A parallel preceding vehicle speed prediction method combining deep belief networks and gated recurrent units

In autonomous driving systems, an accurate and rapid prediction of the speed of the preceding vehicle is vital for both the safety of the vehicle and the real-time performance of decision-making systems. In this paper, a parallel bit-hreaded vehicle speed prediction method is proposed that integrates a deep belief network (DBN) and a gated recurrent unit (GRU) is proposed to achieve efficient online prediction of the speed of the preceding vehicle. First, a DBN network is trained offline using large-scale historical data to automatically extract multi-level abstract features from raw vehicle speed data. Next, a GRU network is employed to model the temporal evolution of these feature vectors and generate preliminary predictions. Finally, real-time sensor data is received in the vehicle via the online model, and online prediction is performed using the trained model. Meanwhile, the data collected online are fed into a storage buffer, and event trigger conditions are designed so that upon receiving the trigger signal, parameter adjustment and updates are executed in parallel, followed by a switch to the updated model. The experimental results demonstrate that the parallel architecture proposed in this paper, which integrates DBN and GRU networks, achieves adaptive model optimisation in dynamic scenarios while ensuring real-time prediction of the preceding vehicle's speed. Furthermore, it provides a scalable solution for online learning in autonomous driving and other real-time systems.