NeuroSync: A Novel Neural Network Architecture for Time Series Forecasting of Vehicle Traffic Data Over 5G and Beyond

Abstract

The efficient management and prediction of urban traffic flow are paramount in the age of beyond 5G smart cities and advanced transportation systems. Traditional methods often fail to handle the nonlinear and dynamic nature of traffic data, necessitating more advanced solutions. This paper introduces NeuroSync, a novel neural network architecture designed to leverage the strengths of spiking neuron layers and gated recurrent units (GRUs) combined with temporal pattern attention mechanisms to effectively forecast traffic patterns. The architecture is specifically tailored to address the complexities inherent in nonstationary urban traffic datasets, capturing both spatial and temporal relationships within the data. NeuroSync not only outperforms traditional forecasting models such as ARIMA and exponential smoothing but also shows significant improvement over contemporary neural network approaches like LSTM, CNN, Seq2Seq, RNN, GRU, Transformer, and Autoencoder in terms of mean squared error (MSE) and mean absolute error (MAE). The model's efficacy is demonstrated through extensive experiments with real-world traffic data, underscoring its potential to enhance urban mobility management and support the infrastructure of intelligent transportation systems.