SIA OpenIR  > 工业控制网络与系统研究室
基于能量树模型的工厂能源管理系统感知与控制
其他题名Sensing and control of factory energy management systems based on the model of Energy Tree
汪扬1,2
导师曾鹏
分类号TP393.4
关键词需求响应 工厂能源管理系统 多智能体优化 分散控制 能量树
索取号TP393.4/W29/2014
页数132页
学位专业机械电子工程
学位名称博士
2014-12-03
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门工业控制网络与系统研究室
摘要当前的能源管理系统主要基于多介质/载体子网架构、集中式管理的离线优化、基于KPIs驱动的监控和对标,实现工厂能流网络的能效管理。为了满足节能减排的精细化能源感知和自适应实时控制,实现工厂同智能电网交互过程当中的能量需求响应自主调控;亟需以设备为中心的分散架构、能量泛在感知、高性能计算支持、以及实时优化控制的新型能源管理系统。为此希望通过物联网的信息通讯技术改变目前能量管理方式,包括:SOA架构的分散架构、无线传感器网络的泛在感知、分散云提供的高性能计算、智能电网的能源需求管理。使得从能量产生、转换、存储进行采集、建模、和分析,建立一种能够清楚的描述工厂能源网络中能量时空动态性的工厂能源管理系统。 考虑工厂能源网络的双向传输、价格波动、多载体能量、分散型能量资源的跨时空复杂性,通过信息通讯技术实现能够体现工厂能流分配关系和网络服务功能的能量认知模型;并设计出基于多智能体的优化分解协调方法,实现基于模型驱动的工厂能量需求响应调度控制。立足于工厂能效管理对物联网技术的重大需求,针对影响能源网络的架构、感知、计算瓶颈制约问题,研究了模型驱动的能量经济调控所需的多智能体架构、自适应感知、和分散云计算的基础理论。(1) 针对以设备为中心、兼顾机理和服务的细化能流的认知需求,提出了一种对能流机理和业务服务实现有效认知的能量树模型。该模型作为一种用于描述工厂能源网络面向多载体能量的全局模型,能够清晰的描述工厂能流在设备工序单元中输入输出关系、产生、转换、存储、传输、消费、丢失的机理,并用平衡方程和联接方程形式化表达其能量机理。同时为了支撑起能量树动态能量信息服务,定义了面向服务的架构7层网络协议参考模型,为过程单元提供可持续性的管理和维护。(2) 针对扁平化分散控制和3层分解协调优化的需求,提出了基于多智能体系统MAS能源网络架构。通过部署多智能体系统MAS,形成基于工序过程单元的多个子系统。多个过程单元子系统通过约束条件弱关联,从结构上能够实现工厂能源管理系统的需求响应任务的分解部署。提出基于多智能体目标优化算法,以及自顶至下的分解调度方法,使得系统获得快速简化,能够实现面向设备能效管理的分散控制。(3) 针对基于能量树模型的KPIs信息泛在感知需求,提出了基于感知压缩的自适应数据采集方法。通过将原始数据中冗余的部分丢弃掉,利用自适应的采样率形成有效的采样数据,减少通讯资源并降低能耗。采用非周期性LCS采样方法,并在满足误差容限的基础上假设采样频率小于数据率,有效降低数据的发送次数,达到自适应感知。(4) 针对多智能体高实时性大数据量分散云计算需求,提出了基于延迟敏感的分散云计算模型。由SCADA系统内部控制器和汇聚服务器、以及SCADA系统外部远程计算中心共同组建为分散型云计算架构。提出了一种基于空间随机过程的分散型云计算溢出流丢失问题分析模型--多M/M/m/m+M排队系统。该模型基于丢弃率参数能够实现延迟敏感的服务请求准入控制,使得不能满足延时需求的计算服务快速溢出、快速切换,满足高实时性、大数据需求的计算任务。(5) 针对定义多智能体之间主导关系和能源需求响应实时控制需求,提出了基于HJB贪婪策略的工厂能源需求响应分散控制。通过为工业设施组件建立数学物理方程,提出工厂能源管理系统需求响应的优化控制。集中式的优化控制依照不同的优化目标被分割为多个分散式的优化单元。工厂能源管理系统将这些优化单元部署为多智能体系统,为之建立优化单元之间基于斯塔克尔贝格领导者模型的主导型关系,并求解出各个优化单元基于哈密尔顿-雅克比-贝尔曼贪婪策略的最优控制律。基于能量树模型及其基于多智能体优化控制,设计了面向工厂能源网络基于多智能优化模型的能量需求响应算法。将模型驱动的分散控制应用于以设备为中心的能量动态感知和决策计算之中,提出基于能量树模型的能源管理系统认知模型与优化控制,并通过工业设施能量需求响应数值仿真做出了模型验证。
其他摘要Current energy management system is mainly based on multi-media/carrier subnet architecture, centralized management of offline optimization, and KPIs drived monitoring and benchmarking to achieve factory energy management of energy flow network. In order to meet fine energy-aware perception and adaptive real-time control for energy saving, and to achieve the autonomous ability of scheduling and controlling for energy demand response in the interactions between factories and smart grids, an urgent need for a novel energy management system demands device-centric decentralized architecture, ubiquitous energy-aware perception, high-performance computing support and real-time optimal control. For this reason the current way of energy management is expected to be changed through information and communication technologies based Internet of Things, including: SOA based decentralized architecture, ubiquitous sensing using WSNs, high-performance computing through decentralized cloud, and energy demand management by smart grid. The energy process of generation, conversion, storage can be sampled, be modeled, and be analyzed; furthermore, a clear description about the spatio-temporal dynamics of factory energy networks can be demonstrated by the factory energy management system.Considered the inter-spatio-temporal complex of factory energy network incurring bi-directional transmission, energy price volatility, multi-carrier energy, and decentralized energy resources, energy cognitive model embodies the allocation of energy flow and the functions of networked services through information and communication technology; meanwhile, the decomposition and coordination of multi-agent optimization is designed to achieve the model-driven scheduling and controlling for factory energy demand response. Based on major demand of factory energy management for networking technology, forcusing on the bottleneck problem that architecture, perception, and computation impact energy network, research on model-driven economic dispatch on energy demand is done around these fundamental theories on multi-agent architecture, adaptive perception, and decentralized cloud computing. (1) Focused on the device-centric cognitive demand for refined energy flow both considering on mechanism and services, a cognitive model of energy tree is proposed to denote the mechanism of energy flow and the services of business process. As the global model describing the