Research was conducted on task heterogeneity and hierarchical dependencies, in order to enhance the real-time performance of multi-task processing, improve resource scheduling efficiency, and control energy consumption in smart vehicular networks. An efficient task offloading and resource allocation algorithm based on deep deterministic policy gradients was proposed. A hypergraph structure was adopted to extract task dependency features, through which complex inter-task relationships were characterized. A queueing mechanism integrating task priority and latency constraints was introduced, and a centralized training with decentralized execution architecture was employed to improve policy stability, enabling real-time offloading decisions and resource allocation. A cooperative optimization model was further constructed based on a multi-agent deep deterministic policy gradient framework, where edge computing load, wireless network conditions, and task deadlines were jointly incorporated into the design of the state space and reward function. Simulation results showed that the proposed algorithm can detect node computing power bottlenecks in real time based on network changes and generate scheduling strategies that meet the requirements of low latency, high reliability, and low energy consumption. Compared with existing algorithms, the proposed method reduced task scheduling latency by 18.17%, increased resource utilization by 13.21%, and lowered average energy consumption by 21.73% on average. These results demonstrate that the method exhibits superior performance in multi-task scheduling and resource optimization.