Escolar Documentos
Profissional Documentos
Cultura Documentos
k (MW)
Abstract — In an isolated power system (rural microgrid), energy capacity of BESS (kWh)
distributed energy resources (DERs), such as renewable energy η efficiency of BESS
resources (wind, solar), energy storage and demand response, can minimum SOC of BESS
be used to complement fossil fueled generators. The uncertainty maximum SOC of BESS
and variability due to high penetration of wind makes reliable ( ) threshold used as control input to compensate for
system operations and controls challenging. In this paper, an wind or net load variability
optimal control strategy is proposed to coordinate energy storage ( ) actual load power at time step k
and diesel generators to maximize wind penetration while ( ) actual wind power at time step k
maintaining system economics and normal operation
( ) predicted power of diesel generator i at time
performance. The problem is formulated as a multi-objective
optimization problem with the goals of minimizing fuel costs and
step k
changes in power output of diesel generators, minimizing costs ( ) predicted SOC of BESS at time step k
associated with low battery life of energy storage, and ( ) predicted threshold value
maximizing the ability to maintain real-time power balance ( ) forecasted load power at time step k
during operations. Two control modes are considered for ( ) forecasted wind power at time step k
controlling the energy storage to compensate either net load set of all wind generators
variability or wind variability. Model predictive control (MPC) is
used to solve the aforementioned problem and the performance is I. INTRODUCTION
compared to an open-loop look-ahead dispatch problem under
high penetration of wind. Simulation studies using different
prediction horizons further demonstrate the efficacy of the
closed-loop MPC in compensating for uncertainties in the system
I SOLATED power systems are typically small distribution
systems in remote areas, which lack support from larger
interconnected power grids. In these systems, electricity is
caused by wind and demand. often supplied by small fossil fueled generators that tend to be
very expensive to operate. Integrating distributed energy
Index Terms— model predictive control, coordination of resources (DERs), such as renewable resources and energy
distributed energy resources storage, can allow for economical and environmentally
friendly operation. However, there is significant variability
NOMENCLATURE and uncertainty associated with high penetration of renewable
( ( )) fuel cost for diesel unit i at time step k ($) resources like wind and solar. Energy storage devices have
( ) scheduled output level of diesel generator i at inter-temporal constraints associated with their operation, and
time step k (kW) it can be difficult to predict the state of charge (SOC) during
minimum rated power of generator i operation of some energy storage devices. Due to these
maximum rated power of generator i inherent characteristics of wind and energy storage, real-time
maximum ramp rate of generator i operations and control coordination becomes challenging.
set of all diesel generators Many centralized/decentralized control strategies have been
( ( ) cost of operating the Battery Energy Storage and are being developed to integrate DERs in power system
System (BESS) operations. Examples of control strategies already proposed
( ) State of Charge of BESS at time step k and/or developed such as the ‘Grid Friendly Appliance’
penalty factor on low State of Charge (SOC) technology (decentralized) is given in [1]. A decentralized
reference state of charge droop control is added to disaggregated loads using quasi-
continuous control law to have a desired aggregated response
( ) BESS charge/discharge power level at time step
for frequency and stability control in [2]. In [3], a
decentralized control of voltage profile is proposed in the
This work was supported by the Laboratory Directed Research and distribution system with DGs using reactive power control of
Development (LDRD) program at the Pacific Northwest National Laboratory.
inverters. A centralized AGC-type control of DGs is proposed
E. Mayhorn and K. Buttler-Purry are with Texas A&M University, College in [4]. A combination of centralized and decentralized
Station, TX. coordination strategies for a rural microgrid, containing wind
K. Kalsi, M. Elizondo, S. Lu and N. Samaan are with Pacific Northwest and diesel generators, BESS, and demand response, were
National Laboratory, Richland, WA 99354 USA.
