1

Coordinated micro-generation and load

management for energy saving policies
S. Bertolini, M. Giacomini, IEEE M., S. Grillo, IEEE M., S. Massucco, IEEE M., F. Silvestro*, IEEE M.

Abstract-- The paper describes a joint university-industry Community related to the Technological Platform SmartGrids
project aimed at defining hardware and software requirements [1] which addresses and encourages the concept of microgrids
for the development of a prototypical microgrid platform on and related control and management.
existing sites able to manage generation resources and load. The The paper describes a joint university-industry project aimed
paper describes an architecture developed for integrating at defining hardware and software requirements for the
existing software tools for on-line load monitoring and control
development of a prototypical microgrid platform on existing
and for management of in-site generation. The system acquires
data from field and stores information on a server, from which
sites able to manage the generation resources. An architecture
an optimization tool gets data in order to perform its calculations has been developed for the integration of existing software
and give the optimal set points to the programmable generation tools for on-line load monitoring and control and for
resources. management of in-site generation. The system acquires data
Results are reported about experimentation at test sites with from field and stores information on a server and in a
different generating units, such as cogeneration of heat and database, from which an optimization tool gets data in order to
power (CHP) and/or renewable power generating units equipped perform its calculations and to give the optimal set points to
with a real-time monitoring tool. the programmable generation resources.
Index Terms— Distributed generation, Energy Management, II. TECHNICAL ASPECTS CONCERNING DISTRIBUTED
Energy efficiency, Load forecasting, RES optimization.
GENERATION (DG)
I. INTRODUCTION
In order to realize an adequate development of DG compatible

T he significant changes introduced by the deregulation of

with the present status of power systems, it will be necessary
to develop two typologies of structures: microgrids and
Virtual Power Plants.
electricity markets, by the need for a sustainable development
and by the consequent use of less polluting fuels, the A microgrid is a low voltage grid with distributed sources,
development of a technology for small and medium size with the presence of storage and load control devices [2]. It is
generations and the investments in the renewable source sector a controlled entity which can be managed as an aggregate of
are addressing the development of electric power systems generators and/or loads as a response to technical and
towards the concept of active electric distribution grids against economic perspectives. A microgrid can contribute to
the old passive paradigm. optimize the management of the electric power system thanks
With “active networks”, one means electric distribution to a hierarchical control architecture made up of three levels:
networks with a significant presence of distributed generators (a) Control at the distribution grid level; (b) Microgrid control;
of small and medium size, often with randomly variable (c) Local control of generators
production profiles typical of renewable sources (wind, solar, Virtual Power Utilities or Virtual Power Plants (VPP) are a set
etc.). In this kind networks, generators and loads can of conventional and/or renewable generators, distributed
participate to electricity market as main actors and a System resources, on wider geographical areas, which are able to
Operator may dispatch individual components as well as an participate to the electricity market by means of intensive use
aggregation of them. of information and communication technologies.
Future developments for electric distribution networks are
well identified by a qualified document of European
III. THE PROJECT
A. Project description
* reference author The goal of the project was the study and the formulation of
S. Grillo, S. Massucco, F. Silvestro are with the Electrical Engineering guidelines for energy management and saving by means of
Department of the University of Genoa, Italy. Emails:[email protected];
[email protected]; [email protected] intelligent management of in-site available sources and of
M. Giacomini, S. Bertolini are with the Department of Communication loads.
Computer and System Sciences of the University of Genoa, Italy. Emails: In the previously mentioned context, for which both
[email protected]; [email protected] generators and loads (either individually or as an aggregate)
can be considered by the grid as suppliers of main services
 2

