Master–slave Based Hierarchical Control for a Small Power Dc-distributed Microgrid System with a Storage Device

In this paper, we analyze one of the main drawbacks of droop control-based DC microgrid systems, and propose a novel control method to overcome this problem. Typically, DC microgrid systems use droop control techniques to enable communication independency and expandability. However, as these advantages are based on bus quality and regulation abandonment, droop-based schemes have limitations in terms of high bus impedance and bus regulation. This paper proposes a novel master–slave based hierarchical control technique for a DC distribution system, in which a DC bus signaling method is used to overcome the communication dependency and the expandability limitations of conventional master–slave control methods. The concept and design considerations of the proposed control method are presented, and a 1 kW simulation under a Powersim (PSIM) environment and hardware prototype—built to verify the system—is described.


Introduction
Technological development and environmental protection concerns have led to the introduction of distributed energy resources (DERs)-such as gas-and wind-turbines and photovoltaic (PV) energy source-into microgrids [1][2][3][4][5].Furthermore, the growth of PV into the most popular renewable energy source, as a result of its installation convenience, has helped to further the concept of the DC distribution system [6][7][8][9][10].Not only do such systems bring the advantages of PV-to-DC grid energy conversion efficiency, but they also allow for the utilization of storage devices and digital loads [11].For these reasons, DC microgrids have been studied under various application scenarios in an effort to find more efficient functionalities, structures, and control strategies.
As the technique for interfacing multiple energy sources at a DC distribution bus, the droop-based control method is widely used for energy management and bus voltage control [12][13][14][15][16][17][18][19][20][21][22][23][24][25].Droop controls have many advantages, including communication independency and expandability, therefore, they can be widely used in microgrid environments requiring stable and reliable control.However, these benefits come at the cost of decreased bus quality owing to voltage regulation and transient response on the bus and, accordingly, incur high bus impedance [14][15][16][17]26].These problems may result in difficulties of overall system optimization in terms of efficiency and stability.
Master-slave controls have many advantages in terms of bus quality and control, but face a critical drawback in terms of a communication link requirement.A conventional master-slave control depends on the power sharing and control ability of a communication link, and thus needs a high-speed communication link to a control system simultaneously [18,19].Moreover, any communication delay or omission can reduce bus quality and system stability in general.This paper proposes a novel master-slave control method that eliminates the need for a communication link.The proposed system utilizes the DC bus signaling (DBS) method [20] to replace additional communication, and controls the input power of device-stored energy instead of a load condition.A detailed concept, analysis, and design consideration are presented, and a 380 V, 1 kW simulation under a Powersim (PSIM) environment and hardware prototype were built to verify the suitability of the proposed system.This system is intended for use in DC home applications such as net zero energy houses [27,28] or nanogrid systems.Although the system has a small power capacity, it has the similar structure as a large-scale microgrid system, allowing for simplified system analysis and verification.As bus voltage regulation is more difficult with a single-phase power factor correction (PFC) circuit, thus such a circuit is used here to demonstrate the effectiveness of the proposed control method.

Structure and Operational Principles of the Proposed Master-Slave Method
Figure 1 shows the structure of the proposed DC distributed microgrid system.The DER consists of a PV panel, utility grid, battery, and load, and is connected to a DC bus using a renewable interface unit (RIU), grid interface unit (GIU), storage interface unit (SIU), and a load converter.Detailed parameters and system design are presented in Section 4.
Energies 2016, 9, 880 2 of 14 This paper proposes a novel master-slave control method that eliminates the need for a communication link.The proposed system utilizes the DC bus signaling (DBS) method [20] to replace additional communication, and controls the input power of device-stored energy instead of a load condition.A detailed concept, analysis, and design consideration are presented, and a 380 V, 1 kW simulation under a Powersim (PSIM) environment and hardware prototype were built to verify the suitability of the proposed system.This system is intended for use in DC home applications such as net zero energy houses [27,28] or nanogrid systems.Although the system has a small power capacity, it has the similar structure as a large-scale microgrid system, allowing for simplified system analysis and verification.As bus voltage regulation is more difficult with a single-phase power factor correction (PFC) circuit, thus such a circuit is used here to demonstrate the effectiveness of the proposed control method.

