Neural Network Demand Forecasting for Transportation Services
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing travel industry systems rely on conventional big data analytics for price forecasting and itinerary recommendation, which fail to accurately predict non-linear demand fluctuations in transportation services, particularly for services with limited supply and seasonal variations.
Innovation Solution
Implementing a machine-learning architecture with neural network models tailored for specific dates before a departure date, allowing for accurate prediction of non-linear demand changes by training unique models for each day using historic data and adjusting configuration values for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional big data analytics and statistical analysis are used for price forecasting, then the system complexity is reduced and ease of operation is improved, but the measurement precision of demand forecasting deteriorates due to inability to capture non-linear fluctuations
Solution Approach 1:
The patent divides the forecasting period into multiple segments (different numbers of days before departure) and creates separate neural network models for each segment. This segmentation allows each model to be optimized for its specific time window, improving forecasting accuracy for non-linear demand patterns while keeping the overall system manageable through modular architecture.
Solution Approach 2:
The system dynamically selects and applies different neural network models based on the current forecasting horizon (number of days before departure). This dynamic adaptation allows the system to capture non-linear demand fluctuations at different stages of the booking cycle, significantly improving measurement precision without requiring a single overly complex static model.
2Measurement precision
If a single neural network model is used for all dates before departure, then the device complexity is reduced, but the measurement precision of demand prediction deteriorates due to inability to account for date-specific patterns
Solution Approach 1:
The patent segments the forecasting task by creating separate neural network models for different time windows before departure (e.g., 30-60 days, 10-30 days, 1-10 days). Each model is trained on historic data specific to its time window, enabling precise capture of date-specific demand patterns while maintaining individual model simplicity.
Solution Approach 2:
Each neural network model is locally optimized for its specific time window before departure, with unique parameters and training data. This local quality approach allows each model to specialize in capturing the specific demand patterns relevant to its time window, improving overall prediction accuracy without requiring every model to handle all temporal patterns.
3Reliability
If historic data from multiple years is used for training, then the reliability of demand forecasting is improved through better capture of seasonal and event-driven fluctuations, but the loss of time increases due to extended data processing requirements
Solution Approach 1:
The patent segments the training process by organizing historic data into time windows corresponding to different numbers of days before departure. This segmentation allows efficient processing of multi-year historic data by training separate models on relevant time windows, capturing seasonal and event-driven patterns without requiring processing of the entire historical dataset for each model.
Solution Approach 2:
The system performs preliminary training of neural network models using historic data before actual forecasting is needed. By pre-training models on multi-year historic data organized by time windows, the system captures seasonal and event-driven fluctuations in advance, enabling reliable forecasting without time-consuming real-time processing during actual demand prediction.
Data Source
AI summary
Embodiments described herein are related to systems and methods for forecasting demands for a transportation service. In one aspect, a set of neural network models may be implemented, where each neural network model can be configured to predict a booking status of a category of carriers on a corresponding date from a range of dates before a departure date. In one aspect, for each neural network model, a corresponding set of configuration values can be determined. Examples of the corresponding set of configuration values includes at least one of a number of layers, a number of neurons, and an activation function of the each neural network model. The set of neural network models can be constructed, according to corresponding sets of configuration values, and the constructed neural network models can be trained.