【目的】 伴随数据中心绿色供能需求的持续攀升，如何有效协调风光不确定性与数据中心算力负荷需求间的匹配性面临严峻挑战。对此，本文提出一种考虑风光不确定性与算力灵活性的数据中心两阶段随机优化调度方法。【方法】 首先，针对风光出力的随机波动与时序耦合特征，构建基于一阶自回归与科列斯基分解的场景生成模型，捕捉风光时序相关性与互补性。其次，考虑延时容忍度差异特性，建立基于离散时间任务流的算力灵活响应模型，通过队列状态方程量化不同时延容忍度业务在时间维度上的迁移能力与积压约束。最后，以系统期望运行成本最小为目标，构建包含算力调度与多能协调的两阶段随机优化模型，决策算力任务的最优时序运行策略。【结果】 算例分析表明，所提策略可有效引导延迟容忍型任务从电价高峰时段向风光资源充裕的电价低谷时段平移，系统期望运行成本较传统刚性调度模式降低了4.1%。【结论】 所提方法能够有效挖掘算力负荷的调节潜力，降低运行成本，实现数据中心经济高效运行。

[Objective] The increasing demand for green energy supply in data centers poses a critical challenge in matching the wind and photovoltaic （PV） power uncertainty with computational load requirements. To address this issue， this paper proposes an optimized two-stage stochastic scheduling method to coordinate wind and PV power uncertainty with computational flexibility. [Methods] First， considering the stochastic fluctuations and temporal coupling of wind and PV power outputs， a scenario generation model integrating first-order autoregressive process and Cholesky decomposition is developed to capture the temporal correlations and complementarity between wind and PV power outputs. Second， considering the heterogeneity of delay tolerance， a flexible computational response model is developed based on discrete-time task flows， in which the queue state equations quantify the time-dimensional migration capability and backlog constraints of workloads with different delay tolerances. Finally， in order to minimize the expected system operating cost， a two-stage stochastic optimization model incorporating computational load scheduling and multi-energy coordination is formulated to determine the optimal time-sequential operating strategy for computational tasks. [Results] Numerical case studies demonstrate that the proposed strategy shifts delay-tolerant workloads from peak price periods to off-peak periods with abundant wind and PV power. This coordination reduces the expected system operating cost by 4.1% compared with traditional rigid scheduling approaches. [Conclusions] The proposed method effectively leverages the demand response potential of computational loads to enable cost-effective data center operation.