Applied Energy 158 (2015) 12–25

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Energy management strategy for a renewable-based residential

microgrid with generation and demand forecasting
Julio Pascual a,⇑, Javier Barricarte b, Pablo Sanchis a, Luis Marroyo a
a
Department of Electrical and Electronic Engineering, Public University of Navarre (UPNa), Edificio de los Pinos, Campus Arrosadía s/n, 31006 Pamplona, Spain
b
Ingeteam Power Technology S.A., Avda. Ciudad de la Innovación 13, 31621 Sarriguren, Spain

h i g h l i g h t s

An energy management strategy is proposed for a renewable-based microgrid.

The strategy seeks to minimize power peaks and fluctuations in the grid.

The energy management strategy makes use of demand and generation forecasting.

Analysis by means of simulation has been carried out using one-year measured data.

Experimental validation has been done in a real size microgrid.

a r t i c l e i n f o a b s t r a c t

Article history: This paper proposes an energy management strategy for a residential microgrid comprising photovoltaic
Received 16 March 2015 (PV) panels and a small wind turbine. The microgrid is connected to the main grid, allowing for a con-
Received in revised form 17 July 2015 trolled power exchange through a battery system and its control strategy. As input data, the proposed
Accepted 14 August 2015
control strategy uses the battery state of charge (SOC), the power at each microgrid node as well as
Available online 27 August 2015
the load and renewable generation forecasts. By using forecasted data and correcting any forecasting
errors according to the SOC of the battery, the strategy manages to make a better use of the battery
Keywords:
resulting in a better grid power profile. The simulation of the system using a one-year data period shows
Microgrids
Renewable energy
that the proposed energy management strategy results in a better grid power profile for a given storage
Energy management system when compared with other state-of-the-art strategies. Finally, the proposed strategy was exper-
Forecasting imentally validated in the microgrid built in the Renewable Energy Laboratory at the UPNa.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction configurations and goals, which will determine the energy

management strategy governing each particular microgrid.
Microgrids, defined as power systems which include loads, dis- The energy management strategy of a microgrid is responsible
tributed generation, energy storage, and which are managed as a for setting the operating point of every element in the microgrid,
single unit in order to exchange power with the grid through a sin- such as the power delivered by the battery or the generators, the
gle coupling point [1], are becoming a way of integrating renew- power used by manageable loads and the power exchanged with
able energies, lowering costs and providing better grid quality all the grid. Having more than one degree of freedom implies an infi-
around the world [2,3]. nite number of energy management strategies that could be
Microgrids cover a broad range of scenarios and can be classi- designed to comply with the physical restrictions, yet performing
fied as either grid-connected or stand-alone, or according to use: very differently. It is therefore essential to design the right strategy
military, industrial, residential, etc. Furthermore, according to their to meet the set goals in the most effective manner.
generation sources, they will comprise either renewable genera- In the field of stand-alone microgrids, when the only power
tors (typically, PV panels and wind turbines) or fossil fuel-based sources are non-dispatchable renewable energies, the main goal
generators (typically diesel generating sets or gas microturbines). is to manage the energy management system in order to keep
The diversity of microgrids encompasses a wide range of the microgrid running, limiting the power output when necessary
and sometimes using demand side management (DSM) techniques
⇑ Corresponding author. Tel.: +34 948168934. in order to avoid battery depletion ([4,5]). However, when dis-
E-mail address: [email protected] (J. Pascual). patchable power units are present in the microgrid, the main goal

http://dx.doi.org/10.1016/j.apenergy.2015.08.040
0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 13

is usually to reduce operating costs by optimally scheduling the In [24], a battery is used in order to minimize the power peaks
different dispatchable units in the system, using different opti- absorbed from the grid by a microgrid. In order to do so, the system
mization routines, as seen in [6–9]. forecasts the day-ahead power generation and the load in the
The case of grid-tied microgrids is not very different from the microgrid and, by means of a linear optimization routine, the
latter case, because, from an energetic point of view, the grid acts day-ahead set point of the battery is calculated. As mentioned in
like a dispatchable power unit, although in this case, it can also this paper, this system has some limitations. The main one being
absorb energy, unlike diesel generating sets. Furthermore, electric- that, by programming the day-ahead SOC to reach both 0% and
ity market prices have to be taken into account in grid-connected 100% based on power prediction, a deviation between the actual
microgrids when operating costs are to be minimized as in [10– power output and the prediction may lead to a loss of control when
16]. the battery is either fully charged or discharged. As noted by the
Most papers found in the literature have in common that they author, this issue may be handled by using some security coeffi-
pursue an economic goal. However, economic incentives and cients that would underuse the battery capacity but would guaran-
penalties set by grid operators are a reflection of technical issues tee a lower loss of control rate.
such as line congestion or grid stability. For this reason this paper This paper proposes a new control strategy that addresses
focuses directly on minimizing the peaks and fluctuations of the directly the problems of other state-of-the-art strategies. Firstly,
power exchanged with the main grid, as in [17–24] resulting in the strategy focuses on the lag that is introduced in the power
the following benefits: (1) avoidance of overvoltage events in exchanged with the grid by SMA strategies, which results in an
low-voltage grids when power is injected into the main grid, oversized battery system. This is due to the fact that the battery
[25,26]; (2) better use of the transmission lines thanks to local has to handle the energy imbalance in the microgrid due to power
power balancing [27]; and (3) better grid quality and stability variations until the grid power reaches its new working point,
due to the reduction in power fluctuations [28,29]. needed to restore the energy balance in the system. By making
A common way to reduce power peaks and fluctuations in a use of forecasting data, an SMA strategy is converted into a central
power profile is to apply a low-pass filter [17,18]. Such a strategy moving average (CMA) strategy, eliminating any lag in the
would separate the low and high frequencies of the power profile response and reducing the battery needs or, in other words, getting
by using, for example, a simple moving average (SMA) filter, and a better grid power profile with the same battery system. This
would assign the high frequency component to the battery while approach, however, is subject to the accuracy of the forecasted
the grid would only handle the filter output, i.e., the low frequency data. In order to handle forecasting errors, the proposed strategy
component. However, the low frequency component lags behind makes corrections in the power exchanged with the grid not only
the power profile and a larger battery would be necessary, than taking into account the forecasting error, but also the SOC of the
in the case of using other strategies. Moreover, such a simple strat- battery. By combining the SOC and the forecasting error in a single
egy overlooks important state variables of the system such as the control rule, the strategy manages to minimize the fluctuations in
state of charge (SOC, percentage over useful capacity) of the bat- the grid power, by ignoring any forecasting errors as long as they
tery, making poor use of it, and making it necessary to oversize it can be handled by the battery. By thorough analysis of the system,
in order to attain certain power profile requirements. the strategy parameters can be optimized without the need of
An enhanced strategy is designed in [20] which is based on a complex optimization routines. Furthermore, due to the simplicity
series of sequential steps that seek to minimize the power peaks of the strategy, it can be readily adapted to be used under different
and fluctuations in the power exchanged with the grid. In its calcu- scenarios. This strategy has been experimentally validated in a
lations, this strategy includes control coefficients that are depen- microgrid that emulates a single-family home with PV panels, a
dent on the battery SOC and that determine the share of power small wind turbine and a lead-acid battery. However, the philoso-
to be managed by the grid and the battery. This same control idea phy of this control may be applied to most microgrids.
is carried out in [21] by means of fuzzy logic. In Section 2, the microgrid and the problem statement are
Ref. [22] proposes an enhanced strategy for that seen in [20] described. Section 3 explains the control strategy for the SMA strat-
and [21] which uses a variable set point for the SOC. In order to egy. In Section 4, two enhanced strategies are explained, which
do so, the SOC set point is defined as a function of the moving aver- make use of persistence prediction in one case and a mixed fore-
age of Pnet, in an attempt to compensate the SOC displacement due casting technique using persistence for the load and NWP weather
to seasonal fluctuations in the microgrid energy balance. In short, forecasting models in the other. In Section 5, the experimental val-
persistence forecasting of Pnet is used to preempt the effects of a idation is analyzed and finally, in Section 6, the main conclusions
varying energy balance in the microgrid. Using this technique, are discussed.
the SOC is held around 50% throughout the year, making better
use of the battery. 2. System description and problem statement
In the case of load forecasting, persistence models or other
methods based on historical data may be used. However, even The scenario analyzed herein consists of a grid-tied microgrid
though these methods can be useful for renewable energy forecast- with the usual electric loads for a single-family home including a
ing in the short term (as in [11,13], where artificial neural net- heating, ventilation and air conditioning (HVAC) system. The
works are used to predict renewable output for up to 1 h) or for yearly average consumption of these loads is 1.27 kW. For the
spotting seasonal behavior, they may lead to considerable errors power generation, the microgrid is equipped with a PV array and
for a time horizon of about one day. In this case, numerical weather a wind turbine both rated at 6 kW, which cover 87.0% of the home
prediction (NWP) models are preferred [30,31], like those used in energy consumption. A 35 kW h battery is used for the energy
[23,24]. management of the power exchanged with the grid. which, being
In [23], the authors propose a system to program the day-ahead at a 50% SOC, gives an approximate 12 h capacity to either absorb
output of a hybrid renewable system comprising a wind-turbine power or supply the yearly average consumption of the home. The
and PV panels based on output prediction, using a battery to cor- diagram of this microgrid is represented in Fig. 1.
rect any hourly deviation. Although hourly production is matched In order to simulate and evaluate the performance of the differ-
with that programmed the day before, the strategy does not man- ent strategies, data was recorded at 15 min intervals during one
age power peaks and fluctuations. year from the PV array and wind turbine installed at the
14 J. Pascual et al. / Applied Energy 158 (2015) 12–25

