TRANSACTIONS ON ENVIRONMENT AND ELECTRICAL ENGINEERING ISSN 2450-5730 Vol 1, No 4 (2016)
© G. Belli, G. Brusco, A. Burgio, D. Menniti, A. Pinnarelli, N. Sorrentino and P. Vizza
Abstract – The increase of renewable non-programmable
production and the necessity to locally self-consume the produced
energy led to utilize ever more storage systems. To correctly
utilize storage systems, an opportune management method has to
be utilized. This paper implements a multi-period management
method for storage systems, using different management
strategies. The method aims to minimize the total absorbed and
supplied energy or the peak power exchanged with the grid. The
results show the effectiveness of the method in diminishing the
energy exchanged with the grid and also the possibility to
optimize the performance of the storage systems.
Keywords — Storage management, multi-pediod scheduling,
prosumer, optimal energy management
I. INTRODUCTION
he common tendency to produce on site the necessary
energy to the end user by means of small size plants,
generally from non-programmable renewable sources, led to a
rapid increase of installed photovoltaic (PV) power. Although
PV generation offers economic and environmental benefits, its
non-programmability can reduce these benefits. To increase
the possibility to self-consume the produced energy and
increase the profit for the user, storage systems should be
utilized.
Otherwise, it is worth to underline that storage systems have
a limited lifetime, related to their charge and discharge cycles.
For this reason, it is opportune to manage storage systems with
a specific strategy that can ensure them a long lifetime [1-2].
In order to better manage storage systems, realizing an
accurate scheduling, generation and load forecasting systems
would be useful to help the management of the grid.
Several methods carrying out storage management,
considering different strategies, have been proposed in
literature [3-7]. Such methods focus on an economical
optimization, or a real time management without considering
This work was financed by the Italian Ministry of Economic Development
(MISE) and the Ministry of Education, University and Research (MIUR)
through the National Operational Program for Development and
Competitiveness 2007-2013, Project DOMUS PON 03PE_00050_2.
G. Belli, G. Brusco, A. Burgio, D. Menniti, A. Pinnarelli, N. Sorrentino,
and P. Vizza are with Department of Mechanical, Energy and Management
Engineering (DIMEG) University of Calabria Via Bucci 42C, Arcavacata di
Rende - CS, Italy (e-mail: name.surname@unical.it).
the optimization of storage lifetime. In [3] the “economical”
optimal management of a storage system is carried out in a
single period (24 hours ahead); if the optimization is not
required 24h ahead, the method utilizes a real time approach.
In [4] an appropriate management to reduce losses and
increase distribution grid capacity is implemented and to this
aim, a distributed storage is utilized; such a method allows to
improve the utilization of renewable energy resources
minimizing the energy adsorption from the grid.
Both methods [3,4] are implemented on a 24h period and
they consider an economical optimization, moreover a multi-
generation/multi-storage scheduling is realized.
It would be interesting to observe the benefits for a single
user equipped with a PV generation system and a storage
system, if the considered period is greater than 24h and a
multi-period optimal management method is adopted.
In [5] users equipped of energy storage systems are
considered; it explains how the interaction of different storage
systems can be harmful for the grid, although such storage
systems have been introduced to protect it. So a novel
management technique for distributed storage systems is
implemented. The utilization of this technique led to saving up
to 13% of the electricity bill for each consumer with a 4-kWh
storage system.
In [6] a day ahead optimization algorithm is implemented to
provide the optimal storage and/or production scheduling
strategy for a single user. The real load profile is considered
exactly as the programmed 24h ahead profile.
At the user level, that is a prosumer level, it is advantageous
to know generation and load profiles forecast to choose the
better storage power scheduling.
Indeed, a goal for the user should be to make “himself-
sustainable”, satisfying his total energy demand by means of
local energy production, minimizing the exchanges of energy
with the grid.
This paper presents a multi-period storage management
method and considers a prosumer equipped with a storage
system. In particular basing on weather forecast, generation
and load forecasts are obtained; Artificial Neural Networks are
utilized to implement the two forecasting models. Starting
from these load and generation forecasts, different simulations
are performed: the method minimizes the overall energy
exchanged with the grid, the power peaks between prosumer
A multiperiodal management method at user
level for storage systems using artificial neural
network forecasts
G. Belli, G. Brusco, A. Burgio, D. Menniti, Member, IEEE, A. Pinnarelli, Member, IEEE,
N. Sorrentino, Member, IEEE, and P. Vizza, Student Member, IEEE
T
and grid or the energy in a particular time interval; moreover,
a limit for the power exchanged with the grid is considered.
