I.INTRODUCTION

W ITH the rapid development of economy,electrical power demand has become continuously stronger year by year.A vast amount of fossil fuels utilized in power generation results in energy crisis and environmental deterioration around the world[1].One possible solution is to adopt clean and renewable energy instead of fossil fuel for power generation.Wind power is now the fastest growing energy source around the world because of its zero emission[2].The percentage of wind generation in power systems increases with years.Wind energy has become one of the central research themes in energy science.

Concerning real applications,a constant supply of Electricity in some remote areas cannot be guaranteed by power grids.In these areas,wind energy may be inexhaustible and convenient.Therefore,wind power has been paid more and more attention and some control problems rise up in power systems with wind turbines[3].An advocacy of wind power is due to its sustainable and renewable status.However,wind power affected by climate changes is intermittent.Its intermittence also has impressive effects on operation and control of renewable power systems.

Consider a power system with wind turbines.The load in the power system is random and the power output of wind power is fluctuating.The power-output fluctuation and the load change would pose a reliability supply challenge.The challenge is displayed by power imbalance and frequency deviation in the power system[4].Consequently,frequency control strategies must be adopted to overcome the challenge.

Load frequency control(LFC)is one of the most profitable auxiliary services to guarantee the stable operation of power systems with the objective of preserving the balance between power generation and power consumption[5].Recently,the LFC problem of renewable power systems has been paid more and more attention[6],[7].In order to attack the generation intermittency of renewable power systems,some LFC methods have been investigated,such as fuzzy control[8],[9],predictive control[10],[11],and adaptive control[12],[13].

Invented by Utkin,the sliding mode control(SMC)is recognized as a powerful design tool[14].On the sliding-mode stage,an SMC system is completely insensitive to parametric uncertainties and external disturbances under certain matching conditions,which exhibits better performance than the conventional robust control methods[15].This property inspires some researches on SMC for the LFC problem[16]−[22].However,the SMC-based LFC methods in pervious works[16]−[22]do not consider the complexity and challenge of renewable-energy sources.The method of terminal sliding mode control(T-SMC)[23],[24]guarantee the convergence of the SMC system within finite time.The T-SMC method can be considered for the LFC problem of renewable power systems.

1.6 统计学处理 应用SPSS 20.0软件进行数据分析，正态分布计量资料以x±s表示，采用t检验；偏态分布计量资料以 M（P25，P50）描述，采用秩检验；计数资料比较采用χ2检验。P＜0.05为差异有统计学意义。

Power systems are inherently nonlinear[25].The two main nonlinear factors are the governor dead band(GDB)and the generation rate constraint(GRC).The existence of GRC has adverse effects on the system robustness,the system performance as well as the system stability[26].A common technique to deal with the GRC nonlinearity is to design a controller for the linearized nominal model;then the controller is directly imposed on the original nonlinear system.In a sentence,the linearized model is adopted to achieve the control design for the original nonlinear system.Although it is available in most cases,the technique has some potential hazards because there is no theoretical guarantee on the stability of the control system.Concerning the applications of SMC on LFC[16]−[22],some works[16],[17],[21]only consider linear power systems and other works[18]−[20],[22]adopt the linear-model-based design.However,a series of drawbacks may be induced to the linear-model-based control systems because of lack of theoretical supports.

The methodology of radial basis function neural networks(RBF NNs)has a universal approximating feature[27].The RBF NNs technology has been widely adopted to solve nonlinearities and uncertainties of complex systems[28].In[20],RBF NNs are employed to approximate and compensate the GDB nonlinearity of power systems.However,how to conquer the GRC nonlinearity by RBF NNs remains untouched and problematic.To turn the T-SMC into practical accounts on the LFC problem of renewable power systems,it is urgent to solve the GRC nonlinearity by the RBF NNs technique.

