Abstract: In order to solve the problem caused by the traditional load forecasting of cloud computing resources method’s low prediction accuracy of sequence high frequency components and weak generalization ability, this paper proposes a method of combining wavelet packet decomposition and Whale Optimization Algorithm based on Chaotic Sine Cosine operator（CSCWOA） to optimize the MultiLayer Perceptron（MLP） neural network for short-term cloud computing resources forecasting. Load sequence is decomposed into multiple band series by wavelet packet. While the CSCWOA-MLP optimized neural network is enrolled to predict the sub-band series. And by superimposing and reconstructing all the forecasting series, it gets the actual forecasting results. The experimental results show that the method can grasp the change rule of the impact burr perfectly, which has good prediction accuracy and generalization ability.

Key words: load forecasting of cloud computing resources, whale optimization algorithm based on chaotic sine cosine operator, wavelet packet transform, multilayer perceptron neural network

摘要： 为了解决传统云计算资源负载预测方法对负载序列高频分量预测精度不高和泛化能力弱的缺点，提出一种混合小波包变换和正余混沌双弦鲸鱼优化（CSCWOA）算法优化多层感知器神经网络（MLP）的短期云计算资源负载预测方法。通过小波包变换对负载序列进行多频段预处理分解，然后采用CSCWOA算法优化的MLP神经网络，对单支重构所得的负载子序列进行预测；最后叠加各子序列的预测值来获取实际预测结果。实验结果表明，该方法能掌握负载序列各频段冲击毛刺的变化规律，具有较好的预测精度和泛化能力。

关键词: 云计算资源负载预测, 正余混沌双弦鲸鱼优化算法, 小波包变换, 多层感知器神经网络