Technical Field
The present invention relates to an intelligent energy management method and system
for efficient energy management of an energy supply system which is equipped with an energy
storage and a generator having PDE (power dependent efficiency) such as monotonically
decreasing PDE.
Background Art
Energy storages such as batteries are combined with power generators to constitute
energy supply systems which supply electric power to various equipment. Such energy
supply systems can operate without electric power from other sources such as grids ( i.e.,
commercial power supplies) and renewable power sources (also known as "renewables").
Alternatively, energy supply systems can be supplied with power from grids and/or renewables.
In some cases, grids are unreliable grids which often suffer blackout. Recently, energy
storage systems are becoming cheaper, more compact and reliable, they can help to improve
overall efficiency of energy supply systems by increasing the generator loading. However,
various factors affect the efficiency of energy supply systems, and therefore, an energy
management system which aids more efficient operation of the energy supply system is
required.
The goal of the energy management system is to reduce cost and fuel consumption of
the generator and keep generator wear within reasonable bounds by use of intelligent
prediction, optimization and control. Also the inclusion of a renewable power source into the
energy supply system can be managed by the energy management system.
A typical example would be the energy management for an energy supply system for a
base transceiver station (BTS) which is equipped with a battery and a Diesel generator and
connected with a grid subjected to blackouts, as it occurs typically in rural areas of developing
countries. Typically in such applications, when the grid is available, the equipment such as
BTS equipment and air conditioning take their energy from the grid, but during blackout, the
energy for the load is taken from the energy storage and the Diesel generator (and sometimes
also renewables such as PV (photo voltaic) generation and/or wind generation). Optimal
charge/discharge cycles for the energy storage and appropriate on/off commands for the Diesel
generator during off grid time lead to optimal operation of the energy supply.
Examples of the energy storage-based energy supply systems with no or sporadic grid
supply are described in related art references [NPL1], [NPL2]. These systems do not use load
prediction, and in the case that renewable generation is used, they do not use renewable
generation prediction. Especially, the systems neither use blackout duration probability
function prediction technology nor provide means to include uncertainty in the optimization of
the charge/discharge pattern and the generator on/off command.
Therefore, the systems of the related arts, fuel and cost optimality by choosing the right
charge/discharge pattern of the battery and the generator on/off command cannot be achieved.
The systems of the related arts do not use the available information as best as possible.
The problem with related-art systems is that, for example, it is not guaranteed that at
the end of the blackout the energy storage (typically, a battery) is not unnecessarily charged
and that the generator runs most efficiently considering the optimal generator loading and a
low number of generator starts (it depends on the type of generator if the latter is an issue).
The unnecessary charging of a battery results in a loss of efficiency, since the battery can be
charged at lower cost with electric grid energy.
The related-art system described in [NPL1] relies on a priority based switching logic
without prediction and optimization, where the different power sources are given priorities.
For example, priorities of 1, 2, 3 and 4 are assigned to renewables, a battery, a grid and a
generator, respectively. However, the system is not able to achieve fuel or cost optimality.
In the method described in [NPL2], during blackout, the battery is fully discharged and
charged and this cycle repeated. However, also this method does not achieve full optimality.
A system described in [PL1] which is provided with a PV generator and a battery uses
day-ahead prediction and decides to charge or not to charge the battery during the night
depending if the needed energy of the loads is higher or lower than the predicted generated
energy by the renewable source. However, this system does not foresee the connection of a
generator with PDE as, for example, a Diesel engine generator represents.
A system described in [PL2] relates to an electric vehicle (EV) including a fuel cell
(generator), a motor (load) and a battery and controls the EV to improve overall efficiency.
However, the solution is not applicable to the type of PDE generator, since for fuel cell more
loading reduces efficiency. Additionally, the system of [PL2] works with fixed charging and
discharging limits, which would amount to suboptimal operation in the general application.
A certain hardware topology for the connection of a grid, a renewable power source and
a battery is proposed in [PL3]. However, the proposal in [PL3] does not consider blackouts or
non linear generators.
Related-art systems that deal with blackout prediction are described in, for example,
[PL4]-[PL8]. These systems either predict the concrete start time of the blackout or calculate
the blackout probability of a certain time instant but do not consider the blackout duration
explicitly. In [PL9], only a planned blackout, i.e., a blackout whose start time and end time
are known, is considered, so it is assumed that the blackout information is perfectly known.
This is generally not the case.
By the way, for some type of applications, the knowledge of the concrete time instant
of the blackout is not necessary for optimal operation. For optimal operation of a energy
storage and a generator the blackout duration is important. Knowing the blackout duration,
and the future load and renewable generation allows the determination of optimal
charge/discharge cycles of the pattern together with the right on/off command for the generator.
Citation List
Patent Literature
[PL1]: US 2012/0181864 Al
[PL2]: US 2001/0024105 A1
[PL3]: US 2011/01 15295 Al
[PL4]: JP 2004-355219 A
[PL5]: JP 2005- 184890 A
[PL6] : JP 2006- 01604 A
[PL7]: JP 2007-034639 A
[PL8]: JP 2012 235541 A
[PL9]: WO 2013/002155 Al
Non-patent Literature
[NPL1]: Technical Whitepaper -SolarConnect™ Energy management solution for Base
Stations -A Solar Powered Cellular Base Station [online]. Solar Semiconductor, 2010 [retrieved
on 2014-02-26]. Retrieved from the Internet:
.
