Real-Time Energy Management Algorithm for Mitigation of Pulse Loads in Hybrid Microgrids

IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 4, DECEMBER 2012 1911

Real-Time Energy Management Algorithm forMitigation of Pulse Loads in Hybrid Microgrids

Ahmed Mohamed, Student Member, IEEE, Vahid Salehi, Student Member, IEEE, andOsama Mohammed, Fellow, IEEE

Abstract—This paper presents a real-time energy managementalgorithm for hybrid ac/dc microgrids involving sustainable en-ergy and hybrid energy storage. This hybrid storage system con-sists of super capacitors (SC) for ultra-fast load matching besidelithium-ion batteries for relatively long term load buffering. Theenergy management algorithm aims mainly at managing the en-ergy within the system such that the effect of pulsed (short dura-tion) loads on the power system stability is minimized. Moreover,an average annual saving of around 7% is achieved by shiftingloads to off-peak hours. The expected energy needed during a fu-ture peak, the time of its occurrence and the current state of chargeof both elements of the hybrid storage system are all examples ofthe inputs to the algorithm. A nonlinear regression technique isused to obtainmathematicalmodels for the uncertain quantities in-cluding load and sustainable energy curves. The results show a sig-nificant improvement for the system in terms of voltage and powerstability by applying the proposed algorithm.

Index Terms—Energy management, fuzzy controller, hybridsuper capacitor/battery storage, nonlinear regression, pulsed loadmitigation, state of charge.

I. INTRODUCTION

I T IS EXPECTED that the rapidly growing implementationof smart grids and microgrids will continue to change cur-

rent systems in terms of design and operation. New designs mayinclude much larger local generation, storage elements, hybridac/dc distribution systems and more extensive involvement ofpower electronic converters and pulsed loads [1]–[4]. An ex-ample would be a shipboard power system, which resemblesthe concept of a microgrid operating in a smart grid systemwhere the system is capable of self-diagnosing, self-healing,and self-reconfiguring [5], [6]. In these systems, some partic-ular loads draw very high short time current in an intermittentfashion such as electromagnetic rail weapon launch systems andfree electron lasers. Henceforth, they will be referred to collec-tively as pulsed loads [7]. Such current behavior can potentiallycause the system voltage and frequency to drop in the entire mi-crogrid, momentarily. This disturbance can trip other sensitivecontrol loads offline.

Manuscript received July 30, 2011; revised November 29, 2011; acceptedMay 05, 2012. Date of publication June 11, 2012; date of current version De-cember 28, 2012. This work is supported in part by the Office of Naval Research.Paper no. TSG-00304-2011.The authors are with the Energy Systems Research Laboratory, Department

of Electrical and Computer Engineering, Florida International University,Miami, FL 33174 USA (e-mail: [email protected]; [email protected];[email protected]).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TSG.2012.2200702

In the shipboard example, when a large magnitude, andprolonged voltage/frequency sag occurs, the propulsion systemmay shut down, or perhaps the fighting loads themselves maybe thrown offline. Therefore, there is a great concern abouthow these loads can coexist in the same electrical environmentand share the same energy storage systems while allowing adiverse range of operational scenarios [8], [9]. However, pulsedloads are not limited to shipboard power systems. For instance,a plug-in hybrid electric vehicle (PHEV), or a group of PHEVsduring their charging process, or a big machine during itsstarting can be considered as pulsed load in residential andindustrial applications, respectively.Loads based on hourly average variations can be considered

as low-frequency variations, whereas power transients that sus-tain for minutes, seconds, or milliseconds come under the high-frequency segment. To buffer out the low-frequency oscillationsand to compensate for the intermittency of the renewable en-ergy sources, energy storage elements with high energy densityis required. To provide the high-frequency component of powerand also to supply or absorb the high-power transients, energystorage with high power density is required [10]. Recently, high-power capability of super capacitors and high energy capabilityof batteries or fuel cells are exploited in pulse operating modesfor portable power systems, electric vehicle and digital telecom-munication systems [11]–[13]. Advanced storage technologiesnow allow extraordinary energy densities where the load drawslarge power impulses. This deficiency can be solved by usingmore batteries in parallel. Technically, hybrid power sourcesthat utilize batteries with advanced charge/discharge strategiesin parallel with super capacitors can overcome the power defi-ciency problems and increase the operating time [14], [15].Real time or dynamic energy management in smart grids

whether directed towards microgrids or electric vehicle appli-cations was investigated in several publications [10], [15]–[18].These papers generally aim at handling renewable energy andits uncertainty, managing the demand side in an intelligent wayin order to enhance the performance of the microgrid as well asthe main grid, and/or achieving an optimal economic operationof the system. However, all these energy management algo-rithms do not take into consideration the occurrence of pulsedloads. In this paper, an energy management algorithm that aimsat handling the energy in a system involving renewable energysources such that pulsed loads are mitigated is developed. Fur-thermore, this developed algorithm assures economic operationof the microgrid.The paper is organized as follows, in Section II: the non-

linear regression modeling technique used to obtain mathemat-

1949-3053/$31.00 © 2012 IEEE

1912 IEEE TRANSACTIONS ON SMART GRID, VOL. 3, NO. 4, DECEMBER 2012

ical models for the PV and load data is presented. In Section III:a discussion of the operation and modeling of the hybrid supercapacitor/battery storage systems is detailed. In Section IV: thedeveloped real time energy management algorithm is explained.In Section V: the results are presented and discussed. Finally, inSection VI: a summary of the main conclusions that can be de-ducted out of this paper is presented.