SMALL
PHOTOVOLTAIC WIND
BATTERY GENERATOR TURBINE

+
LOADS
_

UTILITY
GRID

DC BUS AC BUS

Fig. 1. Microgrid block diagram.

Renewable Energy Laboratory at UPNa and from the loads con-

sumption of a nearby home. Note that this consumption data will 1
10
also be used for the experimental validation as it is explained in 24 hour component
12 hour component
Section 5. The difference between consumption and generation, i. 0
e., the net power, Pnet, is the power that would be exchanged with 10
the grid in the absence of either an energy storage system or a con-
-1
trol strategy. This is the baseline scenario, hereinafter referred to as 10
Power (kW)

Strategy 0. Fig. 2 shows the net power profile, which has maximum
peaks of 5.7 kW and 6.4 kW, values that are far higher than the -2
10
yearly average for this profile, 0.165 kW. This power profile also Seasonal fluctuation
shows considerable fluctuations due to both the load profile and
-3
the power generation. 10
Fig. 3 shows the frequency analysis of Pnet. It is possible to
observe a seasonal component primarily due to the HVAC system, -4
10
which has a maximum consumption during summer and winter
and minimum during spring and fall. The highest peak is that -5
related to daily fluctuations, due to the daily pattern of both con- 10 -8 -7 -6 -5 -4 -3
10 10 10 10 10 10
sumption and generation. Furthermore, the spectrum shows peaks
at frequencies which are multiples of the daily one, due to the Frequency (Hz)
shape of the daily pattern.
Fig. 3. Spectrum analysis of the net power profile.

3. Simple moving average strategy

on a simple moving average to Pnet. The low frequency component,
3.1. Without SOC control Pnet,LF, is then assigned to the grid, Pgrid, while the battery, Pbat,
absorbs the high frequency component, as long as the battery is
The most obvious way to smooth out power fluctuations in the able to either absorb or feed the required power (Fig. 4). Otherwise,
power exchanged with the grid is to apply a low pass filter based the grid will assume the unfiltered net power. Given that the high-
est component of the spectrum analysis is the daily one, the SMA
window is set to 24 h to completely remove them. Hereinafter
6 referred to as Strategy 1.
As can be observed in the frequency analysis of the power
4 exchanged with the grid once Strategy 1 has been applied (blue
in Fig. 5), the daily component and its multiples have been greatly
2 reduced while the components of lower frequencies remain almost
equal to those of Pnet (in grey).
Power (kW)

0 Fig. 6 shows the evolution of Pnet, Pgrid and SOC for 5 days in July
2013, applying Strategy 1. Over the first few days, the daily profile
-2

-4
+
-6 Pnet Pnet,LF = Pgrid Pnet,HF = Pbat
24 h SMA
-8
–
Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun
Fig. 4. Block diagram of the simple moving average energy management strategy
Fig. 2. Net power, Pnet in the microgrid from July 2013 to June 2014. (Strategy 1).
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 15

1 6
10

0
10 4

-1
10
2

Power (kW)
Power (kW)

-2
10
0
-3
10
-2
-4
10

-5 -4
10

-6 -6
10 Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun
-8 -7 -6 -5 -4 -3
10 10 10 10 10 10
Frequency (Hz) Fig. 7. Pgrid evolution from July 2013 until June 2014 using Strategy 1.