岩溶地区桩基一般采用冲击钻成孔方式，也有采用旋挖钻利用长套筒成孔的先例，但相比之下，采用冲击钻成孔便于进行溶洞回填冲击造壁处理，在岩溶地区适用性强。另外对于岩溶区上岩土层厚的区域，也可在上部非岩溶区域采用旋挖钻或回旋钻成孔，再在岩溶区域采用冲击钻成孔的组合方式，以提高成孔效率。本工程引桥段采用全程冲击钻成孔，主桥主墩处采用旋挖钻+冲击钻组合成孔方式。

This paper focuses on the T-SMC method for LFC of nonlinear power systems with wind turbines.To deal with the GRC nonlinearity,the linearization method is adopted at first.Then,a terminal sliding mode controller is designed based on the linearized nominal system.The uncertainties of the LFC problem have three components,i.e.,the intermittency of wind power,the uncertainties of power systems and the error of linearization.The components mix together and worsen the uncertainties of the LFC problem.To theoretically guarantee the system stability,the sliding-mode-based neural networks are designed to suppress the entire uncertainties.Weight update formulas of the neural networks are derived from the Lyapunov direct method.The neural-network-based T-SMC scheme is employed to accomplish the LFC problem.To illustrate the feasibility and validity of the presented scheme,some numerical simulations are conducted by a nonlinear interconnected power system with wind turbines.

The remainder of this paper is organized as follows.Section II formulates the system configuration.Section III presents the T-SMC method,the RBF NNs design and the system stability.Simulation results are demonstrated in Section VI.Finally,conclusions are drawn in Section V.

II.SYSTEM CONFIGURATION

A.Component Dynamics

This paper considers the LFC problem of a multi-area interconnected power system.The power system is composed of N control areas which are interconnected by tie-lines.Fig.1 represents the block diagram of the ith control area in the multi-area power system.In Fig.1,variables∆Pgi(t),∆Xgi(t)and∆fi(t)are the incremental changes of generator output,governor valve position,and frequency,respectively.∆PLi(t)is the load disturbance,∆Pci(t)is the control input.Tgi,Ttiand Tpiare the time constants of governor,turbine and electric system governor,respectively.Bi=1/Ri+1/Kpiis the frequency bias factor where Riis the adjustment deviation coefficient and Kpiis the electric system gain.Tijis the synchronizing power coefficient between area i and area j,i,j=1,...,N and N is the number of control areas.

According to the LFC objective,not only should the frequency of the control area return to its nominal value,but also the net interchange through the tie-line should return to the scheduled values.To achieve the composite goal,a measure,named area control error(ACE),is introduced.In Fig.1,the measure in the ith control area is defined by

where∆Ptie,i(t)is the tie-line active power deviation,determined by

where∆Vi(t)is employed to represent the control area interface,defined by

To force the composite measure(1)to zero,the integral of ACEi(t)is used as an additional state,determined by

When because C2B2is non-singular.In(38),replace sss˙ by(26)and take(31)into consideration.Then,(38)becomes

Define a vector xi(t).The vector is described by xi(t)=[∆Xgi(t)∆Pgi(t)∆fi(t)∆Ptie,i(t)∆Ei(t)]T.Then,the following state equations can be deduced from(1)−(4)and Fig.1.

where ui(t)= ∆Pci(t)is the control input,∆Pdi(t)=[∆PLi(t)∆Vi(t)]Tis the disturbance vector.For a nominal system,the matrices Ai,Biand Fiin(5)are formulated in Appendix A.The system model(5)will be employed for the LFC design of the ith control area in the multi-area power system.

B.Simplified Wind Turbine Model

The doubly-fed induction generator(DFIG)system has been proven a proper renewable energy conversion system.A simplified frequency response model of a DFIG based wind turbine unit[10]is illustrated in Fig.2.

The model of the DFIG-based wind turbine unit can be described by the following equations.

当前,世界面临石油、天然气的短缺,需要生产一种节能的汽车产品[1].据不完统计,以目前对油气开采的速度,大约能够维持60～100年,这对汽车工业的发展影响较大.因此,需要生产一种节能汽车.目前,节能汽车主要包括电动汽车和混合动力汽车[2].但是,目前技术发展水平导致电动汽车制造成本高,充电设备还不够完善,从而使混合动力汽车占据主导市场.混合动力汽车可以在制动过程中回收能量并储存起来,在车辆起动和加速过程中释放能量,从而降低了燃油消耗.

where Vqr(t)and iqr(t)are the q-axis components of the rotor voltage and the rotor current,respectively.w(t)is the rotational speed,Tm(t)is the mechanical power,Htis the equivalent inertia constant,Pe(t)is the active power.Other parameters in Fig.2 are explained as X2=1/Rr,X3=Lm/Lss,T1=L0/wsRs,L0=Lrr+L2m/Lss,and Lss=Ls+Lm,Lrr=Lr+Lm,here Lmis the magnetizing inductance,Rrand Rsare the rotor and stator resistances,respectively.Lrand Lsare the rotor and stator leakage inductances,respectively.Lrrand Lssare the rotor and stator self-inductances.wsis the synchronous speed.