W. Zhang is with the Ohio State University, Columbus, OH studied in [5]. The objectives for the coordination strategies
were to maintain system frequency close to nominal and to
reduce fossil fuel generator movement by allowing energy
storage devices to compensate wind variability. Arbitrary diesel generators and energy storage are dispatched based on
control inputs were selected only to show the effectiveness of wind and load forecasts over an entire horizon. The
the control coordination strategies. The authors recognized the optimization problem is formulated as follows:
need for an optimally coordinated control scheme between ( ), ( ) ∑ ∑ ( ) +∑ ( ( )) (1)
different DERs. subject to
Several coordination strategies of DERs, to provide ( )+∑ ( )+ ( )= ( )
∑ (2)
ancillary services (i.e., scheduling, dispatch, balancing,
( )= ( − 1) − ( − 1) (3)
contingency response, etc.) have been explored in [6]-[11]. In
[12], an energy management system is proposed that is ≤ ( )≤ , = 1, 2, … (4)
divided into several modules: forecasting, energy storage | ( + 1) − ( )| ≤ , = 1, 2, … (5)
management, and an optimization module. The optimization ≤ ( )≤ (6)
module performs day-ahead unit commitment that uses The above constraints (2-6) are calculated for = 1, … , ,
information from load and distributed generation (DG), power where is the length of the prediction horizon. The fuel cost
forecasting module, market information, and energy storage of each generator ( ( )) is assumed to be linear and is
management system to economically allocate generation in a given by
microgrid. A power management strategy for wind-diesel- ( ) = + ( ) (7)
BESS systems is presented in [13]. Diesel and energy storage where, , are the fuel cost coefficients. The cost associated
power setpoints are dispatched, using day-ahead wind and with operating the BESS, ( ( ), is given by the following
load forecasts, to minimize diesel generator operating costs, as expression which is adapted from [13]:
well as, costs related to battery lifetime. A conceptual idea for ( ( ) = ( ) (8)
multi-stage economic load dispatch in island microgrids is In (3), is a constant given by = η⁄( ) ∆ where, ∆
presented in [14]. To address the issues of variability and is the time step duration (hr). The objective function defined in
uncertainty, in [15]-[18], a model predictive control (MPC) (1) is convex, and hence, any standard quadratic programming
approach is introduced. The strategy is based on dispatching solver can be used to obtain the optimal solution. The decision
power at minimal cost, assuming that energy storage is not variables in the optimization problem are the power setpoints
available, that renewable sources are dispatchable, and that of energy storage and generators. The basic power balance
only short term wind forecasts are reliable. equation is given by (2), which must be satisfied at every time
In this work, a centralized MPC based coordination strategy step over the prediction horizon. The evolution of the state of
is proposed for dispatch of DERs in an isolated system. One charge at every time step is given by (3). Furthermore, at
key difference between this work and that proposed in [15]- every time step, the current state of charge is a function of the
[18] is that performance objectives are incorporated in state of charge of the previous time step, the storage
addition to economics. The goal of this work is to maximize charge/discharge power, and the energy capacity. The output
the amount of wind generation in the system while considering power of the generators and state of charge of the storage are
system economics and the individual controls of the DERs. constrained with the limits defined in (4), (5) and (6).
This can be done by formulating a look-ahead dispatch
problem and casting it in a multi-objective framework. The III. OPTIMAL CONTROL OF DISTRIBUTED RESOURCES USING
objectives are to: minimize fuel costs of diesel generators, MODEL PREDICTIVE CONTROL
minimize changes in power output of diesel generators
The look-ahead dispatch problem discussed earlier has
(reducing wear and tear), minimize costs associated with low
inherent increased uncertainty with high penetration of
battery life of energy storage, and to maximize the ability for
renewable energy resources in the system and is implemented
generators to provide real-time balancing. Two control modes
in an open-loop manner. The optimization problem is solved
are adopted depending on whether the energy storage system
over an entire horizon once and the resulting sequence of
used to compensate for wind or net load variability.
control inputs are implemented at the corresponding time
Simulation studies are used to evaluate the performance of the
steps. Even though day-ahead forecasts for load demand are
different control strategies and to demonstrate the
reliable, day-ahead forecasts for wind are not. One possible
effectiveness of the closed loop MPC in compensating for
technique to solve this problem is to use MPC, where at every
uncertainties in wind and load forecasts.
