Energy Management for a Grid-tied Photovoltaic-Wind-Storage System—Part II: Operation Strategy

Ala A. Hussein Department of Electrical Engineering

United Arab Emirates University Al Ain, United Arab Emirates

ala.hussein@uaeu.ac.ae

Issa Batarseh Department of Electrical Engineering and Computer Science

University of Central Florida Orlando, USA

batarseh@eecs.ucf.edu

Abstract—Renewable energy has unique characteristics such as it is sustainable, clean and free. However, renewable generation systems have two major limitations: they are strongly dependent on the weather conditions; and they have unsynchronized generation peaks with the demand peaks, in general. In a series of two papers, an energy management strategy for a distributed photovoltaic (PV)-wind-storage system is proposed. This second paper proposes a strategy to control the operation of the energy storage to overcome the limitations of renewable sources and forecasting models uncertainty. The proposed operation strategy is advantageous in terms of it allows a highly efficient and profitable operation of the system especially in an electricity spot market. Simulation results that shows the effectiveness of the proposed control strategy are provided.

Index Terms—Artificial neural networks (ANNs), distributed generation (DG), load forecasting model, energy storage system (ESS), photovoltaic (PV), solar radiation forecasting model.

I. INTRODUCTION

Renewable energy sources have two major drawbacks beside their invaluable advantages: they strongly depend on the weather conditions, and they have unsynchronized generation peaks with the demand peaks. To overcome those two limitations, renewable distributed generators (DGs) are usually backed up with energy storage to minimize the fluctuations in their generated power and to synchronize the generation peaks with the demand peaks. However, energy storage devices are expensive. A key element in increasing the profitability of the battery (in case a battery is used for energy storage) is to extend its lifetime by developing a battery management system (BMS) that can manage the battery operation carefully under all circumstances without violating its design limits, [1]. Another key element in increasing the overall system profitability is to optimize the scheduling and operation of the system when it operates in either stand-alone or grid-connected mode. The focus of this paper is on developing an operation strategy for a PV-wind-storage system. Thus, the BMS design and operation will not be discussed since it is beyond the scope of this paper.

In [2], an energy management system (EMS) is proposed. However, in this proposed EMS, it is assumed that the forecasting models and the state-of-charge (SOC) estimation are 100% accurate, which is unrealistic. In [3], some load forecasting models were analyzed and evaluated. However, the battery operation was not discussed. In [4]-[7], although load and wind power forecasting methods were presented, no discussion on the battery model or operation were given.

The operation of the storage can be controlled to allow a highly efficient and profitable power exchange with a spot market system. This can be achieved by:

• minimizing the usage of energy storage; • maximizing the usage of renewable energy sources;�• scheduling power exchange plans in an electricity

spot market system. �The forecasting models presented in the first paper of this

series, [8], show good matching between the actual and predicted output for the DG system. However, it is obvious that there are different levels of uncertainty in those models. In a spot market, if the actual generated power don’t match the predicted generated power, the owner of the renewable DGs is penalized. However, with a proper control of the energy storage system, even if the actual and predicted power don’t closely match each other, the energy storage can supply the difference and thus avoid penalties.

In this paper, an operation strategy is proposed to control the operation of the energy storage in a PV-wind-storage system. Section II presents a description of the DG system. In Section III, scheduling and real-time operation strategies are presented. Simulation results are given in Section IV. Finally, conclusions are drawn in Section V.

II. SYSTEM DESCRIPTION

The distributed system is shown in Fig. 1, where PL is the load power, PPV is the solar PV power, PW is the wind power, PS is the energy storage power, and PE is the exchanged power with the utility grid. The load (which is assumed to be a house or any other electricity consumer), renewable DG

This work was partially supported by the U.S. Department of Energy under the SEGIS project and Florida Energy Systems Consortium.

978-1-4799-1303-9/13/$31.00 ©2013 IEEE

sources (solar PV and wind) and the energy storage system (ESS) are connected to the main utility grid through a common AC bus as shown. Basically, the EMS must minimize the usage of storage and conventional power plants (utility) and maximize the usage of renewable sources. Further, the EMS must optimize the scheduling of the ESS operation by making smart decisions that allow maximizing the economic return, which can be achieved by utilizing the variation in the spot electricity prices, maximizing the battery lifetime, and transacting power with the utility.