Linearizing the wind turbine model at a certain operating point,we can rewrite(8)as

Since C2is nonsingular,(46)can be obtained

C.Dynamics for Wind Turbine Load Frequency Control

Defineand ∆Pwdi(t)=[∆PLi(t)∆Vi(t)∆Tmi(t)]T.Then,the frequency response model for the ith control area with one aggregated generator unit and one aggregated wind turbine can be combined in the following state space model from(1)−(10)and Fig.2.

Equation(11)will be employed for the LFC design for wind turbines in the multi-area power system.The details about Awi,Bwiand Fwiare available in Appendix A.

Remark 1:Equation(5)presents the mathematical model for the LFC problem of the conventional generating system in control area i.Equation(5)does not consider the effects of wind turbines on the system frequency because the capacity of the wind turbine unit only shares a small proportion in the power system.On the contrary,to design a frequency controller for the wind turbine unit,some variables of the renewable power system are considered in(11).

D.Analysis About System Models

From(5)and(11),these system models can be described by a uniform expression.Without loss of generality,the expression has a form of

By considering the parameter uncertainties and the modelling errors,equation(12)can be written as

where A0,B0and F0denote the nominal constant matrices,∆A xx(t),∆B (t)and∆F∆P(t)denote the parameter uncertainties and the modeling errors.

It is noted that the above discussions just consider the uncertainties existing in the linear power system.Power systems actually cover the GRC nonlinearity.The existence of GRC has adverse effects on the system stability.Inherently,the GRC nonlinearity acts as a limiter to limit the rate of change in the power generation.Taking the turbine of generating units as an example,the limiter output remains unchanged while reaching its top or bottom.But the turbine output keeps increasing or decreasing at the extreme rate of change.This fact indicates the nonlinear power system becomes linear before the limiter output reaches its limit value.Having been triggered by the two critical points of the limiter,the turbine output is still changing but the change is at its extreme rates determined by the limiter.From Fig.1,these discussions can be formulated by

where σ(t)= ∆Xgi(t)∆Pgi(t)and δ＞ 0 is the maximum output of the GRC nonlinearity.

As far as the LFC problem of the nonlinear power system is concerned,the effect of the extreme rates only exists on the outset because∆Xgi(t)must change to make∆Pgi(t)resist the load disturbance∆PLi(t).From this perspective,the effect of the GRC nonlinearity can be treated as a kind of uncertainties.Different from the uncertainties in(13),this kind of uncertainties can attenuate their effects in the dynamic process.

其三，许多的“文学流派”以地域命名，有鲜明的地域色彩，如宋代的“江西诗派”、“四灵诗派(永嘉四灵)”，明代诗文的“公安派”、“竟陵派”、“茶陵派”、戏曲的“吴江派”、“临川派”，清代的“桐城派”、“浙西词派”、“常州词派”等。

From(13)and(14),the following general model can be derived for the power system.

where φ(t)denotes the uncertainties due to GRC.

From(15),the LFC design of the nonlinear power system with wind turbines can be divided into two parts.One is to design a T-SMC-based load frequency controller for the nominal system.The other is to consider how to suppress the system uncertainties.

III.CONTROL DESIGN

A.Mathematical Descriptions of the System for Terminal Sliding Mode Control

To aggregate all the uncertain terms in(15), ddd (t)is defined and formulated by

where1,2,...,n−m.

Assumption 1:The uncertain term (t)in(16)satisfies ddd(t)=B0H(t)where ‖H(t)‖ ≤ ‖ ·‖ denotes the Euclidean norm andis constant but unknown.

在本文中，MCS结果（使用传统的非线性潮流研究）假定为给出的准确结果。为了评估MCS的准确性和收敛性，令输入变量采样为x1，x2，…，xNs，结果变量为y=V（x）。所需的MCS采样数可以估算为：

According to the principle of matrix controllability decomposition,we adopt the nonsingular transformation of Z(t)=[29],[30].Then,some matrices in(17)becomes

In(18),Rm×(n−m)and A22∈ Rm×m.B2∈ Rm×mis a nonsingular matrix. n is the of x (t) and m is the dimension of u(t).

1.2.3.1 实验对象 以北京大学体育舞蹈选修课学生为实验对象，共计65人，在实验前对2组人员进行协调性与灵敏素质前测，数据表明两组之间没有显著性差异，可以进行试验（表2）。

According to(18),the following equations(19)can be derived from(17).

where

[zn−m+1(t) ···zn(t)]T ∈ Rm.In(19),H(t)means the system uncertainties.