Fig. 5. Spectrum analysis of the profile of Pnet profile (grey) and Pgrid (blue) using
Strategy 1. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.) 100

90

6 100 80

90 70
4 SOC (%) 60
80

70 50
2
Power (kW)

40
SOC (%)

60

0 50 30

40 20
-2 10
30

20 0
-4 Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun
10
Fig. 8. SOC evolution from July 2013 to June 2014 using Strategy 1.
-6 0
24/07/2013 25/07/2013 26/07/2013 27/07/2013 28/07/2013
Fig. 8 shows the SOC evolution throughout the year, and proves
Fig. 6. Net power (red), power exchanged with the grid (blue) and SOC (green)
during four days of July 2013 using Strategy 1. (For interpretation of the references how the positive and negative power spikes seen in Fig. 7 corre-
to color in this figure legend, the reader is referred to the web version of this spond to moments when the battery is fully discharged or charged
article.) respectively. This figure also shows how this problem is aggravated
due to the fact that the previously mentioned daily energy balance
variations make it difficult for the SOC to return to the 50% value
after it has been charged or discharged.
of Pnet is almost identical, and, as such, the output of the 24 h SMA
of Pnet, Pnet,LF, and in turn, Pgrid, remain almost constant. In short, the
daily energy balance in the microgrid is compensated by absorbing 3.2. With SOC control
an almost constant power from the grid, while Pnet variations, that
is, the power imbalance in the microgrid, are compensated by the In order to maintain the SOC at around 50%, allowing for daily
battery. As can be observed in this figure, during these first days, fluctuations and without causing perturbations in the grid power
the SOC lies within its operating range. profile, a proportional control of the 24-h SMA of the SOC, SOC24h,
However, when the net power profile changes from one day to is used, as shown in Fig. 9, where power is expressed in kW and
the next, in this case when a sunny day is followed by an overcast SOC as a %. In this case, the set point for the power exchanged with
day, then the energy balance in the microgrid changes and has to the grid is obtained as the sum of the low frequency component of
be compensated at first by the battery. During this time, the output Pnet, Pnet,LF, and a component proportional to the deviation of
of the SMA filter, i.e., Pgrid, evolves slowly up to the new operating SOC24h with respect to its reference point, set at 50%.
point and, as a result of this slow dynamic, the battery becomes The equations that define this control of SOC24h are:
completely discharged, hence losing the management capability
over Pgrid, and resulting in a strong power peak absorbed from 1 XN
SOC24h ¼  SOC iN ð1Þ
the grid. N i¼1
Fig. 7 shows the evolution of Pgrid throughout the year analyzed
and, as expected from the frequency analysis of Fig. 5, it is rela- P24 ¼ K 24  ð50  SOC 24h Þ ð2Þ
tively smooth. Nevertheless, it can be observed how the loss of
control seen in Fig. 6 occurs several times during the year, resulting where N is the number of samples in 24 h, i  N is the sample
in undesirable power spikes. number within the time series (a value of zero represents the
16 J. Pascual et al. / Applied Energy 158 (2015) 12–25

Pnet

24 h
SOC control SMA

+
Pnet,LF +
refSOC Pgrid Pbat
+
P24 Battery SOC
K24
- -
SOC24h 24 h SOC
SMA

Fig. 9. Block diagram of the SMA strategy including the control of the average SOC.

current sample and a negative value represents a past sample) path that continuously controls the SOC value as shown in Fig. 11.
and K24 is the proportional gain. Hereinafter referred to as Strategy 2.
Note that in the real system, the SOC is an input parameter pro- Naturally, the output of this new branch of the control, Pp, will
vided by the battery management system (BMS) and the strategy introduce fluctuations in the power exchanged with the grid, fol-
will not need to calculate it. However, in order to test the strategy lowing the evolution of the SOC. In order to minimize such fluctu-
by simulation, this feedback needs to be estimated. For this pur- ations, this paper proposes using a proportional control that uses a
pose, a simple model of the battery has been used, which is variable proportional gain as a function of SOC, as shown in Fig. 12.
depicted in Fig. 9 as the block ‘Battery’. This block represents the In this way, the control has a zero output when the SOC is 50%, and
battery charging and discharging process, which is modeled as an responds with more strength as it approaches 0% or 100%, that is,
integrator multiplied by the charging or discharging efficiency as the probability of loss of control rises.
and constant 100/Cbat, where Cbat is the useful capacity of the bat- The equations that define this part of the control are:
tery rated for a 10 h discharge (C10), which is in this case 35 kW h.
j50  SOCj
In order to carry out a simplified analysis of this control struc- K P ¼ K P max  ð4Þ
50
ture, the SMA filter is approximated by a first-order low-pass filter
with a time constant of 12 h. This approximation is valid when PP ¼ K P  ð50  SOCÞ ð5Þ
working at frequencies below the filter cut-off frequency,
3.68e06 Hz. Also note that the efficiency of the charging and dis- Keeping K24 = 0.025 kW/%, a KPmax = 0.05 kW/% is chosen, a
charging process has not been considered at this point, as its influ- value slightly larger than the minimum needed to keep the SOC
ence on the stability analysis can be neglected. The resulting within its working range at all times throughout the year.
transfer function is finally: Fig. 13 shows the evolution of Pnet, Pgrid and SOC during the
same days seen in Fig. 6, using Strategy 2. As can be seen, the bat-
tery is not fully depleted in this case, although there now are fluc-
    tuations in Pgrid that did not appear when using Strategy 1.
1 100
TF ¼ ½K 24    ð3Þ In Fig. 14, which shows the evolution of Pgrid during the year
12  s þ 1 C bat  s
analyzed, it can be noted that by using Strategy 2 to keep the
Imposing a phase margin of 55°, the resulting K24 is 0.025 kW/%. SOC within its working limits, the new power profile Pgrid has only
As can be seen in Fig. 10, after adding this control, the SOC is slightly smaller power peaks that those seen when using Strategy 1
successfully centered around the 50% value, although there are still and much higher fluctuations.
days in which the SOC reaches 0% and 100%, hence losing control For the purpose of quantifying the quality of the power profile
over Pgrid. exchanged with the grid, Pgrid, the following quality criteria have
With the objective of avoiding these moments in which the con- been defined:
trol over Pgrid is lost, this paper proposes including another control

Positive grid power peak (P+): Maximum value of the power
absorbed from the grid during one year (kW);
100
Negative grid power peak (P): Minimum value of the power
90 absorbed from the grid during one year (kW);

Maximum power derivative (MPD): The maximum value of the
80
ramp-rate, evaluated every 15 min, of the power exchanged
70 with the grid in absolute value during one year expressed in
W/h;
60
SOC (%)

Average power derivative (APD): The average of the absolute

50 value of the ramp-rate, evaluated every 15 min, of the power
40 exchanged with the grid during one year, expressed in W/h;

Power profile variability (PPV): This is defined as the square
30 root of the sum of the squares of the power harmonics evalu-
20 ated during one year, relative to the constant component, i.e.,
the yearly mean power absorbed by the microgrid. Because
10
the energy management strategy does not seek to compensate
0 seasonal variations, which would require a vast amount of stor-
Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun
age capacity, the PPV only evaluates frequencies above
Fig. 10. SOC evolution from July 2013 until June 2014 using Strategy 1 with the 1.65e06 Hz, i.e., the components with variation periods of
control of the 24 h SMA of the SOC. one week or less. Furthermore, because the sample period is
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 17

Pnet

24h SMA
SOC control +
- P24 Pgrid Pbat SOC
K24 + -
Battery

+
PP

Kp
refSOC +
SOC24h - SOC
24h SMA

Fig. 11. Block diagram of the SMA strategy including the SOC control (Strategy 2).

