A Model Predictive Control Based Distributed Coordination of Multi-microgrids in Energy Internet
-
Abstract: This paper focuses on the development of optimization-based distributed scheduling strategies for the coordination of an energy internet (EI) with multi-microgrids with consideration of forecast uncertainties. All microgrids have flexible loads, schedulable loads and critical loads; some microgrids have distributed generators, such as micro-turbines, wind turbines, photovoltaic panels; besides, a few microgrids have energy storage devices, such as battery storage. Each microgrid is considered as an individual entity and has its individual objective, these objective functions of microgrids are formulated by mixed integer programming (MIP) models. A game theory based parallel distributed optimization algorithm is proposed to coordinate the competitive objectives of the microgrids with only a little information interaction. A model predictive control (MPC) framework which integrates the distributed optimization algorithm is developed to reduce the negative impacts introduced by the uncertainties of the EI. Simulation results show that our method is flexible and efficient.
-
Key words:
- Energy internet (EI) /
- game theory /
- model predictive control (MPC) /
- multi-microgrids /
- parallel optimization
摘要: 研究了预测不确定性条件下含多个微电网的能源互联网分布式协同调度策略.各微电网都拥有多种智能负荷,如功率可调负荷、可调度负荷和关键负荷;部分微电网含有分布式电源,如微型燃气轮机、风电机组、光伏发电系统等;且部分微电网还拥有储能设备,如电池储能系统.每个微电网都可当做一个独立的实体,拥有自己的运行目标,这些运行目标可表示成混合整数规划模型.提出了基于并行分布式优化的博弈模型以较小的信息通信量协调各微电网带有竞争性的运行目标.在此基础上,引入模型预测控制(MPC)机制以降低能源互联网中风、光等可再生能源输出、负荷需求及电价波动的不确定性产生的不利影响.算例证明了本文所提方法的可行性和有效性. -
磨浆过程主要为后续造纸过程提供满足相应物理特性的纸浆纤维, 同时也是保证纸品质量的前提.但由于磨机物理结构复杂性以及磨浆运行过程中受外界不确定因素的影响, 导致纤维长度具有较强的随机分布特性, 而传统的均值或者方差并不足以表征整个纤维长度随机分布(Fiber length stochastic distribution, FLSD)特征, 即具有非高斯分布特性[1].而事实上, FLSD形状作为衡量磨浆过程中最主要的生产指标之一, 不仅影响到纸浆脱水效率和后续造纸过程的电耗, 而且直接影响到整个制浆和造纸工业的能耗和最终的纸品质量[2-6].在这种情况下, 就必须寻找一种能够对磨浆过程输出FLSD进行有效控制的方法.
虽然一些学者较早地意识到FLSD在纸浆生产过程中的重要作用, 但目前仍利用离线获得的纤维长度的均值和方差作为衡量纸浆纤维质量的生产指标[3-9].如文献[3$-$6]采用长纤维百分含量作为评价纸浆质量的工艺指标, 而事实上, 这种以统计纤维束长度的均值和方差并足以有效表征整个纤维长度的分布特征, 甚至会导致FLSD信息的缺失, 这主要因为木片等富含纤维的纸浆原料经过盘磨机的横向挤压和纵向帚化后, 虽然使得纤维束逐渐分解成单根纤维, 然而, 也导致纤维长度的分布形状具有很强的随机性和不确定性.此外, 加之纤维长度分布在线检测仪器缺失, 使得通过离线纤维长度的均值和方差来调整过程操作变量, 其检测的滞后性难以使纸浆纤维质量稳定在工艺要求范围内, 致使长纤维百分含量的控制大多过度依赖操作人员的经验, 然而人工调整主观性较强, 常常造成工况波动大, 严重影响磨浆过程生产指标的稳定性.
另一方面, 现有针对随机动态系统的最小方差控制、自校正控制、随机线性二次型控制, 均假设系统服从高斯分布, 主要集中针对过程输出随机变量的均值和方差进行建模和控制, 然而, 针对输出变量服从非高斯分布的有界动态随机系统, 1996年王宏教授提出了直接设计控制器以使输出PDF形状跟踪期望PDF形状的控制策略, 并系统地提出了多种建模和控制方法[10-18].这类控制策略包含了传统以输出均值和方差为目的随机控制方法, 具有更为广泛的应用.近些年, 随机分布控制理论已成功应用于各类具有随机分布动态特性的工业过程, 如造纸过程的絮凝粒径分布[10, 12]、燃烧过程的火焰分布[18-19], 聚合过程的分子量分布[20], 铜粗选过程的泡沫尺寸分布[21]等, 这些过程输出随机变量均不能满足高斯分布特征, 并且具有较强的随机分布动态特性.
另外, 在制浆和造纸工业领域, 目前已有多种用于测量纸浆纤维各种形态参数的在线自动化检测和分析仪器, 这些检测仪器常采用数码CCD摄像机获取的纤维图像经数字化后传输到计算机系统进行处理, 通过实时二维图像分析软件将每根纤维从图像信号中识别出来, 逐一测量纤维的形态参数, 能迅速准确地获得纤维形态参数是测量结果.如加拿大Optest公司FQA-360、芬兰Kajaani公司FS-300、丹麦Fiber-Visions等[8-9], 这些先进测量仪器为研究基于磨浆过程输出FLSD的建模及控制提供了技术支撑.
磨浆过程输出FLSD具有典型的非高斯分布动态特性, 采用传统纤维长度的均值和方差难以有效描述其分布特征, 本文根据随机分布相关控制理论[10], 利用RBF神经网络逼近输出FLSD的PDF, 为了改善传统线性权值模型[14-19]精度不高、泛化能力不强等问题, 采用随机权神经网络(Random vector functional-link networks, RVFLNs)[22-24]建立表征输入变量和权值向量之间的非线性模型, 基于磨浆过程输出FLSD模型提出了一种预测PDF控制方法, 实现了对输出FLSD形状的跟踪控制, 基于工业数据实验表明了所提方法的有效性.
1. 磨浆过程描述
典型的磨浆过程工艺流程如图 1所示, 磨浆过程即是将植物原料经盘磨机反复研磨后, 经汽浆分离后获得造纸所需的纸浆纤维.其主要包括喂料系统, 供水系统, 液压伺服系统和磨盘调速系统.当磨机运行时, 首先将经过将被筛选木片在蒸煮仓里进行高温预处理, 经清除杂质后的木片在螺旋喂料器作用下送入磨室.磨机作为磨浆过程中的核心设备, 主要有定盘、动盘、电液伺服装置和主电机等组成.当预处理后木片和稀释水注入磨区后, 利用电液伺服装置可以实时调节磨盘间隙, 动盘在主电机带动下通过机械摩擦、剪切、撕裂、切割等作用, 最终将预处理后木片分解为单根纤维.然后, 纸浆通过送入旋风分离器实现汽浆分离, 最终获得满足造纸过程所需的纸浆纤维.可以看出, 由于工艺流程长及现场环境恶劣等原因, 若操作变量调节不当, 即便通过磨机的反复研磨, 也难以获得满意的FLSD形状, 这样不但导致纸浆质量无法满足工艺要求, 而且也极易造成原料浪费和过程能耗过大.