B.Design of Global Fast Terminal Sliding Mode Control

In[24],a kind of sliding surfaces is entitled global fast terminal sliding surface.However,this kind of surfaces in[24]is defined in the scalar form such that it cannot be directly employed for the LFC problem of renewable power systems.To develop a global fast terminal sliding surface for the LFC problem,the general sliding surface vector(20)is extended for multi-variable systems[31].The extended global fast terminal sliding surface vector for the nominal part of(19)is formulated by

where (t)∈ Rm×1,C1 ∈ Rm×(n−m),C2 ∈ Rm×m,C3∈ Rm×(n−m).Both p and q are positive and odd numbers,andis a vector,defined byC2is non-singular to guarantee C2B2invertible.Matrices C2and C3are selected to satisfy the following equations.

In(21)and(22),αi＞ 0,βi＞ 0,i=1,2,...,n−m.When A12is row full rank,the right inverse of A12denoted byexists and has a form of

From(21)and(22),C1and C3can be selected as follows.

Differentiating s(t)in(20)with respect to time t and substituting(19)into the derivative of(20),we get(26).

Then,the general system model(15)has a form of

231 Coil combined with Onyx-34 glue in treatment of coronary artery perforation: a case report

When the system trajectory enters the sliding mode stage and keeps on the sliding surface, sss˙ (t)=Omexists,here Om∈Rm×mis an m-dimensional vector of zeros.Without regard to uncertainties,the equivalent control law of the T-SMC system can be deduced from (t)=Om,determined by

Consider(19)and define the global fast terminal sliding surface(21).Then,the final T-SMC law[31]can be formulated by

广佛线，是国内第一条跨越两个城市的全地下城际快速轨道交通线路，横跨广州的海珠区、荔湾区和佛山的禅城区、南海区，呈东西走向。线路规划西起佛山市魁奇路，东达广州市沥滘，全长32.16公里，全线共设车站21座，共有9座换乘站。

where η ＞ 0.

Remark 2:Since the T-SMC law(28)contains¯h0,the value of¯h0must be known in advance to guarantee the system stability.In other words,the control law(28)is available for the LFC problem under the condition that the uncertainties are with a known boundary.Unfortunately,the boundary value is rather difficult to know in practice.As displayed in Assumption 1,the LFC system is designed under the assumption that the uncertainties have an unknown boundary.Consequently,it is necessary to consider how to deal with the issue.

C.Design of RBF Neural Networks

To fill the gap between the system stability and the boundary value of uncertainties,RBF NNs are employed because such a kind of neural networks owns the ability to approximate complex nonlinear mapping directly from input-output data with a simple topological structure.RBF NNs are a kind of three-layer feed-forward networks,where the mapping from the input layer to the output layer is inherently nonlinear but the mapping from the hidden layer to the output layer is linear.

Fig.3 displays the designed RBF NNs.Concerning the LFC problem,each element of the state vector xxx(t)in(17)is employed as an input element,where γ=1,...,n.Accordingly,the network output y is defined as the estimated value of the system uncertainties.In other words,the designed RBF NNs have n inputs and 1 output.Illustrated in Fig.3,there are l neurons in the hidden layer.

From Fig.3,the output of the RBF NNs is determined byThen,the network output can be calculated by

对于事业单位而言，把好人员进口关十分关键，必须不断完善公开招聘、岗位竞聘制度，规范的招聘程序，根据“用进废退”规律可知，人力资源可以通过使用实现增进，闲置只会让人力资源一步步丧失资本存量，同时只有在人力资源开发中引入竞争机制，方可令员工在竞争中不断发展，让有能力、有才华的人脱颖而出。

In(29), ∈Rlis the weight vector of the RBF NNs, hhh( xxx)∈Rlis the Gaussian function vector and the λth element hλ( xxx)of the vector is defined by

where λ =1,2,...,l,cλ ∈ Rnis the center vector of the λth Gaussian function,bλ is a scalar,indicating the width of the λth Gaussian function.Both cλ and bλ are predefined parameters.

Adopting the RBF approximation technology,the control law(28)can be rearranged by

Remark 3:The term h(t)depicts the system uncertainties.Not only is h(t)hard to be formulated,but also its boundary value is difficult to be obtained.To guarantee the network output can match the change of uncertainties,each element of the weight vector ω has to be renewed by an update law.It is necessary to obtain the update law of the weight vector.On the other hand,the update law makes a difference to the system stability.Consequently,this topic will be investigated in the sense of Lyapunov.