KP KPmax 6

2
Power (kW)

0
SOC
-2
0% 50% 100%

Fig. 12. Proportional factor for the SOC control. -4

-6
6 100 Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun

90 Fig. 14. Pgrid evolution from July 2013 until June 2014 using Strategy 2.
4
80

70 where Pf is the grid power harmonic at frequency f, fi = 1.65  106 -

2
Hz (one week) and ff = 5.55  104 Hz (30 min).
Power (kW)

60
SOC (%)

Evaluating these quality criteria to the power profile exchanged

0 50 with the grid when using Strategies 0, 1 and 2, gives the results
shown in Table 1. By using Strategy 1 the average value of power
40
ramp rates, APD as well as the PPV are greatly reduced as the grid
-2
30 power profile obtained using this strategy is much smoother than
20 that of Pnet. However, there is not such a notable improvement with
-4 regard to power peaks or to the maximum power ramp rate, due to
10 the occasional loss of control. By using Strategy 2, which prevents
-6 0 this loss of control, power peaks are slightly reduced and the max-
24/07/2013 25/07/2013 26/07/2013 27/07/2013 28/07/2013 imum power ramp rate is also smaller than that of Strategy 1.
Fig. 13. Net power (red), power exchanged with the grid (blue) and SOC (green)
However, the daily power fluctuations that now appear in Pgrid
during four days of July 2013 using Strategy 2. (For interpretation of the references throughout the year yield to higher APD and PPV.
to color in this figure legend, the reader is referred to the web version of this
article.)
4. Central moving average (CMA) strategy: power forecasting

15 min, the maximum frequency considered to calculate the Given that the lag introduced in Pgrid due to the SMA of Pnet
PPV is half the sampling frequency, i.e., the Nyquist frequency, means that the energy imbalance in the microgrid needs to be
in this case 5.55e04 Hz, corresponding to 30 min variations. compensated temporarily by the battery, causing the SOC to drift,
This leads to the following equation to calculate the PPV: it is logical to think that, if the central moving average is used
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi instead of the SMA, the lag would disappear, thereby eliminating
Pf f 2ffi the need for a SOC control, and leading to a better grid power pro-
f ¼f i P f
PPV ¼ ð6Þ file. However, in order to compute the 24 h CMA in real time,
Pf ¼0 power data for the following 12 h is needed. This is obviously not
18 J. Pascual et al. / Applied Energy 158 (2015) 12–25

Table 1
Quality criteria evaluated for Strategies 0–2.

Strategy P+ (kW) P (kW) MPD (W/h) APD (W/h) PPV

0 (Net power) 5.75 6.45 18,468 1122 13.28
1 (SMA w/o SOC control) 4.71 2.40 12,839 44.42 2.51
2 (SMA + SOC control) 2.83 3.45 1241 104.46 4.88

available, but can be estimated by means of different forecasting Note that, in order to identify consistent errors in the net power
techniques. forecasting, and not to actuate with unnecessary variability, the
If net power forecasting is perfect, then the power supplied by cumulative 3-h error, FE, is used instead of the instantaneous error,
the grid will be equal to the average value of Pnet but without the as shown in the following equation.
lag that was introduced by the SMA, in such a way that the grid
1 XN  
will compensate the energy imbalance at all times, while the bat- FE ¼ 3  PnetiN  P net;foreiN ð7Þ
tery will only compensate the daily power fluctuations of Pnet. N i¼1

However, the inevitable error in the forecasting of Pnet (defined where N is the number of samples in the last 3 h and Pnet,fore is the
as the difference between the measured and forecasted power) will net power forecasting. The equations that determine the variable
mean that the energy imbalance in the microgrid will not be fully proportional gain, KP1, are:
compensated by the grid and that the battery will also be required,
causing the SOC to drift, in a potentially cumulative way. For this if ½FE > 0 & SOC < 50% K P1 ¼ K P1;max  ð50  SOCÞ=50 ð8Þ
reason, the SOC control will still be needed.
This paper proposes the implementation of a SOC control strat- if ½FE < 0 & SOC > 50% K P1 ¼ K P1;max  ðSOC  50Þ=50 ð9Þ
egy composed of two branches. Firstly, in order to avoid the poten-
tially cumulative effect of the forecasting error, the strategy will if ½FE > 0 & SOC > 50% or ½FE < 0 & SOC < 50% K P1
include a SOC24h control like the one used in Strategy 2. Secondly,
¼0 ð10Þ
an additional branch that continuously controls the SOC is pro-
posed to avoid depleting or fully charging the battery whenever On the other hand, the control action of the SOC control is atten-
there are large errors in the net power forecasting. As seen in the uated through factor KP2, whenever there are small errors in the
analysis of Strategy 2, the control of the SOC will introduce power net power forecast. Whenever this error is zero, the battery will
fluctuations in Pgrid proportional to the SOC itself. Seeking to min- not need to compensate any energy imbalance and therefore,
imize these fluctuations, this second branch of the control is mod- regardless of the SOC, the control action will be overridden by set-
ified to only respond when it is really necessary, by adding new ting KP2 to zero. Conversely, if the magnitude of the net power fore-
control rules that will override the control action in the event of casting error, |FE|, relative to the battery capacity, Cbat, is
small net power forecasting errors or whenever this error is large significant, e.g., 25%, then the value of KP2 will be 1, so the output
but could be managed by the battery yet without compromising of this branch of the control is not modified and the SOC control
the system. To do so, proportional gain KP, is now defined as the can fully actuate. In the intermediate cases, KP2 will take a propor-
product of two gains, KP1 and KP2 which are functions of the SOC tional value to the ratio |FE|/Cbat as seen in Fig. 16.
and the forecasting error. In the end, the output of the second branch of control, PP, is:
On one hand, whenever the forecasting error is positive, i.e.,
whenever there is more consumption than predicted, so that the
PP ¼ K P1  K P2  ð50  SOCÞ ð11Þ
battery tends toward depletion, the control action is overridden This value, added to Pnet,LF and P24, is the set point for the power
as long as the battery SOC is over 50%. Conversely, if the SOC is to be exchanged with the grid, as seen in Fig. 17, which shows the
below 50%, then KP1 will take a positive value that will increase block diagram of the strategy using the CMA which uses the net
as the SOC approaches 0% (blue function in Fig. 15). In the event power forecast. Given that the SOC24h control branch has not chan-
of a negative error, the inverse logic will apply and the red function ged from that used in Strategy 1, the value for K24 in this case
in Fig. 15 will be used. remains the same, while the new value of KP1,max will be selected
in order to maintain the SOC within its working limits.
Finally, Fig. 18 shows the normalized value of the product KP1-
KP2 which will determine the value of PP. In this figure it can be
observed that the lower is the forecasting error value, the lower
KP1 KP1,max will be the perturbations introduced in Pgrid. In this paper, two fore-
casting techniques will be analyzed: one based on the persistence
forecasting of Pnet and the other, referred to as the mixed

KP2

SOC
|FE|/Cbat
0% 50% 100%
0,25
Fig. 15. Proportional factor for the enhanced SOC control when the forecasting
error is positive (blue) and negative (red). (For interpretation of the references to Fig. 16. Gain of the SOC control as a function of the forecasting error magnitude, |
color in this figure legend, the reader is referred to the web version of this article.) FE|, relative to the useful capacity of the battery, Cbat.
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 19

Pnet

Pnet,fore

SOC control
-
24h
CMA

∫3h Pnet,LF

FE +
+ PP Pgrid Pbat SOC
KP1·KP2 + Battery
-
- P24
SOC
K24

refSOC SOC24h 24h

+ - SMA

Fig. 17. Block diagram of the CMA strategy including the enhanced SOC control using the forecasting error.

