Novel Linear Iteration Maximum Power Point Tracking Algorithm for Photovoltaic Power Generation

  • Published on
    14-Mar-2017

  • View
    212

  • Download
    0

Transcript

  • IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 24, NO. 5, OCTOBER 2014 0600806

    Novel Linear Iteration Maximum Power PointTracking Algorithm for Photovoltaic

    Power GenerationWei Xu, Chaoxu Mu, and Jianxun Jin

    AbstractA novel maximum power point tracking (MPPT)algorithm is proposed in this paper based on the linear iterationmethod for photovoltaic (PV) power generation for improvingsteady-state performance and fast dynamic response simultane-ously. In this MPPT algorithm, linear prediction estimates theMPP of PV with a very high accuracy due to the constant-voltageand constant-current characters of PV. Then, further iterations forerror correction are executed to get a better approximation to theMPP. Experiments verify theoretical analysis and demonstrate itsfast dynamic response and high working efficiency.

    Index TermsConvergence, dynamic response, linear iteration,maximum power point tracking (MPPT) algorithm, photovoltaic(PV) power generation.

    I. INTRODUCTION

    R ECENTLY, there has been an increasing interest in usingPV systems to supply electricity for various consumerssince it is clean, noiseless, inexhaustible, and little maintenance[1]. However, the PV power generation performs a nonlinearvoltagecurrent (UI) curve, and the output power of a PVsystem largely depends on the array temperature and the solarinsulation, so it is necessary to constantly track the maximumpower point (MPP) of the solar array. Several methods, such asconstant-voltage tracing method, perturbation and observation(P&O) method, incremental conductance method (INC), etc.,have been used to draw the maximum power of the solar arrayin the past couple of years. Generally, each method mainlyincludes three criteria, i.e. tracking velocity, tracking accuracy,and stability. The constant-voltage method can track the MPPrapidly, but its accuracy is poor. The P&O method has beenwidely used in practice due to its simplicity and easy implemen-tation, but the tracking speed is slow and always has oscillationsduring the MPP tracking period [2], [3]. The INC method couldbe regarded as an improved P&O method. It can track the

    Manuscript received May 22, 2014; accepted June 18, 2014. Date of pub-lication June 30, 2014; date of current version July 31, 2014. This work wassupported by theNational Natural Science Foundation under Grant 51377065,Grant 61301035, and Grant 61304018. (Corresponding author: Chaoxu Mu.)

    W. Xu is with the School of Electrical and Electronic Engineering,Huazhong University of Science and Technology, Wuhan 430074, China(e-mail: weixu@hust.edu.cn).

    C. Mu is with the School of Electrical and Automatic Engineering, TianjinUniversity, Wuhan 300072, China (e-mail: cxmu@tju.edu.cn).

    J. Jin is with the School of Automation Engineering, University of Elec-tronic Science and Technology of China, Chengdu 610054, China (e-mail:jxjin@uestc.edu.cn).

    Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

    Digital Object Identifier 10.1109/TASC.2014.2333534

    MPP more accurately, but the fixed step size is determined bythe requirements of steady-state calculation accuracy and theMPPT dynamic response. Hence, its tracking speed is also veryslow [4], [5]. The variable step size algorithms can track MPPfast and accurately, but some essential parameters, such as themaximal step size and scaling factor, are difficult to estimateaccurately [6]. In general, fast tracking can be achieved withbigger increments, but the system might not operate exactly atthe MPP, resulting in oscillations around it [3]. This situationcould turn to the contrary when the MPPT operates with asmaller increment [6].

    To overcome the limitations of the traditional MPPT meth-ods, a novel MPPT algorithm based on linear iteration isproposed in this paper to achieve much faster tracking speedand higher efficiency. In the proposed algorithm, the step sizecan modify its response speed intelligently without step sizereference.