II. PV AND LOAD DATA FORECASTING

A. Nonlinear Regression Modeling and Model EvaluationIndices

In order to design the energy management algorithm suchthat mitigation of pulsed loads is achieved, prior knowledgeabout the total energy generation available from renewablesources should be known. The solution to the energy commit-ment problem involving renewable energy sources also requiresthis information in advance. Hence, the mathematical modelingfor these uncertain quantities is obtained using a nonlinearregression technique. Different model evaluation indices wereused to validate the developed mathematical models.The nonlinear regression model used in this paper has the

ability to cope with the nonlinearity of the data and form anaccurate model. It is based on the idea of transformation of thedata using a predefined set of nonlinear functions in order toachieve linearity [19].The nonlinear model designated as has the following

form:

(1)

(2)

where:

total number of nonlinear functions;

total number of variables to be includedin the model;

.nonlinear model for each variable andis the summation of all terms resultingfrom transforming the input through apreselected set of nonlinear functions;

constants to be determined,;

preselected set of nonlinear functions thatwill be used for transformation of inputs.The set of nonlinear functions may contain, , and ;

numerical values for a given input to beused for deducing the model.

A detailed description of the way to calculate the nonlinearregression model parameters is given in the Appendix [20]. Al-though, nonlinear regression was selected in this paper to modelthe PV and load data due to its relatively short computation time

with acceptable accuracy [20], any other modeling techniquethat can be implemented effectively in real-time such as artifi-cial neural networks can also be used. A drawback of the math-ematical modeling technique used here is that it is site-specific.Generally, since the algorithm developed is independent fromthe modeling technique utilized as long as it is capable of pre-dicting PV and load data quickly enough and with acceptableaccuracy, the modeling stage is not the main focus of the paper.Various model evaluation indices were implemented to mea-

sure the accuracy of the proposed mathematical models. One ofthe indices is the mean absolute percentage error cal-culated by (3) and the other is the coefficient of determinationcomputed by (4) given as

(3)

(4)

where and are the vectors of the real and predicted data,respectively.The value of for the model ranges from 0 to 1 and it im-

plies that of the sample variation is attributable to or ex-plained by one or more of the variables as long as it approachesunity. The better regression fits the data the closer the value ofis to one.

B. Mathematical Modeling Results

The mathematical models for PV output power in addition tothe load demand were deduced. These mathematical models aregiven by (5) and (6), respectively.

(5)

(6)

where and are the hour and month, respectively.The data forecasting process was based on scaled-down PV

data collected over 15 years on an hourly basis for an examplePV system, namely, Texas Energy Park in Dallas, TX. Thissystem consists of 288 modules; four rows with 72 modules ineach row. Each row has a length of 104 m and each module israted 430 Wp. The power data was used as output data to beforecasted, whereas the day of the year (1–365) and the hour ofthe day (1–24) were used as inputs [21]. The PV mathematicalmodel was trained using the sets of data of the 14 previous years.However, the model was tested using the data of the most recentyear, which were not included during the training process. The

of this model is 4.65%, which is a reasonable valuetaking into consideration that we are minimizing the inputs tothe model (variables of the nonlinear functions) to only timebases. However, if we were to take other inputs related to envi-ronmental variations corresponding to sun radiations, we couldobtain a more detailed model as these inputs are more correlatedto the output power of the PV than just time.Moreover, the value

MOHAMED et al.: REAL-TIME ENERGY MANAGEMENT ALGORITHM FOR MITIGATION OF PULSE LOADS IN HYBRID MICROGRIDS 1913

Fig. 1. Battery/SC hybrid storage system. (a) Passive hybrid. (b) Active hybrid.

of is 0.951, which means that the transformed inputs usedare representative to the output power of the PV system. In thispaper, we try to count on only time to predict the output power.Nonlinear regression is helpful in this case as it transforms thesets of inputs into other forms that are more correlated to thedesired output [20].The load data of four consequent years were used to model

the load duration curve and they were categorized in such a waythat data of three years were used as training data whereas dataof another year were used as testing data. Values of theand are 6.45% and 0.934, respectively. The value ofis relatively small, whereas the value of is close to one. Thesetwo facts support our conclusion that the mathematical model iswell representing the actual data.

III. HYBRID STORAGE SYSTEM

According to Ragone Plot [22], which is usually employedto classify the available energy sources according to theirpower/energy density, high energy Li-ion rechargeable bat-teries have the highest energy density of all modern batteries50–500 Wh/kg and low power density of 10–500 W/kg. Onthe other hand, super capacitors have a high power densityrange of 1000–5000 W/kg and very low energy density of 1–10Wh/kg. Moreover, the internal resistance of super capacitorsis much lower than that of a battery resistance. Therefore, ithas much higher charging/discharging efficiency [23], [24]. Inorder to possess the benefits of high specific power and highspecific energy, a hybrid Li-ion/super capacitor storage systemis utilized in this paper in order to mitigate pulsed loads and usethe batteries for relatively longer time for normal loads feedingas well.The hybrid storage system is modelled with the circuit shown

in Fig. 1(a). The battery is modelled by an ideal voltage sourcein series with its internal resistance whereas the super capac-itors are modelled by the nominal capacitance in series withan equivalent series resistance .The Thevenin equivalent voltage and impedance of the par-

allel system in the frequency domain are described by the fol-lowing set of equations [25]:

(7)

(8)

where is the complex frequency, is the initial voltage ofthe super capacitor, and

(9)

(10)

Now, assuming a pulsed load current with pulse frequency(the period ) and pulse duty ratio, the current for thefirst pulses can be expressed as

(11)

where is the amplitude of the current and isa unit step function at . The current in frequency domaincan be found by performing the Laplace transform operation on(11), yielding the result

(12)

The average value of the load current can be expressed as theproduct of the pulse amplitude times the duty factor as

(13)

