Learning-Guided Energy and Task Coordination for Distributed Edge Control Systems

The use of distributed edge control systems is becoming common to serve real time, load sensitive applications at strict energy budget limitations. The coordination mechanisms used today are mostly based on policies that are not dynamic enough or scheduling is done heuristically, which do not accommodate the dynamic load of work, resource of heterogeneous edges and time-consuming control requirements. The present paper attempts to overcome these drawbacks by proposing a learning-based energy and task coordination model of distributed edge control systems which optimises the energy consumption, task execution, and control performance jointly. The proposed methodology combines an adaptive learning architecture, energy-conscious task distribution and control-conscious feedback systems, which support autonomous and context responsive coordination in distributed edge nodes. An integrated system model is developed that is able to represent the computation and communication and control-loop energy expenditures under the latency and stability restrictions. According to this formulation, a learning based-coordination algorithm is formulated to readjust task assignment and resource allocation dynamically, based on state change of the system. Results of high simulation show that the proposed framework greatly decreases total energy consumption and still controls performance by meeting strict task deadlines and control performance, which are superior to traditional static and learning-blind coordination schemes in terms of energy consumption, convergence time, and scale. The results mean that learning-based coordination has the potential to be a viable and scaleable solution to next-generation compressed-energy distributed energy-constrained edge control systems.