Worth noting that the benefits of the storage systems have
been also demonstrated by the same authors in [8]; in
particular, they underline the suitability of the Li-Ion batteries
compared with lead-acid batteries. Indeed, in this paper the
economic aspects concerning the storage systems and in
particular the management of the storage system are not
examined. In this paper, an energetic and power analysis is
carried out.
The rest of the paper is structured as follows: in the second
section, the implemented forecasting models are illustrated; in
the third part, Storage Management Method is described; in
the last part, simulation results are shown and the benefits in
the use of the method are underlined.
I. PV AND LOAD FORECASTING MODELS
The Multi-Period Storage Management (MPSM) Method
has for input the renewable production profile and the load
profile of a user. To generate these profiles, the MPSM
method uses renewable production and load forecasting
models. Several production and load forecasting models exist
in literature; they present different accuracy which depends on
several factors, such as utilized input data, utilized
methodology, and so on. Independently from the used method,
the results of the generation and load power forecast can be
evaluated in different way.
A. Accuracy Estimation
Many performance parameters are used according to
forecast purpose. In general, all these parameters represent the
error between the forecasted and the actual value.
The Mean Absolute Error (MAE) [9] measures how close
forecasts are to the outcomes. It is more sensitive to high value
of the discrepancy between the forecasted and the actual
power profiles, so it is used to underline the presence of
discrepancy peaks (emphasizing the high error). It allows to
know the consequent average power imbalance; the MAE is
given by:
MAE =
1
N
∑|fi − yi| =
N
i=1
1
N
∑|ei|
N
i=1
(1)
where fi is the forecast value, yi is the real value. Moreover,
the evaluation of the Maximum Absolute Error (MaxAE) is
important to know the maximum difference between fi and yi,
in particular it is useful to know the maximum power
imbalance resulting from the forecast; it is given by:
MaxAErr = Max(|fi − yi|) (2)
The Mean Squared Error (MSE) [9] measures the mean
square discrepancy between fi and yi. The MSE is more
sensitive than the MAE to high error value. The MSE square
root provides another statistical quantity, that is the Root
Mean Squared Error (RMSE). The MSE and the RMSE are
defined as:
𝑀𝑆𝐸 =
1
𝑁
∑(fi − yi )
2
𝑁
𝑖=1
(3)
𝑅𝑀𝑆𝐸 = √𝑀𝑆𝐸 = √
1
𝑁
∑(fi − yi )
2
𝑁
𝑖=1
(4)
Another important statistical quantity is the Mean Absolute
Percentage Error (MAPE) [9]. It measures the forecasting
accuracy and expresses this accuracy as a percentage. It is
calculated by:
𝑀𝐴𝑃𝐸 =
1
𝑁
∑ |
fi − yi
fi
|
𝑁
𝑖=1
(5)
Moreover, the Maximum Percentage Error (MAxPE) allows
knowing what the maximum percentage discrepancy between
the forecasted and the actual value is:
MaxPE = max (|
fi − yi
fi
|) ⩝ i = 1 … N (6)
Such statistical parameters allow to estimate the accuracy of
the several forecasting models, to find the best suited to the
case under estimation.
B. PV forecasting model
Referring to renewable production forecasting model, it is
supposed that the power source is a photovoltaic (PV) plant.
Several PV forecasting models exist in literature: they often
present accurate results. However, such models are very
sophisticated and require information not always available, as
those presented in [10-12]; the PV forecasting model
implemented by the authors uses available input data and for
this reason it is defined a “practical” forecasting model.
Such a PV generation forecasting model is implemented by
an Artificial Neural Network (ANN).
For the implementation of the ANN, the Neural Network
Toolbox of Matlab is utilized. The chosen ANN typology is a
Multi-Layer Perceptron (MLP), with a supervised training
algorithm; in particular, the back-propagation algorithm is
used. Moreover, only a hidden layer is utilized, the activation
function of all the neurons is tan-sigmoidal, while the training
function is the Levenberg-Marquardt method [13].
