Traffic Fluctuation Prediction Using Sparse Dictionary Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting unsteady traffic fluctuations, such as ARIMA and deep learning-based approaches, face challenges in achieving high accuracy with a small amount of calculation and are limited by constraints like causality and the need for frequent parameter updates.
Innovation Solution
A traffic fluctuation prediction apparatus and method that divides time-series data into two sets, learns estimation and prediction dictionaries simultaneously, and uses sparse representation with Lyapunov optimization to correct prediction errors, allowing for accurate forecasting of unsteady traffic with reduced computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning-based methods (RCLSTM) are used to predict unsteady traffic fluctuation, then prediction accuracy is improved, but computational resources and calculation time increase significantly
Solution Approach 1:
The patent segments the traffic prediction problem into two distinct dictionary learning tasks: one for capturing short-term fluctuations and another for modeling long-term trends. This segmentation allows each dictionary to specialize in specific temporal patterns, achieving high prediction accuracy while using computationally efficient sparse representation instead of resource-intensive deep learning models.
Solution Approach 2:
The patent transforms the prediction approach by changing from learning complex neural network parameters to learning dictionary atoms and sparse codes. This parameter transformation enables the system to achieve comparable prediction accuracy with significantly reduced computational resources by using iterative optimization algorithms rather than gradient descent on large neural networks.
2Power
If ARIMA model is used for traffic prediction, then calculation amount is reduced, but prediction accuracy deteriorates due to difficulty in parameter selection
Solution Approach 1:
The patent implements self-service through automated dictionary learning that adapts to the specific characteristics of the input traffic data without requiring manual parameter tuning. The algorithm automatically determines the optimal number of dictionary atoms and their structures based on the data itself, eliminating the need for analyst discretion while maintaining low computational complexity similar to ARIMA.
Solution Approach 2:
The patent introduces dynamics by using time-varying dictionaries that can adapt to changing traffic patterns. Unlike static ARIMA models with fixed parameters, the dictionary learning approach allows the basis functions to evolve and adjust to new traffic conditions, improving accuracy for non-stationary traffic fluctuations while maintaining computational efficiency.
3Reliability
If deep learning methods are used to follow unsteady network traffic, then predictive performance is maintained, but parameters must be changed frequently requiring customized hardware and software
Solution Approach 1:
The patent employs computationally lightweight dictionary learning that can be rapidly retrained on new data without requiring specialized hardware. The simple iterative optimization algorithm can be implemented on standard processors and retrained frequently with minimal computational overhead, enabling the system to adapt to changing traffic patterns using ordinary hardware rather than customized deep learning infrastructure.
Data Source
AI summary
The present disclosure has been made in view of such a problem, and an object of the present disclosure is to make it possible to predict fluctuation of unsteady traffic with a small amount of calculation.A traffic fluctuation prediction apparatus (91) according to the present disclosure includes: a data division unit (12) that divides time-series data XN(t) in a certain period of the traffic into estimation and prediction data sets; a learning unit (13) that learns a dictionary Dr(t) using one of the two divided data sets, and learns a dictionary Dp(t) using the other of the two divided data sets; a prediction unit (14) that obtains a sparse code YN(t) in representing the time-series data using the learned dictionary Dr(t) and obtains a predicted value XN{circumflex over ( )}(t) of future traffic using the obtained sparse code YN(t) and the dictionary Dp(t); and a prediction correction unit (15) that corrects a prediction error of the predicted value XN{circumflex over ( )}(t) obtained by the prediction unit (14) using search for a stable region of a double queue of overestimation and underestimation.