With the widespread adoption of embedded systems in industrial control, intelligent driving, and edge computing, their architectures are increasingly evolving toward distributed and heterogeneous platforms. In safety-critical applications, ensuring the secure and predictable execution of tasks on multiple processing units has become a key challenge. Such tasks are typically modeled as Directed Acyclic Graphs (DAGs) to accurately capture inter-task dependencies. To address these challenges, this paper proposes MPDES, an energy-efficient scheduling algorithm for multiple periodic DAGs in safety-critical embedded systems. MPDES ensures deadline satisfaction while enhancing scheduling predictability, resource isolation, and overall energy efficiency. The algorithm adopts a two-phase scheduling strategy: (1) In the pre-allocation phase, tasks are mapped to the fastest processors to guarantee timing constraints; (2) In the energy-aware phase, idle time is exploited and processor frequency is dynamically adjusted via DVFS technology to reduce energy consumption. This hierarchical mechanism not only lowers energy overhead but also mitigates unpredictable behavior caused by scheduling anomalies. Experimental results demonstrate that MPDES achieves both timing compliance and energy efficiency in scenarios such as Fast Fourier Transform, Gaussian elimination, and randomly generated task sets. Moreover, it provides a predictable and stable execution environment, showcasing its practical value for safety-critical embedded systems.