Chosen the ANN typology, it is necessary to provide to the
method historical input data and historical output (target) data
(in the case under examination the collect of the PV power
production data really generated by the PV plant).
The number of neurons of the input layer is chosen
considering the input data, while the number of neurons of the
hidden layer has been determined empirically. In fact, a
sensitivity analysis has been conducted: the input data are kept
constant and the number of neurons of the hidden layer are
changed; under this condition, several tests to calculate the
MAPE have been made. The optimum number of neurons for
the hidden layer, to minimize the MAPE is 30, as shown in
Figure1.
Fig. 1. Sensitivity analisys
So, summarizing, the implemented ANN consists in (Fig. 2):
5 input neurons (meteorological condition of the
considered hour h, meteorological condition of the
hour h+1, meteorological condition of the hour h-1,
the hourly irradiance, the considered hour);
30 hidden layer neurons;
1 output neuron (hourly forecasted power production).
Fig. 2. ANN PV Forecasting Representation
Meteorological data about sky conditions are transformed to
be used as input data for the ANN, in a number. A “1” to “5”
scale has been used to represent the increasing of cloudiness.
The minimum value (“1”) indicates “clear-sky” condition
while the maximum value (“5”) indicates “storm” (Tab.1).
TABLE I. WEATHER CONDITIONS CODING
Code Weather condition
1 Clear sky
2 Nearly cloudless, scattered clouds
3 Few clouds
4 Partly cloudy
5 Covered, Storm
The model has been tested on a specific a PV plant and
good results are obtained both for clear and non-clear sky
conditions.
Indeed, the results show that: in clear sky conditions the
MAPE is less than 10%, while in non-clear sky conditions the
MAPE is about 42%. The MAE, compared to PV power plant,
is equal to 2.6% for clear sky conditions and 6.8% for non-
clear sky conditions.
Figure 3 depicts the results of the PV forecast for the days
from 30 August to 2 September; for these days, clear-sky
conditions were predicted. The MAPE is calculated for all the
four days: it is equal to 9.8%. Whereas, the percentage error is
maximum and equal to 45% in the hours close to sunrise and
sunset; while the percentage error becomes minimum (less
than 1%) for the hours of higher production.
Fig. 3. Clear-Sky Days Results
C. Load forecasting model
Referring to the load forecasting model, in [14] the
relevance to use load forecast for several purposes is
underlined, especially for islanded operation, because it is
necessary to guarantee grid stability, in addition to allow
generation programmable system to work at a maximum
efficient point.
In literature, there are a large number of load forecasting
models as those presented in [15]. Reference [16] highlights
that is more difficult to predict the individual load than an
aggregate of loads. Nevertheless, in this paper a predictive
model for an individual load is implemented.
The load forecasting model implemented in this paper
utilizes accessible data; such a model predicts also individual
and aggregate loads. Although its simplicity the results
demonstrate the good performances of the model.
A feed-forward Multi-Layer Perceptron (MLP) ANN,
supervised by a back-propagation algorithm, has been
implemented (Fig. 4). For ANN training, the collection of
consumption data of the considered user is necessary.
Similarly than PV forecasting model, also for load forecasting
model a sensitivity analysis has been carried out, so to detect
the number of the hidden layer neurons which lead to a better
accuracy of the model.
The implemented ANN consists in:
7 input neurons (month, day, day type, hour, daily
maximum temperature, daily minimum temperature
and daily average temperature);
30 hidden layer neurons;
1 output neuron (hourly forecasted power
consumption).
9
11
13
15
17
19
10 15 20 25 30 35 40 45
MAPE[%]
Number
of neurons
Fig. 4. ANN Load Forecasting Representation
The input data are so defined: the day, month and hour
identify the period which the forecast is required; the day type
identify if the considered day is a workday or a holiday, if it is
a day before or after a holiday. Moreover, the minimum,
maximum and average temperature are utilized; these are
useful specially if the electric air conditioning is utilized, in
particular maximum temperature is useful for cooling,
whereas minimum temperature is useful for heating.
The load forecasting model has been tested on real data of
a typical residential user. In Fig. 5, the forecasted and the real
load profiles are shown, for three days (Thursday, Friday and
Saturday). The obtained Mean Absolute Error (MAE),
compared to the rated power of the considered user’s contract
is less than 6%, that is an acceptable error for the purpose of
the method.