[NPL2]: Ashok Jhunjhunwala, Bhaskar Ramamurthi, Sriram Narayanamurthy, Janani
Rangarajan, and Sneha Raj, Powering Cellular Base Stations: A Quantitative Analysis of
Energy Options [online]. Telecom Center of Excellence (RiTCOE), Indian Institute of
Technology, Madras 2012 [retrieved on 2014-02-26]. Retrieved from the Internet:
.
Summary of Invention
Technical Problem
The problems with the related-art technology are that it is not possible to achieve
optimality or not even sub-optimality of operation. Since there is no information about future
load demand, renewable generation and blackout duration (or blackout duration probability
function), optimal charge/discharge patterns and generator on/off commands cannot be
calculated. In order to derive optimal decision, prediction of future development is necessary.
More concretely, following problems I to IV should be considered to derive the optimal
decision.
Problem I: The fuel consumption of an energy supply system equipped with a battery
and a generator with PDE is high. Therefore, the fuel consumption of the generator must be
reduced as much as possible. Without the above-described comprehensive type of prediction
and optimization, wrong charging/discharging and generator decisions may be taken which
lead to increased fuel consumption.
The related-art technology has often a switching logic which is priority based as
described in [NPL1]. However, with this type of technology, even if priority based switching
is better than some other approaches, optimality cannot be guaranteed in most of the cases.
Low loading of the generator leading to a waste of fuel cannot be avoided in the general case.
The related-art technology often works with a fixed upper charge limit and a fixed lower
discharge limit for the energy storage. It can be shown analytically that there are a lot of
cases where only suboptimal operation and not optimal operation can be achieved with this
technology.
In order to minimize the fuel consumption of the generator with PDE (monotonically
decreasing), future load, future renewable generation (if available), and blackout duration (if
there is an unreliable grid connection) is necessary to know, in order to guarantee that the
generator runs always in its most efficient mode and unnecessary battery charging is avoided.
The related-art technology that does not provide this functionality needs prediction technology
and special optimization and control method. Without prediction of load, generation and
blackout duration and without time variant charge and discharge limits, it is not possible to
determine the optimal charge/discharge cycle for the battery and the optimal generator
start/stop command.
Problem II: The cost (or TCO (total cost of ownership) ) of an energy supply system
equipped with an energy storage and a generator with PDE is high. Therefore, the running
cost of the system should be reduced. Cost optimization is strongly related with the fuel
consumption minimization.
Problem III: Guaranteeing reasonable number of PDE generator starts.
Without sophisticated prediction, cost/fuel optimality and reasonable number of PDE
generator starts, both, cannot be guaranteed at the same time.
Problem IV: Blackout probability or prediction of exact blackout start time does not
help to find the optimal charge/discharge cycle for the battery and the optimal generator
start/stop command.
A lot of previous proposals such as [PL4]-[PL9] related with blackout events deal with
prediction of the start time of the blackout, or with the blackout start probability of future time
instants. However, this type of prediction is not needed and does not help to improve
efficiency of this type of energy supply system. If a blackout happens, the energy storage can
automatically take over to feed the load for a certain period, while the parameter of optimal
future charge/discharge pattern for the energy storage and the parameter of optimal future
generator starts can be only be computed (exactly or approximated) if some kind of knowledge
about the blackout duration is available.
Problem IV is addressed by a method to predict the blackout duration probability
function. This means that not the single blackout duration is predicted, but different blackout
durations are given a probability of occurrence and the optimization method tries to derive
energy storage charge/discharge commands and generator off/on commands which satisfy
different blackout durations.
An exemplary object of the present invention is to solve the problems of the related-art
technology and to provide an energy management method and system which can reduce cost,
fuel consumption of the generator and keep generator wear within reasonable bounds.
Solution to Problem
According to one exemplary aspect of the present invention, an energy management
method for an energy supply system which includes at least an energy storage, a load and a
generator with power dependent efficiency includes: calculating two time variant parameters
indicating a discharge lower limit and an upper charge limit, respectively, of the energy storage,
based on optimization using different kinds of prediction; and controlling, in a real time
manner, charging and discharging of the energy storage and operation of the generator, with a
certain priorities given to various power sources, such that state of charge of the energy storage
is controlled within a region between the discharge lower limit and the upper charge limit.
According to another exemplary aspect of the present invention, an energy management
system for an energy supply system which includes at least an energy storage, a load and a
generator with power dependent efficiency includes: prediction means for performing different
kinds of prediction; optimization means for calculating two time variant parameters indicating
a discharge lower limit and an upper charge limit, respectively, of the energy storage, based on
optimization using different kinds of prediction; and control means for controlling, in a real
time manner, charging and discharging of the energy storage and operation of the generator,
with a certain priorities given to various power sources, such that state of charge of the energy
storage is controlled within a region between the discharge lower limit and the upper charge
limit.