D.Stability Analysis

(1)原料处理。冻猪皮自然解冻，洗净，去脂肪，清洗后煮制3～5min，捞出，自然冷却至室温，去毛，切分成长5～8cm,宽0.5～1cm的条状，放入泡制液中泡制12h，捞出，加入调味料。

Assumption 2:There exists an optimal weight vector ω∗such thatsatisfies

Assumption 3:The boundary valueminus the Euclidean norm of the system uncertainties satisfies

Theorem 1:Take Assumptions 1−3 into account,consider the system model(17),define the global fast terminal sliding surface(20),and adopt the control law(31).Then,the T-SMC-based LFC system is of asymptotic stability if and only if the update law of the network weight vector has a form of

where ξ is a constant and satisfies

Proof:Consider the candidate Lyapunov function

where is determined by

Differentiate V with respect to time t in(36).The derivative of V can be formulated by

where KEiis the gain of this additional state.

Taking Assumptions 2 and 3 into account,we can conclude that the inequalityexists.Then,(39)becomes

这套程序是不是和新闻中某些镜头很像？像就对了。实际上我就是按着新闻中的某些镜头描写的，想来场景应该不差。虽说雷同，但却无意抄袭。

首先，从静态层面来看，土地用途管制的决策依据是土地利用总体规划，土地利用总体规划是土地用途管制的基本依据。国家编制土地利用总体规划，在土地利用总体规划中，将土地分为农用地、建设用地和未利用地。使用土地的单位和个人必须严格按照土地利用总体规划确定的用途使用土地，这是土地用途管制权概念静态化之表达，这一点在上述学者的概念界定中基本没有争议。

In(40),it is obvious thatat sss (t)=Om.

From(36)and(40),we have V≥0 and˙V≤0.Further,(41)can be deduced from(36).

Defineand substitute(41)into(40).Then,(40)can be rearranged by

Integrating both sides in(42)yields

The time τ can be calculated by

In the sense of Lyapunov,V will reach zero from the state of V0in the finite time τ.Meantime,the global fast terminal sliding surface sss (t)reaches zero.This also indicates that the state variables of the proposed power system will enter the sliding mode stage in a finite time.

On the sliding surface,there is

where woptis the rotational speed at the operating point and Te(t)is the electromagnetic torque.

糖尿病周围神经病变（DPN）是糖尿病常见的并发症，患病率高达50%［1］。以往DPN常被认为是仅累及外周神经系统的疾病，近年有研究［2-3］证实DPN患者存在中枢神经系统的结构性异常，包括：脊髓横截面积及体积减少、躯体感觉区域灰质体积减少等。静息态功能磁共振（rs-fMRI）相关研究［4-5］发现DPN患者可能存在中枢脑功能活动的异常，但其确切的中枢脑功能改变仍然未明。

Substituting(46)into(19)yields

By selecting C1,C2and C3in(24)and(25),the derivative of ZZZ1can have the form of

From(48),each element in bothwill rapidly converge to zero in a finite time.Further, ZZZ2is the linear combination ofin(46)such that each element in ZZZ2will converge to zero in a finite time as well as ZZZ1.Since ZZZ is defined by the form of,each element in ZZZ and x will definitely converge to zero in a finite time.

Consequently,the closed-loop LFC control system is asymptotically stable in the sense of Lyapunov.In this regard,not only can the designed update law(34)guarantee the convergence of the RBF NNs,but also the whole LFC system can earn the asymptotic stability by such a control scheme.■

Remark 4:From(19), uuueqin(27)can be rewritten as

In(49),the ith element in GWhen zi=0 and˙zi/=0,there will be a singular condition and an in finite control input.Due to˙zi/=0,the state of zi=0 cannot be kept unchanged.The singular point(zi=0,˙zi/=0)is not a stable state and occurs rarely.Therefore,zihas an assignment of a pretty small value when it reaches the singular point to avoid the singularity problem during the calculation.As a result,this may cause a pretty big control input.

Remark 5:An inherent drawback of the sliding mode control technique is chattering[14].To reduce the chattering,the smooth processing should be adopted for the relay characteristics in(31)[32].The amendment has the form of

where δ is a small positive constant.