(energy and capability) and auxiliary services (regulation, collecting measurement data on a central server, in order to
reserve, etc.), the investigated problem is the optimal allow the treatment and the diffusion of information to the
programming and generation and load management, users through Internet. This solution does not disturb the
performed by means of integrated intelligent devices for the existing installations and it assures the interoperability and the
real-time measurement of energy consumptions and evolution capacity of the system.
production. Such a structure may thus be able to manage the The technical architecture of the platform is structured on
bidirectional energy flow with adequate optimization and three functional levels: (1) sensors; (2) data transmission
control algorithms, by means of a supervisor which proposes network; (3) information architecture.
the optimal strategy. The activity related to sensors has performed the definition of
A prototype platform has been realized by integrating existing the needs in terms of measurements, wiring points and the
hardware and software tools. The result is a structure which is identification of the type of signals provided by each different
able to manage different sources in a microgrid and to sensors. For this energy management application, the sensors
communicate among different sites. are usually integral energy meters. The most common solution
In particular, the proposed system acquires all the signals from for the acquisition of these data is based on impulsive outputs
field and stores the information on a server and on a database, which are proportional to consumption. Fiscal counters are
which is queried by an optimization program which produces normally equipped with this type of signal. The real-time
the set-points of the power production for the programmable monitoring system architecture is represented in Figure 2.
generators. The network architecture is realized so that it presents a good
Figure 1 shows a conceptual scheme of the proposed usability in measurement collection and it does not disturb the
architecture. existing installations. Moreover it uses a “non-proprietary
bus” with transportation of accessible data, thus assuring a
microturbine good interoperability and evolution capacity of the system.

boiler
grid battery

photovoltaic
thermal load

wind

PLC and
measurement electric load
devices

Figure 1 – Overview of Data acquisition system from field

The system integrates the load monitoring module with the

microgeneration programming. A decision support system
allows to obtain an integrated system for the intelligent
monitoring and control of electric consumptions for
decentralized medium-large size structures.
The proposed platform operates in a microgrid hierarchical
structure. A central controller operates both a medium-long
term management (intended as the day-ahead optimization of
the subsequent 24 hours) and a short term optimization (with a
time horizon up to 15 minutes), providing the set-points of
local generators [3],[4].
Figure 2 – Architecture of the real-time monitoring system
IV. PROJECT IMPLEMENTATION
The information architecture is realized so that it assembles
The project has been developed by integrating two different and saves the information collected by data loggers, visualizes
tools for the real-time monitoring of Distributed Energy costs and energy consumptions, it offers a remote access, via
Resources (DER) i.e. electrical loads and small/medium size Internet, to different users, it creates an open system which
generating units [5],[6]. The tools are named Nextep and allows the implementation of new analyses, it gives the user
ECOMP. the possibility to exploit tables and graphs in his own
The overall architecture is presented below. Some results from documents.
the experimental activities are reported in the next section. The main characteristics of the platform can be summed up in
the following points: Centralized DB on Mysql; Inferential
A. Platform Nextep for the monitoring of DER engine for the techno-economic calculation with on-line
The monitoring tool Nextep allows the continuous control and updating (at variable step, 15 min, 1 hour, 1 day); Web-based
management of energy consumptions in the examined sites, by interface. It is Multi-user, Multi-fluid and multi-energy
 3

(electric energy, gas, water, heat), Multi-plaatform (Windows, eliminating the factor of uncertain nty due to setting of the
Linux). It allows multi-site and aggregated caalculations. weights. Each network was trained d for a number of times
equal to 20000 and the functiion chosen to evaluate
performance during training was th he mean square error. The
B. Day-ahead Load Forecasting
w set at 10-4.
error on the desired outputs (goal) was