Structure and Operational Principles of the Proposed Master-Slave Method
Figure 1 shows the structure of the proposed DC distributed microgrid system.The DER consists of a PV panel, utility grid, battery, and load, and is connected to a DC bus using a renewable interface unit (RIU), grid interface unit (GIU), storage interface unit (SIU), and a load converter.Detailed parameters and system design are presented in Section 4. The primary purposes of a microgrid system are (1) to effectively operate and manage the DER, and (2) to maintain a stable bus voltage for supplying energy to the load.To satisfy these conditions, the master unit controls the bus voltage and slave units in order to supply power to the bus.Under a conventional control scheme, the slave units must be able to sense power-sharing information from the master unit or central controller.Using the DBS method, it is possible to eliminate the need for such communication links in the master-slave control; instead, the bus voltage is controlled using a predetermined DBS curve to change the power-sharing ratio of each interface unit.Thus, the master unit controls bus voltage using energy stored in a battery that is applied by the slave units using bus voltage information.
Figure 2 shows a control flow chart of the proposed system and key operation of the DBSadapted communication-less master-slave control method.Figure 2a is a simplified control flow chart.The controllers are separated by a master unit and slave units, and the master unit controls bus voltage using the battery state-of-charge (SoC) level [22].The slave units can be classified as controllable slave unit and uncontrollable slave unit.The renewable energy sources (e.g., PV panels) are used to control input voltage rather than bus voltage in order to generate maximum power, which eliminates arbitrary input power control.Thus, renewable energy sources requiring a maximum power point tracking (MPPT) algorithm are considered to be uncontrollable in this study.The slave units that can control generating power using the input or output inductor current are considered to be controllable slave units.
Figure 2b illustrates the operational principles of the proposed system.The master unit controls the bus voltage following Equation (1): The primary purposes of a microgrid system are (1) to effectively operate and manage the DER, and (2) to maintain a stable bus voltage for supplying energy to the load.To satisfy these conditions, the master unit controls the bus voltage and slave units in order to supply power to the bus.Under a conventional control scheme, the slave units must be able to sense power-sharing information from the master unit or central controller.Using the DBS method, it is possible to eliminate the need for such communication links in the master-slave control; instead, the bus voltage is controlled using a predetermined DBS curve to change the power-sharing ratio of each interface unit.Thus, the master unit controls bus voltage using energy stored in a battery that is applied by the slave units using bus voltage information.
Figure 2 shows a control flow chart of the proposed system and key operation of the DBS-adapted communication-less master-slave control method.Figure 2a is a simplified control flow chart.The controllers are separated by a master unit and slave units, and the master unit controls bus voltage using the battery state-of-charge (SoC) level [22].The slave units can be classified as controllable slave unit and uncontrollable slave unit.The renewable energy sources (e.g., PV panels) are used to control input voltage rather than bus voltage in order to generate maximum power, which eliminates arbitrary input power control.Thus, renewable energy sources requiring a maximum power point tracking (MPPT) algorithm are considered to be uncontrollable in this study.The slave units that can control generating power using the input or output inductor current are considered to be controllable slave units.
Figure 2b illustrates the operational principles of the proposed system.The master unit controls the bus voltage following Equation (1): The bus voltage (V Bus ) changes linearly as the product of the battery SoC and the master coefficient K m , which determines the correlation between the DC bus voltage and the battery SoC, and can be expressed as Equation (2): The minimum and maximum range of the DC bus voltage and battery SoC level can be determined from the system design levels.Detailed parameters and a design guide are presented in Section 4.
Energies 2016, 9, 880 3 of 14 The bus voltage (VBus) changes linearly as the product of the battery SoC and the master coefficient Km, which determines the correlation between the DC bus voltage and the battery SoC, and can be expressed as Equation (2): .max .minmax min The minimum and maximum range of the DC bus voltage and battery SoC level can be determined from the system design levels.Detailed parameters and a design guide are presented in Section 4. The slave coefficient Ks determines the power-sharing ratio of the system.Each slave unit senses bus voltage and controls its input power using its own predetermined Ks curve, which can be expressed as Equation (3): .max .max .min The master and slave coefficients Km and Ks can be designed as nonlinear functions.The voltage level signaling for the master unit and the power level signaling for a slave unit can take a nonlinear form that allows the DERs to be scheduled in various conditions.Figure 3 illustrates various Km and Ks design cases for scheduling different power ratings of the master and slave units in a DC distributed microgrid system.
Using voltage level signaling, the DER energy scheduling can be designed to follow the battery SoC level.Similarly, the slave units can control their generating power based on the battery status. Figure 3 shows small and large Ks cases, respectively, for handling the battery depending on the SoC range.In case (a), the battery is freely used over the 30%-70% SoC level, and the slave units charge and discharge the battery in states 1 and 3, respectively.In case (b), the slave units supply energy to The slave coefficient K s determines the power-sharing ratio of the system.Each slave unit senses bus voltage and controls its input power using its own predetermined K s curve, which can be expressed as Equation (3): The master and slave coefficients K m and K s can be designed as nonlinear functions.The voltage level signaling for the master unit and the power level signaling for a slave unit can take a nonlinear form that allows the DERs to be scheduled in various conditions.Figure 3 illustrates various K m and K s design cases for scheduling different power ratings of the master and slave units in a DC distributed microgrid system.
Using voltage level signaling, the DER energy scheduling can be designed to follow the battery SoC level.Similarly, the slave units can control their generating power based on the battery status. Figure 3 shows small and large K s cases, respectively, for handling the battery depending on the SoC range.In case (a), the battery is freely used over the 30%-70% SoC level, and the slave units charge and discharge the battery in states 1 and 3, respectively.In case (b), the slave units supply energy to the battery more sensitively, but the battery use range is more limited.Thus, a large K s design can be adopted to maintain battery energy for standalone operation or islanding condition, while a small K s design can be used for net zero energy building in order to decrease interference from the utility grid.
Energies 2016, 9, 880 4 of 14 the battery more sensitively, but the battery use range is more limited.Thus, a large Ks design can be adopted to maintain battery energy for standalone operation or islanding condition, while a small Ks design can be used for net zero energy building in order to decrease interference from the utility grid.