Remark 6:According to the designed control scheme,the GRC nonlinearity is transformed into a part of system uncertainties.The RBF NNs can be treated as a compensator to approximate and compensate the entire system uncertainties.The compensator and the controller cooperate to overcome the LFC problem for the nonlinear power system with wind turbines.

Remark 7:The generating unit with GRC in Fig.1 and the wind turbine in Fig.2 simultaneously exist in a control area.Both of them can affect the frequency of the considered control area.Hence,two load frequency controllers and their compensators in one control area should be designed to achieve the LFC task according to(5)and(11),respectively.

IV.SIMULATION RESULTS

Consider the LFC problem of renewable power systems.An interconnected power system with wind turbines is employed to illustrate the effectiveness and feasibility of the proposed TSMC scheme.The power system is composed of two control areas.Each control area has an aggregated generating unit with GRC in Fig.1 and an aggregated wind turbine unit in Fig.2.As mentioned above,the generating unit and the wind turbine mean all generator units and all wind turbines in the control area are aggregated together.The proposed control scheme will be carried out by the LFC solution of such a nonlinear interconnected power system,where the system schematic is illustrated in Fig.4.

Some parameters and data from the interconnected power system[10]are listed in Table I.The power system consists of two 800MVA-scale control areas.Wind turbine parameters[10]are listed in Table II,where the wind speed vwind=11m/s,the rotational speed vr=1.17m/s,Xm is the magnetizing reactance,(wr)base= 1.15rad/s.The wind turbine parameters originate from the 247MW operating point,where the turbine power composed of 200 units of 2MW wind turbines is 400MVA-scale.The gains of the two additional states in the two control areas are selected as KE1=KE2=1.According to the listed parameters,the nominal models of the interconnected power system in Fig.4 can be obtained.Subsequently,the models can be employed for the LFC design.

As mentioned above,some controller parameters are pre-defined.Concerning the T-SMC-based controllers in Areas 1 and 2,thetransformation matrices Tr(r=1,...,4)for each controller design are specified by

For Controllers 1 and 3,C1=C3=[1 2 3 4],C2=1.For the other two controllers in Fig.4,their parameters are determined by

Other controller parameters are determined by p=9,q=7,η=0.1 and δ=0.04.

Concerning the designed RBF NNs,some network parameters should also be set up.For RBF NNs 1 and 3,place 6 neurons in their hidden layers,select the initial weights between the hidden and output layers as[0.1 0.1 0.1 0.1 0.1 0.1]T,and determine the widths of the Gaussian function vector as[0.2 0.2 0.2 0.2 0.2 0.2]T.For RBF NNs 2 and 4,place 8 neurons in their hidden layers,select their initial weights as[0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]Tand determine the widths of the Gaussian function vectors as[0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2]T.For the four RBF NNs,all their centers take random numbers between−1 and 1.Other parameters for the four RBF NNs are defined by ε0=0.002 and ε1=0.001.

To show the performance of the presented method,two step load disturbances PL1=PL2=1%are simultaneously applied to the interconnected power system at t= 5s.Fig.5 illustrates the comparisons of frequency deviations,area control errors and deviations of the tie-line active power,where the blue solid lines mean the LFC system has RBF NNs and the red dashed lines mean the LFC system has no RBF NNs.Under the condition of no RBF NNs,the boundary of system uncertainties is assumed to be known such that the control law(50)is directly adopted.The other controller parameters are kept unchanged.

From Fig.5,∆f,ACE and∆Ptieare damped to zero after disturbances with small oscillations in a finite time.Without doubt,the system state changes of the LFC system with RBF NNs are a little smoother and the settling times are slightly shorter.Anyway,the control performances of the two control methods can just make a little difference on the aspect of these state variables.

Fig.6 illustrates the control inputs in the two control areas,where the red solid lines indicates the LFC system without RBF NNs and the blue solid lines indicates the LFC system with RBF NNs.To guarantee the system stability,the LFC system without RBF NNs needs to predefine a large¯h0 since the LFC system suffers from the severe chattering phenomenon.On the other hand,the LFC system with RBF NNs can adaptively approximate the boundary of the system uncertainties such that the curves of the control inputs are much smoother.In this sense,the designed RBF NNs contribute to a significant reduction of the chattering phenomenon.Further,the outputs of the designed RBF NNs are illustrated in Fig.7.As proven in Theorem 1,not only can the update law(34)make the networks convergent,but also the LFC control system possesses the asymptotic stability.