A Neural network load forecasting tool to bee integrated in the

prototype has been developed. The obtained results report the
ability to forecast energy consumption of a site with a good
degree of confidence.
The set of available data of energy consumpption (15 minutes
samples) was divided into an 'training set' to train a neural
network and 'test set'. In particular the 'trainiing set' was made
from the consumption data of year 2006 (froom 01/01/2006 to
31/12/2006), instead the 'test set' was built with the
consumption of 2007 (data from 1/1/2007 to 31/10/2007).
Three types of input with which to train thhe networks were
tested. The input vector therefore has a ddifferent structure
depending on the type considered. In the firsst case it contains
96 (24x4) values of the day before pow wer consumption,
minimum and maximum temperatures of tthe previous day.
The information is provided on the date oon the day of the Fig. 3 – Error Histogram
week (the holiday is treated as if it was S Sunday) and this
information is represented with a 3-digit binnary number (i.e. Fig. 3 shows that approximately 46% % of the errors fall within
001 = Sunday, Monday = 010; 011 = Tuesdday, Wednesday = the percentage ranges 0-5% and ab bout 25% error rate in the
100; 101 = Thursday, Friday = 110; 111 = Saturday). In the range 5-10%.
second case the input vector consists of tthe 96 values of Moreover, these errors show an average
a of 9.72% and a
electricity consumption of the day of the prrevious week, the standard deviation of 13.35%.
maximum and minimum temperatures of pprevious day. The Fig. 4 illustrates the comparison bettween the prediction made
third case represents a union of the two typess discussed above, by the neural network described above (in blue) and real
in the sense that the input vector contains bboth the set. The consumption (in red) in February.
vector also contains the minimum and maxim mum temperatures
the day before. 45
February

From the mathematical point of view simplle differences are Forecast

Real consumption
made between each predicted value from thee network and the 40

value of real consumption forecasted. Inn particular, the 35

difference between predicted consumptionn and the actual
consumption is further divided by the latter so as to obtain a 30
Consumption (kWh)

relative error. Then mean and standard devviation errors are 25

calculated. Errors are represented graphicallly by a histogram
which indicates the probability of the error too fall in one of the 20

selected bands. The neural networks considerred in the analysis 15

are composed exclusively of one hidden layeer. The neurons of
output layer are obviously ninety-six, that as many as the 10

power values to be predicted for a given day. One of the 5

differences that exist between the networks created is instead 1 7 14
Days
21 28

the number of neurons in the hidden layer thaat has been varied Fig. 4 – Forecast for February
F
from 20 to 60.
For these types of neurons belonging to thhe hidden layer a Forecasting electricity consumption using neural networks has
sigmoidal transfer function 'tansig' were chosen to force the provided quite different results deepending on the type of
output to take values within the range [-1, + +1]. It was instead training used. The best results were obtained in the case of the
simulated different transfer functions of the ooutput layer and in “Resilient backpropagation”.
particular “pureline”, “logsig “ and “tansig”. Furthermore, the The prediction of power consum mption made during the
analysis was also performed by varying the trraining algorithm: summer months does not offer as it can be seen from Fig. 5
backpropagation, Conjugate gradient, G Gradient descent, satisfactory results as those obtained
d in the case of forecasting
Conjugate gradient with Polak-Ribiere uppdates, Conjugate the winter months.
gradient with Fletcher-Reeves updates and Scaled conjugate
gradient algorithms. It important to emphasizze that the results
achieved by different algorithms are fully ccomparable as the
same networks were always used to perform the workout, thus
 4

100
August
• Energy inputs, formed by fuel costs and purchased
forecast electric energy cost and by the related contractual
90 real costs for distribution and supply;
80 • Cogeneration system, defined by plant configuration,
70
single component propriety and the relationships with
the external grids which supply electric energy and
fuel;
Consuption (kW)

60

50 • Energy management, which is influenced by the

different load levels at which a component can
40
operate (component off-design);
30 • The environmental conditions where the cogeneration
20
system operates;
• Energy outputs, formed by electric and thermal users
10

0 The code refers to three different libraries. In the first one data
1 7 14 21 28
Days are loaded from the input file and are associated with a matrix
Fig. 5 – Forecast for August
which stores their values. In the second library it is possible to
evaluate the fixed costs of components and to associate them
This different behavior in the forecast performance is due to with the goal function. In the third library the machine
the fact that summer time presents a rather different shape due performances are calculated in design and off-design
to holidays and due to the presence of air conditiong. conditions, thus it is possible to calculate the related variable
costs for each significant period, measured in seconds. The
code structure is reported in Fig. 7.
C. Sensors and communication architecture
A specific datalogger (see Fig. 6) has been built in order to
monitor energy consumption and Photovoltaic production of
one site.