The present invention focuses on the energy management method and system and is not
restricted to a certain hardware topology. The present invention may consider also the
inclusion of non-linear generators (from the view point of efficiency) others than Diesel engine
generators.
These and other objects, features and advantages of the present invention will become
more apparent from the following description, taken in conjunction with the accompanying
drawings, illustrating by way of example the principles of the invention.
Brief Description of Drawings
Figure 1 is a block diagram illustrating an example of a basic arrangement of an energy
supply system to which an energy management method according to an exemplary
embodiment of the present invention is applied.
Figure 2 is a block diagram illustrating another energy supply system with unreliable
grid supply.
Figures 3Ato 3C are graphs illustrating possible inefficiency characteristics of a fuel
engine generator with PDE (power dependent efficiency).
Figure 4 is a block diagram illustrating an example of a prediction server.
Figure 5 is a block diagram illustrating another example of the prediction server.
Figure 6 is a block diagram illustrating an example of a local controller.
Figure 7 is a diagram explaining the meaning of policy parameters p io (t), Phigh(t) with a
concrete charge/discharge cycle.
Figure 8 is a graph illustrating an example of a blackout duration probability function.
Figure 9 is a diagram explaining the meaning of policy parameters pio (t), Phigh(t) with a
concrete charge/discharge cycles.
Figure 10 is a bar graph explaining the composition of the unnecessary cost (UC)
consisting of an unnecessary expensive produced energy cost (UEEC), an engine start cost
(ESC) and an excessive fuel cost (EFC).
Figure 1 is a diagram explaining the meaning of the reference (in-)efficiency.
Figure 2 is a flowchart of operation of a real time control module implementing the
policy based on the parameters pio (t), Phigh(t).
Figure 13 is a block diagram of an example of a blackout duration probability function
prediction unit.
Figure 4 is a schematic block diagram of a specific example of application of the
exemplary embodiment in the region of energy management for energy supply for a base
transceiver station (BTS).
Figure 15 is a detailed block diagram of the example shown in Figure 14.
Figure 16Ato 16H are graphs illustrating simulation results of the operation of the
system shown in Figures 14 and 15.
Description of Embodiments
Exemplary embodiments of the present invention will be next described with reference
to the accompanying drawings.
As shown in Figure 1, an energy supply system to which an energy management
method according to an exemplary embodiment of the present invention generally includes:
energy storage 11 such as a Li-ion (lithium-ion) battery, nickel metal hydride battery and other
secondary cells; single or aggregated load 12 to which electric power is supplied; and energy
generator 13 that needs some kind of fuel. Efficiency (in kWh/L: kilowatt hour per liter) or
inefficiency (in L/kWh: liter per kilowatt hour) of generator 13 is dependent on the power that
the generator produces. In other words generator has a characteristic of PDE (power
dependent efficiency). Energy storage 1, load 12 and generator 13 are electrically connected
by means of physical connection apparatus 20 which is typically some kind of a wiring board.
For conversion between direct current (DC) and alternate current (AC), physical connection
apparatus 5 may contains switches, a DC/AC converter and an AC/DC converter.
In the exemplary embodiment, charging and discharging power of energy storage 1
and on/off mode of generator 13 can be controlled. For the purpose of this control, physical
connection apparatus 5 is equipped with local controller 20. Local controller 20 is a key
component of an intelligent energy management system (IEMS) which performing the energy
management method of the exemplary embodiment.
The arrangement shown in Figure 1 is a fundamental arrangement of the energy supply
system, and the energy supply system can also be equipped with single or aggregated
renewable generation (i.e., renewable power source) 6 such as PV power, wind power, and
PV-wind combined power. The energy supply system can be supplied with power from
unreliable grid supply 17. Term "unreliable" in this circumstance means that there are grid
blackouts where no energy can be provided from the grid. It is assumed that the electricity
price from the grid is constant and cheaper than the electricity produced by the generator, no
feed-in is allowed. In addition, local controller 20 in physical connection apparatus 15 can be
communicated with external prediction server 30 via communication link 18. In case of
arranging prediction server 30, prediction server 30 also will functions as a key component of
the IEMS.
In Figures 3A to 3C, some inefficiency characteristics (i.e., inefficiency p in L/kWh
versus power P generated by the generator) are shown. The method of the exemplary
embodiment can cope with all these different characteristics and also other monotonically
decreasing characteristics.
The purpose of the method of the exemplary embodiment is to solve Problems I to IV,
especially Problems I and III (i.e., fuel minimization of the PDE generator with a reasonable
number of DG starts within a day), described above in the best possible way under reduced
information to achieve optimal operation of the energy supply system with/without unreliable
grid electric power and/or renewable power generation. The optimal operation is understood
as fuel minimization of generator 3 with PDE and cost minimization under consideration of a
reasonable number of generator starts per day. This optimal operation has been considered
especially difficult in the case where the blackout duration of unreliable grid supply 17 is not
known.