Measuerd data, used for the 24 h SMA Persistence forecasting

6 assuming a daily pattern
(KP1·KP2)/KP1,max
1 4

0.8
Power (kW)

0.6
2

0.4
0
0.2
100
0 80
0.5 60
-2
0.25
40 SOC
0
-0.25 20
-0.5 0 -4 Data used for the 24 h CMA
FE/Cbat
t=-24h t=-12h t=0 t=+12h
Fig. 18. Normalized product of factors KP1 and KP2.
Fig. 19. Equivalence of the SMA and the CMA using persistence forecasting
assuming a daily pattern. Measured Pnet data (solid blue line) and predicted Pnet data
(dashed blue line). (For interpretation of the references to color in this figure legend,
forecasting technique, is based on the combination of persistence the reader is referred to the web version of this article.)
forecasting of the load and prediction of renewable generation
using weather forecasting by means of an NWP model.
6 100
4.1. Persistence forecasting
90
4
Persistence forecasting is one way to forecast the net power 80
profile (Strategy 3), assuming that Pnet has a daily pattern and 70
hence estimating that the next 12 h of Pnet will be equal to those 2
Power (kW)

60
SOC (%)

between hours (t  24) h and (t  12) h. Note that, by making such

an assumption, the result of computing the 24 h CMA using persis- 0 50
tence is exactly the same as computing the 24 h SMA of Pnet, as
40
seen in Fig. 19.
-2
As such, it can be said that the effect of the lag introduced by the 30
SMA on the SOC is equivalent to the effect of the forecasting error
20
when using persistence. For this reason, and given that the branch -4
of the SOC control that returns PP is equal to that of Strategy 2, 10
except that it is now overridden when not required, the same value -6 0
for its proportional gain is maintained, i.e., KP1,max = KP,max = 0.05 24/07/2013 25/07/2013 26/07/2013 27/07/2013 28/07/2013
kW/%.
Fig. 20. Net power (red), power exchanged with the grid (blue) and SOC (green)
Fig. 20, shows the evolution of Pnet, Pgrid and SOC during the during four days in July 2013 using Strategy 3. (For interpretation of the references
same days seen in Figs. 6 and 13, using Strategy 3 in this case. It to color in this figure legend, the reader is referred to the web version of this
can be observed how the fluctuations of Pgrid are considerably article.)
20 J. Pascual et al. / Applied Energy 158 (2015) 12–25

reduced, so that, during the first few days, the fluctuations that is now less variable, although the maximum power ramp is slightly
appeared using Strategy 2 are now strongly damped. This is due higher. The power peaks, however, as observed in the previous fig-
to the fact that the new control has been redesigned to be attenu- ure, are only slightly reduced.
ated or even overridden when the forecasting error can be handled
by the battery, resulting in a smoother Pgrid profile than using Strat- 4.2. Mixed forecasting
egy 2.
Fig. 21 shows how the SOC still lies within its working limits Although the new control offers slightly lower quality criteria
even though the control action has been weakened. values, persistence forecasting is not recommended for predicting
Finally, Fig. 22 shows the evolution of Pgrid during the year ana- the net power for a 12 h horizon. In order to make a better predic-
lyzed. Compared to the power profile obtained using Strategy 2, tion and therefore to improve the power profile of Pgrid, a mixed
the new power profile is much smoother, although the power forecasting technique is to be used, based on persistence forecast-
peaks remain almost the same. ing for the prediction of the demand and weather forecasting using
Table 2 shows the quality criteria evaluated for Strategies 0–3. an NWP model for the wind turbine and PV panel power output
When comparing Strategies 2 and 3, APD and PPV can be seen to forecasting. This is the proposed strategy, Strategy 4.
have been reduced, which means that the new Pgrid power profile For the load forecasting, given that, in this scenario, the daily
pattern is very distinct and steady, persistence forecasting has
been used. Because there are no significant differences between
weekdays and weekends, no grouping of days has been carried out.
100 For the generation forecasting, weather forecasting data is com-
90 bined with the model of the wind turbine and the PV panels. The
weather forecasting data is downloaded from Meteogalicia’s
80
THREDDS server [32], and consists of a set of weather data for
70 the Iberian Peninsula obtained with the Weather Research and
Forecasting (WRF) model. Through this web site, this free service
60
SOC (%)

offers weather forecasting data with a 1 h resolution and a

50 12  12 km grid resolution for the area analyzed. Every 12 h the
weather report is updated with a time horizon of 3 h. This server
40
offers many variables, from which the proposed strategy uses hor-
30 izontal irradiance, G(0), ambient temperature, Ta, wind speed at
20 10 m, v10, and atmospheric pressure for the next 12 h. In order to
combine this weather data with the model of the panels and the
10 wind turbine, it is necessary to carry out some transformations
0 such as calculating the irradiance in the plane of the PV array or
Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun estimating the cell temperature from the irradiance and ambient
temperature. The equations and methods used are as follow.
Fig. 21. SOC evolution from July 2013 to June 2014 using Strategy 3.
In order to estimate the PV array’s output, the following equa-
tion can be used:
6 Gðb; aÞ
PFV ¼ P ½1 þ c  ðT C  25Þ ð12Þ
1000
4 where PFV is the estimated output power of the PV array, P*, the out-
put power under standard test conditions (STC), G(b, a), the irradi-
2 ance in the plane of the panels, c, the power temperature coefficient
Power (kW)

and TC, the cell temperature in °C. As the weather report gives the
horizontal irradiance, G(0), and air temperature, Ta, it is necessary
0 to transform G(0) into G(b, a), using the method seen in [33] and
the air temperature, Ta, into Tc using the following equation:
-2
T C ¼ T a þ Gðb; aÞ  ðNOCT  20Þ=800 ð13Þ
where NOCT is the Nominal Operating Cell Temperature, which is
-4
given by the panel manufacturer. Similarly, P* and c are obtained
from the panel datasheet.
-6 In the case of the wind turbine, its power output may be esti-
Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun
mated using the following equation [34].