Fig. 5. Forecasted load profile and real load profile
Figure 5 depicts the results of the load forecast model; the
MAPE is calculated and it is less than 20%. Whereas, the
percentage error is maximum in the hours and day with a low
consumption, and the minimum percentage error occurs for
the workdays, when the consumption is high, and it is less
than 1%.
II. STORAGE MANAGEMENT METHOD DESCRIPTION
The most important variables used in the Multi-Period
Storage Management (MPSM) method are reported in Table 2.
TABLE II. MPSM METHOD VARIABLES
Nomenclature
𝑃𝑔 𝑡
𝑑
Grid Transferred Power; (time t, day d)
𝑃𝐿 𝑡
𝑑
Load Power; (time t, day d)
𝑃𝑆 𝑡
𝑑
Storage Transferred Power; (time t, day d)
𝐸𝑆 𝑡
𝑑
Storage Energy Level; (time t, day d)
𝐸𝑆 𝑚𝑖𝑛, 𝐸𝑆 𝑚𝑎𝑥 Minimum and maximum Storage Energy Level
𝑃𝑆 𝑚𝑖𝑛, 𝑃𝑆 𝑚𝑎𝑥 Minimum and maximum Storage Power
𝐸𝑆 𝐼𝑛𝑖𝑡 Initial Energy stored
The main equations which describe the MPSM method are as
follow:
𝑂𝐹: min (∑ 𝑓(𝑃𝑔 𝑡
𝑑 )
𝐷
𝑇
𝑡=1
𝑑=1
) (7)
s.t.
𝑃𝑔 𝑡
𝑑 = 𝑃𝐿 𝑡
𝑑 − 𝑃𝑃𝑉 𝑡
𝑑 − 𝑃𝑆 𝑡
𝑑
(8)
𝐸𝑆 𝑡+1
𝑑 = 𝐸𝑆 𝑡
𝑑 + 𝑃𝑆 𝑡
𝑑 ∗ 𝑡 (9)
𝐸𝑆 𝑚𝑖𝑛 ≤ 𝐸𝑆 𝑡
𝑑 ≤ 𝐸𝑆 𝑚𝑎𝑥 (10)
𝑃𝑆 𝑚𝑖𝑛 ≤ 𝑃𝑆 𝑡
𝑑 ≤ 𝑃𝑆 𝑚𝑎𝑥 (11)
𝑠𝑖𝑔𝑛(𝑃𝑆 𝑡
𝑑 ) = 𝑠𝑖𝑔𝑛(𝑃𝑃𝑉 𝑡
𝑑 − 𝑃𝐿 𝑡
𝑑 ) (12)
𝐸𝑆 𝑡=1
𝑑=1 = 𝐸𝑆 𝐼𝑛𝑖𝑡 (13)
The MPSM method can be utilized for more days (d) and
every day is divided in more time intervals (t); such time
intervals are the same of that used in the forecasting models.
In Objective Function (OF) (7), 𝑓(𝑃𝑔 𝑡
𝑑 ) indicates different
goals of energy exchange optimization. Indeed, the user can
require to: minimize overall the energy exchanges with the
grid, minimize the power peaks or minimize the energy
exchanged for a particular time period with the grid
The MPSM method is subjected to the constraints from (8)
to (13). Constraint (8) is used to calculate the power
exchanged with the grid, 𝑃𝑔 𝑡
𝑑
. The constrains (9), (10), (11),
(12) and (13) concern the storage. In (9) the variation of the
stored energy (between two time intervals) is calculated, in
(10) the stored energy is limited between a minimum and
maximum value (depending on the used storage); in (11) the
charge and discharge storage power is limited, (12) indicates
that the storage can charge only if there is a power surplus (PV
power 𝑃𝑃𝑉 𝑡
𝑑
is greater than the load power 𝑃𝐿 𝑡
𝑑
), vice versa
storage can only discharge. In (13) the initial stored energy is
defined as 𝐸𝑆 𝐼𝑛𝑖𝑡 .
Once load and PV production power forecasts for the user
are obtained, the difference between the two profiles is
calculated. This difference profile represents the input of the
MPSM method, which solves the Objective Function (OF),
taking into account the constraints. The method returns the
storage power exchange profile and the consequent grid power
exchange profile.