Fig 7 – Structure of the ECOMP code

Fig. 6 – Datalogger module

ECOMP is divided into two levels (an upper and a lower
Further in the project and extensive activity on communication level).
solutions between sensors, site and remote server have been At the lower level the optimization is carried out by
done, studying network security and adequacy. minimizing the variable costs by means of a genetic optimizer,
which varies the exploitation percentage of the nominal power
D. Platform ECOMP for energy system planning and of the prime mover. The optimal management condition over
management the year will be represented by the set of optimal
managements in the individual periods. After carrying out the
ECOMP is a modular and semi-empirical calculation code calculation in each period, the goal is achieved.
devoted to the evaluation of energy systems and applications
in distributed cogeneration and trigeneration sectors, in order At the upper level a non linear optimizer varies the rating of
to verify the adequacy of the operated choices and to guide the the desired component at each iteration. The goal function is
optimization of different constructive parameters. The factors the sum of variable and capital (fixed) costs.
which influence the method, aimed at determining the
configuration, the component sizing and the operation strategy The minimization of the goal function is carried out over
and based on an energy saving criterion, are various: defined time horizons. The number of significant periods is
 5

variable and it is chosen by the user. Some diistinctive days are A. Example of user intervention and d optimization of a site
typically chosen as representative (e.g. summmer or winter day, The consumption of the Facullty of Economics was
week or week-end day). The request value foor each hour in the characterized by night hourly averagea consumption of
examined day is also characterized by thhe number which 100kWh against day average of abou ut 400 kWh.
represents how often this situation occurs oveer the year. This 1:4 ratio was immediately y detected, even by a
The ECOMP code has been created with a m modular structure comparison with other studied casess, as unusual and therefore
so that its development is as simple as posssible. There is a requiring an intervention.
library with built-in components; each of theem presents some Furthermore, there was a peak at about 6 am immediately
gates through which information is exchangedd. followed by a significant drop in co onsumption. This behavior
appeared to be disadvantageous in n terms of power and not
V. TEST SITE CONSIDERE
ED coordinated with the public opening of the building.
The experimental activities of the project arre oriented to the In general, the overall trend of consuumption presented a series
analysis of the technologies and the solutionss proposed by the of periodic peaks probably related tot switching of large loads
market of intelligent systems for energy management and not coordinated in their operation.
savings for civil and industrial plants, especially with An analysis and a measurement campaignc were made on
reference to cogeneration technology represented by different switchboards to identify the
t behavior of individual
microturbines, with economic indicationns of the main loads and tried to reconstruct a 15 minutes
m power curve to see
typologies of conventional and renewable generation. In if the monitoring system was conssistent with the measures
particular, six test sites with different (exiisting and future) made the field in order to clarify the repetitiveness of the
generating plant solutions, have been examined. pattern of recorded consumptio on. Figure 9 shows a
For each site the existing and the future plaant solutions have representation of consumption meaasured for the Faculty of
been examined by evaluating their advantages and Economics.
disadvantages in terms of appropriateness too the experimental
phase of the present study.

Fig. 9 – Overall and single load shape

As it can be seen in Fig. 9, saw tooth night consumption

detected by the system is associateed to air conditioning that
are activated for 15 minutes and theen switched off for the rest
Fig. 8 – One of the test sites considered for experimentattion of the hour.
In general the overall trend of consu
umption presented a series
Evaluation criteria have been set-up bassed on proposed of periodic spikes related to large appliances
a not coordinated
parameters for the comparison of the differeent solutions. The in their operation.
results have been reported in an evaluationn matrix aimed at
outputting the best solution(s) which are achievable for the
experimental monitoring.
At the end of this process, three sites have beeen chosen for the
experimental phase:

1. ACAM trigeneration plant in La Spezia. The site

wass made up of a conventional enngine (190 kW) in
cogeneration asset with an absorpttion chiller and a
PV plant (35 kWp).
2. RGM photovoltaic plant in Genova. The
experimentation was focused on loaad monitoring and
PV characterization (see Fig. 8).
3. University of Genova (Unige) load mmanagement. The
experimentation was focused on lload management Fig. 10 – Overall optimized load shape
and optimization. The site is made up of 19 MV/LV
substations.
 6

Following the analysis and the relative countermeasures VI. CONCLUSIONS AND PERSPECTIVES
regulating power circuits clocks the absorption lines of the
main distribution were optimized. Fig. 10 presents the new The paper has presented some of the present problems related
shape of the building power curve. The black line represents to the penetration of DG into power systems. The
the pattern of consumption before intervention. transformation of electric distribution networks from relatively
Table 1 summarizes the detected anomalies and the actions simple and mainly radial structures towards more complex
undertaken to achieve the new consumption pattern. structures with a strong presence of active components, like
distributed generators, implies the need to develop studies for
Tab. 1. Summary of interventions
the integration of distributed resources with complex and
Anomaly Action results uncertain characteristics, to indentify or adapt control and
Withdrawals clock setting of the Flattening of the curve monitoring methodologies, the design and the realization of
periodic sawtooth groups of Air and reducing overall adequate communication and protection structures.
peaks handling units consumption in 24h. The great interest in the control of electric load – intended as a
(AHU) Synchronizing the true distributed resource – is located in this line. The
operation of facilities availability of adequate tools for (even real-time) monitoring
with the operating hours
of the building
of electric loads could pave the way to the systematic use of
system architectures and devices for the coordinated control of
High consumption Installation of new Decreased night of generated power and of load which realize the widespread
ratio night / day clocks for AHU energy usage by interest among medium-small users in real applications for the
approximately 50% optimization of consumptions coordinated with the installation
Operation of air Better room Drastic reduction in of distributed generators.
conditioning in occupacy planning consumption
rooms not used ACKNOWLEDGEMENT
temporarily (for
example This work has been financed by PSTL – Parco Scientifico
classrooms during Tecnologico Ligure for the feasibility study “Innovative
vacation) Integrated Technologies for Energy Management and micro-
Consumption at Performed Further reduction of generation management also from renewable sources for
night due to feasibility analysis consumption by about energy saving in civil and industrial sites”.
electrical circuits and identification of 30% Night.
and non-automated the work required Contemporary REFERENCES
for better optimization of operating
management and maintenance costs of
the lamps [1] European SmartGrids Technology Platform, “Vision and Strategy for
Europe’s Electricity Networks of the future”, Document EUR 22040,
2006
In Fig. 11 the daily energy costs of the building are [2] DISPOWER: “Distributed generation with high penetration of
represented. It can be noticed that the action performed have renewable energy sources”, Progetto EU Framework V, Energy - n.
led to a significant saving (see point (a) with respect to point ENK6-CT2001-00522, 2002-2005
[3] A. Bertani, C. Bossi, F. Fornari, S. Massucco, A. Morini, F. Silvestro,
(b) in Fig. 10). Moreover night and holidays cost (see point “Combined Use of Simulation and Test Site Electrical Networks for
(c)) have been reduced. Assessing and Evaluating Distributed Generation Possibilities and
Performances”, CIRED 2005, Turin, 6-8 June 2005
[4] A. Bertani, A. Borghetti, C. A. Nucci, M. Paolone, S. Massucco, A.
Morini, F. Silvestro, L. De Biase, E. Quaia, “Management of Low
b Voltage Grid with High Penetration of Distributed Generation:
a concepts, implementations and experiments”, CIGRE 2006, Paris,
August 26-31, 2006
[5] M. Giacomini, S. Grillo, L. Marigo, S. Massucco, C. Torti, and F.
Silvestro, “Innovative Integrated Technologies For Energy Micro-
c Generation management And Energy Saving Policies,” in Proceedings
4th European Conference PV-HYBRID and MINI-GRID, May 29–30,
2008, ISBN: 978-3-934681-72-9, pp. 132–139
[6] A. Borghetti, M. Bosetti, C.A. Nucci, M. Paolone, S. Massucco, F.
Silvestro, S. Scalari, “A procedure for the automatic scheduling of
distributed energy resources in medium voltage networks”, Proc. of the
20th Int. Conf. on Electricity Distribution, CIRED 2009, Prague, 8-11
June 2009.