�������

�� �����

���

����

��������

��

���

��

��

Fig. 1. A general distributed PV-wind-storage system. Arrows’ directions

indicate positive values of variables.

The assumptions made in the proposed EMS are as follows:

• The reactive power is ignored. • The transmission lines are assumed to have an

infinite capacity. • All quantities in Fig. 2 have positive values (except

for PS and PE, which can be positive or negative depending on whether they act as a source or load).

The power balance equation is:

(t) (t) (t) (t) (t) (1)E PV W S LP P P P P= + + −

The daily return equation that must be maximized is given as follows:

For PG > PL:

( ) ( ) ( ) ( )

( ) ( ) ( ), (2)

chH disch L

G L

G L

R SP . C SP . CMax. P P

SP . P P . t

= −� �� �>� �

+ − Δ� �� �For PG < PL:

( ) ( ) ( ) ( ){ } , (3)ch G LH disch LMax. R SP . C SP . C P P= − < (5.3)

where PG is the renewable generated power and it is equal to PPV + PW, �t is the time increment which is assumed to be one hour, and SPH and SPL are the peak and off-peak spot market electricity prices, respectively. The power exchange constraints that must be met under any circumstances are [1]:

(5.4)

( ) ( )min maxSOC SOC SOC (4)≤ ≤

(5)p minR R>

where SOCmin and SOCmax are the lower and upper state-of-charge (SOC) cutoffs which are determined by the BMS, and

Rp is the predicted return, which is determined by a dynamic economic model, according to [9]. Fig. 2 shows a functional block diagram of the EMS.

Fig. 2. A functional block diagram of the EMS.

III. OPERATION STRATEGY

To increase the system’s profitability, the energy storage operation must be optimized by charging or discharging the storage device to an optimum state-of-charge that will result in a maximum return. In fact, there are two scenarios of operation; ESS only and DG with ESS, [9]. Those two scenarios are discussed in this section.

A. ESS Only

In this case, the operation of the ESS is based on dynamic economic-based decisions. Economic-based decisions can be performed for operating or not operating the ESS once the expected return is determined from the given input data. Fig. 3 shows the scheduling operation of the system. There are three possible states for the system: charge, discharge and idle. The scheduling operation strategy shown in Fig. 3 is summarized as follows [9]:

• The operation scheduling is initiated on daily basis for day 1 (n = 1) at a specific hour (t = tsch).

• The initial SOC and state-of-health (SOH) for the scheduled day are imported from the BMS.

• The hourly spot prices (SP) for the scheduled day are received from the utility provider.

• If SP is between SPmin and SPmax, the system turns to “idle” state.

• If SP is below SPmin, then, if the system is not fully charged, the system will go to “charge” state; otherwise, if the system is fully charged, it will go to “idle” state.

• If SP is greater than SPmax, and the system SOC is above 20%, then, if the predicted return is above Rmin (Rmin is a function of electricity prices and SOH), the system will discharge, and otherwise, if the system SOC is at 20%,

the system will go to “idle” state (assuming an 80% depth-of-discharge or DOD).

The timer (t) is incremented until it reaches 23 (last hour in the scheduled day).

Fig. 3. Flowchart of the day-ahead operation scheduling of the ESS.

The lower and upper cutoffs (SPmin and SPmax) of the spot market electricity prices can be calculated on daily basis as given in (6) and (7).

( )min L7 1

( ) Max ( ) (6)i n ,...,n

SP n SP i= − −

=

max diff min min( ) ( ) ( ) (7)SP n p SP n= +

1diff min( ) ( ).(1 cycle) (8)Np L −

=

where n refers to the day, LSP is the average off-peak price

between 12:00 AM and 6:00 AM (SPL � SPmin) and (pdiff)min is the minimum differential price at which the system starts paying back (return of investment or ROI=0%), and LN is the normalized cycle-life per dollar of capital cost, [9]. During operation in real-time, the goal is to match the scheduled plan for power exchange, i.e., where EP is the power delivered to

the grid, and SP̂ is the estimated (forecasted) power supplied

by the ESS. To find an expression for Rmin, the charging cost as well as the battery total capacity or SOH must be considered. Since the discharging time is constant, which is four hours based on the previous analysis, the return in one hour must be greater than a minimum threshold. To meet these constraints, the average of the total charging cost and the normalized cycle-life, LN, including the SOH factor over four hours are considered. Accordingly, the minimum hourly return, Rmin, is given as [9]:

4

min1

1 1( ) ( ) 2 (9)

4 L Lt N

R SP t E t DL=

= + × ×� �

��

Once Rmin is determined, the expected return can be calculated and the exchange plan can be scheduled accordingly.