III. SIMULATION
To test the effectiveness of the MPSM method, some
simulations are carried out, considering as prosumer a build of
University of Calabria: this is a business user, equipped with
photovoltaic plants.
The considered build has a maximum power consumption
of 25 kW and the installed PV plants power is 45 kW.
The test considers a time period of 7 days, from the 10th to
16th October 2015. In Fig. 6, load and PV power forecast
profiles of the considered 7 days are reported.
After determining load and PV power profiles, it is necessary
to sizing storage system for the required function.
A. Non optimized PV power
First of all, the storage capacity is calculated to supply
loads and limit the exchange of energy with the grid.
Storage capacity will be the smallest between the resulting
average daily energy purchased and supplied to the grid,
which are calculated as the difference between PV production
and load profiles. This analysis is carried out for profiles of a
typical day. In the present case, the daily purchased energy is
almost 170 kWh, whereas the energy supplied to the grid is 80
kWh; so the storage would have a capacity of 80 kWh.
After calculating storage capacity, an overestimation to be
conservative will be necessary: an increase of 20% will be
considered. In addition, in order to safeguard the storage
useful life, a residual state of charge (SOC) of 40% has to be
considered as a further increase of the estimated storage
capacity.
Considering the previous estimated storage capacity (80
kWh) and the increases of 20% and 40%, the obtained storage
capacity is about 140 kWh.
Starting from the calculation of the difference between load
and PV power forecasts, it is used for two groups of
simulations. The first group of simulation aims to minimize
the total exchange of energy with the grid, trying to make the
prosumer self-sustainable and to avoid congestions on the
grid. Instead, the second group of simulations aims to
minimize only the peaks of energy during the day, trying to
reduce the costs of energy supply and to avoid worthless
oversize of the generation plants. For both the groups of
simulations, the comparison is made between the condition
with storage working in “real time”, that is no storage
management is taking into account, and the condition with
storage managed by MPSM method.
The starting point is the value of the overall energy
exchanged between the user and the grid without using a
storage system: this value is equal to 1.68 MWh, where 1.13
MWh is the energy adsorbed by the user and 0.55 MWh is the
energy left to the grid.
For the first groups of simulations, when none
management method is utilized, the total energy exchange
with the grid decreases until 0.69 MWh, where 0.60 MWh is
the purchased energy by the user and 0.09 MWh supplied to
the grid. If the storage is managed by MPSM method, the total
energy exchanged with the grid is equal to 0.68 MWh.
Respect the previous case, the difference is very limited.
Although this difference is only of 0.01 MWh, the positive
effect of the MPSM method consists in the possibility to
maximize the performances of the storage. Indeed, using a
“real time” operation strategy, the charge and discharge cycles
are not optimized because they are partial cycles, while with
MPSM method, the storage executes always full cycles of
charge and discharge (Fig.7). Only in a few hours, a distorted
trend is visible in Fig. 7, due to the high variability of weather
conditions on the 5th day, that involves to have a partial cycle
of charge and discharge.
For the second group of simulations, minimizing only the
peaks of power exchanged with the grid, the power exchanged
in “real time” reaches 23 kW, while using MPSM method, it is
about 8 kW. In Fig. 8 the exchanged energy profile with and
without MPSM method are depicted.
Fig. 6. Load and PV profiles
Fig. 6. Stored Energy Profile with and without MPSM
Fig. 7. Exchanged power profiles with and without MPSM
B. Optimized PV power
In this section, starting from the average daily load profile
and the monthly average daily PV production profile, to
minimize the exchange of energy with the grid, the PV plant is
sized to cover the daily energy demand. Considering this, the
obtained rated PV power is about 56 kW.
Similarly to the previous subsection A, the storage system
is properly sized and the obtained capacity is about 240 kWh.
With such data the method is utilized to carry out the same test
of the previous case.
First of all, the MPSM method is utilized to minimize the
exchange of energy with the grid; the obtained result of the
total energy exchanged with the grid is equal to 0.41 MWh,
where 0.25 MWh is the purchased energy by the user and 0.16
MWh is the energy supplied to the grid.
In this case, as the rated PV power and the storage capacity
are optimized, the use of the MPSM method, compared to the
real time management, does not contribute to many
advantages in the management of the charge/discharge storage
cycles.