step a finite horizon optimal control problem is solved using
This paper is organized as follows. In Section II, a brief
feedback from the system. However, the control sequence is
description of the standard look-ahead dispatch problem is
implemented for only one step ahead. In this manner, MPC is
given. An optimal control coordination scheme using MPC is
considered closed-loop and has the ability to compensate for
presented in Section III. In Section IV, case studies are
additional uncertainty in demand variability caused by high
presented that demonstrate the effectiveness of the optimal
penetration of renewable energy resources. The MPC based
control coordination strategy. Finally, conclusions are given in
optimal control problem can be viewed as a multi-objective
Section V.
optimization problem with goals to: 1) minimize fuel costs of
diesel generators, 2) minimize changes in power output of
II. CLASSICAL DISPATCH PROBLEM FORMULATION
diesel generators reducing mechanical wear and tear, 3)
In a typical dispatch formulation with conventional minimize costs associated with low battery life of energy
generation, wind generation, and BESS, the objectives are to: storage, and 4) minimize the inability of isochronous
1) minimize fuel costs of diesel generators and 2) minimize generators to provide real-time balancing. Isochronous control
operating costs of energy storage. The power outputs of the
3
= 1 0 0 0 1 0 0 0 0
< ⇒ = − 0 1 0 0 0 1 0 0 0
= = =
= 0 0 1 0 0 0 1 0 0
≥ ⇒ =0 0 0 1 0 0 0 −
When there is excessive wind generation ( < ), = [0 −1 −1 0] = [0 0] (28)
the dump load takes the balancing power. The dump load The choice of the disturbance prediction models, , is also
represents wind curtailment applied in real-time to maintain very important [20]. Autoregressive integrated moving
generator 1 balancing control. This dump load is useful to average (ARIMA) and seasonal ARIMA (SARIMA) models
quantify the performance of the different coordination were used for the prediction models of wind and load,
strategies being compared; the control strategy that minimizes respectively. These univariate time series models allow
the dump load is making better use of the wind resource. forecasted values to be calculated as a linear function of
Remark: Notice that there is a difference between the power previous values. The Box-Jenkins [24] approach was used to
balance of the actual system and the power balance constraint select the parameters and orders of the models and to evaluate
used in optimization (either classical or MPC). In optimization model adequacy.
the power balance constraint is met for forecasted values of Wind disturbance is modeled as an autoregressive integrated
load and wind generation. In the actual system operation, there moving average (ARIMA) model
is a mismatch due to forecast errors that will be picked up by 1−∑ (1 − ) ( )= 1−∑ ( ) (29)
the generator/s in charge of system balance (e.g. by secondary where p,d, and q are the identified orders of the autoregressive
frequency control). Therefore the actual power output of these (AR), integrated (I), and moving average (MA) parts,
generators will deviate from the optimal scheduled values. In respectively. is a lag operator, are the parameters of the
the real time operation, if the generator/s in charge of AR part, are the parameters of the moving average part,
frequency control reaches their minimum or maximum limits, and ( ) is the error term.
frequency control can be lost and the system could exhibit Since the load demand has a seasonal pattern, a seasonal
frequency deviations (poor performance) as shown in [5] or ARIMA (SARIMA) model was used for the disturbance
frequency instability. In this paper we avoid generator 1 to model
reach the minimum output (avoiding loss of balancing control) 1−∑ (1 − ) (1 − ∑ )(1 − ) ( )=
by using a dump load to represent wind curtailment. 1−∑ 1−∑ ( ) (30)
Predictive Model where P,D,Q are the identified orders of the seasonal AR, I,
and MA parts, respectively. is the seasonal lag operator,
In the predictive model, actual values at = 0 are measured
is the seasonal period, are the parameters of the seasonal
from the system and used as for initialization. States and
AR part, are the parameters of the seasonal moving
disturbance predictions are made for times = 1, … − 1.
average part.