factory energy network incurring multi-carrier energy, energy tree is able to clearly describe the input-output relationships of factory energy flow in device process units and the mechanism of generation, conversion, storage, transmission, consumption, loss with balance equations and interconnection equations as mechanism formal expression. Meanwhile, in order to prop up the information services of energy tree dynamics, a 7-layer network protocol reference model defines sustainable communication services, management and maintenance for the process units. (2) Focused on the flat decentralized control demand for 3-tiers optimization of decomposition and coordination, multi-agent systems are used in the architecture of energy network. Through the deployment of multi-agent system MAS, multiple subsystems are re-established based on step process units. Multiple subsystems of process units establish weak association by constraints; hence, the decomposition deployment of demand response tasks is achieved by factory energy management systems. Using the multi-agent optimization and the top-to-bottom decomposition, the system is fast simplified, devices energy management achieves decentralized control. (3) Focused on the ubiquitous perception demand on KPIs information of energy tree model, adaptive data sampling method is proposed based on sensing compression. The redundant portion among original data is discarded, adopting adaptive sampling rate to form a more effective sampling data reduce communication burden and power consumption. On the condition that error tolerance can be satisfied, and on the hypothesis that the sampling frequency is less than the data rate, LCS aperiodic sampling method reduces the number of transmission data, and to achieve adaptive perception. (4) Focused on the decentralized cloud computing demand on high real-time and big data computation, a delay-sensitive computational model on decentralized cloud computing is proposed. Controllers and aggregation servers inside SCADA system, and remote computing centers outside SCADA system jointly constitutes the architecture of decentralized cloud computing. A Performance Modeling of Decentralized Cloud Computing Based on Multiple M/M/m/m+M Queuing Systems is proposed based on the stochastic process analysis model of the overflow loss problems. The admission control of delay-sensitive service requests could be implemented based on abandonment rate, while waiting service requests are fast overflowed at a certain abandonment probability, and to be quickly switched to meet the computing task demand on high real-time big data.(5) Focused on the demand on the definition of the leading relationship among multi-agents and real-time control of energy demand response, decentralized control based on HJB greedy strategy is proposed. Industrial facility components are modeled into several mathematical physics equations, the demand response of a factory energy management system is formalized into an optimal control. Furthermore, the centralized optimal control is divided into multiple decentralized optimal units in term with different optimal objectives. Factory energy management systems deployment the optimal units as a multi-agents system, which builds the leading relationship between optimal units according to Stackelberg Leader model. The optimal control laws of the optimal units are solved by Hamilton-Jacobi-Bellman Greedy Strategy.Based on energy tree model and correlated multi-agent optimization and control, an energy demand response algorithm is designed for energy network optimization based on multi-agent systems. The model-driven decentralized control is applied to the device-centric computing including power dynamic perception and decision-making. Energy management systems and optimal control is proposed based on the cognitive model of energy tree. A numerical simulation of energy demand response in a specific industrial facilitie is used for model verification.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/16744
专题工业控制网络与系统研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
汪扬. 基于能量树模型的工厂能源管理系统感知与控制[D]. 沈阳. 中国科学院沈阳自动化研究所,2014.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
基于能量树模型的工厂能源管理系统感知与控(13413KB)学位论文 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[汪扬]的文章
百度学术
百度学术中相似的文章
[汪扬]的文章
必应学术
必应学术中相似的文章
[汪扬]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。