Edge Intelligence with Distributed Processing of DNNs:A Survey

Withthe rapiddevelopment of deep learning,the size of data sets anddeepneuralnetworks(DNNs)models are also booming.As a result,the intolerable long time for models’training or inference with conventional strategies can not meet the satisfaction of modern tasks gradually.Moreover,devices stay idle in the scenario of edge computing(EC),which presents a waste of resources since they can share the pressure of the busy devices but they do not.To address the problem,the strategy leveraging distributed processing has been applied to load computation tasks from a single processor to a group of devices,which results in the acceleration of training or inference of DNN models and promotes the high utilization of devices in edge computing.Compared with existing papers,this paper presents an enlightening and novel review of applying distributed processing with data and model parallelism to improve deep learning tasks in edge computing.Considering the practicalities,commonly used lightweight models in a distributed system are introduced as well.As the key technique,the parallel strategy will be described in detail.Then some typical applications of distributed processing will be analyzed.Finally,the challenges of distributed processing with edge computing will be described.

Boosting efficiency in state estimation of power systems by leveraging attention mechanism

Ensuring stability and reliability in power systems requires accurate state estimation, which is challenging due to the growing network size, noisy measurements, and nonlinear power-flow equations. In this paper, we introduce the Graph Attention Estimation Network (GAEN) model to tackle power system state estimation (PSSE) by capitalizing on the inherent graph structure of power grids. This approach facilitates efficient information exchange among interconnected buses, yielding a distributed, computationally efficient architecture that is also resilient to cyber-attacks. We develop a thorough approach by utilizing Graph Convolutional Neural Networks (GCNNs) and attention mechanism in PSSE based on Supervisory Control and Data Acquisition (SCADA) and Phasor Measurement Unit (PMU) measurements, addressing the limitations of previous learning architectures. In accordance with the empirical results obtained from the experiments, the proposed method demonstrates superior performance and scalability compared to existing techniques. Furthermore, the amalgamation of local topological configurations with nodal-level data yields a heightened efficacy in the domain of state estimation. This work marks a significant achievement in the design of advanced learning architectures in PSSE, contributing and fostering the development of more reliable and secure power system operations.

A New Solution Architecture for Online Power System Analysis

The current DSA system used in the dispatching control centers in China is a near real-time analysis system with response speed in the order of minutes.Based on a review of the state-of-the-art in online analysis and discussion of distributed data processing and computation architecture patterns,a new online analysis architecture is proposed.The primary goal of the new architecture is to increase the online analysis response speed to the order of seconds.A reference implementation of the proposed online analysis architecture to validate the feasibility of implementing the architecture and some performance testing results are presented.

Graph Computing Based Distributed Parallel Power Flow for AC/DC Systems with Improved Initial Estimate

The sequential method is easy to integrate with existing large-scale alternating current(AC)power flow solvers and is therefore a common approach for solving the power flow of AC/direct current(DC)hybrid systems.In this paper,a highperformance graph computing based distributed parallel implementation of the sequential method with an improved initial estimate approach for hybrid AC/DC systems is developed.The proposed approach is capable of speeding up the entire computation process without compromising the accuracy of result.First,the AC/DC network is intuitively represented by a graph and stored in a graph database(GDB)to expedite data processing.Considering the interconnection of AC grids via high-voltage direct current(HVDC)links,the network is subsequently partitioned into independent areas which are naturally fit for distributed power flow analysis.For each area,the fast-decoupled power flow(FDPF)is employed with node-based parallel computing(NPC)and hierarchical parallel computing(HPC)to quickly identify system states.Furthermore,to reduce the alternate iterations in the sequential method,a new decoupled approach is utilized to achieve a good initial estimate for the Newton-Raphson method.With the improved initial estimate,the sequential method can converge in fewer iterations.Consequently,the proposed approach allows for significant reduction in computing time and is able to meet the requirement of the real-time analysis platform for power system.The performance is verified on standard IEEE 300-bus system,extended large-scale systems,and a practical 11119-bus system in China.

A distributed spatial computing prototype system in grid environment

Digital Earth has been a hot topic and research trend since it was proposed,and Digital China has drawn much attention in China.As a key technique to implement Digital China,grid is an excellent and promising concept to construct a dynamic,inter-domain and distributed computing environment.It is appropriate to process geographic information across dispersed computing resources in networks effectively and cooperatively.A distributed spatial computing prototype system is designed and implemented with the Globus Toolkit.Several important aspects are discussed in detail.The architecture is proposed according to the characteristics of grid firstly,and then the spatial resource query and access interfaces are designed for heterogeneous data sources.An open-up hierarchical architecture for resource discovery and management is represented to detect spatial and computing resources in grid.A standard spatial job management mechanism is implemented by grid service for convenient use.In addition,the control mechanism of spatial datasets access is developed based on GSI.The prototype system utilizes the Globus Toolkit to implement a common distributed spatial computing framework,and it reveals the spatial computing ability of grid to support Digital China.