The output voltage is a linear combination of the Theveninvoltage source and the internal voltage drop. The inverseLaplace transform of the Thevenin voltage source, accordingto (7), is

(14)

The second term of (14) is due to the energy redistributionbetween the super capacitor and the battery at the beginningof the discharge. When , . For the currentwaveform it is defined by (11), the internal voltage dropas

(15)

The corresponding expression in the time domain is

(16)

From the circuit shown in Fig. 1, the output voltage can befound as

(17)


Applying the linear property of the Laplace transform to (17)and using (14) and (16), one obtains

(18)

The currents from the battery and the ultra-capacitor can befound once the output voltage is resolved. These currents are

(19)

(20)

A hybrid storage system with battery and super capacitorsonly, without utilizing any dc-dc converters as shown in Fig. 1.b, is called a passive hybrid. However, if a dc-dc converter isused, as an interface between the batteries and the super capac-itor, a significant improvement can be achieved. This was ex-perimentally verified in [26]. For instance, the battery currentcan be controlled to a constant value. Moreover, there will beno need for voltage matching between the battery and the load.Furthermore, the active hybrid yields a peak power that is 3.2times that of a passive hybrid, and a specific power of 2.7 timesthat of a passive hybrid [26].These features allow the active hybrid storage systems to mit-

igate pulsed loads efficiently if properly designed.

IV. REAL TIME ENERGY MANAGEMENT ALGORITHM

The main objective of the real time energy management al-gorithm developed in this paper is to mitigate pulsed loads. Be-sides, the total cost of energy is to be reduced using this al-gorithm by minimizing the energy drawn from the main gridand/or shifting it to off-peak hours. Therefore, we can definetwo main modes of operation namely; the pulsed load mitiga-tion mode and the normal operation or cost minimization mode.A flow chart of this algorithm is shown in Fig. 2. The prioritysetting of the algorithm can be listed as follows:

A. Super Capacitors Are Always Fully Charged to MitigatePossible Pulsed Loads

In order to achieve this in real time, the amount of energyavailable in the super capacitor must be monitored andcompared to the energy that would be available in fully chargedsuper capacitors to assure having all the super capaci-tors initially charged and ready to operate. If the super capacitorsare not fully charged, in case they are connected to the dc busthrough a dc-dc converter [24], they are immediately chargedusing the batteries and/or the grid power according to the avail-ability of energy in the battery.

Fig. 2. A flow chart of the developed energy management algorithm.

B. Li-ion Batteries Have Enough Energy to Help SuperCapacitors Mitigate Pulsed Loads

Li-ion battery is used along with super capacitors to mitigatethe effect of pulsed loads. Therefore, the batteries are controlledsuch that they are 100% charged, according to the design of thestorage system in this paper, before themoment of occurrence ofthe pulsed loads. However, if the future pulsed load is predictedto take place after time that is more than the time neededto fully charge the battery , the battery is used normallyin the cost minimization mode.


C. Normal Loads on the Smart Microgrid are Supplied UsingPV System Operating at its Maximum Power Point

Loads on the dc microgrid are supplied by the PV generation.If there is a surplus in power, which means that the difference

, where

(21)

There will be two options: either to charge the battery or tosell this surplus power to the grid. In order to minimize the an-nual overall cost of energy and maximize the saving, the deci-sion is made based on the energy tariff. Utility grids increasethe tariff during peak hours. Hence, if it is a peak period, whichis defined as 2 h around the peak of the daily load curve, pri-ority is given to selling the power to the utility grid. Otherwise,batteries are charged. However, this all depends on the state ofcharge (SoC) of the batteries as shown in Fig. 2. If there is asurplus power during a peak period, the percentage of the sur-plus power that goes to the battery follows the formulaproposed in [27] as

(22)

However, if the surplus power is available during an off-peakperiod, the percentage of the surplus power that goes to the bat-tery follows the following formula [27]:

(23)

The batteries’ SoC can be estimated as follows:

(24)

where (Ah) is the full charge or total energy capacity of thebattery, whereas the hour at which the next peak load is takingplace can be calculated using the predicted load curve ofthe day. At the peak instance, we have

(25)

where is the day in which we are calculating. The solutionof (25) yields the peak hour . The peak period is defined as

(26)

D. Any Load Deficiency on the Microgrid Is Supplied Fromthe Battery if It Is During a Peak Period or From the Grid if ItIs an Off-Peak Period

If there is a deficiency in power defined by , pri-ority is given to satisfy this deficiency using the energy storedin the batteries. However, the energy tariff is again taken intoconsideration. If lies within the peak period, the energy storedin the battery system is used to satisfy the biggest possible share

of the load demand. The rest is satisfied from the grid. The per-centage of the power deficiency that will be shared by the bat-tery system is implemented mathematically using the followingexponential curve [27]:

(27)

If lies within an off-peak period, is covered partiallyby the battery system based on . If there is enough timeto recharge the batteries and have them ready during the futurepeak period, the batteries are more involved in satisfying theload deficiency. However, if the time needed to charge the bat-teries is less than , the batteries are carefully discharged.This algorithm is implemented using a fuzzy agent within thealgorithm. Inputs to the fuzzy agent are and , which isthe ratio between the energy available in the batteries to the en-ergy needed during the future peak period. The fuzzy system isexplained in details in the following subsection by the batterysystem. This fuzzy system was based on the forecasted data ofthe expected next peak period and the energy needed within it.The energy demand during the peak period can be cal-

culated by substituting in (6) with the value of and. The energy demand during the peak period is

(28)