As described below, Problem I (i.e., fuel efficiency) is solved by use of prediction of
load (and generation), and of, especially introduced, a blackout duration probability function to
optimize the charge/discharge sequence of the battery and the generator operation. Since the
reduction of fuel leads to reduction of overall cost, Problem II (i.e., cost minimization) is also
solved. The fuel efficiency and cost minimization is guaranteed by a special optimization
criteria which uses the unnecessary cost (UC), which combines penalizing poor generator
operation (as it occurs in the low efficiency region of the generator), generator starts and
charged battery at the end of the blackout. Since the optimization criteria include the
generator starts, the problem of optimal operation with reasonable number of generator starts
(i.e., Problem III) can be also solved by the exemplary embodiment.
Since the exact blackout occurrence is difficult or impossible to predict (Problem IV),
the exemplary embodiment introduces an optimization method which is based on the blackout
duration probability function, and outputs time variant discharge and charge limits which are
used by a real time control module. For predicting the blackout duration probability function,
the exemplary embodiment uses different available variables and introduces a special
prediction method which is based on a database and artificial intelligence (e.g., ANN (artificial
neural network), SVM (support vector machine), etc.) based predictors whose information is
combined to get the blackout duration probability function.
In this way, the method of the exemplary embodiment relies on different kind of
prediction (i.e., load, renewable, and blackout duration probability function) and a special
optimization method taking into account the properties of the blackout duration distribution.
The output of the optimization method are generally a time variant lower parameter and
especially for a special variation discharging limit and a time variant upper charging limit for
the real time control module which implements a policy considering these parameters.
Due to reduced computing power, it is preferable to use a system based on both local
controller 20 with restricted computing resources and prediction server 20 with ample
computing resources connected with local controller 15 via communication link 18. In
prediction server 30, the computationally heavy tasks of the prediction and optimization can be
carried out. If communication link 18 is weak or not available at all (which implies also no
prediction server), the prediction and optimization functionality is partly or completely carried
out in local controller 20 which asks for more computing power of the hardware on which local
controller 20 is executed.
Next, details of the intelligent energy managements system (IEMS) according to the
exemplary embodiment will be explained.
As described above, the IEMS consists of local controller 20 and, if available,
prediction server 30. Prediction server 30 runs typically on sophisticated hardware (e.g., a
cloud solution) and has the capacity to run sophisticated prediction algorithms demanding
speed and computing power. It is connected with communication link 18 for bidirectional
communication (e.g., Internet, GPRS (General Packet Radio Service), GSM (Global System
for Mobile communications), etc.) with local controller 20. If prediction server 30 is
available, demanding self-learning prediction algorithms on sophisticated hardware are
executed by the prediction server and the results send to local controller 20. If prediction
server 30 is not available, the prediction is realized in local controller 20.
In Figure 4, the simplest construction of prediction server 30 is shown. Prediction
server 30 includes: load prediction module 3 1 predicting the power consumption of aggregated
load 12; and communication module 32 compressing the prediction data and transmitting the
compressed data to local controller 20 via communication link 18. If renewable generation 16
is available, renewable power generation prediction module 33 predicting the generated power
of renewable generation 16 may be integrated to prediction server 30. In the case of grid
supply, blackout duration probability function prediction module 34 predicting the blackout
duration probability function is also integrated to prediction server 30. Figure 5 illustrates
prediction server 30 with renewable power generation prediction module 33 and blackout
duration probability function prediction module 34.
In prediction server 30, the prediction data predicted by modules 31, 33, 34 are
combined and compressed by communication module 32 and sent to local controller 20. How
large the prediction horizon and how the information compression in communication module
32 is carried out, depends on the physical and reliability, availability properties of
communication link 18. Depending on the quality and reliability of communication link 8,
communication module 32 may also contain the optimization method according to the present
exemplary embodiment.
In Figure 6, the construction of local controller 20 is illustrated. Local controller 20
includes: real time control unit 2 1 controlling energy storage 1 1 and generator 13; local
renewable generation prediction module 22 predicting the generated power of renewable
generation 16; local load prediction module 23 predicting the power consumption of
aggregated load 12; local blackout duration probability function prediction module 24
predicting the blackout duration probability function; and local optimization module 25
optimizing the predicted data to send the result to real time control module 21. The
functionality of these three prediction modules 22 to 24 depends on the quality of the
communication link 18. In addition, real time control unit 2 1 includes a function of collecting,
in real time, various measurement values related to the energy supply system.
If communication link 8 is of high quality, local prediction module 22 to 24 and local
optimization module 25 are not needed. In this case, the optimization is carried out in
prediction server 30, and local controller 20 consists basically of real time control module 21.
Hardware requirements for implementation of local controller 20 are low in this case, since all
computationally heavy tasks are carried out in prediction server 30.
If communication link 18 is of low quality and unreliable, local prediction module 22 to
24 are used to correct the long term prediction obtained by server server 30 based on the local
measurements. Based on the corrected prediction, the optimization is carried out by local
optimization module 25. The hardware requirements for implementation of local controller
20 are intermediate in this case.
If there is no communication link 18 (and therefore no prediction server 30), the
prediction functionality is completely realized in local controller 20. Hardware requirements
for implementation of local controller 20 are very high in this case since computationally heavy
prediction algorithms must run on the local controller hardware.