Master-Slave-Based Hierarchical Control Method
In a controlling DC distributed microgrid system, a hierarchical control structure is needed to satisfy ANSI/ISA-95.A detailed structure should be divided into more than three levels [13,21,23], but here, we will use two levels for simplicity (Figure 4).The controllers are separated into a local (low level) and central controller (high level), which are both connected with a low-speed communication link with the following functionalities:

Master-Slave-Based Hierarchical Control Method
In a controlling DC distributed microgrid system, a hierarchical control structure is needed to satisfy ANSI/ISA-95.A detailed structure should be divided into more than three levels [13,21,23], but here, we will use two levels for simplicity (Figure 4).The controllers are separated into a local (low level) and central controller (high level), which are both connected with a low-speed communication link with the following functionalities: Energies 2016, 9, 880 4 of 14 the battery more sensitively, but the battery use range is more limited.Thus, a large Ks design can be adopted to maintain battery energy for standalone operation or islanding condition, while a small Ks design can be used for net zero energy building in order to decrease interference from the utility grid.

Master-Slave-Based Hierarchical Control Method
In a controlling DC distributed microgrid system, a hierarchical control structure is needed to satisfy ANSI/ISA-95.A detailed structure should be divided into more than three levels [13,21,23], but here, we will use two levels for simplicity (Figure 4).The controllers are separated into a local (low level) and central controller (high level), which are both connected with a low-speed communication link with the following functionalities:

•
Local Controller (inner control loop + primary control): bus voltage regulation, current and voltage control loop design, system stability issues, protection of each module, etc.

