Attention mechanism‐based model for short‐term bus traffic passenger volume prediction

Abstract

Abstract To explore the relevance between bus stops and make the real‐time prediction of bus passenger flow more accurate, this paper proposes a Traffic Forecast Model based on the Attention mechanism (TFMA). The model combines data preprocessing with bus stops’ information coding to predict short‐term bus passenger flow based on the real‐time relevance of the bus stops. First of all, the paper conducts a statistical analysis of the actual public transportation card data of Suzhou, China, and obtains the characteristics of real‐time relevance of different bus stops. Secondly, bus route and bus stop information, the passenger flow rate of change, weather, date, and other related factors are integrated into the coding information of the bus stops. Then the method relies on the Attention mechanism to calculate the real‐time relevance of the bus stops parallelly; the core algorithm also uses a multi‐headed mechanism to increase the connection of the channel and the residual error, further improving the prediction ability. Finally, this article uses actual data from Suzhou's public transport for verification. The results show that: In terms of accuracy, TFMA outperforms multiple linear regression, GRU, and LightGBM, reaching a very high accuracy of nearly 90%.

Keywords

• attention mechanism
• bus stop information encoding
• intelligent transportation
• multi‐headed mechanism
• real‐time relevance of bus stops
• short‐term bus traffic passenger flow prediction