Following TABLES 1 to 4 illustrate task sharing between local controller 20 and
prediction server 30 for various cases based on existence/inexistence of each of renewable
generation 16 and unreliable grid supply 17 in the energy supply system.
TABLE 1: In case of energy supply system consisting of energy storage, load and generator
Type of
Prediction server Local controller
communication link
High quality Load prediction, Real time control
communication link + optimization
Low quality Load prediction, Local load prediction,
communication link + data compression optimization, and real time control
No communication none Local load prediction,
link optimization, and real time control
TABLE 2: In case of energy supply system consisting of energy storage,
load, generator and renewable generation
Type of
Prediction server Local controller
Communication link
Load prediction, and renewable
High quality
power prediction, Real time control
communication link
+ optimization
Load prediction, and renewable Local load prediction, local Low quality
power prediction, renewable power prediction,
communication link
+ data compression optimization, and real time control
Local load prediction, local No communication none renewable power prediction,
link
optimization, and real time control
TABLE 3: In case of energy supply system consisting of energy storage,
load, generator and unreliable grid supply
Again, the use of local blackout duration probability function prediction module 24
characterizes the exemplary embodiment together with local optimization module 25. As
mentioned before, depending on the quality and availability of communication link 18 and the
used hardware, the optimization can be carried out in local optimization module 25 in local
controller 20 or communication module 32 in prediction server 30. However, the
functionality of them is the same. In the exemplary embodiment, the output of the
optimization are two time variant parameters as shown in Figure 7: lower discharge limit p 0w(t)
and upper charge limit phigh(t). Both are inputs for real time control module 25.
In Figure 7, temporal change of SOC (state of charge) of energy storage 1 such as a
Li-ion battery during blackout of the grid is illustrated. Allowable minimum SOC and
allowable maximum SOC of energy storage 1 are indicated by SOCmin and SOCmax,
respectively. According to the optimization of the exemplary embodiment, both limits p io (t)
and Phigh(t) vary in a range between SOC j and SOCma , and the variation range of the SOC of
energy storage 1 is controlled within a region defined by time variable limits pio (t) and
Phigh(t). For example, during the blackout of the grid, energy storage 1 is first set to a
discharge mode (shown by "A" in the figure) and then, when the SOC reaches piow(t), energy
storage 1 is set to a charge mode (shown by "B" in the figure) by starting generator 13.
When the SOC reaches phigh(t), generator 3 is stopped and energy storage 11 is set to the
discharge mode again. The charging and discharging are repeated to constitute
charge/discharge cycles.
Next, the optimization method will be explained in detail.
By using the prediction parameters and a model of the energy supply system, these
parameters can be find by using optimization technology (e.g., simulation based genetic
algorithm based optimization) in order to guarantee optimal operation. Simulation based
optimization allows for considering battery charging and discharging efficiency and AC/DC or
DC/AC conversion losses in the optimization.
The consideration of the blackout duration probability function which is the result of
the prediction needs a special method. As shown in Figure 8, the blackout duration
probability function gives an occurrence probability r(T ) of each blackout having blackout
duration time T . The blackout duration probability function shown in Figure 8 indicates that
occurrence probabilities of blackouts having duration time of 7 hours and 10 hours are 0.5 and
0.5, respectively.
Figure 9 describes the basic principle of the optimization using the blackout duration
probability function by using a simple case to make the concept clear. The underlying
assumption is that, there is no renewable generation, constant load and the probability function
is only different from zero at time t = 7 hours and 10 hours. In both cases, the probability is
50%. So, it is equally probable that the blackout lasts 7 hours or 10 hours. The energy
management according to the exemplary embodiment optimizes the time variant lower
discharging limit p iow(t) and upper charging limit ph t in a way that optimum operation is
achieved regardless of the duration of the blackout. In this case, it can be shown analytically
that this is achieved if the battery is discharged at the end of the blackout which holds true for
the 7 hour-blackout (case 1) and 0 hour-blackout (case 2). The result of the local
optimization of the exemplary embodiment is shown in Figure 9. The dashed double-dotted
line and the dashed dotted line indicate the time variant lower discharging and upper charging
limits p ow(t) and high(t), respectively, resulting from optimization considering the blackout
duration probability function. This can be achieved by using expectation based optimization
or min-max optimization:
Plow t ,Phigh .t
or
max{UC(pl0W t),phigh t),TB }.
Plow t ,Phig t TB
The criterion for efficiency evaluation of the charge/discharge pattern is calculated in
the following way by introducing the unnecessary cost UC as following:
UC = UEEC + GSC + EFC,
where UEEC means Unnecessarily Expensive produced Energy Cost and is related to the
energy EB end in the battery at the end of the blackout in the following way:
UEEC — EB e a,
GSC means Generator Start Cost and is related to the number of generator starts n in the
following way:
GSC = a2 nD,
EFC means Excessive Fuel Cost which is related to a certain reference Diesel engine generator
inefficiency pre in the following way:
EFC = a3 /(p(P(t)) - pref ) P{t)dt,
where
W o x < 0.
