<![CDATA[The power business load has significant periodicity and suddenness characteristics, and has high requirements for operational reliability. Traditional container resource static allocation methods and general elastic scheduling strategies are difficult to achieve efficient and energy-saving resource utilization while ensuring service quality. To address the above issues, research and analyze the load patterns and various constraints of typical power business scenarios, construct a mixed integer programming model with the goal of minimizing service response time, resource fragmentation rate, and system energy consumption, and propose a hierarchical elastic scheduling framework. This framework integrates a threshold based passive scaling mechanism and a pre scheduling mechanism based on bidirectional long short-term memory network load prediction. For high proportion new energy power scenarios, the wind and photovoltaic output cycle encoding is embedded into the model, and grid load constraints are introduced to achieve deep coupling between power system energy flow and container computing power flow. In a typical mixed load scenario, comparative experiments were conducted with Kubernetes' default horizontal container auto scaling and classic best fit algorithms. The results showed that the optimized scheduling scheme proposed in this paper reduced the average response time of applications by 31.2%, increased the average resource utilization rate of the cluster from 58.72% to 86.63%, and controlled the service violation rate from 8.55% to below 1.00%; Through the integration of intelligent nodes, the overall energy consumption of the system has decreased by about 22.3%. This article provides effective theoretical methods and engineering practice references for the dynamic management and optimization of cloud native infrastructure resources in the power industry.]]>
