Virtual Machine Consolidation with Multi-Step Prediction and Affinity-Aware Technique for Energy-Efficient Cloud Data Centers

Virtual machine(VM)consolidation is an effective way to improve resource utilization and reduce energy consumption in cloud data centers.Most existing studies have considered VM consolidation as a bin-packing problem,but the current schemes commonly ignore the long-term relationship between VMs and hosts.In addition,there is a lack of long-term consideration for resource optimization in the VM consolidation,which results in unnecessary VM migration and increased energy consumption.To address these limitations,a VM consolidation method based on multi-step prediction and affinity-aware technique for energy-efficient cloud data centers(MPaAF-VMC)is proposed.The proposed method uses an improved linear regression prediction algorithm to predict the next-moment resource utilization of hosts and VMs,and obtains the stage demand of resources in the future period through multi-step prediction,which is realized by iterative prediction.Then,based on the multi-step prediction,an affinity model between the VM and host is designed using the first-order correlation coefficient and Euclidean distance.During the VM consolidation,the affinity value is used to select the migration VM and placement host.The proposed method is compared with the existing consolidation algorithms on the PlanetLab and Google cluster real workload data using the CloudSim simulation platform.Experimental results show that the proposed method can achieve significant improvement in reducing energy consumption,VM migration costs,and service level agreement(SLA)violations.

Video Frame Prediction by Joint Optimization of Direct Frame Synthesis and Optical-Flow Estimation

Video prediction is the problem of generating future frames by exploiting the spatiotemporal correlation from the past frame sequence.It is one of the crucial issues in computer vision and has many real-world applications,mainly focused on predicting future scenarios to avoid undesirable outcomes.However,modeling future image content and object is challenging due to the dynamic evolution and complexity of the scene,such as occlusions,camera movements,delay and illumination.Direct frame synthesis or optical-flow estimation are common approaches used by researchers.However,researchers mainly focused on video prediction using one of the approaches.Both methods have limitations,such as direct frame synthesis,usually face blurry prediction due to complex pixel distributions in the scene,and optical-flow estimation,usually produce artifacts due to large object displacements or obstructions in the clip.In this paper,we constructed a deep neural network Frame Prediction Network(FPNet-OF)with multiplebranch inputs(optical flow and original frame)to predict the future video frame by adaptively fusing the future object-motion with the future frame generator.The key idea is to jointly optimize direct RGB frame synthesis and dense optical flow estimation to generate a superior video prediction network.Using various real-world datasets,we experimentally verify that our proposed framework can produce high-level video frame compared to other state-ofthe-art framework.

Chaotic time series multi-step direct prediction with partial least squares regression

Considering chaotic time series multi-step prediction, multi-step direct prediction model based on partial least squares （PLS） is proposed in this article, where PLS, the method for predicting a set of dependent variables forming a large set of predictors, is used to model the dynamic evolution between the space points and the corresponding future points. The model can eliminate error accumulation with the common single-step local model algorithm~ and refrain from the high multi-collinearity problem in the reconstructed state space with the increase of embedding dimension. Simulation predictions are done on the Mackey-Glass chaotic time series with the model. The satisfying prediction accuracy is obtained and the model efficiency verified. In the experiments, the number of extracted components in PLS is set with cross-validation procedure.

Attention-based spatio-temporal graph convolutional network considering external factors for multi-step traffic flow prediction