To demonstrate the control performance of the presented control scheme,the comparisons between the presented scheme and the SMC method with RBF NNs are shown in Fig.8.In Fig.8,the RBF NNs parameters in the two LFC control systems make no difference.The control performance is just decided by the two control methods.

Compared with the results of the SMC method with RBF NNs,the performance of the presented control scheme is better without doubt.Especially,the overshoot of the presented method in Fig.8 is zero which shows that the presented control scheme is more robust against load disturbances in power systems with wind turbines.

V.CONCLUSIONS

This article has addressed the LFC problem for renewable power systems in the presence of GRC.The control scheme is designed by means of T-SMC.To suppress the uncertainties of the LFC problem,RBF NNs are adopted.The theoretical analysis in the sense of Lyapunov proves that the T-SMC-based LFC system is asymptotically stable.The presented control scheme has solved the LFC problem of an interconnected renewable power system composed of two control areas.Some numerical simulation results have demonstrated the performance of the presented method against uncertainties of nonlinear power systems with renewable sources.

APPENDIX PARAMETER MATRICES

REFERENCES

[1]S.R.Bull,“Renewable energy today and tomorrow,”Proc.IEEE,vol.89,no.8,pp.1216−1226,Aug.2001.

[2]S.Behera,S.Sahoo,and B.B.Pati,“A review on optimization algorithms and application to wind energy integration to grid,”Renew.Sustain.Energy Rev.,vol.48,pp.214−227,Aug.2015.

[3]T.Basbous,R.Younes,A.Ilinca,and J.Perron,“Optimal management of compressed air energy storage in a hybrid wind-pneumatic-diesel system for remote area power generation,”Energy,vol.84,pp.267−278,May2015.

[4]C.X.Wang and J.D.McCalley,“Impact of wind power on control performance standards,”Int.J.Electr.Power Energy Syst.,vol.47,pp.225−234,May2013.

[5]S.P.Wen,X.H.Yu,Z.G.Zeng,and J.J.Wang,“Event-triggering load frequency control for multiarea power systems with communication delays,”IEEE Trans.Industr.Electron.,vol.63,no.2,pp.1308−1317,Feb.2016.

[6]P.K.Ray,S.R.Mohanty,and N.Kishor,“Dynamic load-frequency control of hybrid renewable energy based power system with HVDC-link,”J.Electr.Eng.Theory Appl.,vol.1,no.1,pp.24−31,Feb.2010.

[7]F.Díaz-González,M.Hau,A.Sumper,and O.Gomis-Bellmunt,“Participation of wind power plants in system frequency control:review of grid code requirements and control methods,”Renew.Sustain.Energy Rev.,vol.34,pp.551−564,Jun.2014.

[8]D.Rangaswami and P.Sennappan,“Load frequency control using multistage fuzzy logic controller for wind-micro hydro-diesel hybrid power system,”J.Vib.Control,vol.19,no.7,pp.1004−1014,May2013.

[9]H.Bevrani and P.R.Daneshmand,“Fuzzy logic-based load-frequency control concerning high penetration of wind turbines,”IEEE Syst.J.,vol.6,no.1,pp.173−180,Mar.2012.

[10]T.H.Mohamed,J.Morel,H.Bevrani,and T.Hiyama,“Model predictive based load frequency control-design concerning wind turbines,”Int.J.Electr.Power Energy Syst.,vol.43,no.1,pp.859−867,Dec.2012.

[11]T.H.Mohamed,J.Morel,H.Bevrani,A.Abd-Eltawwab Hassan,Y.Sayed Mohamed,and T.Hiyama,“Decentralized model predictive-based load-frequency control in an interconnected power system concerning wind turbines,”IEEJ Trans.Electr.Electron.Eng.,vol.7,no.5,pp.487−494,Sep.2012.

[12]M.H.Kazemi,M.Karrari,and M.B.Menhaj,“Decentralized robust adaptive load frequency control using interactions estimation,”Electr.Eng.,vol.85,no.4,pp.219−227,Sep.2003.

[13]L.R.Chang-Chien,N.B.Hoonchareon,C.M.Ong,and R.A.Kramer,“Estimation of β for adaptive frequency bias setting in load frequency control,”IEEE Trans.Power Syst.,vol.18,no.2,pp.904−911,May 2003.

[14]V.I.Utkin,Sliding Modes in Control and Optimization.Berlin Heidelberg,Germany:Springer,1992.

