriorated
Solution Approach 1:
The route is divided into multiple road segments, and the model learns varying travel speeds for each segment independently. This segmentation allows the system to capture local traffic patterns and conditions without requiring an overly complex global model, thereby improving arrival time accuracy while managing model complexity through structured decomposition.
Solution Approach 2:
The patent transforms the problem from a single-dimensional average speed calculation to a multi-dimensional approach by incorporating road segment identifiers, historical travel times, and contextual features. This dimensional expansion enables the model to capture complex patterns in arrival time data without requiring excessive model complexity through careful feature engineering.
2Reliability
If historical data is utilized for training, then the reliability of estimation is improved, but the loss of time for data processing is increased
Solution Approach 1:
The model is trained offline using historical arrival time data before deployment. This preliminary action allows the system to learn from extensive historical patterns without incurring data processing delays during actual arrival time predictions. The training phase consolidates knowledge from historical data, enabling fast and reliable real-time predictions without reprocessing the historical data each time.
Solution Approach 2:
The system creates a trained model that copies the patterns and relationships learned from historical data. This model copy can then be deployed and used for predictions without requiring access to or processing of the original historical data, thereby maintaining high reliability while minimizing data processing time during operation.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Systems and methods are provided for estimating arrival time associated with a ride order. An exemplary method may comprise: inputting transportation information to a trained machine learning model. The transportation information may comprise an origin and a destination associated with the ride order, and the trained machine learning model may comprise a wide network, a deep neural network, and a recurrent neural network all coupled to a multiplayer perceptron network. The method may further comprise, based on the trained machine learning model, obtaining an estimated time for arriving at the destination via a route connecting the origin and the destination.