•
Central Controller (secondary control + tertiary control): energy management of system, supply and demand prediction, bus quality control using bus voltage regulation, etc.
The local controller include all functions in each interface unit not involving communication, while the central controller manages additional algorithms for improving power quality and energy management using a communication link.Detailed designs of the local controller and central controller functions are presented as follows

Local Controller
As mentioned in Section 2, the local controller can be classified as a master, a controllable slave, or an uncontrollable slave unit.The design and function of an uncontrollable slave unit are identical to those of a conventional MPPT controller and, therefore, will be ignored in this paper.
Figure 5 shows a controller block diagram of the master and slave units.External signals v ref_ext and i ref_ext from the central controller direct voltage restoration and other functions.The master controller is similar to a conventional droop control, except that the droop gain is replaced with K m based on the battery SoC level.The battery SoC level could be delivered from the central controller, but this would require additional communication by the master controller; hence, the SIU is preferred as a master unit.The SIU is directly connected to the battery, and therefore senses the battery output voltage in order to instantly calculate the battery SoC level [22].Moreover, a DC-DC converter that compares with the GIU can be designed to have a faster voltage controller bandwidth than an AC-DC rectifier.The voltage loop bandwidth of a PFC circuit is often selected to be under 20 Hz in order to avoid 120 Hz distortion from the utility grid; as the performance of a master-slave system is determined by the controller bandwidth of the master unit, it is advantageous to select the SIU as a master in order to obtain higher DC bus quality.
Energies 2016, 9, 880 5 of 14  Central Controller (secondary control + tertiary control): energy management of system, supply and demand prediction, bus quality control using bus voltage regulation, etc.
The local controller include all functions in each interface unit not involving communication, while the central controller manages additional algorithms for improving power quality and energy management using a communication link.Detailed designs of the local controller and central controller functions are presented as follows

Local Controller
As mentioned in Section 2, the local controller can be classified as a master, a controllable slave, or an uncontrollable slave unit.The design and function of an uncontrollable slave unit are identical to those of a conventional MPPT controller and, therefore, will be ignored in this paper.
Figure 5 shows a controller block diagram of the master and slave units.External signals vref_ext and iref_ext from the central controller direct voltage restoration and other functions.The master controller is similar to a conventional droop control, except that the droop gain is replaced with Km based on the battery SoC level.The battery SoC level could be delivered from the central controller, but this would require additional communication by the master controller; hence, the SIU is preferred as a master unit.The SIU is directly connected to the battery, and therefore senses the battery output voltage in order to instantly calculate the battery SoC level [22].Moreover, a DC-DC converter that compares with the GIU can be designed to have a faster voltage controller bandwidth than an AC-DC rectifier.The voltage loop bandwidth of a PFC circuit is often selected to be under 20 Hz in order to avoid 120 Hz distortion from the utility grid; as the performance of a master-slave system is determined by the controller bandwidth of the master unit, it is advantageous to select the SIU as a master in order to obtain higher DC bus quality.The slave controller in Figure 2 requires an emergency loop for fault operation; it operates only when the output voltage cannot be controlled by the master unit, and thus does not operate under the normal operating condition.

Central Controller
Nearly all functions of the central controller under the proposed method are similar to the functions of a conventional droop-based hierarchical controller [13]; they differ only in terms of the standard point compensation and modified bus voltage regulation functions.
Standard point compensation is needed to amend daily load and generation profile changing.In this method, the daily average power of the slave units affects the nominal bus voltage VBus_nominal (Figure 2b).Thus, nominal voltage restoration is required for the daily profile, which is changed by The slave controller in Figure 2 requires an emergency loop for fault operation; it operates only when the output voltage cannot be controlled by the master unit, and thus does not operate under the normal operating condition.