Hence, for the case when storage compensates for wind
variability, the following dynamics of predictive values are Optimizer
defined: The look-ahead dispatch problem for optimal coordination of
( + 1) = ( )+∆ ( ) (22) DERs can be re-formulated as a multi-objective optimization
( + 1) = ( )+∆ ( ) (23) problem with the following cost function
( + 1) = ( ) + ∆ ( ) (24) =∑ ( )− + ( )+
( + 1) = ( )− ( ) − ( ) (25)
( ) + ( )− + ∆ ( )+
( )=− ( )− ( )+ ( ) (26)
where (.̂ ) are the predicted values of the quantities defined in ∆ ( ) (31)
(15) – (17). Hence, the predicted state variables, control input where penalizes the movement of the generator, in charge
and disturbance vectors can be defined as of real-time balancing, from a reference value to discourage
( ) minimum and maximum output values; is the penalty
∆ ( )
( ) ( ) associated with fuel costs of diesel generators; is the
= , = ∆ ( ) and = penalty associated with low battery life; and is the weight
( ) ( )
∆ ( ) penalizing the movement associated with the diesel
( )
generators. The conflicting objectives are to minimize the cost
and the predicted output is taken to be = ( ). The
associated with maintaining the real-time balancing control,
dynamics (22) - (26) can be re-written in state space form the movement of energy storage SOC, and the diesel generator
= + + movements and fuel costs. Rewriting (31) in terms of the state
= + variables and control inputs of the predictive model gives
1 0 0 0 1 0 0 0 0 ( , ) = ∑ [( − ) ( − ) + + ] (32)
0 1 0 0 0 1 0 0 0 where
= = =
0 0 1 0 0 0 1 0 0 0 0 0
0 0 − 1 0 0 0 0 0 0 0 0 0
= [0 −1 −1 0] = [1 0] (27) = = =
0 0 0 0 0 0
When the energy storage unit is used to compensate for net 0 0 0
0
load variability, the above matrices are modified to
5
wear and tear of resources? The weights in this work are 3500
Actual Load
Actual High Wind Power
chosen arbitrarily by assuming the relative importance is 3000 Actual Low Wind Power
Power (kW)
about the preferences of the decision maker and that can 1500
0
0 5 10 15 20 25
Time (hr)
IV. SIMULATION STUDIES Fig. 2 Total Load and Wind Production
A. Description of Test Cases
The test system consists of one diesel generator rated at
4MW, a diesel generator rated at 2.5 MW, a 3.6MWh rated
BESS, a wind power plant (data obtained from [23]), dump
load and an aggregate load of 1500 houses. System losses are
neglected. One generator is in charge of real-time system
balance, compensating for wind and load uncertainty not
covered by the dispatch algorithms. The other generator
operates at the given set points defined by the dispatch
algorithm. The BESS can operate in two different control
modes to compensate for: 1) net load variability or 2) wind
variability given a threshold set point.
6
Load Demand Furthermore, the BESS neither fully charges nor fully
6000
Actual
discharges.
Power (kW) Predicted Generator 1 Power Output
4000
4000
Power (kW)
Actual
2000 Predicted
2000
0
0 5 10 15 20 25 0
0 5 10 15 20 25
Time (hr)
Time (hr)
Wind Power
4000 (a)
Generator 2 Power Output
Power (kW)
Actual
3000 1000
Power (kW)
Predicted Actual
2000 Predicted
500
1000
0
0 5 10 15 20 25
0
0 5 10 15 20 25 Time (hr)
Time (hr) (b)
State of Charge (SOC)
Fig. 3 Load and Wind Forecasts (24hr prediction horizon) 1
Actual
SOC
Predicted
Load Demand 0.5
6000
Power (kW)
Actual
0
4000 Predicted 0 5 10 15 20 25
Time (hr)
2000 (c)
Fig. 5 Response to open loop dispatch when BESS compensates wind
0
0 5 10 15 20 25 variability under high wind (a) Isochronous generator power output (b) power
Time (hr) output of diesel generator 2 (c) BESS state of charge
Wind Power
4000 Generator 1 Power Output Actual
1500
Power (kW)
Actual
3000 Power (kW) Predicted
Predicted
1000
2000
1000 500
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (hr) Time (hr)
Fig. 4 Load and Wind Forecasts (1hr prediction horizon) (a)
Generator 2 Power Output
3000
Power (kW)
