Travel Time Prediction Using Cyclostationary Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting travel time between multiple locations, such as Support Vector Regression, Association Rule Mining, and Wavelet-based techniques, underperform in city-road scenarios, especially during rare high congestion and struggle to accurately translate traffic volume predictions into travel time forecasts.
Innovation Solution
A method and system that predict travel time by determining a deterministic component and a random fluctuation component using historical data, where the deterministic component is calculated based on wide-sense cyclostationarity and the random fluctuation component is computed using linear mean square error parameters, and then combining these to estimate the total travel time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Support Vector Regression is applied to forecast travel times, then the method can handle time series forecasting, but it underperforms in city-road scenarios and fails to handle rare high congestion effectively
Solution Approach 1:
The patent segments the travel time prediction problem into multiple components: deterministic component (mean travel time) and random fluctuation component (deviations from mean). This segmentation allows the model to handle different aspects of travel time variation separately, improving both accuracy and adaptability to rare high congestion events.
Solution Approach 2:
The patent employs dynamic modeling techniques including ARIMA for the deterministic component and quantile regression for the random fluctuation component. These dynamic models adapt to changing traffic conditions and can capture rare high congestion events, resolving the contradiction between prediction accuracy and adaptability.
2Quantity of substance
If Association Rule Mining is used to forecast traffic volumes, then the method can identify influential roads, but it is hard to translate traffic volume prediction into travel time prediction
Solution Approach 1:
The patent introduces mean travel time as an intermediary parameter that bridges traffic volume prediction and travel time prediction. By first predicting the deterministic component (mean travel time) using ARIMA and then adding the random fluctuation component using quantile regression, the model accurately translates traffic volume forecasts into travel time predictions.
3Device complexity
If Wavelet Transformation is applied to predict traffic volumes at a junction, then the method can decompose time series into trend and variation components, but it grossly underestimates travel time evolution characteristics in city road networks
Solution Approach 1:
The patent changes the modeling parameters by using ARIMA for the deterministic component and quantile regression for the random fluctuation component, rather than relying solely on Wavelet Transformation. This parameter change allows the model to capture travel time evolution characteristics more accurately in city road networks.
4Measurement precision
If deterministic component is determined using historical data, then the prediction can capture periodic patterns, but the random fluctuation component adds uncertainty to the prediction
Solution Approach 1:
The patent segments the prediction into deterministic and random fluctuation components, allowing the deterministic component to capture periodic patterns while the random fluctuation component (modeled using quantile regression) accounts for uncertainty. This segmentation resolves the contradiction by handling both periodicity and uncertainty separately.
Solution Approach 2:
The patent uses historical data to estimate the deterministic component and then uses this information to guide the prediction of the random fluctuation component. This feedback mechanism allows the model to maintain accuracy despite the presence of uncertainty in the random component.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method and system is provided for predicting at a current time "t", a time that may be taken to travel between plurality of locations, at a future time-point "t +t ". The method includes determining deterministic component "µ t + t" and predicting random fluctuation component "y1 t +t ", of the time that may be taken to travel between the plurality of locations at the future time-point "t + t". The deterministic component "µ t + t" and the random fluctuation component "y1 t +t " are added to predict the time that may be taken to travel between the plurality of locations, at the future time-point "t +t "