PEO ¼ 0:5  A  q  C p  v 320

Fig. 22. Pgrid evolution from July 2013 to June 2014 using Strategy 3.
ð14Þ
where PEO is the estimated output power of the wind turbine, A, the
Table 2 area swept by the blades, q, the air density, Cp, the power coefficient
Quality criteria evaluated for Strategies 0–3. of the wind turbine and v20, the wind speed at hub’s height.
Strategy P+ P MPD (W/ APD (W/ PPV Because the hub is at a height of 20 m, the 10 m wind speed v10,
(kW) (kW) h) h) needs to be converted, to v20. To do so, the logarithmic wind profile
0 (Net power) 5.75 6.45 18,468 1122 13.28 equation [34] will be used resulting in the following transforma-
1 (SMA w/o control) 4.71 2.40 12,839 44.42 2.51 tion equation:
2 (SMA + SOC control) 2.83 3.45 1241 104.46 4.88
lnð20=z0 Þ
3 (SMA + SOC/FE 2.19 3.43 1554 71.39 4.33
v 20 ¼ v 10  ð15Þ
control) lnð10=z0 Þ
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 21

where vi is the wind velocity at height i m and z0 is the roughness 1

length, which is typically 0.7 m within cities.
0.9
To obtain the air density, the air has been assumed to behave
like an ideal gas, as in [35]. Rearranging the equation of state of 0.8
ideal gases results in the following equation: 0.7

0.6
Mp
q¼ ð16Þ
Ta  R 0.5

0.4
where M is the molar mass of the air and R, the universal gas 0.3
constant.
Finally, the Cp curve and the area swept by the blades, A, are 0.2
obtained from the wind turbine datasheet. 0.1
Fig. 23 shows the energy balance in the microgrid, for the same
5 days of July analyzed above, using the measured value of the net 0
0 5 10 15 20
power, Pnet (blue), the forecasted value using persistence (red) and FE (kWh)
the mixed forecasting technique (green). As can be seen, the fore-
casting error in the first three days is sometimes larger when using Fig. 24. Cumulative distribution function of the forecasting error of Pnet using the
the persistence method but is also sometimes larger for the mixed persistence model (red) and mixed forecasting (blue). (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of
technique. Nevertheless, these errors are always under 10 kW h, this article.)
and can be withstood by the battery. However, on the fourth day,
the energy balance changes suddenly, causing the persistence
model to fail, resulting in a 20 kW h error, which is more than half 6 100
the size of the battery. In this case, the mixed forecasting also has
90
an accuracy error, but this is under 10 kW h, and can therefore be
4
withstood by the battery. 80
Fig. 24 shows the cumulative distribution function for the fore-
70
casting error, FE, for the analyzed year, using the persistence model 2
Power (kW)

60

SOC (%)
(red) and mixed forecasting (blue). This figure shows the overall
error reduction when using the mixed forecasting model. For 0 50
example, when using the mixed model, the error is, in 95% of cases,
under 5.1 kW h but can be as high as 6.8 kW h when using the per- 40
sistence model. -2
30
Given that the error using mixed forecasting is in general lower,
20
the fluctuations introduced in Pgrid through Pp will also be lower. -4
Furthermore, the improved forecasting would make it possible to 10
reduce KP1,max from 0.05 kW/% required in Strategy 3 in order to
-6 0
maintain control, down to 0.002 kW/%, thereby obtaining a better 24/07/2013 25/07/2013 26/07/2013 27/07/2013 28/07/2013
grid power profile. Nevertheless, a constant of 0.01 kW/% is to be
Fig. 25. Net power (red), power exchanged with the grid (blue) and SOC (green)
used in order to provide a safety margin on the manageability of
during four days of July 2013 using Strategy 4. (For interpretation of the references
the system. to color in this figure legend, the reader is referred to the web version of this
Fig. 25 shows the evolution of Pnet, Pgrid and SOC during the article.)
same days analyzed above, but using Strategy 4. By using this
new strategy, the fluctuations during the first days almost disap-
pear. Furthermore, the power peak that used to appear on the Finally, Table 3 compiles the quality criteria values for all the
fourth day, when the energy balance suddenly changes, is consid- strategies analyzed. Comparing Strategies 3 and 4, it can be noted
erably reduced. how, by using mixed forecasting, all quality criteria values have
Fig. 26 shows Pgrid evolution over the analyzed year, using Strat- been reduced, not only in terms of power peaks, as could be
egy 4. Comparing this figure with Fig. 22, it can be seen that the derived from Fig. 26, but also in terms of power fluctuations. In
power peaks have been reduced compared to Strategy 3. particular, positive and negative maximum power peaks have been
Daily net energy balance (kWh)

25

20

15

10

0
24/07/2013 25/07/2013 26/07/2013 27/07/2013 28/07/2013

Fig. 23. Net daily energy balance in the microgrid: measured (blue), forecasted using persistence (red) and using the mixed forecasting technique (green). (For interpretation
of the references to color in this figure legend, the reader is referred to the web version of this article.)
22 J. Pascual et al. / Applied Energy 158 (2015) 12–25

6 possible to combine different modules in order to form a microgrid,

as shown in Fig. 27.
These modules are:
4

PV array: Formed by 48 BP585 solar panels installed on the lab-

2 oratory’s roof, facing south and tilted at an angle of 30°. They
Power (kW)

have a rated power of 4080 W, although the actual peak power

0 is calculated to be 3600 W mainly due to degradation.

Small wind turbine: Bornay INCLIN6000 with a rated power of
6 kW is installed near the laboratory on top of a 20 m high
-2
tower.

Battery: The battery is composed of 120 FIAMM SMG300 series-
-4 connected cells. They are stationary lead-acid cells with a rated
voltage of 2 V and a C10 capacity of 300 A h and are installed in
the basement of the laboratory building. During normal opera-
-6
Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun tion they can store up to 72 kW h, of which 36 kW h are used to
achieve a 50% depth of discharge (DOD), the typical recom-
Fig. 26. Pgrid evolution from July 2013 to June 2014 using Strategy 4.
mended value for lead-acid batteries [36].

Programmable electronic load: The demand power profiles used
in the simulation of the microgrid are reproduced in real time
Table 3
by AMREL’s PLA7.5K-600-400 electronic load.
Quality criteria evaluated for all the Strategies 0–4.

Power converter: Both the generators and the battery are man-
Strategy P+ P MPD (W/ APD (W/ PPV aged by Ingeteam’s INGECON HYBRID MS30 central power con-
(kW) (kW) h) h)
verter located at the Renewable Energy Laboratory. It comprises
0 (Net power) 5.75 6.45 18,468 1122 13.28 a battery charger, two converters for the wind turbine and the
1 (SMA w/o control) 4.71 2.40 12,839 44.42 2.51 PV array and a reversible converter for grid connection. This
2 (SMA + SOC control) 2.83 3.45 1241 104.46 4.88
3 (SMA + SOC/FE 2.19 3.43 1554 71.39 4.33
device is responsible for absorbing or injecting power from
control) the grid, and its power reference can be controlled either man-
4 (CMA + SOC/FE 1.90 1.56 619 52.65 2.98 ually or through the automatic energy management strategy. It
control) can also work off-grid, and can seamlessly shift from one state
to the other. All the modifications made to this power converter,
reduced by 13.2% and 54.5% respectively, and the PPV value has to enable it to operate in the microgrid, have been carried out as
been reduced by 31.2%. Also MPD and APD have been reduced by a cooperative task between Ingeteam Power Technology and
60.2% and 26.2% respectively, which confirms that the new Pgrid UPNa’s Research Group in Electrical Engineering, Power Elec-
power profile is less variable. tronics and Renewable Energy (INGEPER in Spanish).