B. DG with ESS

Integrating renewable DGs with energy storage is advantageous since energy storage devices can absorb any excess generated energy from the DGs instead of wasting this energy, which means that the ESS can be partially or fully charged at zero cost. On the other hand, storage devices can reduce the swing in the energy generated by renewable DGs, which is extremely beneficial especially when exchanging power with a market system. Ideally, if a 100% accurate DG power forecast is provided, the DG can work without energy storage. However, it is impossible to have a very accurate forecasting model that can predict the weather conditions without uncertainties. So, the advantages of integrating the DGs with the ESS are basically to avoid penalties in a market system due to forecasting errors and to partially/fully charge the ESS at no cost. In this case, it is desired to sustain an SOC within a certain limits to insure that the energy storage can sink/supply energy all the time. The scheduling and real-time operation equations are given in (10) and (11), respectively.

Scheduling operation: ( ) ( ) ( ) (10)E,sch G Lˆ ˆP t P t P t= −

Real-time operation: ( ) ( ) ( ) + (11)E,act G Lˆ ˆP t P t P t S= − Δ

where EP is the power exchanged with the utility, GP̂ is the

forecasted DG power, LP̂ is the predicted local load power (in

case the system supplies a local load), and ΔS is the power forecasting error supplied by the storage device. The ESS must be designed in such a way to absorb any excess power generated by the renewable DGs and deliver required power when necessary without sacrificing the battery life or violating the system design constraints or utility regulations (whenever exist). It is substantial, however, to properly size the ESS according to the DG size and to carefully design the DOD and SOC limits in such a way to achieve an optimum performance.

In the real-time operation, the objective is to enforce the distributed system including the energy storage to match the scheduled plan. Fig. 4 shows 24- and 48-hour scheduling time horizons.

Fig. 4. Scheduling time horizons.

The real-time operation of the ESS is shown in Fig. 5, where Iconst is a positive constant (dis)charge current drawn from the grid or DG, and K is a correction gain, which is defined as follows:

, > 0

0 , 0

1 2 2

2

e e eK =

e

−���

≤��

(5.8)

where e1 and e2 indicate the difference between the real and estimated DG and load power, and are defined as follows:

e1 = PG,real – PL,real (5.9)

e2 = PG,est – PL,est

Fig. 5. Real-time ESS operation schemes vs. SP. The term (K. �t) represents the error in the DG forecast which is compensated by the ESS.

Since the focus in this paper is on the operation itself, no discussion on estimation the battery SOC and SOH is given. However, more details on how to estimate the SOC and SOH can be found in [9], [10].

IV. SIMULATION RESULTS

To examine the proposed operation strategy via simulation, the following parameters, which can fit a typical residential application, were selected.

Table I

System sizing for a simulation example. Component Size

ESS Capacity 10 kWh ESS Power Converter 2.5 kW

PV system 2.4 kW Wind system 1.5 kW

In the simulation performed in this section, the following assumptions were considered:

• The scheduling is performed at the scheduling time one day before the operation day (24-hour time horizon).

• The spot prices, weather and load forecasts as well as the estimated initial battery SOC and SOH (charge/discharge resistance and capacity) for the operation day are all given.

• The maximum possible return for exchanging power with the grid is given and based on it, the operation schedule is determined.

The hourly spot electricity prices in Fig. 6 are obtained from [1]. The load and DG (PV and wind) power values (Figs. 7 and 8) were derived from the data obtained to derive the forecasting models. The forecasting models in [8] were used to estimate the load and DG power for four different days. The SOC was calculated using coulomb-counting (any other technique can be used). The real-time operation of the energy storage in terms of its SOC for the four days is shown in Fig. 9. The normalized hourly return for the four days is

shown in Fig. 10. Purposefully all the variables were normalized to generalize the results for any system.