Figure 10 illustrates relationship among UC, UEEC, GSC and EFC. Figure 1 is a
graph representing typical change of inefficiency p against change of generated power P and
illustrates relation instantaneous inefficiency (t) and the reference inefficiency parameter pr .
The introduction of the unnecessary cost is necessary; otherwise the particular
optimization approach would always focus on minimizing the cost for long blackouts.
Unnecessary cost means the additional cost that should be avoided by choosing an appropriate
charge/discharge pattern.
With the three parameters a ,a2,a and the reference inefficiency parameter pr f, the
different cost can be weighted and traded off.
The logic behind real time control module 2 1 is shown in Figure 12. The single steps
in the flow chart of Figure 12 are described in detail:
First, in box 101, it is determined whether the grid is available or not. If the grid is
available, the energy storage will be charged to its highest level and the load will be supplied
with energy by the renewable power source (i.e., renewables) in box 102. If the renewable
power source does not suffice or is not available at all, the grid power is also used to achieve
this goal (i.e., energy storage charging, supplying load). If the grid is not available at box 101,
then it is determined, in box 103, whether the renewable power source can supply the load or
not.
If the renewable power source is able to supply the load, it is used to supply the load,
and excessive power is used to charge the energy storage until reaching maximum charge limit
Pmax(t) if possible, in box 104. If the generator with PDE has been turned on at the time of
this action, the generator is turned off. If the renewable power source is not able to supply the
load at box 103, further decision is done in box 105 whether the generator is operating or not.
If the generator is operating at box 105, the next decision in box 106 is based on the question if
the energy storage has reached the time variant upper charging limit Phigh(t) . If the energy
storage has reached the time variant upper charging limit Phigh(t), the generator is turned off,
and by use of renewable power and, if necessary, discharging the energy storage, the load is
supplied with power in box 107. If the energy storage has not yet reached the time variant
upper charging limit Phigh(t) at box 106, the generator supplies the load and charges the energy
storage with maximal rate in box 108.
If the generator is not operating at box 105, the next decision in box 109 is based on the
question if the energy storage can still be discharged or not, in other words if the SOC of the
energy storage is larger than the lower limit piow(t) or not. If the energy storage can be
discharged, renewable power together with battery discharge power is used to supply the load
in box 110. If the energy storage cannot be discharged anymore due to the time variant
discharging limit pi0w(t), the generator is started, the energy storage charged with max rate and
the load supplied with power, in box 111.
The operations from box 101 to box 111 are successively or periodically repeated until
overall system is halted. Through the operation shown in Figure 12, real time control module
2 1 entirely controls the energy supply system to which electric powers are supplied from
various power sources such that cost and fuel optimization based on the predictions is achieved.
Especially, certain priorities are given to the various power sources and the optimization is
carried out by using the priorities. For example, priorities of 1, 2, 3 and 4 are assigned to the
renewables, the grid, the energy storage and the generator, respectively.
Generally, the time variant discharge rate is a consequence of the load to feed and, if
available, power generated at the renewable power source such as a PV generation. The
charge rate resulting from the load to feed and the energy storage charging power which for
efficiency is set at highest rate that does not compromise battery life.
An example of blackout duration probability function prediction module 34 is
illustrated in Figure 13. The inputs to module 34 are variables such as month, day of the
week, hour, outside temperature, season and other parameters which describe the current
environmental situation of the grid and concrete site to be considered. Blackout duration
probability function prediction module 34 includes: various and at least two artificial
intelligence based predictors (AI predictors) 4 1 predicting specific blackout durations with an
indication of the reliability of the prediction; database 42 storing historical data necessary for
predicting the blackout durations; and distribution generator 43. Artificial intelligence based
predictors 4 1 may be, for example, an artificial neural network (ANN), support vector machine
(SVM), and so on. With the results of prediction by artificial intelligence based predictors 4 ,
distribution generator 43 creates a blackout duration probability function which is used for
optimization as explained before.
In Figure 14, a practical example of the application of the present exemplary
embodiment is illustrated in form of an energy management system for energy supply of a base
transceiver station (BTS) for a rural site in a developing country. The illustrated system
includes: Li-ion battery 5 as an energy storage, load 52 including BTS equipment and air
conditioning; Diesel engine generator 53 as an energy generator; physical connection apparatus
55; PV generation 56 as a renewable generation; unreliable grid supply 57 such as one in a
rural area in a developing country with blackout(s) everyday; and prediction server 60. In this
example, prediction server 60 is implemented by using cloud solution. Therefore, the
functionality of prediction server 60 may be realized by various resources in the cloud which
are available via network accesses. Diesel engine generator 53 supplies DC power and has
PDE characteristic shown in Figure 3C. Local controller 54 which is implemented by an
industrial PC (personal computer) is arranged in physical connection apparatus 55. Local
controller 54 and prediction server 60 are connected with each other via communication link 58
which is, for example, a GPRS channel.