磨浆过程的主要操作变量包括磨盘间隙、稀释水流量、动盘转速和螺旋喂料量等.研究表明:长纤维百分含量作为衡量磨浆过程纤维长度均值的工艺指标, 主要与注入磨区的稀释水流量和磨盘间隙密切相关[3, 5-6], 由于磨浆过程主要目的是对纤维进行切断、压溃、吸水膨胀, 最终使纤维束分离为单根纤维, 所以, 磨盘间隙直接影响纸浆纤维被切断和压溃强度.同时, 为使植物纤维能够较好地分离为单根纤维, 需要让纤维获得足够的水份进行膨胀, 纸浆纤维的吸水膨胀程度主要取决于注入磨区的稀释水流量.而植物纤维被切断、压溃强度以及吸水膨胀程度决定了最终获得的纤维长度随机分布形状.另外, 结合实际工程经验, 通常情况下分别通过改变喂料螺旋转速和供水泵转速来调节螺旋喂料量和稀释水流量, 根据工程实践经验, 当产量一定的情况下, 动盘转速是固定不变的, 螺旋喂料量也是恒定的.因此, 稀释水流量和磨盘间隙不但可以看作影响长纤维百分含量的主要变量, 同样也可以作为影响最终FLSD形状的关键变量, 对整个制浆生产流程都起着极为关键的作用.因此, 本文将稀释水流量和磨盘间隙作为影响磨浆过程输出FLSD形状的关键输入变量.
2. 控制策略
根据有界动态随机分布控制相关理论[10]可知, 随机分布系统模型主要由随机变量的PDF输出部分和权值与输入变量之间的动态部分组成.为了表示输入变量和输出PDF之间的动态关系, 常引入一组基函数(如B样条基函数[16-17, 20-21]、RBF基函数[15, 18-19])来逼近输出随机变量的PDF, 通过调节基函数的权值来控制输出PDF形状, 这样将随机分布系统的输出PDF和输入变量之间的动态关系转化为权值向量和输入变量之间的动态关系, 最终通过对相对应权值的控制实现对输出PDF形状的动态调节.为此, 本文针对磨浆过程输出FLSD提出预测PDF控制策略如图 2所示, 具体如下:
1) 首先, 需寻找一组合适的RBF基函数来逼近输出FLSD的PDF.采用RBF神经网络逼近输出PDF的均方根, 基于迭代学习方法实现RBF基函数参数整定, 并对实际输出PDF相应权值向量进行估计.
2) 其次, 针对常规线性权值向量模型精度不高, 泛化能力不强等缺点, 本文基于随机权神经网络[22-24]方法建立输入变量和前$n-1$个权值向量之间的非线性模型, 进而获得磨浆过程输出FLSD模型.
3) 最后, 基于输出FLSD模型设计预测PDF控制器, 使得输出PDF获得良好的目标跟踪能力.
3. 纤维长度随机分布预测PDF控制
3.1 随机分布系统模型
随着数据采集技术和检测仪器的快速发展, 对随机变量的输出PDF等已经有了较为成熟的检测方法.为了方便描述各种随机过程, 假设为描述动态随机系统输出的一致有界随机过程变量, 为$k$时刻控制随机系统分布形状的输入向量, 这表明在任一采样时刻$k$, $z(k)$就可以通过其概率密度函数来描述, 其定义式如下:
$ P({a<z(k)<\zeta}, {u(k)})=\int_{a}^\zeta{\gamma(y, u(k))}\textrm{d}y $
(1) 式中, $P({a<z(k)<\zeta}, {u(k)})$表示随机系统在$u(k)$作用下输出落在区间$[a, \zeta]$内的概率, 即$z(k)$的输出PDF${\gamma(y, u(k))}$形状由输入变量$u(k)$控制.假设区间$[a, b]$为已知, 并且输出PDF${\gamma(y, u(k))}$连续且有界, 采用如下具有高斯型RBF神经网络来逼近输出的PDF均方根
$ R_l(y)=\textrm{exp}\left(-\frac{(y-\mu_l)^2}{\sigma_l^2} \right), l=1, 2, \cdots, n $
(2) 式中, $l$表示第$l$个网络节点, $n$为网络节点总数, $\mu_l$和$\sigma_l$分别表示第$l$个网络节点函数的中心值和宽度.根据RBF神经网络逼近原理, 此时输出PDF$\gamma(y, u(k))$的均方根可以表示为
$ \sqrt{\gamma(y, u(k))}=\pmb C(y)\pmb V(k)+R_n(y){\omega_n(k)}+e_0(y, k) $
(3) 式中, $\pmb C(y)=[R_1(y), R_2(y), \cdots, R_{n-1}(y)]$, , $\omega_{n}(k)$为第$n$个基函数相对应的权值, $e_0(y, k)$为逼近误差.此外, 由于输出PDF需要满足隐含条件
$ \int_{a}^b{\gamma({y, u(k)})}\textrm{d}y=1 $
(4) 为便于分析忽略逼近误差$e_0(y, k)$, 则第$n$个权值$\omega_n(k)$可用权值向量的非线性函数$h(\pmb V(k))$表示为
$ h(\pmb V(k))=\frac{{\sqrt{\Sigma_2-\pmb V^\textrm{T}(k)\Sigma_3\pmb V(k)}}-\Sigma_1\pmb V(k)}{\Sigma_2} $
(5) 式中, , \textrm{d}\textit{y}, $\Sigma_2=\int_{a}^{b}{{R_n^2(y)}}\textrm{d}y$, $\Sigma_3=\Sigma_2\Sigma_0-\Sigma_1^{\rm{T}}\Sigma_1.$
从式(5)可以看出, 若保证非线性函数$h(\pmb V(k))$存在, 需满足如下约束条件:
$ {\pmb V}^\textrm{T}(k)\Sigma_4 \pmb {V}(k)<1 $
(6) 式中, $\Sigma_4=\Sigma_2^{-1} \Sigma_3>0$, 不等式(6)可以看作在动态权值向量$\pmb V(k)$需满足的约束条件.由式(5)可以看出, 一旦所有的基函数确定, 由于有界区间$[a, b]$已知, $\pmb C(y)$和$R_n(y)$均已知, 这就意味着在$n$个权值中有$n-1$个是相互独立的.
当实际工业过程中输出随机变量的PDF可测量时, 可通过如式(7)$ \sim $(9)方法对相应的权值进行估计.结合式(3)和式(5)所示的输出PDF的均方根可以表示为
$ \sqrt{\gamma(y, u(k))}=\left[ \begin{array}{cc} \pmb C(y)&R_n(y)\\ \end{array} \right]\left[ \begin{array}{c} \pmb V(k)\\ h(\pmb V(k))\\ \end{array} \right] $
(7) 对式(7)两边左乘$[\pmb C(y)R_n(y)]^{\rm T}$, 并对两端在区间$[a, b]$上进行积分, 由此可得到:
$ \left[ \begin{array}{cc} A_1\\ A_2\\ \end{array} \right] =\left[ \begin{array}{cc} \Sigma_0 & \Sigma_1 \\ \Sigma_1^{\rm{T}} & \Sigma_2 \\ \end{array} \right] \left[ \begin{array}{cc} \pmb V(k)\\ h(\pmb V(k))\\ \end{array} \right] $
(8) 式中, , $A_2=\int_{a}^{b}{R_n}(y)$ $\sqrt{\gamma(y, u(k))}\textrm{d}y$.
当矩阵非奇异时, 式(8)所示权值可以通过矩阵求逆获得:
$ \left[ \begin{array}{cc} \pmb V(k)\\ h(\pmb V(k))\\ \end{array} \right] =\left[ \begin{array}{cc} \Sigma_0 & \Sigma_1 \\ \Sigma_1^{\rm{T}} & \Sigma_2 \\ \end{array} \right]^{-1}\left[ \begin{array}{c} A_1\\ A_2\\ \end{array} \right] $
(9) 式(9)揭示了输出随机变量的PDF与权值向量之间的关系, 可以看出当RBF基函数确定后, 只要输出PDF可测量, 便可很容易获得相应的权值向量.通常情况下在获得相应的权值向量之后, 采用最小二乘法或子空间辨识法获得权值和输入变量之间的线性动态模型[15, 18-20]. 从式(9)可以看出, 在RBF基函数已知的情况下, 若要获得理想的输出FLSD模型.首先, 需要对不同时刻实际输出PDF进行权值估计.其次, 建立输入变量和权值向量之间动态模型.然而, 由于实际工业过程高度非线性、机理复杂等原因存在, 采用常规线性模型难以有效描述权值向量的动态特性.为此, 必须采用有效手段, 获取基于磨浆过程输出FLSD模型, 以实现FLSD形状的在线连续估计及控制.