Traffic flow prediction is an important part of the intelligent transportation system. Accurate multi-step traffic flow prediction plays an important role in improving the operational efficiency of the traffic network. Since traffic flow data has complex spatio-temporal correlation and non-linearity, existing prediction methods are mainly accomplished through a combination of a Graph Convolutional Network (GCN) and a recurrent neural network. The combination strategy has an excellent performance in traffic prediction tasks. However, multi-step prediction error accumulates with the predicted step size. Some scholars use multiple sampling sequences to achieve more accurate prediction results. But it requires high hardware conditions and multiplied training time. Considering the spatiotemporal correlation of traffic flow and influence of external factors, we propose an Attention Based Spatio-Temporal Graph Convolutional Network considering External Factors (ABSTGCN-EF) for multi-step traffic flow prediction. This model models the traffic flow as diffusion on a digraph and extracts the spatial characteristics of traffic flow through GCN. We add meaningful time-slots attention to the encoder-decoder to form an Attention Encoder Network (AEN) to handle temporal correlation. The attention vector is used as a competitive choice to draw the correlation between predicted states and historical states. We considered the impact of three external factors (daytime, weekdays, and traffic accident markers) on the traffic flow prediction tasks. Experiments on two public data sets show that it makes sense to consider external factors. The prediction performance of our ABSTGCN-EF model achieves 7.2%–8.7% higher than the state-of-the-art baselines.

A Content-Aware Bitrate Selection Method Using Multi-Step Prediction for 360-Degree Video Streaming

A content-aware multi-step prediction control(CAMPC)algorithm is proposed to determine the bitrate of 360-degree videos,aim⁃ing to enhance the quality of experience(QoE)of users and reduce the cost of video content providers(VCP).The CAMPC algorithm first em⁃ploys a neural network to generate the content richness and combines it with the current field of view(FOV)to accurately predict the probability distribution of tiles being viewed.Then,for the tiles in the predicted viewport which directly affect QoE,the CAMPC algorithm utilizes a multi-step prediction for future system states,and accordingly selects the bitrates of multiple subsequent steps,instead of an instantaneous state.Meanwhile,it controls the buffer occupancy to eliminate the impact of prediction errors.We implement CAMPC on players by building a 360-degree video streaming platform and evaluating other advanced adaptive bitrate(ABR)rules through the real network.Experimental results show that CAMPC can save 83.5%of bandwidth resources compared with the scheme that completely transmits the tiles outside the viewport with the Dynamic Adaptive Streaming over HTTP(DASH)protocol.Besides,the proposed method can improve the system utility by 62.7%and 27.6%compared with the DASH official and viewport-based rules,respectively.

Multistep-ahead River Flow Prediction using LS-SVR at Daily Scale

In this study, potential of Least Square-Support Vector Regression (LS-SVR) approach is utilized to model the daily variation of river flow. Inherent complexity, unavailability of reasonably long data set and heterogeneous catchment response are the couple of issues that hinder the generalization of relationship between previous and forthcoming river flow magnitudes. The problem complexity may get enhanced with the influence of upstream dam releases. These issues are investigated by exploiting the capability of LS-SVR–an approach that considers Structural Risk Minimization (SRM) against the Empirical Risk Minimization (ERM)–used by other learning approaches, such as, Artificial Neural Network (ANN). This study is conducted in upper Narmada river basin in India having Bargi dam in its catchment, constructed in 1989. The river gauging station–Sandia is located few hundred kilometer downstream of Bargi dam. The model development is carried out with pre-construction flow regime and its performance is checked for both pre- and post-construction of the dam for any perceivable difference. It is found that the performances are similar for both the flow regimes, which indicates that the releases from the dam at daily scale for this gauging site may be ignored. In order to investigate the temporal horizon over which the prediction performance may be relied upon, a multistep-ahead prediction is carried out and the model performance is found to be reasonably good up to 5-day-ahead predictions though the performance is decreasing with the increase in lead-time. Skills of both LS-SVR and ANN are reported and it is found that the former performs better than the latter for all the lead-times in general, and shorter lead times in particular.

Nonlinear system PID-type multi-step predictive control

A compound neural network was constructed during the process of identification and multi-step prediction. Under the PID-type long-range predictive cost function, the control signal was calculated based on gradient algorithm. The nonlinear controller’s structure was similar to the conventional PID controller. The parameters of this controller were tuned by using a local recurrent neural network on-line. The controller has a better effect than the conventional PID controller. Simulation study shows the effectiveness and good performance.