Figure 15 illustrates in detail the system shown in Figure 14. Physical connection
apparatus 55 includes: DC bus 6 1 for distributing DC power; AC bus 62 for distributing AC
power; DC/AC converter 63 arranged between DC bus 6 1 and AC bus 62; and AC/DC
converter 64 arranged between DC bus 6 1 and AC bus 62. Li-ion battery 5 1 is connected to
DC bus 6 1 via charger/discharger 65 which includes a DC/DC converter. PV generation 56
outputting DC power is connected to DC bus 6 1 via MPPT (maximum power point tracking)
unit 66. Diesel engine generator 53 and unreliable grid supply 57 are connected to AC bus 62.
Load 52 includes: BTS unit 1 with constant DC power consumption; free cooling unit
72 which consumes DC power depending on the temperature; and air conditioning 73 which
consumes AC power depending on the temperature. BTS unit 7 1 and free cooling unit 72 are
connected to DC bus 6 1 while air conditioning 73 is connected to AC bus 62.
Local controller 54 gives commands as charge/discharge power to charger/discharger
65 to control charging and discharging of Li-ion battery 57. Local controller 54 gives Diesel
generator on/off commands to Diesel engine generator 53. The concrete realization of the
charge/discharge power is done by setting the voltage of DC bus 61. This voltage control is
realized by other well-known methods which are not explicitly explained, since they are not
part of the present invention.
Prediction server 60 implements the basic prediction functions for load prediction
based on weather prediction and modelling of the air conditioning, PV generation prediction
and the blackout duration probability function prediction. The prediction data obtained by
prediction server 60 is transferred from time to time to local controller 54. Local controller
54 uses the transferred data to compute ideal discharge and charge limits for the real time
control module. In addition, from time to time data is transferred from local controller 54 to
prediction server 60 to to improve the quality of the prediction by prediction server 60 by selflearning.
Simulation of operation of the system shown in Figures 14 and 15 was carried out to
show the validity of the exemplary embodiment. Figures 16A to 16H show the results for the
various simulation items. Operation within a period of 24 hours, i.e., an entire day, was
simulated. In the simulation, it was assumed that blackout occurs from 03:00 AM to 09:00
PM as shown Figure 16H. The air conditioning, i.e., AC load, is operated from around 06:00
AM to around 08:00 PM. Due to availability of load prediction (Figure 16G), PV generation
prediction (Figure 16H) and blackout duration probability function, the optimization method is
able to compute optimal lower discharging limit pio (t) and higher charging limit Phigh(t) as
shown in Figure 16A. These two limits piow(t) and phi h( are input to the real time control
module which controls the charging/discharging of the battery and the diesel engine operation.
The resultant SOC change of the Li-ion battery, charging/discharging power, Diesel generator
power and grid supplied power are shown in Figure 16A to 16D, respectively. Cost can be
calculated from the Diesel generator power, grid supplied power and the number of the
generator starts, and the results of cost calculation are shown in Figures 16E and 16F.
As describe above, the intelligent energy management system (IEMS) according to the
exemplary embodiment allows for optimal or near optimal operation of the energy supply
system under reduced information by use of the prediction technology, especially the new
prediction method of the blackout duration probability function prediction to deal with
unplanned blackouts, and the special optimization method computing the lower and upper
limits piow(t) and Phigh(t) - The optimization method to include the blackout duration
probability function uses a special optimization criterion in order to gain robust results and
achieving fuel and cost optimality under consideration of a reasonable number of generator
starts over the different highly probable blackout durations. The specific procedure of the
control according to the exemplary embodiment is shown in Figure .
The difference between the exemplary embodiment and the related-art method
described in [NPL2] depends on different parameters: renewable generation pattern, blackout
pattern, load pattern, energy storage parameters.
The most important advantageous effect of the exemplary effect is the reduction of the
used fuel of the PDE generator. At the same time, a reasonable number of PDE generator
starts can be guaranteed. The fuel reduction leads to a cost reduction. In the case of the
application of the exemplary embodiment to the energy management system for BTS energy
supply, it leads to better operational cost of the BTS, i.e., reduced operating expense (OPEX),
and reduced environmental impact due to reduced fuel consumption which means less
production of carbon dioxide (C0 2) exhaust gas.
These advantageous effects results from the features of the exemplary embodiment that
the enhanced prediction capabilities of, for example, load, generation and blackout duration
probability function are introduced, the special optimization and control methods are
introduced, and the charge/discharge pattern and the generator operation is chosen in a way that
the battery is mostly discharged at the end of the blackout, the generator is operated as much as
possible near to its optimal operation mode and at the same time the number of generator starts
is kept reasonably small. The feature that the battery is mostly like discharged at the end of
the blackout is advantageous from cost and fuel consumption perspective. The small number
of generator starts should reduce wear of the generator. The exemplary embodiment further
uses intelligent methods to exploit the reduced information by using self-learning prediction
technology.
Industrial Applicability
The invention can be used for energy supply of various systems including: for example,
base transceiver stations in telecommunication industry; and a small micro grid with a battery,
load(s) and a generator, and, if available, an additional equipment such as renewable power
source and/or unreliable grid supply, in energy supply system industry.
The invention as described herein may comprise one, several, all, or any of the
exemplary embodiments provided above in any combination. The above-described
embodiments of the present invention are intended to be examples only. Alterations,
modifications and variations can be effected to the particular embodiments by those of skill in
the art without departing from the scope of the invention, which is defined solely by the claims
appended hereto.