Fig. 11 – Overall optimized cost per day

VII. BIOGRAPHIES
These savings in energy consumption during the not-working Simona Bertolini was born in 1984. She received the degree in biomedical
hours can be estimated to be about approximately 6-7 €/ h, engineering in 2009. She developed her degree thesis at Department of
Communication, Computer and System Sciences (DIST), Genoa, Italy. She
summing up to the considerable figure of 40 k€ per year. actually attend the Ph. D. course in bioengineering at DIST. Her main
research interests include medical informatics, machine learning, neural
networks and data-mining.
 7

Mauro Giacomini (M’07) was born in 1963. He received the degree in Stefano Massucco (M’1980) was born in Genoa, Italy, in 1954. From 1979 to
electronic engineering in 1987and the Ph.D. degree in bioengineering in 1993. 1987, he had been working at the Electrical Engineering Department of Genoa
He is Aggregate Professor in Bioengineering at Department of University, at CREL - the Electrical Research Center of ENEL (Italian
Communication, Computer and System Sciences (DIST), Genoa, Italy, since Electricity Board) in Milano, Italy, and at ANSALDO S.p.A. in Genoa, Italy.
1999. He has authored or coauthored about 300 scientific papers. His main Since 1987, he has been Associate Professor of Power Systems at the
research interests include medical informatics (medical knowledge based University of Pavia and from 1993 at the Electrical Engineering Department,
systems, development of databases, home care technology) biotechnologies University of Genoa, where he is currently Full Professor since 2000. His
(NMR, quantification in immunoenzymatic assays), the study of biological main research interests are in the field of electric energy systems and
systems by modeling methods, and signal analysis. Also other application of distributed generation modelling, control, and management, power systems
artificial intelligence and data mining techniques are among his interests. Prof. analysis and simulation, intelligent systems application to power systems.
Giacomini is currently Associate Editor of the IEEE TRANSACTIONS ON Prof. Massucco has been scientific responsible for several national research
BIOMEDICAL ENGINEERING. projects and EU projects dealing with distributed generation, energy saving
Samuele Grillo (S’05-M’09) was born in Alessandria, Italy, in 1980. He and smartgrids. He is author of more than 140 scientific papers. Member of
received the “Laurea” degree in electronic engineering in 2004 from the the Italian Electrotechnical Committee. Member of CIGRE Working Group
University of Genoa and, from the same University, a PhD in power systems 601 of Study Committee C4 for “Review of on-line Dynamic Security
in 2008 with a thesis on application of neural networks to power system Assessment Tools and Techniques”. Delegate of Genova University Rector
security assessment. Since February 2008 he is holder of a research grant at for energy saving policies. Member of the Scientific Committee of FIRE the
the University of Genoa. His research interests regard optimization and Italian Federation for Rational Use of Energy.
control techniques, neural networks and their application to power systems Federico Silvestro (S’01-M’03) was born in Genova, Italy, in 1973. He
(i.e., security assessment, load and production forecast, local and small received the degree in electrical engineering from the University of Genoa in
generation management). 1998 and a Ph.D. in electric power systems in 2002. He is now a Research
Assistant at the Electric Engineering Department, University of Genoa.
Member of CIGRE Working Group 04 of Study Committee C1 for
“Application and required developments of dynamic models to support
practical planning”. His research interests are distributed generation and
smartgrids, dynamic security assessment, knowledge based systems applied to
power systems.