Optimal Operation of Data Center Microgrids Based on the Dynamic Surrogate Models of Server Power Consumption
Q L Chen, Shenxi Zhang, Keyan Liu, Heng Zhang, Haozhong Cheng
Shanghai Jiao Tong University China Electric Power Research Institute
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Internet data center microgrids (IDC-MGs) integrate data centers with renewable energy sources, energy storage and conversion devices to enhance sustainability. However, as the core load units of data centers, server power consumption exhibits significant temperature-dependent dynamics, posing significant challenges to the secure and economical operation of IDC-MGs. While existing mechanism models can accurately capture this dynamic behavior, their nonlinear and nonconvex nature precludes effective integration into system-level optimization frameworks. To address this challenge, this paper proposes a novel optimal operation method for IDC-MGs based on the dynamic surrogate model of server power consumption, enabling coordination between device-level dynamics and system-level energy management. Firstly, an input convex neural network (ICNN)-based dynamic surrogate model is developed to accurately capture the nonlinear dynamics of server power consumption, avoiding the complex mathematical expressions of mechanism models. Then, to reduce the model scale while maintaining computational accuracy, a novel response-driven temporal pruning (RDTP) method is proposed. This method effectively reduces the input temporal scope of the dynamic surrogate model by identifying the dynamic response boundaries of server power consumption. Finally, an optimal operation framework for IDC-MGs is developed, incorporating the dynamic surrogate model. This framework can jointly optimize energy device outputs and temporal shifting processes of information loads, while considering the dynamic characteristics of server power consumption. Case studies on both practical and test systems demonstrate the effectiveness of the proposed method. It achieves cost reductions of approximately 5.00% and 5.45%, respectively, compared to existing models that neglect the dynamic characteristics of server power consumption.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
工程Microgrid Control and Optimization
Smart Grid Energy Management · Cloud Computing and Resource Management