Strategies for multi-step-ahead available parking spaces forecasting based on wavelet transform

A new methodology for multi-step-ahead forecasting was proposed herein which combined the wavelet transform(WT), artificial neural network(ANN) and forecasting strategies based on the changing characteristics of available parking spaces(APS). First, several APS time series were decomposed and reconstituted by the wavelet transform. Then, using an artificial neural network, the following five strategies for multi-step-ahead time series forecasting were used to forecast the reconstructed time series: recursive strategy, direct strategy, multi-input multi-output(MIMO) strategy, DIRMO strategy(a combination of the direct and MIMO strategies), and newly proposed recursive multi-input multi-output(RECMO) strategy which is a combination of the recursive and MIMO strategies. Finally, integrating the predicted results with the reconstructed time series produced the final forecasted available parking spaces. Three findings appear to be consistently supported by the experimental results. First, applying the wavelet transform to multi-step ahead available parking spaces forecasting can effectively improve the forecasting accuracy. Second, the forecasting resulted from the DIRMO and RECMO strategies is more accurate than that of the other strategies. Finally, the RECMO strategy requires less model training time than the DIRMO strategy and consumes the least amount of training time among five forecasting strategies.

Remaining Useful Life Prediction for Aero-Engines Combining Sate Space Model and KF Algorithm

The key to failure prevention for aero-engine lies in performance prediction and the exhaust gas temperature margin(EGTM)is used as the most important degradation parameter to obtain the operating performance of the aero-engine.Because of the complex environment interference,EGTM always has strong randomness,and the state space based degradation model can identify the noisy observation from the true degradation state,which is more close to the actual situations.Therefore,a state space model based on EGTM is established to describe the degradation path and predict the remaining useful life(RUL).As one of the most effective methods for both linear state estimation and parameter estimation,Kalman filter(KF)is applied.Firstly,with EGTM degradation data,state space model approach is used to set up a state space model for aero-engine.Secondly,RUL of aero-engine is analyzed,and expected RUL and distribution of RUL are determined.Finally,the sate space model and KF algorithm are applied to an example of CFM-56aero-engine.The expected RUL is predicted,and corresponding probability density distribution(PDF)and cumulative distribution function(CDF)are given.The result indicates that the accuracy of RUL prediction reaches 7.76%ahead 580 flight cycles(FC),which is more accurate than linear regression,and therefore shows the validity and rationality of the proposed method.

Environmental Changes on the Tibetan Plateau:Evaluation and Prediction

A research report on the environmental changes of the Tibetan Plateau from the past 2,000 years to a century ahead has been released by the Institute of Tibetan Plateau Research,Chinese Academy of Sciences.After a three-year investigation into the plateau areas in southwest China’s Tibet Autonomous Region with an average altitude of over 4,500 meters,

NWP辅助复合神经网络预测误差修正的风储系统日前上报策略

新能源电站出力存在强波动性导致巨额偏差考核支出,因此基于数值天气预报(NWP)和复合深度学习算法,提出一种计及误差预测修正的风储系统日前上报策略。首先通过改进的组合数据预处理算法对数据进行清洗以降低后续预测难度,建立基于分段式收敛粒子群算法(PCPSO)参数寻优的长短期记忆网络(LSTM)对分量分别进行预测,重构预测结果获取原预测曲线。其次考虑预测误差及NWP信息导入多输入反向传播神经网络(MIBP)获取误差预测曲线,使用非参数核密度函数修订该预测误差曲线后,以储能跟踪误差最小和储能全局调控能力最高为目的模拟储能运行获取最佳储能动作曲线,且叠加原预测曲线和最佳储能动作曲线获取最终日前上报曲线。最后通过仿真分析验证了上报策略的正确性与可行性。