Reference Signs List
1 1 Energy storage;
12, 52 Load;
13 Generator;
15, 55 Physical connection apparatus;
16 Renewable generation;
17, 57 Unreliable grid supply;
18, 58 Communication link;
20, 54 Local controller;
2 1 Real time control module;
22 Local renewable generation prediction module;
23 Local load prediction module;
24 Local blackout duration probability function prediction module;
25 Local optimization module;
30, 60 Prediction server;
3 1 Load prediction module;
32 Communication module;
33 Renewable generation prediction module;
34 Blackout duration probability function prediction module;
4 1 AI predictor;
42 Database;
43 Distribution generator;
Li-ion battery;
Diesel engine generator;
PV generation;
DC bus;
AC bus;
DC/AC convertor;
AC/DC convertor;
Charger/discharger;
MPPT unit;
BTS unit;
Free cooling unit;
Air conditioning.

CLAIMS
[Claim 1]
An energy management method for an energy supply system which includes at least an
energy storage, a load and a generator with power dependent efficiency, the method
comprising:
calculating two time variant parameters indicating a discharge lower limit and an upper
charge limit, respectively, of the energy storage, based on optimization using different kinds of
prediction; and
controlling, in a real time manner, charging and discharging of the energy storage and
operation of the generator, with a certain priorities given to various power sources, such that
state of charge of the energy storage is controlled within a region between the discharge lower
limit and the upper charge limit.
[Claim 2]
The method according to claim 1 further comprising:
when a grid power is available to the energy supply system, considering a possible
duration of blackout in the grid power to predict a blackout duration possibility function,
wherein the two time variant parameters are calculated by using the blackout duration
possibility function.
[Claim 3]
The method according to claim 2, wherein the two time variant parameters are
calculated using unnecessary cost criteria including an unnecessary expensive produced energy
cost, a generator start cost and an excessive fuel cost.
[Claim 4]
The method according to claim 2 or 3, wherein the prediction of the blackout duration
probability function comprises:
predicting specific blackout durations by different artificial intelligence based
predictors, each specific blackout duration accompanying an indication of reliability of
prediction; and
creating the blackout duration probability function based on results of duration
prediction of different artificial intelligence based predictors.
[Claim 5]
The method according to any one of claims 1 to 4, comprising:
when a remote server with relatively high computing power and a communication link
between a site of the energy supply system and the remote server are available and quality of
the communication link is relatively high, performing the prediction and the optimization on
the remote server, and performing the controlling in the real time manner on a local controller
with relatively low computing power provided on the cite of the energy supply system;
when the remote server is available and quality of the communication link is relatively
low, performing the prediction on the remote server, and, on the local controller with an
intermediate computer power, correcting result of the prediction and performing the
optimization and the controlling in the real time manner; and
when at least one of the remote server and the communication link is unavailable,
performing the prediction, the optimization and the controlling in the real time manner on the
local controller with relatively high computing power.
[Claim 6]
An energy management system for an energy supply system which includes at least an
energy storage, a load and a generator with power dependent efficiency, the energy
management system comprising:
prediction means for performing different kinds of prediction;
optimization means for calculating two time variant parameters indicating a discharge
lower limit and an upper charge limit, respectively, of the energy storage, based on
optimization using different kinds of prediction; and
control means for controlling, in a real time manner, charging and discharging of the
energy storage and operation of the generator, with a certain priorities given to various power
sources, such that state of charge of the energy storage is controlled within a region between
the discharge lower limit and the upper charge limit.
[Claim 7]
The energy management system according to claim 6, wherein the prediction means
includes, when a grid power is available to the energy supply system, calculation means for
considering a possible duration of blackout in the grid power to predict a blackout duration
possibility function,
wherein the two time variant parameters are calculated by using the blackout duration
possibility function.
[Claim 8]
The energy management system according to claim 7, wherein the two time variant
parameters are calculated using unnecessary cost criteria including an unnecessary expensive
produced energy cost, a generator start cost and an excessive fuel cost.
[Claim 9]
The method according to claim 7 or 8, wherein the calculation means includes:
different artificial intelligence based predictors predicting specific blackout durations,
each specific blackout duration accompanying an indication of reliability of prediction; and
a distribution generator creating the blackout duration probability function based on
results of duration prediction of different artificial intelligence based predictors.
[Claim 10]
The energy management system according to any one of claims 6 to 9, further
comprising a local controller provided on a cite of the energy supply system,
wherein, when a remote server with relatively high computing power and a
cornmunicatiori link between the local controller and the remote server are available and
quality of the communication link is relatively high, the prediction means and the optimization
means are arranged on the remote server, and the control means is arranged on the local
controller with relatively low computing power,
when the remote server is available and quality of the communication link is relatively
low, the optimization means and the control means are arranged on the local controller with an
intermediate computer power while the prediction means is arranged on both the remote server
and the local controller, the prediction means on the local controller correcting result by the
prediction means on the prediction server, and
when at least one of the remote server and the communication link is unavailable, the
prediction means, the optimization means and the control means are arranged on the local
controller with relatively high computing power.