[15]D.W.Qian,X.J.Liu,and J.Q.Yi,“Robust sliding mode control for a class of underactuated systems with mismatched uncertainties,”Proc.Inst.Mech.Eng.I J.Syst.Control Eng.,vol.223,no.6,pp.785−795,Sep.2009.

[16]Z.M.Al-Hamouz and Y.L.Abdel-Magid,“Variable structure load frequency controllers for multiarea power systems,”Int.J.Electr.Power Energy Syst.,vol.15,no.5,pp.293−300,Oct.1993.

[17]K.Vrdoljak,N.Peri´c,and I.Petrovi´c,“Sliding mode based loadfrequency control in power systems,”Electr.Power Syst.Res.,vol.80,no.5,pp.514−527,May2010.

[18]Z.Al-Hamouz,H.Al-Duwaish,and N.Al-Musabi,“Optimal design of a sliding mode AGC controller:application to a nonlinear interconnected model,”Electr.Power Syst.Res.,vol.81,no.7,pp.1403−1409,Jul.2011.

[19]R.Hooshmand,M.Ataei,and A.Zargari,“A new fuzzy sliding mode controller for load frequency control of large hydropower plant using particle swarm optimization algorithm and Kalman estimator,”Eur.Trans.Electr.Power,vol.22,no.6,pp.812−830,Sep.2012.

[20]D.W.Qian,D.B.Zhao,J.Q.Yi,and X.J.Liu,“Neural sliding-mode load frequency controller design of power systems,”Neural Comput.Appl.,vol.22,no.2,pp.279−286,Feb.2013.

[21]B.Mohanty,“TLBO optimized sliding mode controller for multi-area multi-source nonlinear interconnected AGC system,”Int.J.Electr.Power Energy Syst.,vol.73,pp.872−881,Dec.2015.

[22]Y.Mi,Y.Fu,D.D.Li,C.S.Wang,P.C.Loh,and P.Wang,“The sliding mode load frequency control for hybrid power system based on disturbance observer,”Int.J.Electr.Power Energy Syst.,vol.74,pp.446−452,Jan.2016.

[23]S.Mobayen,“Finite-time tracking control of chained-form nonholonomic systems with external disturbances based on recursive terminal sliding mode method,”Nonlinear Dyn.,vol.80,no.1−2,pp.669−683,Apr.2015.

[24]S.Mobayen,“Fast terminal sliding mode controller design for nonlinear second-order systems with time-varying uncertainties,”Complexity,vol.21,no.2,pp.239−244,Dec.2015.

[25]A.Bidadfar,H.P.Nee,L.D.Zhang,L.Harnefors,S.Namayantavana,M.Abedi,M.Karrari,and G.B.Gharehpetian,“Power system stability analysis using feedback control system modeling including HVDC transmission links,”IEEE Trans.Power Syst.,vol.31,no.1,pp.116−124,Jan.2016.

[26]R.K.Sahu,S.Panda,and N.K.Yegireddy,“A novel hybrid DEPS optimized fuzzy PI/PID controller for load frequency control of multiarea interconnected power systems,”J.Process Control,vol.24,no.10,pp.1596−1608,Oct.2014.

[27]Y.F.Lv,J.Na,Q.M.Yang,X.Wu,and Y.Guo,“Online adaptive optimal control for continuous-time nonlinear systems with completely unknown dynamics,”Int.J.Control,vol.89,no.1,pp.99−112,Jan.2016.

[28]M.C.Pai,“Dynamic output feedback RBF neural network sliding mode control for robust tracking and model following,”Nonlinear Dyn.,vol.79,no.2,pp.1023−1033,Jan.2015.

[29]S.Mobayen,“An adaptive fast terminal sliding mode control combined with global sliding mode scheme for tracking control of uncertain nonlinear third-order systems,”Nonlinear Dyn.,vol.82,no.1−2,pp.599−610,Oct.2015.

[30]C.M.Dorling and A.S.I.Zinober,“Two approaches to hyperplane design in multivariable variable structure control systems,”Int.J.Control,vol.44,no.1,pp.65−82,Jun.1986.

[31]Z.H.Man and X.H.Yu,“Terminal sliding mode control of MIMO linear systems,”IEEE Trans.Circuits Syst.I Fundam.Theory Appl.,vol.44,no.11,pp.1065−1070,Nov.1997.

[32]S.Xu,J.Ma,and Y.Li,“Dynamic nonsingular terminal sliding mode control of uncertain linear multivariable systems,”ICIC Express Lett.,vol.5,no.9,pp.3033−3038,Sep.2011.