E. Fuzzy Agent Involvement During Power DeficienciesFuzzy control is a rule-based control technique that is ap-

proached by linguistic fuzzy rules, which describe the outputdesired from the system under different operating conditions.The fuzzy rules are in the form of if-then rules that the profi-cient user should design such that they cover all the conditionsthat the system is expected to go through.In this model, a fuzzy system was used only in the case when

the instantaneous load demanded is higher than the instanta-neous available power from the renewable energy sources andthe system is not at the peak period. At this state, the batterywill be operated in the discharge mode. Hence, the fuzzy systemdetermines the amount of power to be drawn from the batterywhile taking into consideration the time left for the peak periodand the ratio between the current energy available in the batteryto the total energy needed during that peak period .Designing a fuzzy logic controller is achieved through three

basic steps: fuzzification, inference mechanism, and defuzzifi-cation. The Sugeno type fuzzy system was used here [28]. Inthe fuzzification step, and are the inputs to the con-trol system which are mapped into certain linguistic values.The output of the fuzzy logic is the percentage of load to beshared by the batteries. Three fuzzy variables, two inputs andan output, were involved as shown in Fig. 3. Each variable hassome membership functions. For the first input, which is thetime left for the future peak period, three fuzzy subsets are used:small (S), medium (M), and big (B). For the second input, whichis the ratio between the current energy available in the batteryto the total energy needed during that peak period, four subsetswere used: very small (VS), small (S), medium (M), and big


Fig. 3. Operation of the system for a 24 h interval, while applying Algorithms1 and 2.

TABLE IFUZZY RULES

(B). On the other hand, the output is represented by four fuzzysubsets: very small (VS), small (S), medium (M), and big (B).These membership functions are used to map the input variableinto a fuzzy set. The operation of the membership functions onthe input variable yields the extent to which that variable is amember of a particular rule.The process of converting control variables into linguistics

rules is called fuzzification. However, in inference engine andrule-based step, the output of the fuzzy controller is managedthrough including certain linguistic rules. These control rules,summarized in Table I, are constructed based on the given con-ditions (inputs) such that the fuzzy controller decides the propercontrol action.Finally, in the defuzzification step, as the output of the fuzzy

controller is in the form of fuzzy set, it must be transformed fromlinguistic form into a number that can be used to control thesystem. The rules used here are given in Table I. Many defuzzi-fication methods such as weighted average or weightedsummation methods were proposed [28]. In this paper,we utilized the method.

V. RESULTS AND DISCUSSION

In order to evaluate the performance of the algorithm underpulsed loads, an example hybrid ac/dc system was simulated.The example system, shown in Fig. 4 resembles a shipboardpower system with scaled down ratings. This system includes

Fig. 4. The example system simulated in this paper.

two 13.8 kW main generators (MTG) and two 10.4 kW aux-iliary generators (ATG) connected in a ring bus configuration.The bulk of the load consists of two 50 kW propulsion mo-tors, modeled as permanent magnet machines supplied by PWMdrives, with hydrodynamic propeller models as the mechanicalload. Each rectifier supplies one of two 0.318 kV dc busses. Thedc distribution zone is supplied by one of the two available rec-tifiers. Although various models for the loads may be used, con-stant impedance models were used in this paper.Furthermore, a photovoltaic (PV) system of 10 kW rated ca-

pacity, lithium-ion batteries with 3000 Ah rated capacity andsuper capacitors with 200 F are included in the dc microgrid. APWM controlled dc-dc converter is used as an interface betweenthe PV system and the dc bus. Moreover, a vector decouplingPWM controlled ac-dc/dc-ac bidirectional converter was usedfor connectivity between the ac and dc sides. In the steady statecase, the system voltages and loadings are within the normallimits.For transient simulations, we considered a pulse train of four

pulses with a rate of 0.2 Hz, a duty ratio of 10%, and amplitudeof 20 kW.

A. Cost Minimization

In order to evaluate the performance of the developed algo-rithm designated (Algorithm 1), its operation is compared withthe operation of another algorithm designated (Algorithm 2).Algorithm 2 does not take into consideration any variation inthe power tariff during the day, and its operation can be sum-marized as follows; if there is a surplus in power ,the batteries will be charged until they are fully charged thenthe extra power will be sold to the grid. However, if there is adeficiency in power, the batteries will be discharged until theyreach their lower discharge limit then power will be drawn fromthe grid.In normal operation with no pulsed loads predicted or taking

place in the system, the developed algorithm aims at minimizingthe total cost of energy. This is achieved by handling the bat-tery and controlling the power flow between the ac and dc sidesbased on the changes of the power tariff. Fig. 5 shows the PVand load data of the 24 h at a certain day of the


Fig. 5. Operation of the system for a 24 h interval, while applying Algorithms 1 and 2.

year. As can be seen, there is a deficiency at thatday until around hour 7 (7:00 A.M.), then there is a surplus inpower until around hour 14.67 (2:40 P.M.), and finally there isa deficiency in power until the end of the day. The peak loadof that day takes place at hour 18, hence . More-over, can be found by integrating the area under the loadcurve within the interval 17–19. The second subplot of Fig. 5shows the power surplus/deficiency during the day, the powershares from the battery and the grid using both algorithms. Thesampling time is 0.1 ms, whereas the initial SoC is 90%.During the power deficiency taking place at the beginning ofthe day, both algorithms tend to satisfy the load using the en-ergy stored in the battery. Moreover, when there is a surplus inpower, the battery is charged to its full capacity and the extrapower is injected back to the grid. However, during the seconddeficiency interval starting at 2:40 P.M., there is a significant dif-ference between the operations of both algorithms. Since the fu-ture peak period is approaching soon, Algorithm 1 saves the en-ergy stored in the battery and satisfies the load demanded usingthe grid power. This stored energy can be used to minimize thepower drawn from the grid during the peak period when thepower tariff is the highest during the day. On the other hand,Algorithm 2 tends to satisfy the load deficiency using the en-ergy stored in the battery without considering the future peakperiod. Consequently, when the peak period starts, there is nomeans to supply the load other than drawing the power from themain ac grid with the high tariff. The difference in the energysupply is the highlighted area between the two curves;and . Hence, a significant saving in the total energy costduring the year can be achieved by applying Algorithm 1. Thesaving during a day can be calculated as

