Traffic flow prediction based on improved deep extreme learning machine.

Area of Science:

• Transportation Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate short-term traffic flow prediction is crucial for intelligent transportation systems.
• Existing models often struggle with the complex, non-linear dynamics of traffic data.
• Hybrid models offer potential for improved prediction accuracy by integrating diverse analytical approaches.

Purpose of the Study:

• To propose a novel hybrid prediction model for short-term traffic flow.
• To enhance prediction accuracy by combining advanced signal decomposition and machine learning techniques.
• To evaluate the model's performance against existing methods using real-world traffic data.

Main Methods:

• Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) for signal decomposition into Intrinsic Mode Functions (IMFs).
• Permutation Entropy (PE) analysis to characterize the randomness of IMFs.
• Hybrid prediction using Sparrow Search Algorithm-Deep Extreme Learning Machine (SSA-DELM) for high-entropy IMFs and ARIMA for low-entropy IMFs.
• Ensemble forecasting by summing the predicted values of individual IMFs.

Main Results:

• The proposed SSA-DELM hybrid model demonstrated the smallest prediction errors.
• The model exhibited the best fitting effect with measured traffic flow data.
• Performance evaluation confirmed the model's effectiveness in improving short-term traffic flow prediction accuracy.

Conclusions:

• The hybrid SSA-DELM model, utilizing ICEEMDAN and PE, offers superior performance for short-term traffic flow prediction.
• Decomposition based on randomness characteristics allows for tailored modeling of different signal components.
• The findings suggest a promising approach for enhancing traffic management and planning through accurate forecasting.