基于网格型数值天气预报的风电集群日前功率预测方法

风电集群日前功率预测是省级及以上电网调控中心制定发电计划、促进风电消纳的重要基础之一。风电日前功率预测(次日0时至24时)本质上是构建数值天气预报与实际功率之间的映射模型。充分挖掘数值天气预报气象信息与功率之间的深层映射关系是提升风电功率预测精度的重要途径。利用网格型的数值天气预报并采用残差网络建立风电集群预测模型,挖掘风电集群所属空间三维网格型的气象分布与功率的关联关系。以实际运行数据进行仿真,结果显示所提方法在先进性和适应性两个方面均优于现有成熟方法。

基于自注意力机制增强的CNN-LSTM的榴弹轨迹多步超前预测

由于榴弹飞行轨迹呈现复杂性、时变性和突变性等特点,给近程防空拦截系统带来了极大的挑战。针对目前轨迹数据时空特征捕捉困难且只能进行较少步数预测的问题,提出一种引入自注意力机制的基于卷积神经网络和长短期记忆神经网络(1dimension Convolutional neural network-Long short-term memory-Attention, 1D CNN-LSTM-ATT)的一维轨迹多步超前预测模型。将所提模型与CNN-LSTM、LSTM模型分别进行单步和多步预测对比分析;实现对于目标轨迹的从T时刻到未来任意T+K时刻的高精度实时多步超前预测。实验结果表明:无论是单步还是多步预测,1D CNN-LSTM-ATT模型预测的评价指标明显优于其他2个模型;1D CNN-LSTM-ATT模型预测500步(即10 s)的累计预测误差在射程方向为82.83 m,高度方向为11.68 m,横偏方向为0.07 m,为实施弹体拦截及时响应提供了重要保障。

基于双重注意力变换模型的分布式屋顶光伏变电站级日前功率预测

分布式屋顶光伏地理位置分散,受地理环境遮挡和多种气象因素影响,导致光伏出力特性存在差异,给变电站级分布式屋顶光伏日前功率预测造成挑战。针对上述问题,提出了一种基于双重注意力变换模型的分布式屋顶光伏变电站级日前功率预测方法。首先,基于动态时间规整算法计算分布式光伏用户出力特性间的相似度,并基于凝聚层次聚类法将其划分成若干类;然后,利用自主注意力网络学习各时间步间的时序关联特性,通道卷积注意力机制学习多特征变量间的相关性,构建日前功率预测模型;最后,将每一类日前预测结果相加,实现变电站级日前功率预测。算例结果表明所提方法在多种天气状况下,较Transformer、长短期记忆神经网络和时序卷积网络,预测精度显著提升。

计及转移效用与不确定性的移动式储能系统日前-日内市场竞标策略

为激励移动式储能系统(mobile energy storage system,MESS)参与电力市场,并在增加自身盈利的同时,在一定程度上缓解电力阻塞,计及转移效用与不确定性,提出一种MESS日前日内两阶段市场竞标策略。首先,在日前阶段,构建MESS参与电力市场双层投标模型,上层旨在决策MESS的时空分布及功率,下层为电力市场出清模型;其次,在日内阶段,采用多场景随机优化方法模拟、分析日内不确定性,并以日前荷电水平和转移计划为参考,基于模型预测控制方法构建MESS参与日内电力市场双层投标模型,上层旨在动态调整MESS实时功率,下层亦为电力市场出清模型;进一步,利用KKT条件和互补松弛理论将双层竞标模型转化为单层线性优化模型,以实现高效求解;最后,以国内某城域互联电力交通网络设计典型仿真案例。仿真结果表明,所提策略能够实现可调配资源的最大化利用,有效缓解电力系统输电阻塞,促进清洁能源消纳。