(29)

where and are the power tariffs duringthe peak and off-peak periods, respectively.Assuming that is 100% higher than

[29], a total annual saving of 7–9% can be achieved. It hasto be noticed here that the developed algorithm was based onforecasted data of PV generation and load demand, which willdefinitely include some prediction error, hence the total annualsaving is dependent on that error. For instance if the PV and loadmodels used are replaced with models having values of10% and 13%, respectively, the total saving drops to 5.8–7.2%.This operation can be also seen in the third subplot of Fig. 5

by noticing the SoC of the battery using both algorithms. It canbe seen that at the beginning of the peak period, Algorithm 1enables the battery to have an SoC of almost 98% whereas Al-gorithm 2 depletes the battery to almost its lower limit.

B. Pulsed Load Mitigation

As explained in a previous section, if there is a pulsed loadpredicted to take place, the real time energy managementalgorithm developed in this paper assures that the battery isfully charged to assist the super capacitors mitigate the effectof pulsed loads on the electrical power system, especially onthe ac side. In order to verify the validity of the algorithm, twocases have been investigated; the case when there is a pulseload while the proposed algorithm is implemented (Case 1)and another case when the occurrence of pulsed loads are notpredicted and/or planned for, while dealing with the charge/dis-charge process of the batteries. The main difference betweenthese two cases is that in the first case, the batteries are readyand fully charged when the pulsed loads take place, whereasin the second case the battery SoC is independent and random;this means that the charge/discharge process of the batteries isbased on other factors (Case 2).1) Case 1: Fully Charged Battery: In this case, the real time

energy management algorithm assures that the batteries arefully charged and will be available during the occurrence of thepulsed loads. Fig. 6 shows the active power of the pulsed loads


Fig. 6. Active power of the pulsed loads and the power sharing among ac gen-erators, super capacitor and full-charged battery (Case 1).

and the power sharing among ac generators, super capacitor andbattery. Initially, the battery is not injecting any power to the dcbus since it is dedicated to mitigate the pulsed load. Hence, thebattery voltage is maintained at the same voltage level of thedc bus. At the beginning of the pulsed load, the super capacitorsatisfies the whole demanded power because of the high rateof discharge. Then, the battery starts to increase the injectedpower to the demanded pulse, while the power share from thesuper capacitor is exponentially decaying. Therefore, the sizingof the battery and the super capacitor is a very crucial subject inthe design of hybrid microgrids with pulsed loads. According toinertia time constant of ac generators, they start to maintain thesystem frequency on 60 Hz and react to the pulsed loads. Thiscan be seen by noticing the oscillations of the ac generators’power. By the end of the first pulse, the battery will dischargeenergy to the dc bus due to the drop of its voltage level, whichneeds some time to be recovered as shown in Fig. 6. Becauseof the oscillations in ac generation, the dc bus voltage oscillatesbetween the pulses. Hence, the super capacitor power chargesand discharges before the consequent pulse as well. This willalso affect the battery power as can be seen in Fig. 6. Afterpassing the four pulses, the system goes back to its initial steadystate condition.Fig. 7 shows the voltage amplitude of the ac buses, which

is almost within the over/under voltage limits. The loading ofthe main generators are also presented in Fig. 8, it can be seenthat the system does not suffer from any overloading condi-tion. Fig. 9 shows the system frequency under pulsed loads andfully charged battery operation, the system frequency remainsbetween 59.5 Hz and 60.4 Hz, which is relatively acceptablefrequency.2) Case2: Half Charged Battery: In this case, we assume

that the energy available in the battery is just enough for the firsttwo pulses of the pulses train, which represents the case whenthe SoC drops to its lower limit after the second pulse. Tech-nically, it means that the battery voltage will dramatically dropafter the second pulse and the converter controller will discon-nect the battery from the dc bus. Fig. 10 shows this situation interms of generation and load levels. After the second pulse, thebattery injected power is zero, and then the ac generators are to-

Fig. 7. Voltage amplitude of ac buses during pulsed loads (Case 1).

Fig. 8. Loading of main generators during pulsed loads (Case 1).

Fig. 9. System ac side frequency during pulsed loads (Case 1).

tally responsible for the next two pulses. As can be seen, thereare large oscillations in ac generation levels and the super ca-pacitor’s power. The dc bus voltage drops to 0.823 p.u. becauseof the effect of the pulsed load as shown in Fig. 11. Moreover,there are over and under voltage violations reported on the acbuses as shown in Fig. 12, which has the same axis length asFig. 7 to facilitate the comparison. Fig. 13 shows the frequencyvariations during the pulse train, after consuming the whole bat-tery energy, during the third and fourth pulses, we can see largefrequency oscillations in the range of 58.4 to 61.12 Hz. The fre-quency of these oscillations is almost the same as the pulsedload frequency of 0.2 Hz. As in the previous case, the loading ofthe main generators is shown in Fig. 14, we can see some over-loading situations. Hence, the design and planning of the gener-ators capacity and their protection settings have to be matchedwith the system pulsed loads characteristics and the availablestorage system.The voltage changes in the buses depend on the pulse-load pa-

rameters such as the magnitude of the pulse, its duration and thenumber of iterations. The voltage controller parameters, such


Fig. 10. Active power of the pulsed loads and the power sharing among acgenerators, super capacitor and half-charged battery (Case 2).