    II. BASIC CHARACTERISTICS AND MODEL OF THE PV

    The PV power generation system has many parameters in-fluenced by irradiance and temperature, which are difficult todetermine accurately. In engineering application, a simple PVcharacteristic equation based on PV cell physical character hasbeen widely used as follows [7]:

    i = ISC

    {1 C1

    [exp

    (u

    C2UOC

    ) 1

    ]}(1)

    where ISC is the PV generation short circuit current relative tothe solar radiation and temperature, UOC is the PV open circuitvoltage, i expresses the solar cell output current and u is thesolar cell output voltage. Two coefficients C1 and C2 can bedescribed as

    C1 =

    (1 Im

    ISC

    )exp

    ( UmC2UOC

    )(2)

    C2 =

    (UmUOC

    1)/

    ln

    (1 Im

    ISC

    )(3)

    where Um is the voltage of the PV model when the output poweris maximal, and Im is the maximum of the PV model outputcurrent. Then the PV array output power is

    P (u) = u ISC{1 C1

    [exp

    (u

    C2UOC

    ) 1

    ]}. (4)

    1051-8223 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

  • 0600806 IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 24, NO. 5, OCTOBER 2014

    Fig. 1. The DC module structure of PV system.

    The efficiency of PV cell depends on the internal shuntresistance, sunlight intensity, array temperature, and load. SetISC_ref , UOC_ref , Um_ref , Im_ref as the PV reference pa-rameters under standard conditions, and solar irradiation isgiven by Sref = 1000 W/m2 and circumstance temperature isTref = 25 centi degree(C), then ISC, UOC, Um, and Im can beexpressed as

    ISC = ISC_ref S

    Sref(1 + a T ) (5)

    UOC =UOC_ref ln(e+ b S)(1 c T ) (6)

    Im = Im_ref S

    Sref(1 + a T ) (7)

    Um =Um_ref ln(e+ b S)(1 c T ) (8)where S = (S/Sref 1), T = (T Tref), and typicallya = 0.025/C, b = 0.5/(W/m2), c = 0.00288/C, respec-tively, e is constants. It is obvious that the PV output changesdepending upon circumstance. Especially, when the system isunder partially shade condition, the output characteristics ofPV array will become more complex. In order to overcomethis shortcoming, a MPPT PV system with DC module isproposed shown in Fig. 1, the PV arrays are connected inparallel through DC/DC converter and the maximum outputpower is controlled by some independent micro programmedcontrol unit (MCU) controllers. A constant-voltage strategy canbe adapted to DC/AC converter to maintain the voltage of DCbus. Therefore, every unit of PV array, converter and controllercan be regard as separated with each other.

    III. LINEAR PREDICTION METHODFOR MPPT ALGORITHM

    A. Parameters Choice for Linear Iteration Method

    Normally, the UI curve of PV includes two regions, i.e. theconstant-current region when u [0, Um ] and the constant-voltage region when u [Um + , UOC], , > 0, as shownin Fig. 2. The derivative of the power to voltage (dP/dU)depicted by Curve (1) in Fig. 2, can be described by

    P (u) = ISC

    (1 + C1 C1 exp

    (u

    C2UOC

    ))

    C1ISCuC2UOC

    exp

    (u

    C2UOC

    ). (9)

    Fig. 2. The curves of P (u), 1/[1 + P (u)2], P (u) and angle .

    H(ux) is set as the tangent line through point (ux, P (ux)) ofthe UP curve, which can be calculated by

    H(ux) = P(ux) (u ux) + P (ux) (10)

    is supposed as a angle between the tangent line H(ux) andx-axis depicted by Curve (2). Its expression is

    =arctandP

    du=arctan

    [iC1ISCu

    C2UOCexp

    (u

    C2UOC

    )]. (11)

    The derivative of arctan(P (u)) can be estimated by

    d arctan (P (u))

    du=

    1

    (1 + P (u)2). (12)

    When u [0, Um] or u [Um+, UOC], 1/[1+P (u)2]shown in Fig. 2 is less than 0.1. It illustrates that i and unearly keep constants in the constant-current region and theconstant-voltage region, respectively. Hence, the estimation ofMPP saves time consumption and achieves high accuracy byusing two first-order Taylor series expansions. From the viewof geometry, is the most intuitive representation of P (u), andit changes very slowly when PV system works on the regionsof constant current or constant voltage.