利用2DGRA-BiLSTM模型的日前光伏功率曲线预测方法

为了克服光伏发电固有的间断性和波动性对电网稳定性的负面影响,提出一种二维灰度关联分析-双向长短期记忆神经网络(two-dimensional grey relational analysis and bidirectional long short-term memory network, 2DGRA-BiLSTM)模型,用于实现日前光伏功率曲线预测,以更好指导电网调度.不同于以往的点预测,本研究将日功率曲线作为整体进行预测.首先用2DGRA实现最佳历史相似日数据的获取;其次,根据日功率曲线的波动性将总数据分为3类;最后,根据3种分类,分别训练3种BiLSTM模型对日功率曲线进行预测.所提出的预测模型通过沙漠知识澳大利亚太阳能中心历史气象和功率数据进行训练,并通过数值天气预报和功率数据进行测试.对比其他几种神经网络模型,实验表明所提出模型具有更好的综合预测性能,在晴空、轻度非晴空和重度非晴空条件下,决定系数(R2)分别为0.994、0.940和0.782.

Non-Minimum Phase Nonlinear System Predictive Control Based on Local Recurrent Neural Networks

After a recursive multi-step-ahead predictor for nonlinear systems based on local recurrent neural networks is introduced, an intelligent FID controller is adopted to correct the errors including identified model errors and accumulated errors produced in the recursive process. Characterized by predictive control, this method can achieve a good control accuracy and has good robustness. A simulation study shows that this control algorithm is very effective.

基于模型预测控制的含新能源多阶段AVC优化策略

针对新能源接入后的无功电压控制问题,基于模型预测控制(model predictive control,MPC)理论,提出一种多阶段自动电压控制(automatic voltage control,AVC)优化策略。在日前优化安排离散无功补偿设备(电容器、有载变压器分接头)投切计划的基础上,日内采用基于MPC的优化控制思路,利用连续无功补偿装置(static var generator,SVG)对电压进行控制。通过建立灵敏度矩阵计算得到未来多个时刻的母线电压预测值;以最小化未来一段时间预测的电压控制偏差为目标函数,建立日内滚动优化控制模型,求解得到SVG的出力序列,并通过反馈校正,完成日内无功电压MPC。在改进的IEEE 30算例的基础上对所提方法进行验证,结果表明,该方法能够有效应对电网电压快速频繁波动的问题,及时追踪电网电压波动,使SVG出力更加平滑、电压控制效果更好。

计及碳排放的综合能源配网日前与日内多时间尺度优化调度

为了降低碳排放量,提高新能源的消纳能力,考虑可削减、可转移、可替代3种需求侧响应负荷,计及碳交易机制,构建以碳排放量最少、综合运行成本最低、弃风弃光量最少、网络损耗最低为综合优化目标的风-光-电-气-储综合能源配网系统日前优化调度模型。以风电、光伏发电预测误差波动最小为日内优化调度优化目标,采用模型预测控制理论对风电、光伏的预测误差量进行反馈矫正,构建综合能源配网系统日内滚动优化调度模型。在修正的33节点配电网和20节点配气网组成的配网系统上进行仿真实验,验证了所提模型能够有效地降低碳排放量,提高风电、光伏的消纳能力,平滑负荷曲线,降低系统的综合运行费用。

考虑传输线动态增容风险的电力系统日前调度模型

为避免输电网传输通道不合理增容所引起的潮流热越限风险,在电力安全传输的保证下,合理提升新能源电力系统中各类资源的跨时空消纳能力,提出一种考虑传输线动态增容风险的电力系统日前调度模型。首先,建立了具有时变结构的ForecastNet输电线动态热容量极限(dynamic thermal rating,DTR)预测模型,该模型可以动态跟踪环境因素影响程度并修正预测网络权值,提高预测精度;其次,基于DTR日前预测结果确定输电线动态增容裕度,引入增容风险成本及风险偏好系数,构建面向电力市场报价机制的日前调度模型;最后,利用辽宁省实网数据在IEEE-39节点系统上对所提模型进行仿真。仿真结果验证了预测模型的准确性及调度模型的有效性,同时表明该调度模型可以充分利用输电网冗余传输空间,大幅提升可再生能源的消纳水平,保证电力市场环境下日前调度策略的安全性和经济性。