Fig. 11. DC bus and battery voltage during pulsed load (Case 2).

Fig. 12. Voltage amplitude of ac buses during pulsed loads (Case 1).

Fig. 13. System ac side frequency during pulsed loads (Case 2).

Fig. 14. Loading of main generators during pulse-loads (Case 2).

Fig. 15. Power-delta curve for main generator during Pulsed Load (Case 1).

as the AVRs settings of the generators also affect the systembehavior during pulsed load conditions, and after its departure.One of the key factors of transient stability is the rotor angle ofeach generator during and after an event such as a pulsed load.During the pulsed load, the power and angle jump to a new oper-ating range for both generators. After the end of the pulse-load,the power and rotor angle return to their normal values withsome oscillations around the steady state values. The magni-tude and duration of these oscillations depend on the systeminertia, the generator voltages, and the power controller param-eters. Figs. 15 and 16 show the transmitted power versus rotorangle of the main generator during the pulsed load for Cases 1and 2, respectively. The amplitude of the transferred power inCase 2 and the rotor angle changes during the pulses are morethan Case 1 due to the absence of battery energy for last twopulses, but the system remains stable and the rotor angles returnto the steady state point.

C. System Performance Under Auxiliary Generator Outage

In order to clarify the system performance under real-timemanagement algorithm, another simulation was performed ina case that one of the auxiliary generators, ATG1, experiencesoutage before the pulsed load and the battery charging statusplays an important role here. The system generators equippedby frequency relays, which are all set to 58 Hz, and wheneverthe system frequency drops, they will disconnect the system im-mediately after 0.1 s. Similar to the first case we assumed thatthe battery is charged totally and at ATG1 is discon-nected and at pulsed load occurs. As shown in Fig. 17,the system can recover after pulsed load and maintain the fre-quency more than 58 Hz. But for half charged battery, after


Fig. 16. Power-delta curve for main generator during Pulsed Load (Case 2).

Fig. 17. Active power of the pulsed loads and the power sharing among acgenerators, super capacitor and full-charged battery (Outage of ATG1 at).

Fig. 18. System ac side frequency and voltages during pulsed loads with half-charged battery (Outage of ATG1 at ).

second pulse the battery’s remained energy becomes zero andthe system frequency drops dramatically to 58 Hz as shown inFig. 18. Therefore, the protection system disconnected the acgenerators at . Fig. 19 shows the active powers ofthe pulsed load and the power sharing among the ac generators,super capacitor and half-charged battery during this simulationcase.

Fig. 19. Active power of the pulsed loads and the power sharing among acgenerators, super capacitor and half-charged battery (Outage of ATG1 at).

VI. CONCLUSION

In this paper we developed a real-time energy managementalgorithm in order to mitigate pulsed loads effects on systemperformance in hybrid microgrids. The main objective of thealgorithm is to manage the energy storage devices in real-timein order to maintain system stability and performance in theshort term operation and minimize the energy cost in the longterm operation particularly for peak shaving purposes. Thealgorithm involves nonlinear mathematical models for PV andload data as well as smart techniques including fuzzy logicand adaptive control concepts. An investigation on the systemperformance under pulsed loads shows that when the battery’sstate of charge is managed by the developed algorithm, thesystem has better stability margin and the battery is sustainedto share all its stored energy to pulsed loads. The comparisonwith the system performance without fully charged batteryresults in more system parameters changes, which may causestability problems or protection reactions due to high drop infrequency/voltage and hence the system may be thrown offline.

APPENDIX

The set of nonlinear functions selected for transforming inputvectors contains , , , and . These func-tions are useful to be used for transforming input vectors toachieve linearity between them and the output vector. The con-stants , , , are determined so as to maximize the cor-relation index given by (A1), which represents the linear rela-tionship between any two vectors

(A1)

where:

the two vectors representingfor

simultaneously;

the average values of .


The trial and error method was used for such determination;the constants were set to a small value (0.01) and varied in awide range until the maximum correlation between the resultingtransformed input vector and the output is obtained. Values ofthe constants , , , are 0.4, 0.8, 0.7 and 100, respec-tively. Hence, the set of nonlinear functions selected to modeleach input vector were There-fore, the nonlinear model for each variable for modeling thepower is as given by (A2):

(A2)

Determination of Constants : The con-stants have just two possible values 0 or 1,so they control the presence of the transformed vectors in themodel. These constants are determined through correlation anal-ysis through two subsequent steps,Firstly, the correlation index between , is deter-

mined by (A1), where is the real output values (experimentaldata which are required to be evaluated by the model). willequal 1 if the absolute value of the correlation index is greaterthan prespecified index ( has been taken as 0.5 for the non-linear model) and will equal 0 if the correlation index is lessthan .Through this step, only the functions having substantial effect

on the output is retained for further processing.The correlation index between and is determined. If

the absolute value of the cross correlation is smaller than a pre-specified value ( has been taken as 0.8 for the nonlinearmodel) both terms are retained otherwise only the term with thegreater correlation with respect to is retained and the other iseliminated to avoid information overlap.Up to this step, the nonlinear basic matrix is updated to

a number of basic columns. The reduced matrix is given thedesignation and any linear combination of any of the vectorsincluded in the reduced matrix forms a basis for the nonlinearmodel.

Determination of Constants : If isone of the combinations of the reduced basic matrix havingdimension so the nonlinear multivariable model has the fol-lowing form:

(A3)

The unknown vector of constant coefficients is given by

(A4)

(A5)

where is the vector of real outputs andis the following weight matrix:

(A6)

and is a weighting factor. has been varied from 0.8 to 1 instep of 0.01 for the nonlinear model [20].The final model has the lowest value of .