The real advantage would occur in the management of the
exchanged energy with the grid for the days with non-clear
sky conditions. In Fig. 9 the exchanged energy profile with
the grid, with and without the MPSM method is depicted.
Moreover, referring to Fig. 9, it is possible to observe that
the energy supplied to the grid is greater than the energy
purchased from the grid; only the 5th day the purchased energy
is greater than the supplied one, because it is not a clear sky
day.
Fig. 8. Exchanged grid power profile with and without MPSM
Worth noting that the PV power is sized to cover the daily
energy demand; so if the rated PV power increase, obviously
the produced energy increase and as a consequence the total
energy exchanged with the grid increases. In fact, for example,
if the rated PV power is 60 kW the total exchanged energy is
0.49 MWh, instead of 0.41 MWh. This shows that before to
use the MPSM method, optimal PV and storage sizing is
necessary.
C. Grid Power Restriction
In this section, it is supposed that the considered prosumer
has a limit for 𝑃𝑔 𝑡
𝑑
. This can be due to different reason, for
example if the prosumer has a contract with the energy
provider for a reduced power, or if the power line is designed
for a limited power.
In fact, this kind of optimization allows to reduce the
problems due to the congestion problem and any restrictions
of the power interface devices.
In particular two cases are examined: in the first case the
maximum 𝑃𝑔 𝑡
𝑑
(Pg_max) is 10 kW, in the second case Pg_max
is 7 kW. Worth noting that the limit for the power is both for
the supplied and delivered energy. The MPSM method is
completed using the sequent equations:
|𝑃𝑔 𝑡
𝑑
| ≤ 𝑃𝑔_𝑚𝑎𝑥 (14)
The OF is implemented to minimize the entire energy
exchanged with the grid, as implemented above. It is worth to
underline that this test is different to the previous
minimization of the peak power, in fact in the previous case a
restriction for 𝑃𝑔 𝑡
𝑑
is not utilized but solely the peaks of 𝑃𝑔 𝑡
𝑑
are reduced.
In the first test Pg_max is equal to 10 kW and the
constraint (12) is relaxed, in this way the storage can be
charged also by the grid and can discharge also if there is a
surplus of energy, this is limited only through the OF.
The utilized storage capacity is 240 kWh and the rated PV
power is 56 kW; the load profile is reported in Fig. 6.
In this first case, the obtained result of the total energy
exchanged with the grid is equal to 0.41MWh, where 0.26
MWh is the purchased energy by the user and 0.15 MWh is
the energy supplied to the grid.
In the second case, Pg_max is equal to 7 kW, the total
energy exchanged with the grid is also equal to 0.41MWh, and
the purchased energy is equal to 0.26 MWh whereas the
supplied energy to the grid is equal to 0.15 MWh.
Such results demonstrate that the constrains on
𝑃𝑔 𝑡
𝑑
are almost irrelevant for the OF, in fact the quantity of
energy exchanged with the grid is the same of the previous
case. The only difference is for the profile of 𝑃𝑔 𝑡
𝑑
: in Figs. 10
and 11 the profiles of power exchanged with the grid for either
cases are reported.
Fig. 9. Exchanged grid power profile with the constrain Pg_max=10 kW
Fig. 10. Exchanged grid power profile with the constrain Pg_max=7 kW
The figures show that the trend of 𝑃𝑔 𝑡
𝑑
is constant in the
area where the power peak occur, this means that for those
time and day the bonds are achieved; this is particularly
observable for Pg_max is 7 kW.
Moreover, worth noting that the maximum power Pg_max
(7 kW), obtained in this case, is less than the maximum power
obtained with the minimization of the peak of power (in
Subsection A) where Pg_max is 8.2 kW.
It is important to observe the behaviour of the storage
system when Pg_max is equal to 7 kW compared to the case
when there is not a constrain for 𝑃𝑔 𝑡
𝑑
: in figure 12 this
comparison is reported.
Fig. 11. Stored Energy Profile with and without Pg_max constrain
It is possible to observe the differences between the two
profiles, especially for the fifth day. In fact, to limit the power
exchanged with the grid, in particular the power drawn from
the grid, the stored energy is maintained as long as it is not
used to decrease 𝑃𝑔 𝑡
𝑑
: the battery is not discharged just when
there is a deficit of energy but when this energy is utilized to
limit the maximum 𝑃𝑔 𝑡
𝑑
.