另外, 对于不能或者难以获得机理模型的复杂工业过程, 基于数据驱动建模方法通常被看作一种非常有效的替代手段.目前, 常见的数据驱动建模方法主要有支持向量机方法[25]、模糊推理方法[26]、案例推理方法[27]以及随机权神经网络方法[22]等.其中, 随机权神经网络作为一种简单易用、有效的单隐层前馈神经网络学习算法, 在保证逼近任意连续函数的前提下, 采用随机给定神经元隐含层权值和偏置, 通过计算隐含层输出矩阵的广义逆建立学习网络, 克服了传统单隐层神经网络的缺点, 由于训练速度快, 模型结构简单、易于实现以及泛化能力强等鲜明特点, 在很多领域获得广泛应用[23-24].本文利用随机权神经网络方法建立输出变量和前$n-1$个权值向量之间非线性预测模型.因此, 最终磨浆过程输出FLSD模型可以表示为
$ \left \{\begin{aligned}\pmb V(k+1)=f \left( \pmb V(k), \pmb u(k) \right)\\ \sqrt{\gamma(y, \pmb u(k))}=\pmb C(y)\pmb V(k)+R_n(y)h(\pmb V(k)) \end{aligned}\right. $
(10) 式中, $f(\cdot)$为表示过程输入和权值之间的非线性表达式, 可以看出在控制输入和输出PDF可测量情况下, 同时在RBF基函数已知时, 在利用式(9)获得权值向量之后, 通过随机权神经网络方法很容易获得输入变量和权值向量之间的非线性模型.
从式(10)明显发现要获得理想的输出FLSD模型, 首先需要选择一组合适的RBF基函数, 若RBF基函数的中心值和宽度选择不当, 不但影响PDF输出部分的近似精度, 而且也难以获得满意的输出FLSD模型.因此, RBF基函数的选择对于磨浆过程输出FLSD的模建模精度显得至关重要.
3.1.1 RBF基函数参数整定
为了提高随机分布模型输出PDF对实际输出PDF逼近精度, 文献[15]中提出了一种基于迭代学习控制原理的RBF基函数参数更新算法, 假设给出初始的RBF基函数参数, 结合实际输出PDF数据, 通过式(9)可以获得初始的权值向量, 并将此初始向量乘以相应的初始RBF基函数, 便可获得近似的输出PDF.基于迭代学习控制原理, RBF基函数的中心值和宽度要根据上一迭代学习周期的逼近误差来调节, 通过对RBF基函数参数的反复调节, 最终使得将近似输出PDF与实际输出PDF之间误差最小.因此, 所采用的性能指标主要考察近似输出PDF跟踪期望的输出PDF能力, 所以所用误差来自迭代周期内的每一个采样点, 采用如下性能指标
$ J_{p}(i)=\int_a^b \left({\sqrt{\gamma_{p+1}(y, u_i)}-{\sqrt{g(y)}}}\right)^2\textrm{d}y $
(11) 式中, $J_{p}(i)$可以表示为第$p$个迭代周期后第$i$个采样时刻RBF神经网络近似输出PDF的均方根与期望输出PDF$g(y)$的均方根之间的逼近性能指标.同时, 定义在第$p$个迭代周期后$M$个采样点近似输出PDF分别与期望输出PDF之间构成性能指标向量可表示如下
$ \pmb E_{p}=[J_{p}(1), J_{p}(2), \cdots, J_{p}(M)]^{\rm T} $
(12) 为了实现式(10)所示的RBF基函数参数整定, 在相邻的第$p$次和第$(p+1)$次迭代学习周期内采用如下P型迭代学习控制率
$ \begin{cases} \mu_{l, p+1}=\mu_{l, p}+{\pmb\alpha}_{\mu}{\pmb E_{p}} \\ \sigma_{l, p+1}=\sigma_{l, p}+{\pmb \beta}_{\sigma}{\pmb E_{p}}\\ \end{cases} $
(13) 式中, 学习参数$\alpha_\mu$和$\beta_\sigma$分别定义如下
$ \left\{ \begin{align} & {{\pmb\alpha}_{\mu}}={{\zeta }_{\mu }}[{{\mathit{\lambda }}_{1}},{{\mathit{\lambda }}_{2}},\cdots ,{{\mathit{\lambda }}_{\mathit{M}}}] \\ & {{\pmb \beta}_{\sigma}}={{\zeta }_{\sigma }}[\mathit{\lambda }{{\mathit{'}}_{1}},\mathit{\lambda }{{\mathit{'}}_{2}},\cdots ,\mathit{\lambda }{{\mathit{'}}_{M}}] \\ \end{align} \right. $
其中, $\lambda$和$\lambda'$分别为学习元素, 参数$\zeta_\mu$和$\zeta_{\sigma}$分别为迭代学习率.从式(13)看出所有元素均为非负, 这表明迭代学习率可以为正数也可以为负数, 这就意味着RBF基函数的中心值和宽度随着迭代学习次数呈现出增加或者降低趋势.
3.1.2 基于RVFLNs权值模型
从式(10)可以看出当前时刻输出PDF不但与磨浆过程输入变量有关, 同时与前一时刻的输出PDF形状也密切相关, 而输出PDF通过调节RBF基函数中心值和宽度以及相应的权值, 所以在获得理想RBF基函数参数之后, 通过式(9)可以获得所有时刻输出PDF相对应权值向量, 然而由于权值向量之间相互耦合, 因此, 权值向量模型可以看作是一个多输入多输出的回归建模问题.为此, 采用基于RVFLNs[22-24]建立输入变量与前$n-1$个权值向量之间的非线性模型.
假设分别有$m$个输入变量与权值向量组成的的样本集合, 其中为磨浆过程的输入变量, 表示$n$个权值中的前$n-1$个权值向量.对于一个有$L$个隐层节点, 若以$g(x)$作为激活函数的RVFLNs输出可以表示为
$ f_{\textrm{R}}({\boldsymbol{u}}_i)=\sum\limits_{j=1}^L {\pmb \beta_j g(\pmb \omega_j \cdot \pmb u_i+b_j)} $
(14) 式中, $\pmb u_i$为磨浆过程的输入变量, 为$m$个输入节点连接第$j$个隐含层的输入权重, 为第$j$个隐含层连接输出节点的输出权重, $b_j$是第$j$个隐层单元的偏置, $\pmb \omega_j \cdot \pmb u_j$表示和$\pmb u_i$的内积.