REFERENCES

[1] A. G. Tsikalakis and N. D. Hatziargyriou, “Centralized control for op-timizing microgrids operation,” IEEE Trans. Energy Convers., vol. 23,no. 1, pp. 241–248, Mar. 2008.

[2] N. Amjady, F. Keynia, and H. Zareipour, “Short-term load forecast ofmicrogrids by a new bilevel prediction strategy,” IEEE Trans. SmartGrid, vol. 1, no. 3, pp. 286–294, Dec. 2010.

[3] M. Zhixin, A. Domijan, and F. Lingling, “Investigation of microgridswith both inverter interfaced and direct AC-connected distributedenergy resources,” IEEE Trans. Power Del., vol. 26, no. 3, pp.1634–1642, Jul. 2011.

[4] P. Tenti, H. K. M. Paredes, and P. Mattavelli, “Conservative powertheory, a framework to approach control and accountability issues insmart microgrids,” IEEE Trans. Power Electron., vol. 26, no. 3, pp.664–673, Mar. 2011.

[5] V. Salehi, B. Mirafzal, and O. Mohammed, “Pulse-load effects on shippower system stability,” in Proc. 36th Annu. Conf. IEEE Ind. Electron.Soc. (IECON), Nov. 7, 2010, pp. 3353–3358.

[6] J. M. Crider and S. D. Sudhoff, “Reducing impact of pulsed powerloads on microgrid power systems,” IEEE Trans. Smart Grid, vol. 1,no. 3, pp. 270–277, 2010.

[7] S. Kulkarni and S. Santoso, “Impact of pulse loads on electric shippower system: With and without flywheel energy storage systems,”in Proc. IEEE Electric Ship Technol. Symp., Apr. 20–22, 2009, pp.568–573.

[8] J. J. A. Burgt, P. Gelder, and E. Dijk, “Pulsed power requirements forfuture naval ships,” in Proc. 12th IEEE Int. Pulsed Power Conf., 1999,vol. 2, pp. 1357–1360.

[9] E. L. Zivi, “Integrated shipboard power and automation control chal-lenge problem,” in Proc. IEEE Power Eng. Soc. Summer Meet., Jul.2002, vol. 1, pp. 325–330.

[10] Z. Haihua, T. Bhattacharya, D. Tran, T. S. Terence, and A. M. Kham-badkone, “Composite energy storage system involving battery and ul-tracapacitor with dynamic energy management in microgrid applica-tions,” IEEE Trans. Power Electron., vol. 26, no. 3, pp. 923–930, Mar.2011.

[11] I. Aharon and A. Kuperman, “Topological overview of powertrains forbattery-powered vehicles with range extenders,” IEEE Trans. PowerElectron., vol. 26, no. 3, pp. 868–876, Mar. 2011.

[12] A. Khaligh and L. Zhihao, “Battery, ultracapacitor, fuel cell, and hy-brid energy storage systems for electric, hybrid electric, fuel cell, andplug-in hybrid electric vehicles: State of the art,” IEEE Trans. Veh.Technol., vol. 59, no. 6, pp. 2806–2814, Jul. 2010.

[13] A. H. Eghbali, B. Asaei, and P. Nader, “Fuel efficient control strategy,based on battery-ultracapacitor energy storage system, in parallel hy-brid electric vehicles,” in Proc. IEEE Veh. Power Propulsion Conf.(VPPC), Sep. 1–3, 2010, pp. 1–5.

[14] L. P. Jarvis, T. B. Atwater, and P. J. Cygan, “Hybrid power sources forland warrior scenario,” IEEE Aerosp. Electron. Syst. Mag., vol. 15, no.9, pp. 37–41, Sep. 2000.

[15] M. Zandi, A. Payman, J. P. Martin, S. Pierfederici, B. Davat, and F.Meibody-Tabar, “Energy management of a fuel cell supercapacitor bat-tery power source for electric vehicular applications,” IEEE Trans. Veh.Technol., vol. 60, no. 2, pp. 433–443, Feb. 2011.

[16] J. Byun, I. Hong, B. Kang, and S. Park, “A smart energy distributionand management system for renewable energy distribution and con-text-aware services based on user patterns and load forecasting,” IEEETrans. Consum. Electron., vol. 57, no. 2, pp. 436–444, May 2011.

[17] H. Kanchev, D. Lu, F. Colas, V. Lazarov, and B. Francois, “Energymanagement and operational planning of a microgrid with a PV-basedactive generator for smart grid applications,” IEEE Trans. Ind. Elec-tron., vol. 58, no. 10, pp. 4583–4592, Oct. 2011.

[18] P. Palensky and D. Dietrich, “Demand side management: Demand re-sponse, intelligent energy systems, and smart loads,” IEEE Trans. Ind.Inf., vol. 7, no. 3, pp. 381–388, Aug. 2011.

[19] D. T. Larose, Data Mining Methods and Models. Hoboken, NJ:Wiley, 2006.

[20] M. A. Wahab, M. M. Hamada, and A. Mohamed, “Artificial neural net-work and non-linear models for prediction of transformer oil residualoperating time,” Elec. Power Syst. Res., vol. 81, pp. 219–227, Jan.2011.


[21] A.Mohamed,M. Elshaer, andO.Mohammed, “Smart operation for ACdistribution infrastructure involving hybrid renewable energy sources,”in Proc. 18th IFAC World Congr., Milan, Italy, Sep. 2, 2011.

[22] T. Christen and M. W. Carlen, “Theory of Ragone plots,” J. PowerSources, vol. 91, no. 2, pp. 210–216, Dec. 2000.