Supervisory and control station: The supervisory and control sta-
5. Implementation and experimental results tion comprises a PXI industrial computer by National Instru-
ments, a network of sensors and power meters, and a general
Experimental validation of different microgrid scenarios is car- purpose PC. The power meters monitor the main electrical
ried out at the UPNa’s Renewable Energy Laboratory, where it is variables (i.e., voltage, current, active power, reactive power,

Small wind turbine and PV generator Energy

Comm.
Supervisory and control station

Batteries Power analyzers and Programmable

INGECON HYBRID MS electronic load
Fig. 27. Microgrid installed at the Public University of Navarre (UPNa).
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 23

3.5
3

2.5
Power (kW)

1.5

0.5

0
01/10/2014 02/10/2014 03/10/2014 04/10/2014 05/10/2014 06/10/2014 07/10/2014 08/10/2014 09/10/2014 10/10/2014

0.8
Power (kW)

0.6

0.4

0.2

0
01/10/2014 02/10/2014 03/10/2014 04/10/2014 05/10/2014 06/10/2014 07/10/2014 08/10/2014 09/10/2014 10/10/2014

2.5

2
Power (kW)

1.5

0.5

0
01/10/2014 02/10/2014 03/10/2014 04/10/2014 05/10/2014 06/10/2014 07/10/2014 08/10/2014 09/10/2014 10/10/2014

Fig. 28. In red: PV power (top), wind power (middle) and consumption (bottom) during ten days measured at UPNa’s microgrid. In green: forecasted values. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

2
Power (kW)

-2

-4

-6
01/10/2014 02/10/2014 03/10/2014 04/10/2014 05/10/2014 06/10/2014 07/10/2014 08/10/2014 09/10/2014 10/10/2014

Fig. 29. Net power profile (red), net power profile forecasting (green) and power exchanged with the grid (blue) measured at UPNa’s microgrid. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)

frequency, etc.) and send them to the PXI. The network of sen-
Wind speed (two anemometers by the wind turbine, and one
sors comprises a series of 4–20 mA current loops to measure the next to the PV array);
following variables:
Irradiance (calibrated cell).

Battery temperature (Pt-100);

Battery room temperature (Pt-100); The PXI is responsible for acquiring and processing all the data.

PV panel temperature (Pt-100); All the programs needed to run the energy management strategy

Outdoor temperature (Pt-100); are run in this computer as well, which also sends the commands
24 J. Pascual et al. / Applied Energy 158 (2015) 12–25