Graph Computing and Its Application in Power Grid Analysis

Approaches to apply graph computing to power grid analysis are systematically explained using real-world application examples.Through exploring the nature of the power grid and the characteristics of power grid analysis,the guidelines for selecting appropriate graph computing techniques for the application to power grid analysis are outlined.A custom graph model for representing the power grid for the analysis and simulation purpose and an in-memory computing(IMC)based graph-centric approach with a shared-everything architecture are introduced.Graph algorithms,including network topology processing and subgraph processing,and graph computing application scenarios,including in-memory computing,contingency analysis,and Common Information Model(CIM)model merge,are presented.

Fast Cycle Structure Detection for Power Grids Based on Graph Computing

The cycle structure in a power grid may lower the stability of the network;thus,it is of great significance to accu-rately and timely detect cycles in power grid networks.However,detecting possible cycles in a large-scale network can be highly time consuming and computationally intensive.In addition,since the power grid's topology changes over time,cycles can appear and disappear,and it can be difficult to monitor them in real time.In traditional computing systems,cycle detection requires considerable computational resources,making real-time cycle detection in large-scale power grids an impossible task.Graph computing has shown excellent performance in many areas and has solved many practical graph-related problems,such as power flow calculation and state estimation.In this article,a cycle detection method,the Paton method,is implemented and optimized on a graph computing platform.Two cases are used to test its performance in an actual power grid topology scenario.The results show that the graph computing-based Paton method reduces the time consumption by at least 60%compared to that of other methods.

基于正序瞬时功率算法的宽频振荡检测技术

现代电力系统表现出含高比例可再生能源和高比例电力电子设备的“双高”特征,电力系统中不同频率的谐波、间谐波与工频量相互作用,将可能导致宽频振荡,威胁电力系统安全稳定运行。针对无法在三相不平衡状态下实现宽频振荡准确检测的问题,提出了基于FPGA(现场可编程门阵列)和正序瞬时功率算法的宽频振荡检测技术。利用三相正序瞬时功率能滤除三相不平衡分量的原理,确保宽频振荡的检测结果不受三相不平衡的影响。利用FPGA并行计算的特性,大幅提升了宽频振荡检测算法的性能。测试结果表明,该技术能够在发生宽频振荡时准确检测出振荡分量,解决了现有技术存在误判和并行计算能力不足的问题。

基于流计算和大数据平台的实时交通流预测

目前交通流预测实时性差,很难满足在线分析和预测任务的需求,基于此提出一种Flink流计算框架和大数据平台结合的实时交通流预测方法。基于流计算框架实时捕捉和预处理数据,包括采用Flink的transform算子对数据进行校验和处理,将处理后的数据sink到大数据的HDFS文件系统,交由下一步的大数据并行框架进行分析建模与训练,实现基于流计算和大数据平台的实时交通流预测。实验结果表明,Flink能够实时捕捉和预处理交通流数据,把数据准时无误送入分布式文件系统中,在此基础上借助大数据框架下的并行分析和建模优势,在实时性数据分析与预测方面取得了较好的效果。

新型省级电网电磁暂态实时仿真建模方法

为了解决大规模省级电网电磁暂态仿真效率低下的问题,研究了一种适用于省级电网的电磁暂态实时仿真建模方法。分析了大电网电磁暂态建模难点,通过子系统建模和任务优化方法建立了省级电网模型,对各子网内部进行功率调节,实现了首次功率平衡;基于输电线路模型法原理完成了各子网之间输电线接口的联络,在预设平衡点的基础上,通过在线调节潮流的方法使得所建模型运行在合理平衡点上。为验证该方法的有效性,基于自主研发的实时仿真器(universal real-time experimental platform,UREP-300)对某新型省级电网进行实时仿真。仿真结果表明,所提方法能提高仿真效率,扩大了系统的仿真规模,减小了实时仿真机的计算负担,可为各类大规模系统的仿真与建模提供一种参考。