[23] G. Sikha and B. N. Popov, “Performance optimization of a battery-capacitor hybrid system,” J. Power Sources, vol. 134, pp. 130–138,2000.

[24] A. Kuperman and I. Aharon, “Battery-ultracapacitor hybrids for pulsedcurrent loads: A review,” Renew. Sustainable Energy Rev., vol. 15, no.2, pp. 981–992, Feb. 2011.

[25] R. A. Dougal, S. Liu, and R. E.White, “Power and life extension of bat-tery-ultracapacitor hybrids,” IEEE Trans. Compon. Packag. Technol.,vol. 25, no. 1, pp. 120–131, Mar. 2002.

[26] L. Gao, R. A. Dougal, and L. Shengyi, “Power enhancement of anactively controlled battery/ultracapacitor hybrid,” IEEE Trans. PowerElectron., vol. 20, no. 1, pp. 236–243, Jan. 2005.

[27] A.Mohamed,M. Elshaer, and O.Mohammed, “Dynamic Energy Com-mitment Scheme for Hybrid Renewable Energy Systems in Smart GridApplications,” J. Energy Syst., to be published.

[28] K. M. Passino and S. Yurkovich, Fuzzy Control. Petaluma, CA: Ad-dison-Wesley Longman, 1998.

[29] S. Deilami, A. S. Masoum, P. S. Moses, and M. A. S. Masoum, “Real-time coordination of plug-in electric vehicle charging in smart gridsto minimize power losses and improve voltage profile,” IEEE Trans.Smart Grid, vol. 2, no. 3, pp. 456–467, Sep. 2011.

Ahmed Mohamed (S’09) was born in Minia, Egypt.He received the B.Sc. and M.Sc. degrees from theCollege of Engineering, Minia University, Minia, in2006 and 2009, respectively. He is currently workingtoward the Ph.D. degree in the Energy Systems Re-search Laboratory at the Electrical and Computer En-gineering Department, Florida International Univer-sity, Miami.From 2006 to 2009, he was a Research/Teaching

Assistant in the College of Engineering, Minia Uni-versity. He is the recipient of the Doctoral Evidence

Acquisition Fellowship from Florida International University. His current re-search interests are smart grids, renewable energy systems, hybrid ac/dc powersystems, and sensorless control of permanent magnet machines.

Vahid Salehi (GS’09) was born in Tabriz, Iran in1980. He received the B.S. degree in electrical en-gineering from University of Tabriz, in 2003 and theM.Sc. degree in power system engineering from Uni-versity of Tehran, Iran, in 2006.During 2003 to 2008 he worked in the Energy

Research Institute (MATN), Tehran, Iran. He iscurrently working toward the Ph.D. degree at FloridaInternational University, Miami. His research in-terests include power system studies, smart grid,renewable energy integration and energy conversion

in power system, distributed energy resource integration, dynamic modelingof power system, power system stability, protection, wide area monitoring,control and protection of power system. His dissertation related to developmentand verification of control and protection strategies in wide area power systemsfor smart grid applications.

Osama A. Mohammed (S’79–SM’84–F’94) re-ceived the M.S. and Ph.D. degrees in electricalengineering from Virginia Polytechnic Institute andState University, Blacksburg.He is a Professor of Electrical and Computer Engi-

neering and the Director of the Energy Systems Re-search Laboratory at Florida International University,Miami. He has published numerous journal articlesover the past 30 years in areas relating to power sys-tems, electric machines, and drives, computationalelectromagnetics and in design optimization of elec-

tromagnetic devices, artificial intelligence applications to energy systems. Heauthored and coauthored more than 300 technical papers in the archival liter-ature. He has conducted research work for government and research laborato-ries in shipboard power conversion systems and integrated motor drives. He isalso interested in the application communication and wide area networks for thedistributed control of smart power grids. He has been successful in obtaining anumber of research contracts and grants from industries and federal governmentagencies for projects related to these areas. He also published several book chap-ters including; Chapter 8 on direct current machinery in the Standard Handbookfor Electrical Engineers, 15th Edition (McGraw-Hill, 2007) and a book chapterentitled “Optimal Design of Magnetostatic Devices: the Genetic Algorithm Ap-proach and System Optimization Strategies” in Electromagnetic Optimizationby Genetic Algorithms (Wiley, 1999).Prof. Mohammed is the recipient of the IEEE PES 2010 Cyril Veinott Electro-

mechanical Energy Conversion Award. He is also a Fellow of the AppliedComputational Electromagnetic Society. He is Editor of IEEE TRANSACTIONSON ENERGY CONVERSION, IEEE TRANSACTIONS ON MAGNETICS, POWERENGINEERING LETTERS, and also an Editor of COMPEL. He is the pastPresident of the Applied Computational Electromagnetic Society (ACES). Hereceived many awards for excellence in research, teaching, and service to theprofession and has delivered numerous invited lectures at scientific organiza-tions around the world. He has been the general chair of several internationalconferences including; ACES 2006, IEEE-CEFC 2006, IEEE-IEMDC 2009,IEEE-ISAP 1996, and COMPUMAG-1993. He has also chaired technicalprograms for other major international conferences including; IEEE-CEFC2010, IEEE-CEFC-2000 and the 2004 IEEE Nanoscale Devices and SystemIntegration. He also organized and taught many short courses on powersystems, Electromagnetics and intelligent systems in the United States andabroad. He has served ACES in various capacities for many years. He alsoserves IEEE on various boards, committees, and working groups at the nationaland international levels.

Date post:	04-Dec-2016
Category:	Documents
Upload:	osama
View:	217 times
Download:	4 times

Real-Time Energy Management Algorithm for Mitigation of Pulse Loads in Hybrid Microgrids

Documents