Traffic Flow Prediction and Analysis in Smart Cities Based on the WND-LSTM Model

Area of Science:

• Intelligent Transportation Systems
• Data Mining
• Pattern Recognition

Background:

• Urban road traffic flow exhibits complex spatio-temporal distribution, leading to prediction challenges and congestion.
• Existing models struggle to accurately capture the dynamic nature of daily travel patterns in intelligent cities.

Purpose of the Study:

• To develop and validate a novel model for analyzing and predicting daily urban vehicle travel patterns.
• To address the challenges of uneven traffic distribution and congestion in intelligent cities through advanced data mining and pattern recognition.

Main Methods:

• Proposed a Weather-Noise-Decomposition-LSTM (WND-LSTM) model integrating data preprocessing, data modeling, and implementation.
• Utilized pattern recognition and big data mining techniques to analyze travel pattern similarities across seasonal changes.
• Established a daily travel model for urban road traffic vehicles based on data mining insights.

Main Results:

• The WND-LSTM model demonstrated superior performance compared to ARIMA, LR, SVR, KNN, SAEs, GRU, and LSTM.
• Achieved a Mean Absolute Percentage Error (MAPE) of 0.651%, indicating an average accuracy improvement of 71.25% over other models.
• Successfully analyzed travel pattern similarities considering seasonal variations.

Conclusions:

• The WND-LSTM model offers a robust and accurate solution for predicting urban vehicle travel patterns.
• This approach can significantly mitigate traffic congestion and improve traffic flow management in intelligent cities.
• The findings highlight the potential of integrating big data mining and deep learning for intelligent transportation systems.