Bi-level planning of data centers with coupled electricity-heat-computation system using data-driven scenario generation for representing uncertainties

Along with the rapidly increasing data traffic, data centers (DTCs) have considered the expansion of their server capacity, which has led to higher energy consumption. To deal with this issue, an effective solution is presented for the coupled electricity-heat-computation system (CEHCS) in DTCs. However, CEHCS uncertainties of loads and energy resources can pose additional challenges for the optimal planning of DTC. Therefore, this work proposes a bi-level expansion planning of DTCs, which considers data-driven scenario generation for representing CEHCS uncertainties. First, a diffusion model is designed using the operational scenario generation (DMOSG) method to characterize the multi-dimensional randomness of multivariate time series, where a one-dimensional U-net works as the denoising network with an interpretable latent output space. Then, a DTC expansion planning model with CEHCS is formulated considering the safe operation temperature of CPUs, where the operational status of each server is individually characterized to increase the proposed model's fidelity. Although the DMOSG helps address the randomness of input scenarios for planning, it is also necessary to consider the uncertainty hedging for energy scheduling. Thus, chance constraints are included for BESSs to cope with risks during CEHCS dispatch. A heuristic solution is proposed for the bi-level scheme to solve the DTC planning model. At the upper level, the number of servers to be added is determined by the particle swarm optimization (PSO) algorithm. The upper-level solution is submitted to the lower level, where the performance cost of DTC planning and scheduling decisions is obtained by the Gurobi solver. Then, the DTC performance cost is returned to the upper level iteratively for calculating the optimal DTC planning results. The proposed DTC planning solution with CEHCS is validated through case studies, where the numerical results confirm the accuracy of the proposed characterization of operational uncertainty, as well as the cost-effectiveness and improved energy efficiency of the DTC expansion planning scheme.