Fig. 6. Normalized spot market electricity prices for four random days.

Fig. 7. Actual and predicted load power (PL).

Fig. 8. Actual and predicted distributed generated power (PG).

Fig. 9. The SOC as a function of time.

Fig. 10. Hourly return for different cases. (normalized return = actual return/maximum return). Negative return indicates charging from the grid.

The hourly return for the proposed system was calculated using (12):

( ) ( ). ( ) (12)h HR i SP i E i=

where Rh is the hourly return, SPH is the peak spot electricity price, and E is the energy exchanged with the grid. Obviously, the return in Fig. 10 is much higher than Rmin (for DG with ESS case) and is slightly above Rmin for a ESS system only.

By comparing the results of the hourly return for different distributed system possibilities (DG only, ESS only, and DG with ESS), it is obvious that the return can be boosted by integrating renewable sources with ESS. With a proper ESS operation and sizing, the system’s economic feasibility can be further increased.

V. CONCLUSION

In this paper, it was shown that regardless how accurate the load and power forecasting models are, the EMS can still be profitable by optimally scheduling power exchange plans between the DG/ESS and utility grid. However, selecting the optimal size of the energy storage is a critical part in maximizing the financial return. That is, the ESS must be designed in such a way to absorb any excess power generated by the renewable DGs and deliver required power when necessary without sacrificing the battery life or violating the system design constraints or utility regulations (if exist).

In general, it is advantageous to oversize the storage device with respect to the DG size in order to maximize the profit especially when the uncertainties in the weather conditions are high. It was shown that the SOC must be kept within certain limits to allow the storage to supply or sink any unexpected deficit or surplus in the DG power due to forecasting uncertainties. The SOC cutoffs and the DOD must be all designed in such a way to ensure an exact matching between the forecasted and actual exchange plan.

REFERENCES [1] A. Hussein “Design and Operation of Stationary Distributed Battery

Micro-storage Systems”, Ph.D. Dissertation, University of Central Florida, Orlando, July 2011.

[2] M. Korpaas, A. T. Holena, R. Hildrum “Operation and sizing of energy storage for wind power plants in a market system”, Electrical Power and Energy Systems 25 (2003) 599–606.

[3] I. Moghram, S. Rahrnan “Analysis and Evaluation of Five Short-term Load Forecasting Techniques”, IEEE Transactions on Power Systems, vol. 4, no. 4, 1989.

[4] M. Monfared, H. Rastegar, H. M. Kojabadi “A new strategy for wind speed forecasting using artificial intelligent methods”, Renewable Energy 34 (2009) 845–848.

[5] M. Lydiaa, S. Suresh Kumar “ Wind Farm Power Prediction: An Overview”, The Fifth International Symposium on Computational Wind Engineering (CWE2010), 2010.

[6] H.S. Hippert, C.E. Pedreira, R.C. Souza “Neural Networks for Short-Term Load Forecasting: A Review and Evaluation”, IEEE Transactions on Power Systems, vol. 16, no. 1, pp. 44-55, 2001.

[7] H. M. Al-Hamadi, S. A. Soliman “Short-term electric load forecasting based on Kalman filtering algorithm with moving window weather and load model”, Electric Power Systems Research 68 (2004).

��� A.A. Hussein, I. Batarseh “Energy Management for a Grid-tied Photovoltaic-Wind-Storage System—Part I: Forecasting Models”, In Press.��

[9] A. A. Hussein, N. Kutkut, J. Shen, I. Batarseh “Distributed Battery Micro-storage Systems Design and Operation in a Deregulated Electricity Market”, IEEE Transactions on Sustainable Energy, Volume 3, Issue 3, pp. 545-556, July, 2012.

[10] A. A. Hussein, I. Batarseh “A Review of Charging Algorithms for Nickel and Lithium Battery Chargers”, IEEE Transactions on Vehicular Technology, Volume 60, Issue 3, pp. 830-838, March, 2011.

DG Only ESS Only DG with ESS