Thanks to the MPSM method the prosumer can employ a
reduced power contract with consequent less costs for the
prosumer. At the same time the Distribution System Operator
(DSO) can design the line for a reduced power, with further
savings.
IV. CONCLUSION
The paper shows the importance of an opportune
management method for storage devices. In fact, the positive
effect resulting from the use of storage systems, particularly in
relation to non-programmable resources, can be increased if an
appropriate management strategy is utilized. One of the
feature of the implemented method is its multi-periodicity. In
fact, if the management is made on more days, there are more
data input and the storage can be managed in a better way.
The presented storage management method implements
different management goals.
First of all, the method minimizes the total energy
exchanged with the grid to make users self-sustainable. This
implies the use of opportune PV and load forecast models.
Secondly the method is also utilized to reduce the peak of
power exchanged with the grid, decreasing from 23 kW to
about 8 kW.
Moreover, it is underlined that an accurate sizing of PV
and storage systems is necessary before to implement and
utilize a management strategy.
The results are compared with the real time storage
management; such a comparison shows the effectiveness of
the method. The results show also the possibility to optimize
the performance of the storage device in terms of charge and
discharge cycles.
At the end, the behaviour of the management method, if a
constrain for the maximum 𝑃𝑔 𝑡
𝑑
is utilized, is evaluated.
Simulations are carried out; they demonstrate that despite a
further constrain is utilized, the entire energy exchanged with
the grid is minimized. This can be a good result both for the
prosumer and for the DSO, indeed they can respectively
reduce the contract power and reduce the line capacity, with
consequent savings.
Moreover, it would be interesting evaluate the behaviour of
the method if a schedulable load is adopted.
REFERENCES
[1] D. Menniti, A. Pinnarelli, N. Sorrentino, A. Burgio, and G. Brusco,
“Energy Management System for an Energy District With Demand
Response Availability”, Smart Grid, IEEE Transactions on, vol. 5(5),
2014, pp. 2385-2393.
[2] D. Menniti, A. Pinnarelli, N. Sorrentino, G. Belli, A. Burgio, “Demand
Response Program in an Energy District with storage availability”,
International Review of Electrical Engineering, in press.
[3] A. Nottrott, J. Kleissl, and B. Washom, “Energy dispatch schedule
optimization and cost benefit analysis for grid-connected, photovoltaic-
battery storage systems”, Renewable Energy, vol. 55, 2013, pp. 230-240.
[4] N. Jayasekara, P. Wolfs, and M.A.S. Masoum, “An optimal management
strategy for distributed storages in distribution networks with high
penetrations of PV”, Electric Power Systems Research, vol. 116, 2014,
pp. 147-157.
[5] P. Vytelingum, T.D. Voice, S.D. Ramchurn, A. Rogers, and N.R.
Jennings, “Agent-based micro-storage management for the smart grid”,
in Proc. of the 9th International Conference on Autonomous Agents and
Multiagent Systems, vol. 1, International Foundation for Autonomous
Agents and Multiagent Systems, pp. 39-46, May 2010.
[6] I. Atzeni, L.G. Ordóñez, G. Scutari,D.P. Palomar, and J.R. Fonollosa,
“Demand-side management via distributed energy generation and
storage optimization”, Smart Grid, IEEE Transactions on, vol. 4(2),
2013, pp. 866-876.
[7] A. Mohamed, and O. Mohammed, "Real-time energy management
scheme for hybrid renewable energy systems in smart grid applications",
Electric Power Systems Research, Vol. 96, 2013, pp. 133-143.
[8] D.Menniti, A. Pinnarelli, N. Sorrentino, A. Burgio, G. Brusco, “The
economic viability of a feed-in tariff scheme which solely awards the
self-consumption for promoting the use of integrated photovoltaic-
battery systems”, Applied Energy, in press.
[9] S. Makridakis, and M. Hibon, “Evaluating accuracy (or error)
measures”, Working paper 95/18/TM, INSEAD, (1995) France.
