This work proposes a data-driven expansion planning solution

aiming at environmentally friendly upgrading and uncertainty modelling of data centers (DCs). Firstly

a multi-channel one-dimensional U-net based diffusion model is designed to generate scenarios of source-load multivariate time series

thus coping with the uncertainties. Then

an electricity-heat-computation coupling system expansion model is established

which considers CPU temperature constraints and individual status of servers with the objective of minimizing performance cost. To deal with the non-linearity of server model

a heuristic bi-level solving scheme is presented. At the upper level

the configured number of servers is determined by heuristic algorithm and transferred to the lower lever as a constant

thus making the planning model linear. At the lower level

the planning model is directly solved by commercial solver

and the performance cost is submitted back to the upper level

thereby iteratively obtaining the optimal capacities of servers

cooling systems and energy equipment. Based on a modified 18-node district heating system and an IEEE-14 bus power system

the numerical results confirm that the proposed solution can effectively characterize the source-load uncertainties

improve the energy efficiency and reduce economic cost of DCs.