Optimization of BP neural network traffic flow prediction model based on improved particle swarm algorithm

With the annual increase in car ownership, traffic congestion and related issues are becoming increasingly severe. Urban traffic flow exhibits time-varying and stochastic characteristics, presenting a major challenge for prediction. Most existing prediction models require manual tuning of hyperparameters, which is often complex, lacks global optimality, and is prone to overfitting. Therefore, this paper applies the CAPSO algorithm to the traditional BP neural network, integrating chaotic mapping to enhance population diversity. It employs adaptive parameter balancing for exploration and development, thereby avoiding convergence to local optima. Additionally, it iterates the relevant parameters of the neural network to identify the global optimal value. The BP neural network for short-term traffic flow prediction at road intersections has been optimized and refined, resulting in a significant improvement in prediction accuracy. The linear correlation has increased by 9.23% compared to the previous BP neural network. Additionally, the speed of iterative convergence has been greatly enhanced, and the sensitivity to initial weights has been substantially reduced. These improvements contribute to a prediction model that demonstrates greater accuracy, timeliness, and robustness. This model can be utilized to optimize traffic management, providing essential data support for enhancing traffic efficiency and contributing to the development of smart cities.