    B. Basic Analytic Steps of Linear Prediction Algorithm

    The new algorithm includes linear prediction and error cor-rection. The procedures are summarized as follows:

    S1: Take any two points uL and uR from the PV curve. LetuL [0, Um ] and uR [Um + , UOC]).

    S2: Define HL(u) is the left tangent line through point(uL, P (uL)), HR(u) is the right tangent line throughpoint (uR, P (uR)). The corresponding slopes are

  • XU et al.: NOVEL LINEAR ITERATION MPPT ALGORITHM FOR PV POWER GENERATION 0600806

    Fig. 3. The procedure of linear prediction algorithm (a) Linear prediction.(b) Error correction.

    P (uL) and P (uR), where P (uL) > 0 and P (uR) 0 and P (uR) < 0 are always true, thetwo tangent lines will intersect at O1, and set the corre-sponding point on the PV curve as P1. Then it considerspoint P1 as the prediction MPP of the PV curve, sothe segment O1P1 is the truncation error referring toFig. 4(a). And the abscissa of point P1 is

    uP1 =P (uR) P (uL) + P (uL) uL P (uR) uR

    P (uL) P (uR).

    (14)S4: The tangent line through point P1 is H(uP1) and its

    slope is P (uP1). If P (uP1) > 0, set uL = uP1; other-wise, set uR = uP1. Referring to (13) and (14), updateP (uL), P (uR), P (uL), P (uR), HL(u) and HR(u).

    S5: Update the intersection point Oi of HL(u) and HR(u)and the corresponding Pi. Calculate uPi referring to(14). If |H(uOi) P (uPi)| , return S3 until theterminal condition is satisfied. If the truncation error|H(uOi) P (uPi)| < , > 0, the iteration stops, andmaintain the operating voltage uPi until the systemrestarts or the irradiance changes.

    Fig. 3(a) and (b) describe the processes of linear predictionand error correction. When |H(uOi) P (uPi)| < , the algo-rithm stops and the operating voltage is maintained until thenext duty cycle.

    Fig. 4. The maximum truncation error of linear prediction.

    C. Estimation of Truncation Error

    The truncation error is defined as R(u) = |H(uoi)P (uPi)|, where H(u) is the tangent line of UP curve, Oi theintersection point of the ith iteration cycle. Due to that H(ui)goes through point (ui, P (ui)), R(ui) = H(ui) P (ui) = 0and R(ui) = H (ui) P (ui) = 0, it can calculate the errorR(u) by

    R(u) = (u) (u ui)2 (15)

    where (u) is the scale factor. In order to get the specificexpression of (u), it should calculate P (u) according toRolle theorem expressed by

    P (u) = 2C1ISCC2UOC

    eu

    C2UOC C1ISCu(C2UOC)2

    eu

    C2UOC . (16)

    It is obvious that P (u) < 0 is always true. When u [0, UOC], there is at least one that satisfies

    P ()H() () ( ui)2 = 0. (17)

    By further investigation, it can get the relationship between(u) and , as illustrated by

    (u) = P ()/2 (18)

    where P () is a boundary function and is a variable duringthe range of [0, UOC]. Then, the estimation maximum error is

    R(u)=P ()

    2[P (uR)P (uL)+P (uL)(uLuR)

    P (uL)P (uR)

    ]2.

    (19)

    As shown in Fig. 4, the maximal absolute value of truncationerror is less than 30, and it could become smaller gradually withthe increasing times of error correction.

    D. Proof of New Algorithm Convergence

    Due to P (u) < 0, the P (u) can be regarded as one con-tinuous convex function. According to the PV curve, uO [uL, uR], the interval [uL, uR] will become [uiL, u

    iR] after i times

    error correction, and the equations are obtained as follows:uiO uiR =Li uiL uiR (20)Li

    uiL uiR = ui+1L ui+1R (21)

  • 0600806 IEEE TRANSACTIONS ON APPLIED SUPERCONDUCTIVITY, VOL. 24, NO. 5, OCTOBER 2014

    Fig. 5. Tracking trajectory in mathematical theory.