to the INGECON HYBRID power converter, to the electronic load [6] Marzband M, Sumper A. Experimental evaluation of a real time energy
management system for stand-alone microgrids in day-ahead markets. Appl
and to the relays. The PXI is connected to the PC, where all the vari-
Energy 2013;106:365–76. http://dx.doi.org/10.1016/j.apenergy.2013.02.018.
ables are monitored in real time and from which all the microgrid [7] Marzband M, Sumper A. Experimental validation of a real time energy
is controlled by the user. Moreover, the PC works as a web server, management system for microgrids in islanded mode using a local day-
allowing for the remote monitoring and control of the microgrid. ahead electricity market and MINLP. Energy Convers Manage 2013;76:314–22.
http://dx.doi.org/10.1016/j.enconman.2013.07.053.
Finally, the PC is also responsible for checking updates in Meteo- [8] Marzband M, Ghadimi M. Experimental validation of a real-time energy
galicia’s web server. Once a new weather report has been pub- management system using multi-period gravitational search algorithm for
lished, the PC downloads it and processes the data in order to microgrids in islanded mode. Appl Energy 2014;128:164–74. http://dx.doi.org/
10.1016/j.apenergy.2014.04.056.
feed it to the PXI, so that the latter can add the new information [9] Zhao B, Zhang X, Li P, Wang K, Xue M, Wang C. Optimal sizing, operating strategy
to the energy management strategy. All the programs running on and operational experience of a stand-alone microgrid on Dongfushan Island.
the PC and PXI have been developed at the UPNa using LabView. Appl Energy 2014;113:1656–66. http://dx.doi.org/10.1016/j.apenergy.2013.09.
015.
This system operates continuously, acquiring and storing all the [10] Xue X, Wang S, Sun Y, Xiao F. An interactive building power demand
data in an automated data base. Fig. 28 shows the actual power management strategy for facilitating smart grid optimization. Appl Energy
profiles (red) and the forecasted output (green) for the PV array 2014;116:297–310. http://dx.doi.org/10.1016/j.apenergy.2013.11.064.
[11] Comodi G, Giantomassi A, Severini M. Multi-apartment residential microgrid
(top), the wind turbine (middle) and the loads (bottom). In the with electrical and thermal storage devices: experimental analysis and
absence of an energy storage system and the appropriate energy simulation of energy management strategies. Appl Energy 2015;137:854–66.
management strategy, the power exchanged with the grid would http://dx.doi.org/10.1016/j.apenergy.2014.07.068.
[12] Velik R, Nicolay P. Grid-price-dependent energy management in microgrids
be the consumption minus the generation curves depicted in
using a modified simulated annealing triple-optimizer. Appl Energy
Fig. 28. This is the net power profile, which is shown in Fig. 29 in 2014;130:384–95. http://dx.doi.org/10.1016/j.apenergy.2014.05.060.
red. Fig. 29 also shows the grid power (blue) and forecasted net [13] Tascikaraoglu A, Boynuegri A, Uzunoglu M. A demand side management
power (green). As can be seen, the profile of the power exchanged strategy based on forecasting of residential renewable sources: a smart home
system in Turkey. Energy Build 2014;80:309–20. http://dx.doi.org/10.1016/j.
with the grid is similar to the one obtained in the simulation. enbuild.2014.05.042.
[14] Niknam T, Azizipanah-Abarghooee R, Narimani M. An efficient scenario-based
stochastic programming framework for multi-objective optimal micro-grid
6. Conclusions operation. Appl Energy 2012;99:455–70. http://dx.doi.org/10.1016/j.
apenergy.2012.04.017.
This paper proposes an energy management strategy for a res- [15] Chen Y, Lu S, Chang Y, Lee T, Hu M. Economic analysis and optimal energy
management models for microgrid systems: a case study in Taiwan. Appl
idential microgrid directed at obtaining a smooth power profile Energy 2013;103:145–54. http://dx.doi.org/10.1016/j.apenergy.2012.09.023.
for the energy exchange with the grid. The proposed strategy [16] Parisio A, Rikos E, Tzamalis G, Glielmo L. Use of model predictive control for
makes use of forecasted power profiles in order to eliminate the experimental microgrid optimization. Appl Energy 2014;115:37–46. http://dx.
doi.org/10.1016/j.apenergy.2013.10.027.
lag in the grid power profile. This approach, however, is subject [17] Kim S, Jeon J, Cho C. Dynamic modeling and control of a grid-connected hybrid
to the accuracy of the forecasted data which may disrupt the generation system with versatile power transfer. IEEE Trans Ind Electron
energy balance in the microgrid. In order to handle forecasting 2008;55:1677–88. http://dx.doi.org/10.1109/TIE.2007.907662.
[18] Zhou H, Bhattacharya T, Tran D, Siew TST, Khambadkone AM. Composite
errors, the proposed strategy makes corrections in the power
energy storage system involving battery and ultracapacitor with dynamic
exchanged with the grid not only taking into account the forecast- energy management in microgrid applications. IEEE Trans Power Electron
ing error, but also the SOC of the battery. By combining the SOC 2011;26:923–30. http://dx.doi.org/10.1109/TPEL.2010.2095040.
and the forecasting error in a single control rule, the strategy man- [19] Koohi-Kamali S, Rahim NA, Mokhlis H. Smart power management algorithm in
microgrid consisting of photovoltaic, diesel, and battery storage plants
ages to minimize the fluctuations in the grid power, by ignoring considering variations in sunlight, temperature, and load. Energy Convers
any forecasting errors as long as they can be handled by the bat- Manage 2014;84:562–82. http://dx.doi.org/10.1016/j.enconman.2014.04.072.
tery. The energy management strategy has been successfully vali- [20] Barricarte JJ, Martín IS, Sanchis P, Marroyo L. Energy management strategies
for grid integration of microgrids based on renewable energy sources. In:
dated at the microgrid installed in the Renewable Energy SET2011, 10th International conference on sustainable energy technologies,
Laboratory at the UPNa. Istanbul, TÜRKIYE; 4–7 September, 2011.
[21] Aviles DA, Guinjoan F, Barricarte J, Marroyo L, Sanchis P, Valderrama H. Battery
management fuzzy control for a grid-tied microgrid with renewable generation.
Acknowledgments In: IECON 2012-38th annual conference on IEEE industrial electronics
society. IEEE; 2012. p. 5607–12. http://dx.doi.org/10.1109/IECON.2012.6389008.
This work was partially funded by the Government of Navarra [22] Pascual J, Sanchis P, Marroyo L. Implementation and control of a residential
electrothermal microgrid based on renewable energies, a hybrid storage
and the ERDF funds under project ‘‘Microgrids in Navarra: design, system and demand side management. Energies 2014;7:210–37. http://dx.doi.
development and implementation” and by the Spanish Ministry org/10.3390/en7010210.
of Economy and Competitiveness under grant DPI2013-42853-R, [23] Marinelli M, Sossan F. Testing of a predictive control strategy for balancing
renewable sources in a microgrid. IEEE Trans Sust Energy 2014;5:1426–33.
as well as by the European Union under project FP7-308468, http://dx.doi.org/10.1109/TSTE.2013.2294194.
‘‘PVCROPS-Photovoltaic Cost reduction, reliability, operational per- [24] Hanna R, Kleissl J, Nottrott A, Ferry M. Energy dispatch schedule optimization
formance, prediction and simulation”. for demand charge reduction using a photovoltaic-battery storage system with
solar forecasting. Sol Energy 2014;103:269–87. http://dx.doi.org/10.1016/
j.solener.2014.02.020.
References [25] Masters C. Voltage rise: the big issue when connecting embedded generation
to long 11 kV overhead lines. Power Eng J 2002;16:5–12. http://dx.doi.org/
[1] Hatziargyriou N. Microgrids: architectures and control. Wiley-IEEE Press; 10.1049/pe:20020101.
2014. [26] Mahmud MA, Hossain MJ, Pota HR, Nasiruzzaman ABM. Voltage control of
[2] Navigant research. Microgrids: commercial/industrial, community/utility, distribution networks with distributed generation using reactive power
campus/institutional, military, remote, grid-tied utility distribution, and compensation. In: IECON 2011-37th annual conference of the IEEE industrial
direct current microgrids: global market. Analysis and forecasts; 2013. electronics society. IEEE; 2011. p. 985–90. http://dx.doi.org/10.1109/
[3] United Nations Foundation. Microgrids for rural electrification: a critical IECON.2011. 6119329.
review of best practices based on seven case studies <http://energyaccess.org/ [27] Black J, Larson R. Strategies to overcome network congestion in infrastructure
images/content/files/MicrogridsReportFINAL_high.pdf> (accessed 16.03.15). systems. J Ind Syst Eng 2007;1:97–115.
[4] Kyriakarakos G, Piromalis D, Dounis A. Intelligent demand side energy [28] Parissis O-S, Zoulias E, Stamatakis E, Sioulas K, Alves L, Martins R, et al.
management system for autonomous polygeneration microgrids. Appl Integration of wind and hydrogen technologies in the power system of Corvo
Energy 2013;103:39–51. http://dx.doi.org/10.1016/j.apenergy.2012.10.011. Island, Azores: a cost-benefit analysis. Int J Hydrogen Energy
[5] Urtasun A, Barrios E, Sanchis P, Marroyo L. Frequency-based energy 2011;36:8143–51. http://dx.doi.org/10.1016/j.ijhydene.2010.12.074.
management strategy for stand-alone systems with distributed battery [29] Shinji T, Sekine T, Akisawa A, Kashiwagi T, Fujita G, Matsubara M. Reduction of
storage. Trans Power Electron 2015;30:4794–808. http://dx.doi.org/10.1109/ power fluctuation by distributed generation in micro grid. Electr Eng Jpn
TPEL.2014.2364861. 2008;163:22–9. http://dx.doi.org/10.1002/eej.20462.
 J. Pascual et al. / Applied Energy 158 (2015) 12–25 25

[30] Remund J, Perez R, Lorenz E. Comparison of solar radiation forecasts for the [33] Lorenzo E. Energy collected and delivered by PV modules. In: Luque A, Hegedus
USA. In: Proc of 23rd European PV Conference, vol. 2; 2008. p. 3–5. S, editors. Handbook of photovoltaic science and engineering. Wiley; 2003.
[31] Foley AM, Leahy PG, Marvuglia A, McKeogh EJ. Current methods and advances [34] Burton T, Sharpe D, Jenkins N, Bossanyi E. Wind energy handbook. Wiley;
in forecasting of wind power generation. Renewable Energy 2012;37:1–8. 2001.
http://dx.doi.org/10.1016/j.renene.2011.05.033. [35] AENOR. UNE-EN-61400-12-1; 2007.
[32] Meteogalicia THREDDS Server <http://www.meteogalicia.es/web/modelos/ [36] Sauer D. Electrochemical storage for photovoltaics. In: Luque A, Hegedus S,
threddsIndex.action?re> (accessed 16.03.15). editors. Handbook of photovoltaic science and engineering. Wiley; 2003.