[10] Y. Zhang, M. Beaudin, Raouf Taheri, H. Zarcipour, and D. Wood,
“Day-Ahead Power PV power production Output Forecasting for Small
Scale Soar Photovoltic Electricity Generators”, IEEE Transactions on
Smart Grid, Vol. 6, no. 5, September 2015
[11] C. Chen, S. Duan, T. Cai, and B. Liu, “Online 24-h solar power
forecasting based on weather type classification using artificial neural
network”, Solar Energy, Vol. 85, no. 11, 2011, pp. 2856-2870.
[12] C. W. Chow, B. Urquhart, J. Kleissl, M. Lave, A. Dominguez, J.
Shields, and B. Washom, “Intra‐hour forecasting with a total sky imager
at the UC San Diego solar energy testbed”, Solar Energy, Vol 85, no. 11,
2011, pp 2881–2893.
[13] J. J. Moré, “The Levenberg-Marquardt algorithm: Implementation and
theory”, in Lecture Notes in Mathematics, No. 630–Numerical Analysis,
Springer-Verlag, 1978, pp. 105–116.
[14] N. Hatziarg, Microgrids: Architectures and Control, Wiley-IEEE Press,
February 2014.
[15] H. S. Hippert, C. E. Pedreira, and R.C. Souza, “Neural networks for
short-term load forecasting: A review and evaluation”, Power Systems,
IEEE Transactions on, vol. 16(1), 2011, pp 44-55.
[16] H. Chitsaz, H. Shaker, H. Zareipour, D. Wood, and N. Amjady, “Short-
term electricity load forecasting of buildings in microgrids”, Energy and
Buildings, vol. 99, 2015, pp. 50-60.
Grazia Belli (Italy, 1985) received her degree in Energetic
Engineering in 2011 and her Ph.D in Science of Complex
Systems in 2016 from the University of Calabria. Her current
research interests concern renewable energy sources,
distributed generation, smart grid technologies and electricity
local market.
Giovanni Brusco (Italy, 1980) received his degree in
Electronics Engineering from the University of Calabria, Italy,
in 2007 and his Ph.D. in Computer and system Engineering in
2013 at the Electronic, Computer and Systems Science
Department of the University of Calabria, Italy. His current
research interests concern renewable energy sources,
distributed generation, harmonic analysis and smart grid
technologies.
Alessandro Burgio (Italy, 1973) received his degree in
Management Engineering from the University of Calabria in
1999 and his Ph.D. in Computer and system Engineering in
2006 at the Electronic, Computer and Systems Science
Department of the University of Calabria, Italy. His current
research interests include electrical power systems, distributed
generation, renewable energy, power electronics and
harmonics, electronic ballast.
Daniele Menniti (Italy 1958) received his degree in
Electrical Engineering from the University of Calabria,
Cosenza, Italy and his Ph.D. degree in Electrical Engineering
from the University of Naples, Italy, in 1984 and 1989
respectively. He is an Associate Professor at the Mechanical,
Energetic and Management Department of the University of
Calabria, Italy. His current research interests concern electrical
power system analysis, real-time control and automation.
Anna Pinnarelli (Italy, 1973) received her degree in
Management Engineering from the University of Calabria in
1998 and her Ph.D. in Electrical Engineering in 2002 from the
Electrical Engineering Department of the University of
Naples, Italy. She is an Assistant Professor at the Mechanical,
Energetic and Management Department of University of
Calabria, Italy. Her current research interests concern FACTS
technology, harmonic analysis, electrical system automation,
decentralized control and smart grid technologies.
Nicola Sorrentino (Italy, 1970) received his degree in
Management Engineering in 1994 and a Ph.D. in Computer
and system Engineering in 1999 at the Electronic, Computer
and Systems Science Department of the University of
Calabria, Italy. He is a Researcher at the Mechanical,
Energetic and Management Department of the University of
Calabria, Italy.
Pasquale Vizza (Italy, 1990) received his degree in
Energetic Engineering in 2014 from the University of
Calabria; he is currently attending the PhD school at the same
University. His current research interests include renewable
energy sources, smart grid technologies, energy storage
economics, generation and load forecasting.
I. INTRODUCTION
I. Pv and Load forecasting Models
A. Accuracy Estimation
B. PV forecasting model
C. Load forecasting model
II. Storage Management Method description
III. Simulation
A. Non optimized PV power
B. Optimized PV power
C. Grid Power Restriction
IV. Conclusion