    Fig. 6. The response speed of Newton iteration and proposed methods.(a) Newton iteration method. (b) The proposed method.

    where Li is a scale factor less than 1. Thus it can get

    Li+k1 ui+k1L ui+k1R = uiL uiR

    i+k1n=i

    (Ln). (22)

    As Li < 1, when k , |ui+kL ui+kR | will get close to0. Due to umax [ui+kL , ui+kR ], the result will finally convergeto the MPP (ui+kP umax), as shown in Fig. 5. Due to theconstant area of PV, it can be seen that the first step (linearprediction) can drive the working point to the vicinity of MPPrapidly, and then others will improve the tracking accuracythough repeated iterations.

    Fig. 6 is the tracking speed comparison between Newtoniteration and the proposed methods. The formula of Newtonsiterative algorithm is

    uk+1=P (uk1)P (uk)[P (uk1)ukP (uk)uk1]

    P (uk)P (uk1).

    (23)

    In Fig. 6(a), Newton iteration method has a character ofquadratic convergence, and it just takes eight calculation cyclesto obtain the optimal solution under standard conditions. But itis less stable for great dependence on its initial iteration value,which can be solved effectively by the new method in this paper.As shown in Fig. 6(b), when the initial point u0 [0, Um], the

    Fig. 7. The accuracy analysis of in one step size. (a) The minimal error.(b) The maximal error.

    Fig. 8. Simulation on the calculation accuracy of the two methods.

    iteration result of Newton method will diverge. However, theproposed method has little requirement of the initial value, andit has almost similar tracking response speed (13 calculationcycles) for MPP.

    E. Accuracy Analysis of New Method and P&O Method

    P (u) is always calculated by P/u, where u is a tinyvoltage increment and P is the corresponding power incre-ment. Compared with other traditional methods with a fixedstep size (uS), the new method has much higher accuracyeven when u = uS.

    In Fig. 7, the minimum error after line iteration is 0, whilethe maximum error is u, < 1 is a scale factor. After thelast iteration, if the voltage uOi of intersection Oi is just on theMPP, the linear prediction algorithm has the minimum error 0,as depicted in Fig. 7(a). Meanwhile, the minimum error of fixstep size method is u, < 1 is also a scale factor. If theslop of H(uL) is zero, as shown in Fig. 7(b), the maximumerror of new method is u, which is different with those oftraditional methods with (1 + u). It can be concluded thatthe new method has higher accuracy even when u = uS.

    Fig. 8 and Table I give the simulation result of the accuracy.It sets the increment of new algorithm u = 1 V which useto calculate the derivative of P (u) and P (u) can be expressby P/u approximately. Set uS = 1 V in P&O method.From Fig. 8, the voltage can converge to 36.59 by the proposedmethod, which is very close to the theoretical UMPP = 36.52even with a big increment, while the voltage oscillates between36 and 38 by P&O method. It is obvious that the proposedmethod is more accurate than the P&O method even with a bigperturbation step. u is much smaller than uS in practice,and hence the new method could have much higher accuracy inthis condition.

    F. Analysis of Stability Under ComplexCircumstance Conditions

    The tracking speed and steady performance of the newmethod are affected by the value of P (ui) and the position

  • XU et al.: NOVEL LINEAR ITERATION MPPT ALGORITHM FOR PV POWER GENERATION 0600806

    TABLE ISIMULATION RESULT OF THE ACCURACY

    Fig. 9. The stability analysis under the solar irradiation step change. (a) Solarirradiation steps up. (b) Solar irradiation steps down.

    TABLE IISIMULATION DATA OF THE RESPONSE SPEED AND GROSS ENERGY

    of tangent line H(ui). When the position of H(u) h...

Recommended

View more >