随机权神经网络和其他单隐层神经网络学习目标一样均是使得模型输出与实际输出$\pmb V_i$之间误差最小, 即有.当存在$\pmb \beta_j$, $\pmb \omega_j$和$\pmb b_j$, 使得
$ \pmb V_i=\sum\limits_{j=1}^L {\pmb \beta_j g(\pmb \omega_j \cdot \pmb u_j+b_j)} $
(15) 此时, 将式(15)可以矩阵表示为
$ {H} {\beta}={Y} $
(16) 式中, $H$为隐含层输出矩阵, $\beta$为输出权重, $Y$为预测模型的权值输出.且有
$ \begin{array}{l} H\left( {{\mathit{\boldsymbol{\omega }}_1}, \cdots ,{\mathit{\boldsymbol{\omega }}_L},{b_1}, \cdots ,{b_L},{\mathit{\boldsymbol{u}}_1}, \cdots ,{\mathit{\boldsymbol{u}}_L}} \right) = \\ \;\;\;\;{\left[ \begin{array}{l} g\left( {{\mathit{\boldsymbol{\omega }}_1} \cdot {\mathit{\boldsymbol{u}}_1} + {b_1}} \right)\;\;\; \cdots \;\;\;g\left( {{\mathit{\boldsymbol{\omega }}_L} \cdot {\mathit{\boldsymbol{u}}_1} + {b_L}} \right)\\ \;\;\;\;\;\;\;\;\; \vdots \;\;\;\;\;\;\;\;\;\;\;\;\;\;\; \cdots \\ g\left( {{\mathit{\boldsymbol{\omega }}_1} \cdot {\mathit{\boldsymbol{u}}_N} + {b_1}} \right)\;\; \cdots \;\;\;g\left( {{\mathit{\boldsymbol{\omega }}_L} \cdot {\mathit{\boldsymbol{u}}_N} + {b_L}} \right) \end{array} \right]_{N \times L}} \end{array}\\ \beta = {\left[ {\begin{array}{*{20}{c}} {\mathit{\boldsymbol{\beta }}_1^{\rm{T}}}\\ \vdots \\ {\mathit{\boldsymbol{\beta }}_L^{\rm{T}}} \end{array}} \right]_{L \times (n - 1)}}\quad ,\quad Y = {\left[ {\begin{array}{*{20}{c}} {\mathit{\boldsymbol{V}}_1^{\rm{T}}}\\ \vdots \\ {\mathit{\boldsymbol{V}}_L^{\rm{T}}} \end{array}} \right]_{N \times (n - 1)}} $
从式(16)可以看出, 当输入权重和隐层偏置$b_j$被随机确定后, 只需调整输出层权值就可以使网络具有较好的逼近性能.为了能够较好地训练上述网络, 希望获得最优的输出权重$\hat{\beta}$, 使得
$ \hat{\beta}=\arg\min\limits_{\beta}\|H{\beta}-Y\|^2 $
(17) 此时随机权神经网络的学习问题就转化为式(16)所示的线性系统${H\beta}=Y$的最小二乘求解问题, 为此隐层输出矩阵$H$就能被唯一确定, 此时可以获得输出权重$\beta$可以表示为
$ \hat{\beta }={{H}^{\dagger }}Y $
(18) 式中, $ {{H}^{\dagger }}$是矩阵$H$的\textrm{Moore-Penrose}广义逆.可以看出此算法只需要设置网络的隐层节点个数, 便可以随机初始化输入权重和偏置并得到相应的输出权重.所以该算法在执行过程中不需要调整网络的输入权值以及隐元的偏置, 便可以获得唯一的最优解.
为了更好地反映磨浆过程输出FLSD动态特性, 将当前采样时刻输入变量, $m$为输出变量个数以及当前时刻相对应的权值向量作为非线性模型综合输入, 即建立的权值动态预测模型用于实现如下的非线性动态映射关系:
$ \pmb V_{\textrm{m}}(k+1)=f_{\textrm{R}}\{\pmb V(k), \pmb u(k)\} $
(19) 式中, $\pmb V_{\textrm{m}}(k+1)$为模型输出的前$n-1$个权值向量.此时, 在第$k$时刻预测输出PDF为
$ {\sqrt{\gamma_{\textrm{m}}(y, k+1)}}= \pmb C(y)\pmb V_{\textrm{m}}(k+1)+R_n(y) {}\nonumber \\ {}h(\pmb V_{\textrm{m}}(k+1))= \pmb C(y) f_{\textrm{R}}\{\pmb V(k), \pmb u(k)\}+{}\nonumber \\ {}R_n(y)h(f_{\textrm{R}}\{\pmb V(k), \pmb u(k)\}) $
(20) 3.2 预测PDF控制
由于实际工业过程中非线性、时变、模型失配和随机扰动等不确定性因素的存在, 模型输出PDF难以与实际输出PDF完全一致, 然而在滚动优化过程中, 需要实际输出PDF与模型输出PDF保持一致, 因此, 通常采用反馈校正来降低过程的不确定性对系统性能的影响, 提高系统的控制精度和鲁棒性.假设模型在$k$时刻第$j$步预测输出PDF为
$ {\sqrt{\gamma_{\textrm{m}}(y, k+j)}}=\pmb C(y)\pmb V_{\textrm{m}}(k+j)+ \nonumber \\ \qquad{}R_n(y)h(\pmb V_{\textrm{m}} {}(k+j))={}\nonumber \\ \qquad{}R_n(y\pmb C(y) f_{\textrm{R}}\{\pmb V(k+j-1), \pmb u(k+j-1)\}+{}\nonumber \\ \qquad{}R_n(y)h(f_{\textrm{R}}\{\pmb V(k+j-1), \pmb u(k+j-1)\}) $
(21) 另外, 在第$k$时刻实际输出PDF和预测输出PDF之间的误差为
$ e(y, k)=\sqrt{\gamma(y, k)}-\sqrt{\gamma_{\textrm{m}}(y, k)}={}\pmb C(y)(\pmb V(k)-\nonumber \\ \qquad{}f_{\textrm{R}}\{\pmb V(k-1)\pmb u(k-1)\})+R_n(y)\cdot\nonumber \\ \qquad{}{(h(\pmb V(k)-h(f_{\textrm{R}}\{\pmb V(k-1), \pmb u(k-1)\}))} $
(22) 利用该误差对第$j$步预测输出PDF ${\sqrt{\gamma_{\textrm{m}}(y, k+j)}}$进行反馈修正, 补偿后预测输出PDF为
$ \sqrt{\gamma_{\textrm{p}}(y, k+j)}=\sqrt{\gamma_{\textrm{m}}(y, k+j)}+\beta_j e(y, k) $
(23) 式中, $\beta_j(0<\beta_j<1)$为校正系数.此时结合式(21)$ \sim $(23)在$k$时刻第$j$步期望输出PDF和补偿后的预测输出PDF之间误差为
$ e_{\textrm{p}}(y, k+j)=\sqrt{\gamma_{\textrm{g}}(y, k+j)}- \sqrt{\gamma_{\textrm{p}}(y, k+j)}= {}\nonumber \\ {} \pmb C(y)[\pmb V_{\textrm{g}}(k+j)-\nonumber \\ {}f_{\textrm{R}}\{\pmb V(k+j-1), \pmb u(k+j-1)\}] +{}\nonumber \\ {} R_n(y)[h(\pmb V_{\textrm{g}}(k+j))-\nonumber \\ h(f_{\textrm{R}} \{\pmb V(k+j-1), {}{} \pmb u(k+j-1)\})]+\nonumber \\ {}\beta_j[\pmb C(y)(\pmb V(k)-f_{\textrm{R}} \{(\pmb V(k-1), {}\nonumber \\ {} \pmb u(k-1)\})]+\beta_j R_n(y)[h(\pmb V(k))-\nonumber \\ {}h(f_{\textrm{R}} \{\pmb V(k-1), {} \pmb u(k-1)\})] $
(24) 式中, ${\gamma_{\textrm{g}}(y, k+j)}$和${\gamma_{\textrm{p}}(y, k+j)}$分别为$k$时刻第$j$步的期望输出PDF和预测输出PDF.
此外, 预测控制作为一种优化控制算法, 常通过最小化系统的性能指标函数来确定未来的最优控制序列, 使得未来预测输出尽可能接近期望的目标输出.在实际工业过程中, 为了保证操作的可行性等要求, 普遍存在着输入变量带约束的情形, 同时对控制作用的大小加以约束, 避免控制作用变化过于剧烈.本文设计预测PDF控制器目的是尽可能地使输出PDF尽可能跟踪期望输出PDF, 所以选取如下所示性能指标函数
$ \min\limits_{\triangle u} J=\nonumber\\ \quad\sum\limits_{j=1}^{N_\textrm{p}}\int_a^b\left({\sqrt{\gamma_{\textrm{g}}(y, k+j)}}-{\sqrt{\gamma_{\textrm{p}}(y, k+j)}}\right)^2\textrm{d}y+\nonumber \\ \quad\sum\limits_{j=0}^{N_\textrm{c}-1}[\lambda_j \Delta u(k+j)]^2 {}\nonumber \\ {\rm s.t.}u_{\textrm{min}}<\Delta u(k+j)+u(k+j-1)<u_{\textrm{max}}, {}\nonumber \\ \quad\Delta u_{\textrm{min}}<\Delta u(k+j)<\Delta u_{\textrm{max}} $
(25) 式中, $N_\textrm{p}$和$N_\textrm{u}$分别为预测时域和控制时域, $\lambda_j$为控制增量加权系数, $u_{\textrm{max}}$和$u_{\textrm{min}}$分别为输入变量的上限和下限值, $\Delta u_{\textrm{max}}$和分别为输入变量变化率的上限和下限值.可以看出对输出PDF跟踪控制最终转化为对前$n-1$个权值跟踪控制.
