Edge cloud-oriented DNN inference model deployment energy consumption optimization method and system

The invention discloses an edge cloud-oriented DNN inference model deployment energy consumption optimization method and system. According to the method, inference combinations of concurrent DNN models are configured according to hardware attributes of heterogeneous edge nodes in each scheduling time slot, historical information of execution time delays of different DNN model combinations in multiple time slots at the heterogeneous edge nodes is accumulated, and a context-based combined multi-arm bandit is utilized to provide a basis for distribution of DNN model combinations at the current moment. Therefore, the DNN inference model deployment strategy of the current time slot is continuously adjusted, the peak load time of heterogeneous nodes is reduced, and the overall execution power consumption is reduced. According to the method, the system energy consumption can be optimized in the edge cloud computing environment, the energy utilization rate is improved, different workloads can be adaptively coped with