中文摘要：

云计算环境下服务用户并发量的预测是云环境自适应资源调整的重要依据,但传统的单值预测所包含的信息量过少,受并发量不确定性影响明显,所以其不足以支持完备的自适应调整策略制定,因而会引发过多无效的调整动作.针对上述问题,该文提出一种云服务用户并发量区间预测模型,通过预测并发量的区间量化其不确定性.该模型利用梯度下降粒子群优化的支持向量机作为主要预测方法.为了更有效地预测不同类型的并发量,提出了一种基于自相关系数以及功率谱分析的AC-PS并发量特征判定规则,并针对不同特征并发量采取不同的区间构造方法.该文通过一个实例分析该区间预测模型对解决自适应资源调整无效问题的有效性,最后利用对比实验验证预测区间的准确性.结果表明,相对于其它方法文中提出的区间预测模型对各类并发量数据的预测精度均达92%以上,其预测效率有76.11%~96.15%的提升,因此提出的并发量区间预测方法能够为避免自适应资源调整无效问题提供可靠支撑.

英文摘要：

User concurrent requests prediction is a critical basis for dynamic resources self-adaptiveadjustment in cloud computing, but with the characteristics of the uncertainty and frequent variation of concurrent requests,traditional single-value prediction is not enough to support an effective self-adaptive adjustment, thereby producing a variety of invalid self-adaptive actions, which brings in the serious degradation of cloud performance and decrease in cloud effectiveness. In this paper, a prediction interval estimation model of user concurrent requests for Cloud service is proposed to quantify the uncertainty of user requests and the probability of variations. The model adopts SVM to obtain the upper and lower bound of the prediction interval, and employs the PSO to train data for the optimal solution. In order to make the model available for different types of user concurrent requests, a rule AC-PS based on autocorrelation coefficient and power spectrum analysis is put forward, which can judge the features of the requests. On account of AC-PS, different concurrent requests consider corresponding construction method of prediction interval.An example of self-adaptive adjustment using proposed prediction interval estimation model is given to analyze the utility in handling the invalidation adjustment. Furthermore, a comparison experiment is implemented to evaluate the accuracy of the proposed model. The experiment results show that compared with other prediction interval estimation models, the model proposed in this work can promise a better prediction performance that the accuracies are all above 92%, and efficiency is promoted sharply by 76. 11% to 96. 15%. Therefore the proposed approach can avoid the invalidation of self-adaptive adjustments in resources effectively, and promote the utility of cloud environments without losing the quality of cloud services.