从式(25)明显可以看出上述预测PDF控制器的设计可以看作是一个带有约束条件的非线性优化求解问题.针对上述求解问题通常采用遗传算法、粒子群算法、序列二次规划算法(Sequence quadratic program, SQP)等优化算法获得非线性最优预测控制率, 其中, SQP算法作为一种求解约束非线性优化问题的有效方法之一, 具有收敛性快、计算效率高、边界搜索能力强, 在实际中受到广泛重视和应用.本文采用SQP方法求解式(25)所示的带约束的非线性规划问题设计预测PDF控制器, 使得磨浆过程输出PDF具有良好的目标跟踪能力.
4. 工业数据验证
本文利用某化机浆磨浆过程的稀释水流量、磨盘间隙以及FLSD PDF生产数据对所提方法进行数据验证, 具体如下:
4.1 RBF基函数参数整定
为了获得磨浆过程输出FLSD模型, 首先, 需要确定一组合适的RBF基函数近似输出PDF, 采用基于迭代学习方法研究RBF神经网络对期望输出PDF的逼近效果, 并将得到RBF基函数作为本批次近似输出PDF的基函数, 此外, 通过大量FLSD的PDF数据分析获得期望输出PDF, 本文选择4个RBF基函数来验证对期望输出PDF的近似效果, 这里中心值和宽度参数迭代学习率分别为$\alpha_\mu=0.01$, $\beta_\sigma=0.005$, 另外, 假设中心值和宽度的初始值如下所示
$ \mu_1=0.3, \mu_2=0.8, \mu_3=1.3, \mu_4=1.8, {}\nonumber \\ {}\sigma_1^2=\sigma_2^2=\sigma_3^2=\sigma_4^2=0.06 $
(26) 基于式(26)所示的初始RBF基函数, 首先, 可以通过式(9)获得期望输出PDF权值估计, 然后利用得到的估计权值与初始RBF基函数相乘便得到对应的逼近值, 以此利用迭代学习方法通过调整中心值和宽度, 直到获得理想的逼近效果.经过100次迭代学习后, 获得中心值和宽度分别为
$ \mu_1=0.40, \mu_2=0.91, \mu_3=1.25, \mu_4=1.46, {}\nonumber \\ {}\sigma_1^2=0.068, \sigma_2^2=0.074, \sigma_3^2=0.098, \sigma_4^2=0.027 $
(27) 另外, 图 3为在迭代学习50次和100次后的RBF基函数位置变化趋势图, 可以看出随着迭代次数的增加, 中心值和宽度逐渐向理想位置移动. 图 4为性能指标函数值随迭代学习次数的变化趋势, 可以看出随着迭代次数的增加, 目标性能函数逐渐减小, 并在迭代学习80次时基本不再变化.中心值和宽度随迭代次数变化趋势分别如图 5和图 6所示, 从图 5和图 6看出在迭代学习100次后, 中心值和宽度均趋于平稳. 图 7为在迭代学习100次后, 近似输出PDF与期望输出PDF的逼近结果, 可以看出本文方法对输出PDF具有满意的逼近效果.同时, 利用式(9)对期望输出PDF数据进行权值估计, 此时获得相对应的期望权值为.
4.2 预测PDF控制效果
在完成RBF基函数参数整定同时利用式(4)对不同时刻输出PDF进行权值估计, 然后利用RVFLNs方法建立前三组权值的非线性预测模型, 利用稀释水流量、磨盘间隙和输出PDF数据, 采用所提方法建立磨浆过程输出FLSD模型, 并基于SQP算法优化式(25)设计预测PDF控制器.
本文取预测时域$N_\textrm{p}=3$, 控制时域$N_\textrm{u}=2$, 控制增量加权系数$\lambda_j=0.05$, 反馈校正系数$\beta_j=0.55$.另外, 根据实际操作经验, 输入变量稀释水流量($u_1$)和磨盘间隙($u_2$)分别满足$70\textrm{L/min}<u_1<75\textrm{L/min}$, $0.8\textrm{mm}<u_2<1.2\textrm{mm}$, $|\triangle u_1|<1$, $|\triangle u_2|<1$.根据上述分析可知, 在4个权值中有3个是相互独立的, 因此, 期望PDF相对的应权值, 则.另外, 假设初始输出PDF所对应的权值, 则$h(\pmb V_0)=1.205$, 输入变量稀释水流量($u_1$)和磨盘间隙($u_2$)的初始值分别为$u_1=71\textrm{L/min}$, $u_2=1.0\textrm{mm}$.
图 8和图 9分别在预测PDF控制器下预测权值输出响应曲线、控制输入的动态响应, 从图 8可以看出预测权值输出能够实现对期望权值的跟踪, 但权值动态模型由于非线性存在, 在一定程度上影响到预测权值输出对期望权值跟踪控制性能.此外, 图 9所示输入变量稀释水流量和磨盘间隙均能较好地稳定在可操作区间内. 图 10和图 11分别为输出FLSD的PDF3D图以及初始时刻、目标、最终时刻输出PDF, 从图 10和图 11明显能够看出在预测PDF控制器作用下, 实际输出PDF从初始输出PDF形状具有很明显逼近期望输出PDF趋势, 并最终实现对输出PDF跟踪控制.
5. 结论
本文从当前磨浆过程实际控制问题出发, 针对具有典型非高斯分布特征的输出FLSD形状提出了一种预测PDF控制方法.采用迭代学习方法获得理想的RBF基函数基础上对不同时刻输出PDF相对应的权值进行估计, 针对权值之间强耦合、非线性强等特点, 采用RVFLNs建立表征输出变量和权值向量之间关系的预测模型, 最终将输出PDF的控制转化为对权值向量的控制, 基于工业数据实验结果表明了所提方法的有效性.
