Researches On Energy Consumption Prediction Of Cloud Data Center Based On Spatiotemporal Long Short-term Memory Network

With the development of the Internet,cloud computing service platform not only provides diversified functions for the Internet,but also brings a lot of energy consumption.The main purpose of this thesis is to mine the deeper potential rules between server indicators and energy consumption data under different tasks through deep learning,so as to improve the accuracy of energy consumption prediction.For the specific energy consumption prediction problem,it can be abstracted as the prediction problem of multi feature time series data.In this thesis,we explore the Recurrent Neural Network(RNN),Seq2 Seq codec network and attention mechanism,and propose a spatiotemporal LSTM prediction model based on two-stage attention mechanism.During the training phase of the model,the self-attention mechanism is used to adaptively obtain the weights of different features at each time,extract hidden information,generate context vectors,and embed them into the circular neural network.In order to verify the performance of the model,our thesis establishes a data collection system for energy consumption under the laboratory environment.By comparing the predicted results,we found that the model performs better.In order to validate and apply the technology of energy efficiency evaluation and prediction for large-scale cloud data centers,this thesis build a front-end development based on Vue.js for assessment of cloud data centers and energy consumption prediction.The overall page structure is built with a front-end framework called Vue.js based on MVVM and the open source component library Element-UI.The continuous integrated automation deployment is implemented with Jenkins and Docker,combined with the dynamic display model Highcharts and Echarts implementation algorithms.The front-end system module effectively improves the development efficiency and has good performance under large-scale integrated testing.