Spatio-Temporal Traffic Speed Prediction Using Segmented ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current traffic state prediction methods fail to accurately account for spatial and temporal dependencies in traffic speed data, leading to inefficiencies in traffic management, particularly in identifying and managing bottlenecks within transportation networks.
Innovation Solution
A traffic speed modeling framework that utilizes a combination of partial least squares regression and deep learning models with a sequence-to-sequence architecture to analyze link-based and weather data, predicting short-term traffic speeds and forecasting future traffic states, including bottlenecks, by integrating spatial and temporal dependencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional traffic prediction models are used, then the system is simple to implement, but the prediction accuracy is insufficient due to failure to account for spatio-temporal dependencies
Solution Approach 1:
The patent segments the traffic network into multiple links and divides the time series into different lags (short-term, medium-term, long-term). Each segment is processed independently through separate regression models, allowing the system to capture spatio-temporal dependencies without requiring a single monolithic complex model. This segmentation enables accurate predictions while maintaining computational efficiency.
Solution Approach 2:
The patent introduces spatial dimensions by incorporating data from multiple links simultaneously and temporal dimensions by using multiple time lags. The model transitions from univariate to multivariate analysis by adding spatial and temporal dimensions to the prediction framework, thereby capturing dependencies across different locations and times without overwhelming complexity.
2Measurement precision
If more data sources including weather data are incorporated, then the prediction accuracy improves, but the computational requirements and data processing complexity increase
Solution Approach 1:
The patent segments the data processing into separate regression models for different time lags and incorporates weather data selectively based on its relevance to each time period. This segmentation allows the system to process multiple data sources efficiently without requiring computational resources for all possible combinations of data sources simultaneously.
Solution Approach 2:
The patent changes the parameters of the regression models dynamically based on the time lag being predicted. Different weights and feature combinations are used for short-term versus long-term predictions, allowing the system to optimize computational efficiency by adjusting model parameters rather than using a fixed complex model for all prediction horizons.
3Reliability
If the model accounts for spatio-temporal dependencies, then the prediction reliability improves, but the difficulty of detecting and measuring increases
Solution Approach 1:
The patent segments the spatio-temporal dependency detection into separate regression models for different time lags and spatial locations. Each model independently captures dependencies for its specific time period and location, making the detection process more manageable and the reliability higher through localized optimization rather than attempting to capture all dependencies simultaneously.
Solution Approach 2:
The patent uses regression models as intermediary components that mediate between the raw spatio-temporal data and the final predictions. These intermediary models simplify the detection and measurement of dependencies by transforming complex multi-dimensional data into manageable intermediate representations that can be processed systematically.
Data Source
AI summary
A framework for modeling traffic speed in a transportation network analyzes both the spatial and temporal dependencies in probe-based traffic speeds, historical weather data, and forecasted weather data, using multiple machine learning models. A decentralized partial least squares (PLS) regression model predicts short-term speed using localized, historical probe-based traffic data, and a deep learning model applies the predicted short-term speed to further estimate traffic speed at specified times and at specific locations in the transportation network for predicting traffic bottlenecks and other future traffic states