-
A. Index $t$ time index $i$ microgrid index $k$ iteration step index $a$ index of schedulable appliances in microgrid $i$ B. Constants $M$ set of microgrids in the EI system ( $i\in M$ ) $T$ number of periods for the control horizon ( $t\in T$ ) $N$ a preset iteration coefficient used for accelerating the convergence speed $A_{i, s}$ set of schedulable appliances in microgrid $i$ ( $a\in A$ ) $\Delta t$ time interval of each period (h) $P_{i, l}^{\rm max}, P_{i, O}^{\rm max}$ the rated power that can be purchased/sold from/to the utility for microgrid $i$ (kW) $E_{i, E}^{\rm max}, E_{i, E}^{\rm min}$ the maximum, minimum available energy level of the ESD unit in microgrid $i$ (kWh) $E_{i, E}^{\rm init}$ the initial energy level of ESD unit in microgrid $i$ (kWh) $P_{i, Ec}^{\rm max}, P_{i, Ec}^{\rm min}$ the maximum, minimum charging power of the ESD unit in microgrid $i$ (kW) $P_{i, Ed}^{\rm max}, P_{i, Ed}^{\rm min}$ the maximum, minimum discharging power of the ESD unit in microgrid $i$ (kW) $\eta_{i, Ed}, \eta_{i, Ec}$ discharging, charging efficiency of the ESD unit in microgrid $i$ (%) $\varepsilon_{i, E}$ self-discharging rate of the ESD unit in microgrid $i$ (kWh/h) $c_{i, E}^{\rm O \ M}$ operation and maintenance cost of the ESD unit in microgrid $i$ ($) $c_{i, E}^{\rm switch}$ status switch cost of the ESD unit in microgrid $ i $ ($) $P_{i, \rm DDG}^{\rm max}, P_{i, \rm DDG}^{\rm min}$ the maximum, minimum allowed power output of the DDG unit in microgrid $i$ (kW) $T_{i, \rm DDG}^{\rm down}, T_{i, \rm DDG}^{\rm up}$ the minimum down, operation time of the DDG unit in microgrid $i$ (h) $c_{i, \rm DDG}^{\rm down}, c_{i, \rm DDG}^{\rm up}$ shut-down, start-up cost of the DDG unit in microgrid $i$ ($) $R_{i, \rm DDG}$ the maximum ramp down/up power rate of the DDG unit in microgrid $i$ (kW) $c_{i, \rm DDG}^{1}, c_{i, \rm DDG}^{2}$ cost coefficients of the DDG unit in microgrid $i$ ( $$/\rm{kW^2}, $/\rm{kW}$ ) $\alpha 1, \alpha 2$ cost coefficients of the utility generator ( $$/\rm{kW^2}, $/\rm{kW)}$ $l_{i, a}^{\rm min}, l_{i, a}^{\rm max}$ the minimum, maximum load demand of appliance a for microgrid $i$ (kW) $l_{i, B}^{\rm max}$ rated capacity of the critical loads in microgrid $i$ (kW) $P_{i, PV}^{\rm max}$ rated power capacity of the PV plant in microgrid $i$ (kW) $P_{i, \rm wind}^{\rm max}$ rated power capacity of the wind farm in microgrid $i$ (kW) $T_{i, a}^{\rm start}, T_{i, a}^{\rm end}$ start time, deadline of appliance a for microgrid $i$ (h) $E_{i, a}$ total energy demand of the appliance a for microgrid $i$ (kWh) $D_{i}$ spinning reserve ratio for microgrid $i$ (%) $\xi_{1}, \xi_{2}, \xi_{3}, \xi_{4}$ preset stopping criteria for the distribution optimization algorithm $\theta_{i, F}^{\rm max}$ the maximum curtailment ratio of flexible loads in microgrid $i$ (%) $c_{i, F}^{\rm curt}$ penalty cost coefficient for curtailing flexible loads in microgrid $i$ $P_{u}^{\rm max}, P_{u}^{\rm min}$ the maximum, minimum power limit of the utility generator (kW) C. Parameters $P_{i, \rm wind}(t)$ power output of the wind turbines in microgrid $i$ at time $t$ (kW) $P_{i, PV}(t)$ power output of the PV plant in microgrid $i$ at time $t$ (kW) $l_{i, B}(t)$ demand of the critical loads in microgrid $i$ at time $t$ (kW) $l_{i, F}(t)$ demand of the flexible loads in microgrid $i$ at time $t$ (kW) $p_u(t)$ base electricity price for the utility company ($/kWh) $p_{i, b}(t), p_{i, s}(t)$ buying, selling electricity price for microgrid $i$ at time $t$ ($) $p_{i, b}(t), p_{i, s}(t)$ buying, selling price coefficient D. Variables $P_{i, l}(t), P_{i, O}(t)$ power imported/exported from/to the utility for microgrid $i$ at time $t$ (kW) $\delta_{i, l}(t), \delta_{i, O}(t)$ purchasing, selling power status for microgrid $i$ at time $t$ (0/1) $P_{i, Ec}(t), P_{i, Ed}(t)$ charging, discharging power rate of the ESD unit for microgrid $i$ at time $t$ (kW) $\delta_{i, Ec}(t), \delta_{i, Ed}(t)$ charging, discharging status of the ESD unit for microgrid $i$ at time $t$ (0/1) $E_{i, E}(t)$ energy level of the ESD unit for microgrid $i$ at time $t$ (kWh) $P_{i, \rm DDG}(t)$ power output of the DDG unit for microgrid $i$ at time $t$ (kWh) $\delta_{i, \rm DDG}(t)$ operation status of the DDG unit for microgrid $i$ at time $t$ (0/1) $\theta_{i, F}(t)$ curtailment ratio of the flexible loads for microgrid $i$ at time $t$ (%) $l_{i, a}(t)$ load demand of appliance a for microgrid $i$ at time $t$ (kW) Table Ⅰ ALGORITHM FOR PARALLEL DISTRIBUTED OPTIMIZATION METHOD FOR EI SYSTEM
$\textbf{Algorithm 1:}$ for utility at time $t$
$\textbf{begin}$
$k$ = 0; ${\%}$ iteration counter
Obtain the initial $P_{i, I}^k(\tau)$ , $P_{i, O}^k(\tau)$ of each microgrid according to the random generation technique; $\tau\in[t, t+1, \ldots, t+T-1]$
Calculate utility cost $\Psi_u^k$ according to (29), the retail buying electricity price $p_{i, b}^k(\tau)$ and selling price $p_{i, s}^k(\tau)$ according to (26) and (27), respectively;
do
Broadcast updated retail prices to all microgrids;
Receive the newly updated $P_{i, I}^{k+1}(\tau)$ , $P_{i, O}^{k+1}(\tau)$ simultaneously from all the microgrids according to Algorithm 2 shown in the following; $i\in[1, M]$
Calculate utility cost $\Psi_u^{k+1}$ , retail electricity price $p_{i, s}^{k+1}(\tau)$ , $p_{i, b}^{k+1}(\tau)$
$k:=k+1$ ;
$\textbf{until}$ $||\Psi_u^k\!-\!\Psi_u^{k-1}||\leq \xi_1, ||l^k(\tau)\!-\!l^{k-1}(\tau)||\!\leq\!\xi_2, $
$|||P_{Ed}^k(\tau)\!-\!P_{Ec}^k(\tau)||- ||P_{Ed}^{k-1}(\tau)-P_{Ec}^{k-1}(\tau)|||\leq\xi_3, $
$||P_{\rm DDG}^k(\tau)-P_{\rm DDG}^{k-1}(\tau)||\leq\xi_4$
$\textbf{end}$
$\textbf{Algorithm 2:}$ for microgrid $i$ at time $t$
$\textbf{begin}$
$k$ = 0; ${\%}$ iteration counter
Initialize $P_{i, I}^k(\tau), P_{i, O}^k(\tau)$ according to the random generation technique;
Report $P_{i, I}^k(\tau)$ , $P_{i, O}^k(\tau)$ to the EI operator; $\tau\in[t, t+1, \ldots, t+T-1]$
While
Update the received retail electricity price $p_{i, s}^k (\tau)$ , $p_{i, b}^k (\tau)$ from the EI operator
Solve the optimization problem (31) and obtain the newly updated $P_{i, I}^{k+1}(\tau)$ , $P_{i, O}^{k+1}(\tau)$ ;
Report $P_{i, I}^{k+1}(\tau)$ , $P_{i, O}^{k+1}(\tau)$ to the EI operator;
$ k:=k+1$ ;
end
endTable Ⅱ POWER LIMITS OF MICROGRIDS AND INDEPENDENT USER
PV plant Wind farm PCC node Critical load Microgrid 1 400 192 1200 672 Microgrid 2 400 0 800 496 Microgrid 3 0 240 800 560 Independent user 0 0 1500 800 Table Ⅲ PARAMETER OF SCHEDULABLE LOADS