Actual
B. Compensating for Wind Variability 2000 Predicted
Actual
0.8 Predicted
(performing real-time balancing) is very different from the
0.6
predicted power output (Fig. 5a). Generator 1 balances power
by compensating for large differences in the forecasted wind 0.4
0 5 10 15 20 25
power and load demand. Generator 2 follows the set points Time (hr)
given as expected (Fig. 5b). The actual SOC of the BESS is (c)
Fig. 6 Response to MPC when BESS compensates wind variability under high
also very different from the predicted values (Fig. 5c). The wind (a) Isochronous generator power output (b) power output of diesel
BESS follows the actual wind variability until it discharges generator 2 (c) BESS state of charge
completely which is the point at which generator 1 supplies The performance index (normalized total cost) is based on
the additional power needed. the specific weights chosen to for each objective defined in
The responses of the different DERs for the closed-loop (31). For these studies, the following set of weights is chosen:
MPC case are shown in Fig. 6. The actual power output of w1=0.2, w2=0.35, w3= 0.15 and w4=0.3. Calculating cost as
generator 1 is, on average, closer to the predicted power defined by equation (31), the open loop cases shown have
output at each time step (Fig. 6a). This is because the much higher costs (~three times greater), than the closed loop
difference between forecasted wind and load is much less in costs, as shown in Fig. 7a. The amount of dump load for the
this case. Furthermore, unlike in the open loop case, generator open and closed loop strategies is compared in Fig. 7b. The
1 meets its control objective of balancing power and oscillates energy of the dumped load is much less in the closed loop case
around the given reference value. Generator 2 follows the as compared to the open loop case, indicating that wind power
optimal set points given as expected (Fig. 6b). In contrast to resource is better utilized in the closed loop MPC coordination
the open loop case, the BESS actual SOC matches closely to strategies.
the predicted values (Fig. 6c). This is because the forecasted
wind and load deviate less from their actual values.
7
Power (kW)
Dump Load - Wind Variability Cases
40 700
500
35 600 Actual
Predicted
0
30 0 5 10 15 20 25
500
Time (hr)
Normalized Cost
25
Energy (kWh)
400 (a)
20 Generator 2 Power Output
3000
Power (kW)
300 Actual
2000
15 Predicted
1000
200
10 0
-1000
5 100 0 5 10 15 20 25
Time (hr)
0 0 (b) ( )
Open Loop Closed Loop Open Loop Closed Loop
Case Case State of Charge (SOC)
1
(a) (b) Actual
SOC (kW)
Fig. 7 Normalized total costs and dump load energy when BESS compensates 0.8 Predicted
Actual
2000
Predicted variability case indicating that the net load variability
coordination strategy is more effective.
0 Performance Index - Net Load Variabilty Cases Dump Load - Net Load Variabilty Cases
0 5 10 15 20 25 40 700
Time (hr)
(a) 35 600
Generator 2 Power Output 30
1000 500
Power (kW)
Actual
Normalized Cost
Energy (kWh)
Predicted 25
500 400
20
300
0
0 5 10 15 20 25 15
Time (hr) 200
10
(b)
State of Charge (SOC) 5 100
1
0 0
Open Loop Closed Loop Open Loop Closed Loop
SOC
Normalized Cost
10
Induction Generator,” IEEE Trans. on Energy Conv, vol. 24, no. 1, pp.
137-145, March 2009.
5 [8] J. A. Peças Lopes, C. L. Moreira and A. G. Madureira, “Defining control
strategies for microgrid islanded operation,” IEEE Trans. on Power
0
4 6 9 12 24
Syst., vol. 21, no. 2, pp. 916-924, May 2006.
Prediction Horizon (Time Steps) [9] A. Mehrizi-Sani and R. Iravani, “Potential-function based control of a
microgrid in islanded and grid-connected modes,” IEEE Trans. on
(a) Power Syst., vol. 25, no. 4, pp. 1883-1891, November 2010.
Dump Load [10] R. Majumder, B. Chaudhuri, A. Ghosh, R. Majumder, G. Ledwich and
600 F. Zare, “Improvement of stability and load sharing in an autonomous
Energy (kWh)