Multi-Section Traffic Flow Prediction Based on MLR-LSTM Neural Network

Area of Science:

• Transportation Engineering
• Data Science
• Artificial Intelligence

Background:

• Road congestion exacerbates traffic crashes, necessitating effective traffic flow prediction.
• Traffic flow in a target section is significantly influenced by adjacent road sections.
• Incomplete historical traffic data, often due to sensor maintenance, poses a challenge for accurate prediction.

Purpose of the Study:

• To propose a novel traffic flow prediction method that accommodates partially missing data in the target section.
• To leverage data from adjacent road sections to improve prediction accuracy for the target section.
• To provide a feasible and widely applicable solution for short-term traffic flow forecasting in real-world scenarios.

Main Methods:

• Developed a hybrid Multiple Linear Regression and Long-Short-Term Memory (MLR-LSTM) model.
• Utilized incomplete historical traffic flow data from the target section.
• Incorporated continuous historical traffic flow data from adjacent road sections for joint prediction.

Main Results:

• The MLR-LSTM model accurately predicts short-term traffic flow changes, even with missing data in the target section.
• Prediction accuracy is comparable to mainstream methods using complete target section data.
• A slight improvement in accuracy was observed with shorter data time intervals.

Conclusions:

• The proposed MLR-LSTM method offers a feasible and robust solution for traffic flow prediction with incomplete data.
• This approach is particularly advantageous given the frequent maintenance of traffic monitoring equipment.
• The method demonstrates potential for widespread application in intelligent transportation systems.