Power demand
(kW)Operation interval
(h)Time window
(h)Duration
(h)Task 1 22 15-21 6 2 Task 2 28 14-23 9 4 Task 3 45 8-18 10 6 Task 4 37.5 6-24 18 8 Task 5 12 2-22 20 12 Task 6 60 8-22 14 7 Task 7 75 6-24 18 9 Table Ⅳ PARAMETER OF ESDS
Max charge/ discharge power Min charge/ discharge power O & M cost Switch cost Max energy level Charge / discharge efficiency Microgrid 1 160 5 0.05 0.06 320 0.95 Microgrid 2 140 8 0.05 0.05 300 0.95 Microgrid 3 120 6 0.05 0.07 260 0.95 Table Ⅴ PARAMETER OF CONTROLLABLE GENERATORS
Max power Min power Ramp rate Min up/down time Startup/shut down cost Cost coefficients Microgrid 2 150 5 100 2/2 1.2/1.2 0.0042/0.32 Utility 4500 50 - - - 0.00048/0.28 Table Ⅵ SCHEDULING COSTS AND THE TOTAL COSTS FOR BOTH DMPC APPROACH AND DDA APPROACH
Cost (×105$) Microgrid
1Microgrid
2Microgrid
3Microgrid
4Scheduling cost with no optimization 0.6661 0.6721 0.6716 1.1479 Scheduling cost with DDA 0.6316 0.5741 0.6376 1.0982 Scheduling cost with DMPC 0.6484 0.5867 0.6537 1.0993 Total cost with no optimization 0.6764 0.6831 0.6856 1.1619 Total cost with DDA 0.6574 0.6016 0.6717 1.1338 Total cost with DMPC 0.6502 0.5878 0.6555 1.1005 Table Ⅶ SCHEDULING COSTS AND TOTAL COSTS FOR DMPC APPROACH WITHOUT SOME DISPATCHABLE ELEMENTS
Cost (×105$) Microgrid
1Microgrid
2Microgrid
3Microgrid
4Scheduling cost without ESDs 0.6522 0.5920 0.6617 1.1186 Total cost without ESDs 0.6536 0.5929 0.6631 1.1201 Scheduling cost without ESDs and DDG 0.6800 0.6961 0.6846 1.1617 Total cost without ESDs and DDG 0.6788 0.6951 0.6831 1.1605 -
[1] H. Farhangi, "The path of the smart grid, " IEEE Power Energy Mag. , vol. 8, no. 1, pp. 18-28, Jan-Feb. 2010. http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=5357331 [2] A. Q. Huang, M. L. Crow, G. T. Heydt, J. P. Zheng, and S. J. Dale, "Energy future renewable electric delivery and management (FREEDM) system: the energy internet, " Proc. IEEE, vol. 99, no. 1, pp. 133-148, Jan. 2011. [3] Y. B. Zha, T. Zhang, Z. Huang, Y. Zhang, B. L. Liu, and S. J. Huang, "Analysis of energy internet key technologies, " Sci. Sin. Inf. , vol. 44, no. 6, pp. 702-713, Jan. 2014. http://en.cnki.com.cn/Article_en/CJFDTotal-PZKX201406004.htm [4] D. Zhang, N. Shah, and L. G. Papageorgiou, "Efficient energy consumption and operation management in a smart building with microgrid, " Energy Convers. Manage. , vol. 74, pp. 209-222, Oct. 2013. http://www.sciencedirect.com/science/article/pii/S0196890413002355 [5] J. Wu and X. H. Guan, "Coordinated multi-Microgrids optimal control algorithm for smart distribution management system, " IEEE Trans. Smart Grid, vol. 4, no. 4, pp. 2174-2181, Dec. 2013. http://ieeexplore.ieee.org/document/6654349/ [6] Z. Y. Dong, J. H. Zhao, F. S. Wen, and Y. S. Xue, "From smart grid to energy internet: basic concept and research framework, " Automat. Electric Power Syst. , vol. 38, no. 15, pp. 1-11, Aug. 2014. [7] T. Zhang, F. X. Zhang, and Y. Zhang, "Study on energy management system of energy internet, " Power Syst. Technol. , vol. 16, no. 1, pp. 146-155, Jan. 2016. [8] R. H. Lasseter, "Smart distribution: coupled microgrids, " Proc. IEEE, vol. 99, no. 6, pp. 1074-1082, Jun. 2011. http://ieeexplore.ieee.org/document/5768104/ [9] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, "Microgrids, " IEEE Power Energy Mag. , vol. 5, no. 4, pp. 78-94, Jul-Aug. 2007. http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=4263070 [10] J. Lee, J. Guo, J. K. Choi, and M. Zukerman, "Distributed energy trading in microgrids: a game-theoretic model and its equilibrium analysis, " IEEE Trans. Ind. Electron. , vol. 62, no. 6, pp. 3524-3533, Jun. 2015. http://ieeexplore.ieee.org/document/7001088/ [11] J. M. Guerrero, M. Chandorkar, T. L. Lee, and P. C. Loh, "Advanced control architectures for intelligent microgrids-Part Ⅰ: decentralized and hierarchical control, " IEEE Trans. Ind. Electron. , vol. 60, no. 4, pp. 1254-1262, Apr. 2013. http://ieeexplore.ieee.org/document/6184305/ [12] J. Silvente, G. M. Kopanos, E. N. Pistikopoulos, and A. Espuña, "A rolling horizon optimization framework for the simultaneous energy supply and demand planning in microgrids, " Appl. Energy, vol. 155, pp. 485-501, Oct. 2015. http://www.sciencedirect.com/science/article/pii/S0306261915007230 [13] M. H. Albadi and E. F. EL-Saadany, "A summary of demand response in electricity markets, " Electric Power Syst. Res. , vol. 78, no. 11, pp. 1989-1996, Nov. 2008. http://www.sciencedirect.com/science/article/pii/S0378779608001272 [14] B. Chai, J. M. Chen, Z. Y. Yang, and Y. Zhang, "Demand response management with multiple utility companies: a two-level game approach, " IEEE Trans. Smart Grid, vol. 5, no. 2, pp. 722-731, Mar. 2014. http://ieeexplore.ieee.org/document/6740887/ [15] W. C. Su and J. H. Wang, "Energy management systems in microgrid operations, " Electricity J. , vol. 25, vol. 8, pp. 45-60, Oct. 2012. http://www.sciencedirect.com/science/article/pii/S104061901200214X [16] C. Chen, J. H. Wang, and Y. Heo, and S. Kishore, "MPC-based appliance scheduling for residential building energy management controller, " IEEE Trans. Smart Grid, vol. 4, no. 3, pp. 1401-1410, Sep. 2013. http://ieeexplore.ieee.org/document/6575202/ [17] A. Parisio, E. Rikos, and L. Glielmo, "A model predictive control approach to microgrid operation optimization, " IEEE Trans. Control Syst. Technol. , vol. 22, no. 5, pp. 1813-1827, Sep. 2014. http://ieeexplore.ieee.org/document/6705582/ [18] D. H. Zhu and G. Hug, "Decomposed stochastic model predictive control for optimal dispatch of storage and generation, " IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 2044-2053, Jul. 2014. http://ieeexplore.ieee.org/document/6839132/ [19] W. C. Su, J. H. Wang, K. L. Zhang, and A. Q. Huang, "Model predictive control-based power dispatch for smart distribution system considering plug-in electric vehicle uncertainty, " Electric Power Syst. Resour. , vol. 106, pp. 29-35, Jan. 2014. http://www.sciencedirect.com/science/article/pii/S0378779613002113 [20] J. W. Cao and M. B. Yang, "Energy internet-Towards Smart Grid 2. 