An AutoEncoder and LSTM-Based Traffic Flow Prediction Method.

Area of Science:

• Urban planning and smart city development.
• Intelligent Transportation Systems (ITS).
• Data science and machine learning applications.

Background:

• Smart cities enhance urban living quality through integrated systems.
• Intelligent Transportation Systems (ITS) are crucial for smart city functionality.
• Accurate, real-time traffic flow prediction is vital for efficient ITS.

Purpose of the Study:

• To propose a novel traffic flow prediction method.
• To enhance the accuracy and stability of traffic flow predictions for ITS.
• To introduce the AutoEncoder Long Short-Term Memory (AE-LSTM) prediction method.

Main Methods:

• Utilizing an AutoEncoder to extract internal relationships and characteristics from traffic flow data.
• Employing a Long Short-Term Memory (LSTM) network for prediction using extracted features and historical data.
• Developing the AutoEncoder Long Short-Term Memory (AE-LSTM) model for complex traffic flow prediction.

Main Results:

• The AE-LSTM method demonstrated higher prediction accuracy compared to previous methods.
• A reduction of 0.01 in Mean Relative Error (MRE) was observed with AE-LSTM.
• AE-LSTM showed good stability with small prediction errors and fluctuations across different stations and dates.

Conclusions:

• The proposed AE-LSTM method significantly improves traffic flow prediction accuracy and stability.
• AE-LSTM is a promising approach for enhancing Intelligent Transportation Systems in smart cities.
• The average MRE of 0.06 for AE-LSTM indicates its effectiveness in real-world traffic prediction scenarios.