Improved genetic algorithm optimized LSTM model and its application in short-term traffic flow prediction

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Data Science

Background:

• Road traffic flow exhibits strong time series correlation, necessitating accurate prediction models.
• Existing neural network algorithms may not fully capture the complexities of traffic flow dynamics.
• Optimizing neural network parameters is crucial for improving prediction accuracy.

Purpose of the Study:

• To propose a novel long-term and short-term memory (LSTM) neural network model for road traffic flow prediction.
• To enhance the prediction accuracy of traffic flow by optimizing LSTM parameters using an improved genetic algorithm (IGA).
• To demonstrate the superiority of the proposed IGA-optimized LSTM model over existing methods.

Main Methods:

• An improved genetic algorithm (IGA) was developed with dynamic adjustment of mutation and crossover rates.
• The IGA was employed to optimize key LSTM parameters, including hidden units, training epochs, gradient threshold, and learning rate.
• Short-term traffic flow data (5-minute intervals) were utilized for model evaluation.

Main Results:

• The IGA successfully optimized LSTM parameters, leading to improved road traffic flow prediction.
• The proposed IGA-optimized LSTM model demonstrated lower Root Mean Square Error (RMSE) compared to other neural network algorithms.
• The model showed adaptability to different datasets, performing well on both weekday and weekend traffic data.

Conclusions:

• The IGA-optimized LSTM model offers a significant improvement in road traffic flow prediction accuracy.
• The method is robust and adaptable to various traffic conditions, outperforming existing approaches.
• This research provides a valuable tool for enhancing traffic management systems through accurate short-term flow forecasting.