0, " in Proc. 4th Int. Conf. Networking and Distributed Computing (ICNDC), Los Angeles, CA, USA, 2013, pp. 105-110. [21] Y. F. Tang, J. Yang, J. Yan, and H. B. He, "Intelligent load frequency controller using GrADP for island smart grid with electric vehicles and renewable resources, " Neurocomputing, vol. 170, pp. 406-416, Dec. 2015. [22] Y. F. Tang, H. B. He, Z. Ni, J. Y. Wen, and T. W. Huang, "Adaptive modulation for DFIG and STATCOM with high-voltage direct current transmission, " IEEE Trans. Neural Networks Learn. Syst. , vol. 27, no. 8, pp. 1762-1772, Aug. 2016. http://www.ncbi.nlm.nih.gov/pubmed/26701900 [23] Y. F. Tang, H. B. He, Z. Ni, X. N. Zong, D. B. Zhao, and X. Xu, "Fuzzy-based goal representation adaptive dynamic programming, " IEEE Trans. Fuzzy Syst. , to be published. http://ieeexplore.ieee.org/document/7346472/ [24] M. H. Amini, R. Jaddivada, S. Mishra, and O. Karabasoglu, "Distributed security constrained economic dispatch, " in Proc. IEEE Innovative Smart Grid Technologiesâ€"â€"Asia (ISGT ASIA), Bangkok, Thailand, 2015. http://ieeexplore.ieee.org/document/7387167/ [25] K. Deng, Y. Sun, S. S. Li, Y. Lu, J. Brouwer, P. G. Mehta, M. C. Zhou, and A. Chakraborty, "Model predictive control of central chiller plant with thermal energy storage via dynamic programming and mixed-integer linear programming, " IEEE Trans. Automat. Sci. Eng. , vol. 12, no. 2, pp. 565-579, Apr. 2015. http://ieeexplore.ieee.org/document/6899700/ [26] M. Huber, F. Sanger, and T. Hamacher, "Coordinating smart homes in microgrids: a quantification of benefits, " in Proc. 20134th IEEE/PES Innovative Smart Grid Technologies Europe (ISGT Europe), Copenhagen, 2013, pp. 1-5. [27] D. E. Olivares, C. A. Cañizares, and M. Kazerani, "A centralized energy management system for isolated microgrids, " IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1864-1875, Jul. 2014. http://ieeexplore.ieee.org/document/6805674/ [28] Y. Z. Zhou, H. Wu, Y. N. Li, H. H. Xin, and Y. H. Song, "Dynamic dispatch of multi-microgrid for neighboring islands based on MCS-PSO algorithm, " Automat. Electric Power Syst. , vol. 38, no. 9, pp. 204-210, May 2014. http://en.cnki.com.cn/Article_en/CJFDTOTAL-DLXT201409030.htm [29] X. Ai and J. J. Xu, "Study on the microgrid and distribution network co-operation model based on interactive scheduling, " Power Syst. Prot. Control, vol. 41, no. 1, pp. 143-149, Jan. 2013. http://en.cnki.com.cn/Article_en/CJFDTotal-JDQW201301026.htm [30] M. Fathi and H. Bevrani, "Adaptive energy consumption scheduling for connected microgrids under demand uncertainty, " IEEE Trans. Power Deliv. , vol. 28, no. 3, pp. 1576-1583, Jul. 2013. http://ieeexplore.ieee.org/document/6510487/ [31] Z. Y. Wang, B. K. Chen, J. H. Wang, M. M. Begovic, and C. Chen, "Coordinated energy management of networked microgrids in distribution systems, " IEEE Trans. Smart Grid, vol. 6, no. 1, pp. 45-53, Jan. 2015. http://ieeexplore.ieee.org/document/6872821/ [32] F. Kamyab, M. Amini, S. Sheykhha, M. Hasanpour, and M. M. Jalali, "Demand response program in smart grid using supply function bidding mechanism, " IEEE Trans. Smart Grid, vol. 7, no. 3, pp. 1277-1284, May 2016. http://ieeexplore.ieee.org/document/7112524/ [33] G. E. Asimakopoulou, A. L. Dimeas, and N. D. Hatziargyriou, "Leader-follower strategies for energy management of multi-microgrids, " IEEE Trans. Smart Grid, vol. 4, no. 4, pp. 1909-1916, Dec. 2013. http://ieeexplore.ieee.org/document/6519333/ [34] P. Yang, P. Chavali, E. Gilboa, and A. Nehorai, "Parallel load schedule optimization with renewable distributed generators in smart grids, " IEEE Trans. Smart Grid, vol. 4, no. 3, pp. 1431-1441, Sep. 2013. http://ieeexplore.ieee.org/document/6576918/ [35] Y. Zhang, T. Zhang, R. Wang, Y. J. Liu, and B. Guo, "Optimal operation of a smart residential microgrid based on model predictive control by considering uncertainties and storage impacts, " Solar Energy, vol. 122, pp. 1052-1065, Dec. 2015. http://www.sciencedirect.com/science/article/pii/S0038092X15005782 [36] E. Camponogara, D. Jia, B. H. Krogh, and S. Talukdar, "Distributed model predictive control, " IEEE Control Syst. , vol. 22, no. 1, pp. 44-52, Feb. 2002. [37] D. P. Bertsekas and J. N. Tsitsiklis, Parallel and Distributed Computation:Numerical Methods. Englewood Cliffs, NJ, USA:Prentice-Hall, 1989. [38] I. Prodan and E. Zio, "A model predictive control framework for reliable microgrid energy management, " Int. J. Electric. Power Energy Syst. , vol. 61, pp. 399-409, Oct. 2014. http://www.sciencedirect.com/science/article/pii/S0142061514001197 [39] J. S. Netz, "Price regulation: A (non-technical) overview, " in Encyclopedia of Law and Economics, B. Bouckaert and G. De Geest, Eds, Cheltenham, Edward Elgar, 2000, pp. 1396-1465. [40] Y. X. Xu and C. Singh, "Power system reliability impact of energy storage integration with intelligent operation strategy, " IEEE Trans. Smart Grid, vol. 5, no. 2, pp. 1129-1137, Mar. 2014. http://ieeexplore.ieee.org/document/6583282/ [41] L. Y. Jia, Z. Yu, M. C. Murphy-Hoye, A. Pratt, E. G. Piccioli, and L. Tong, "Multi-scale stochastic optimization for home energy management, " in Proc. IEEE Int. Workshop on Computational Advances in Multi-Sensor Adaptive Processing, San Juan, Puerto Rico, 2011. http://ieeexplore.ieee.org/document/6135900/ [42] ELIA, Belgium's electricity transmission system operator. Grid data, [EB/OL]. [Online]. Available: http://www.elia.be/en/grid-data. [43] C. Yang and L. Xie, "A novel ARX-based multi-scale spatio-temporal solar power forecast model, " North American Power Symp. (NAPS), Champaign, IL, USA, 2012, pp. 1-6. http://ieeexplore.ieee.org/document/6336383/ [44] R. Hanna, J. Kleissl, A. Nottrott, and M. Ferry, "Energy dispatch schedule optimization for demand charge reduction using a photovoltaic-battery storage system with solar forecasting, " Solar Energy, no. 103, pp. 269-287, May 2014. http://www.sciencedirect.com/science/article/pii/S0038092X1400098X [45] R. Blonbou, "Very short-term wind power forecasting with neural networks and adaptive Bayesian learning, " Renewable Energy, vol. 36, no. 3, pp. 1118-1124, Mar. 2011. http://www.sciencedirect.com/science/article/pii/S0960148110003976 [46] J. Löfberg, "YALMIP: a toolbox for modeling and optimization in MATLAB, " in 2004 IEEE Int. Symp. Computer Aided Control Systems Design, New Orleans, LA, 2004, pp. 284-289. http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=1393890 期刊类型引用(4)
1. 王本斐,张荣辉,冯国栋,Manandhar Ujjal,郭戈. 基于事件触发的直流微电网无差拍预测控制. 自动化学报. 2024(03): 475-485 . 本站查看
2. 郭政,吴武清,刘源,曾志武,杨玎. 基于多任务支持向量机的能源互联网数据深度融合方法. 计算机应用与软件. 2023(03): 22-27 . 百度学术
3. 张华成,陈辉,吴海斌,马海涛,高阳. 基于信息融合的电网运行事件协同感知与交互方法. 电子设计工程. 2023(13): 109-113 . 百度学术
4. 李军,万文军,刘哲. 一种二阶内反馈控制器SO-IFC的研究与应用. 自动化学报. 2019(05): 993-1003 . 本站查看
其他类型引用(5)
-