Get started
Work

LSTM-Attention Based Hybrid SAC-DDPG with Experience Replay for Resource Scheduling in Vehicle-Road-Cloud VEC Systems

Year: 2026
Type: article
Abstract: With the rapid advancement of intelligent transportation systems (ITS), Vehicle Edge Computing (VEC) systems are confronted with prominent challenges stemming from inherent constraints, including limited server resources, diverse task characteristics, fluctuating request patterns, and frequent vehicle handovers. To address the issues of unpredictable traffic flow and dynamic network environments in VEC systems, this paper proposes a collaborative framework integrating cloud servers, multiple roadside units (RSUs), and vehicles, encompassing service caching, task offloading, collaborative computing, and resource allocation. To further enhance communication efficiency, a hybrid algorithm-LDDPGER (a hybrid framework combining Long Short-Term Memory (LSTM), Attention mechanism, Soft Actor-Critic (SAC) and Deep Deterministic Policy Gradient (DDPG) with Experience Replay) -is developed, aiming to support efficient task offloading and processing for vehicle users. Compared with the optimal baseline Attention-SAC, the proposed LDDPGER reduces average system latency by 21.7%, raises system throughput by 18.3%, boosts task completion rate by 12.5% and improves task offloading success rate by 14.1%. Simulation results across various scenarios with different task loads, vehicle quantities, and bandwidth conditions demonstrate that the LDDPGER achieves measurable, consistent performance gains on latency reduction, task completion and offloading success rate promotion, and resource utilization enhancement against all benchmark algorithms.
DOI: 10.1109/jiot.2026.3729650
Language: English
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7204946110
FWCI:
Cites:
Cited by:
Related to: 10