【摘要】 随着信息数据的骤增，数据中心的负载数量逐渐增多，导致数据库负载能耗直线上升。为提高数据库负载能耗预测精确性，提升数据库资源优化配置能力，构建了GA-LSTM时间卷积网络数据库负载能耗预测模型。模型首先采用z-score算法对DLE数据集进行标准化处理，提高数据类交叉计算效率，同时采用LOF局域噪点剔除算法，去除无用数据点，保证数据平滑度，并利用相关系数计算分析，对DLE集进行降维优化，降低模型计算量；然后采用GA算法对DLE数据集进行特征基因提取，计算特征基因的适应度参数，并基于遗传规则，在LSTM反馈误差的基础上，通过F特值计算，构建出LSTM网络的最优初始参数；最后通过提取优化DLE集的隐式特征，构建出LSTM时间卷积网络模型，并采用PSO算法对网络的超参数进行寻优收敛，在最优参数的基础上，构建出GA-LSTM数据库负载能耗预测时间卷积模型。数据库负载能耗预测对比实验的仿真结果显示，较传统LSTM算法相比，上述算法的均方根误差减少了0.244,平均绝对百分比误差减少了0.390,较SVM、AdaBoost和K-Means等数据库负载能耗预测模型相比，上述模型R~2指标增加了3.19%,同时RMSE和MAPE参数分别减少了0.753和0.628,文中所提的GA-LSTM数据库负载能耗预测算法的稳度最优、精度最高。文中算法在数据库负载预测与资源优化运营控制中均有重要的仿真价值。更多还原

【Abstract】 With the rapid increase iin information data, the load on the data center is gradually increasing, resulting in a sharp rise in the energy consumption of database load. In order to improve the accuracy of database load energy consumption prediction and enhance the ability to optimize the allocation of database resources, this paper constructs a GA-LSTM temporal convolutional network database load energy consumption prediction model. Firstly, the model uses z-score algorithm to standardize the DLE data set to improve the efficiency of data class cross calculation, and uses LOF local noise elimination algorithm to remove useless data points to ensure the smoothness of data, and uses correlation coefficient calculation and analysis to reduce the dimension of DLE set and optimize the calculation of the model; Then the GA algorithm is used to extract the feature genes from the DLE data set, calculate the fitness parameters of the feature genes, and construct the optimal initial parameters of the LSTM network based on the genetic rules and on the basis of LSTM feedback errors through the calculation of F;Finally, by extracting the implicit features of the optimized DLE set, the LSTM time convolution network model is constructed, and the PSO algorithm is used to optimize and converge the hyperparameters of the network, and on the basis of the optimal parameters, the GA-LSTM database load energy consumption prediction time convolution model is constructed. The Simulation results of the database load energy consumption prediction comparison experiment show that, compared with the traditional LSTM algorithm, the root mean square error of the proposed algorithm is reduced by 0.244,and the mean absolute percentage error is reduced by 0.390. The R~2 index of the proposed model is increased by 3.19%,while the RMSE and MAPE parameters are reduced by 0.753 and 0.628,respectively. The GA-LSTM database load energy consumption prediction algorithm proposed in this paper has the best stability and the highest accuracy. The algorithm in this paper has important simulation value in database load prediction and resource optimization operation control.更多还原