Abstract: Discuss the characteristics of cloud computing resources load and the role of short-term load prediction. Polynomial regression model is used to optimize the GM（1，1） prediction results. Markov chain is empolyed for secondary optimization. The Cuckoo search algorithm is adopted to re-optimize the gray prediction model again. On this basis, the multi-step optimized GM（1，1） model is estbilished. Experimental results show that, compared with other forecasting models, for short-term resource load forecasting in cloud computing environment, the proposed model has higher prediction accuracy and achieved good predictive performance. The proposed method can provide decision support for efficient resource scheduling and management in cloud computing.

Key words: cloud computing, prediction, GM（1, 1） model, polynomial regression, Markov chain, Cuckoo search

摘要： 论述了云计算资源负荷的特征及其短期预测的作用。首先利用多项式回归模型对GM（1，1）的预测结果进行一次优化，然后使用马尔科夫链对一次优化后的模型进行二次优化，最后运用布谷鸟搜索算法对二次优化后的灰色预测模型进行再度优化，建立基于多步优化的改进GM（1，1）灰色预测模型。实验结果表明，与其他预测模型相比，在云计算环境下的资源负荷短期预测应用中，该模型具有更高的预测精度，表现出良好的预测性能。所提方法能为云计算资源的高效调度和管理提供决策支持。

关键词: 云计算, 预测, GM（1, 1）模型, 多项式回归模型, 马尔科夫链, 布谷鸟搜索算法