Road Speed Prediction Scheme by Analyzing Road Environment Data.

Area of Science:

• Traffic Engineering
• Machine Learning
• Data Science

Background:

• Road speed is a key indicator of traffic congestion.
• Predicting road speed can help mitigate congestion by enabling traffic distribution and providing alternative routes.
• Existing methods may not fully capture the dynamic nature of traffic flow and external influencing factors.

Purpose of the Study:

• To propose a machine learning-based scheme for accurate road speed prediction.
• To enhance traffic congestion management through advanced speed forecasting.
• To develop a model capable of predicting both average and rapidly fluctuating road speeds.

Main Methods:

• Utilized historical average speed data, organized by day and hour, for target and neighboring roads.
• Analyzed speed changes in high-variability road sections to capture dynamic traffic flow.
• Incorporated weather conditions, historical speeds, and event data (accidents, disasters) as weighted inputs.
• Employed Long Short-Term Memory (LSTM) networks, suitable for sequential data learning, for speed prediction.
• Input data included weather and speed data from target and neighboring roads for 30-minute predictions.

Main Results:

• The proposed scheme accurately predicts average road speeds.
• The scheme effectively forecasts rapidly changing road speeds, crucial for dynamic traffic conditions.
• Performance evaluations demonstrated the capabilities of the developed road speed prediction model.

Conclusions:

• The machine learning-based scheme provides accurate road speed predictions by integrating diverse data sources.
• This approach offers a valuable tool for traffic congestion reduction and improved traffic flow management.
• The model's ability to predict short-term (30 min) road speeds with high accuracy has significant practical implications for intelligent transportation systems.