基于MapReduce的并行化电网运行数据处理方法研究

在大规模新能源的新型电力系统中,由于云端电力数据存在结构复杂、数据量大及多维度的特点,因此不利于发挥数据对运行的指导作用,甚至有可能危害电力系统的安全、稳定运行。针对上述问题,文中提出了一种基于MapReduce的电网数据分析方法。其将云计算应用于新型电力系统,并构建了基于MapReduce云计算模型的并行化处理算法,进而提升了系统的响应速度。通过将该方法应用于电网的数据处理结果表明,所提方法可以有效地提高电网运行数据处理的准确性和工作效率。在海量数据的工况下,其处理效率约为30 min,且随着数据量的增加仍可保持稳定性与准确性,实现了网格化的并行分析。

基于MI和EC的新型配电系统源荷监测协调策略

为了高效协调新型配电系统电源和负荷功率平衡,提出了一种基于多元信息和边缘计算的源荷功率监测协调策略。首先,分析了新型配电系统的特点和网格划分方法、协调其内部电源和负荷功率所涉及的多元信息构成;其次,通过在网格内部署具备边缘计算功能的智能终端,给出了前端智能终端和后端监控系统相互配合协调网格内源荷功率平衡的策略,设计了多元信息监测通信方案;最后,选取相关设备对所给策略进行了试验验证,并与传统配电网系统源荷协调策略进行了对比。结果表明:将复杂的新型配电系统进行网格划分,能够利用前端智能终端在1 s内、后端监控系统在10 s内协调网格内源荷功率平衡;网格内电能损耗使后端监控系统协调时间的变化值小于0.5 s、对前端协调策略的影响可以忽略不计;将不同协调手段相结合,具有协调时间短、难度低的优势。所给策略解决了传统配电系统以系统整体进行集中协调源荷功率时信息采集、计算量大和延时长的问题,能够提高源荷协调的效率和可靠性,对快速实现系统整体源荷功率平衡、保障其稳定运行具有重要的参考价值。

基于大数据的电力通信通道智能路由推荐策略研究

面对规模庞大、结构复杂的电力通信网,针对通信调度如何根据业务需求准确、合理、高效地安排通信通道,提出了一种基于分布式图并行计算的大数据智能路由推荐方法。该方法根据电力通信网特点构建电力通信网络拓扑模型、建立优选路由指标模型,并采用基于分布式图并行计算框架实现并行路由推荐算法。实验证明,该方法具有快速迭代与收敛的特性,在大型复杂的电力通信网络路由推荐与规划中具有良好的应用前景。

基于云计算的计算机自动化图像识别方法研究

云计算技术的发展为图像识别任务提供了新的机遇和挑战。基于此,提出了一种基于云计算的计算机自动化图像识别方法(CAIRM),利用分布式存储、并行处理和深度学习等关键技术,实现了高效、准确的图像识别。结果表明,CAIRM在多个公开数据集上取得了优异的识别性能,同时显著提升了计算效率和系统可扩展性,为大规模图像识别任务提供了可行的解决方案。

基于云计算的智能电网信息平台

坚强智能电网是未来电网的发展趋势,而信息平台是支撑坚强智能电网建设的重要基础。为了充分利用计算资源,满足智能电网对全部信息的可靠存储和高效管理的需要,提出基于云计算的智能电网信息平台,给出了该平台的体系结构,并详细分析了新方法的可行性、优势及需要解决的问题。针对智能电网状态监测的特点,结合Hadoop云计算技术,提出智能电网状态监测云计算平台的解决方案。研究云计算中的虚拟化、分布式存储与并行编程模型等问题,实现智能电网海量信息的可靠存储与快速并行处理。

云计算综述及电力应用展望

云计算是当前较为流行的信息新技术之一,技术发展快,已由概念阶段步入到应用实践阶段,全球公用事业用户对云计算的使用也呈快速增长趋势。本文以近几年国内外核心期刊文献为主要依据,针对虚拟化、并行计算、分布式等云计算相关技术,总结了当前技术的研究概况,分析了未来技术的发展趋势。在此基础上,结合智能电网、"三集五大"等业务的发展,对电力企业云计算的应用需求进行了全面梳理与展望,为电力企业云计算的发展与应用提供指导和参考。

大规模电网并行潮流算法

提出了一种大规模电力系统并行潮流算法。该算法将电力网络划分成若干个子网,以子网为计算节点、联络线为支路构造浓缩网格,进行潮流计算时通过双向迭代方法交替实现对网格和网格中计算节点的牛顿法线性增量方程的求解。该算法有效提高了电力系统潮流方程联立求解的效率,为大规模电力系统并行潮流计算提供了方法。在新英格兰测试系统和我国东北电网上进行了验算,结果验证了算法的有效性和合理性。