Central Controller
Nearly all functions of the central controller under the proposed method are similar to the functions of a conventional droop-based hierarchical controller [13]; they differ only in terms of the standard point compensation and modified bus voltage regulation functions.
Standard point compensation is needed to amend daily load and generation profile changing.In this method, the daily average power of the slave units affects the nominal bus voltage V Bus_nominal (Figure 2b).Thus, nominal voltage restoration is required for the daily profile, which is changed by factors such as weather, temperature, and daylight condition, and the central controller must be able to monitor and make daily predictions in order to compensate for such changes.Voltage restoration is also required for bus regulation.Under the droop-based hierarchical control technique used by the proposed method, bus voltage is changed based on the battery SoC and, therefore, external compensation from a central controller is needed.Figure 6 shows the standard point and voltage restoration concept employed by the proposed system.
Energies 2016, 9, 880 6 of 14 factors such as weather, temperature, and daylight condition, and the central controller must be able to monitor and make daily predictions in order to compensate for such changes.Voltage restoration is also required for bus regulation.Under the droop-based hierarchical control technique used by the proposed method, bus voltage is changed based on the battery SoC and, therefore, external compensation from a central controller is needed.Figure 6 shows the standard point and voltage restoration concept employed by the proposed system.The master and slave DBS curve can be changed using Vref_ext and Iref_ext as shown in Figure 6.As the battery SoC is changed discontinuously, it can be easily compensated using Equations ( 4) and ( 5): Using the same reasoning, the standard point compensation can be added to extend Equation ( 5): Applying the values calculated in Equations ( 4) and ( 6) based on the external references in Figure 6, generating power control in the controllable slave units is possible without changing the DC bus voltage.

Simulation and Experimental Results
To investigate the effectiveness of the proposed method, a rated 380 V, 1 kW system was designed and implemented under both a simulation and a hardware environment.The system architecture is shown in Figure 7, and the hardware and controller parameters are specified in Table 1.
The GIU and SIU were designed together as a bidirectional circuit, with the SIU playing the master role.Output capacitors for each subsystem in Figure 1 were designed to have output voltage ripples of less than ±5 V.A detailed battery specification is provided in the Appendix of this paper.The master and slave DBS curve can be changed using V ref_ext and I ref_ext as shown in Figure 6.As the battery SoC is changed discontinuously, it can be easily compensated using Equations ( 4) and ( 5): Using the same reasoning, the standard point compensation can be added to extend Equation ( 5): Applying the values calculated in Equations ( 4) and ( 6) based on the external references in Figure 6, generating power control in the controllable slave units is possible without changing the DC bus voltage.

Simulation and Experimental Results
To investigate the effectiveness of the proposed method, a rated 380 V, 1 kW system was designed and implemented under both a simulation and a hardware environment.The system architecture is shown in Figure 7, and the hardware and controller parameters are specified in Table 1.
The GIU and SIU were designed together as a bidirectional circuit, with the SIU playing the master role.Output capacitors for each subsystem in Figure 1 were designed to have output voltage ripples of less than ±5 V.A detailed battery specification is provided in the Appendix of this paper.

Simulation Results
The simulation was conducted to produce both short-term and long-term characteristics of the proposed control method.As short-term characteristics, the bus voltage ripple and transient response

Simulation Results
The simulation was conducted to produce both short-term and long-term characteristics of the proposed control method.As short-term characteristics, the bus voltage ripple and transient response of the system was illustrated and compared with a conventional droop control system in order to verify the effectiveness of the proposed system in terms of bus regulation and dynamic properties.
As a long-term simulation, the power sharing of the master and slave units was demonstrated in order to verify the validity of the proposed communication-less master-slave control method.
Figure 8 shows steady-state comparisons between a conventional droop control method and the proposed master-slave control method.In both cases, the total load is 1 kW, and the GIU generates 90% power in order to identify maximum bus voltage ripple.At the same controller bandwidth and hardware design, the bus voltage ripple of the proposed method is approximately 4.7 V, while that of the droop control method is approximately 8.3 V.
of the system was illustrated and compared with a conventional droop control system in order to verify the effectiveness of the proposed system in terms of bus regulation and dynamic properties.As a long-term simulation, the power sharing of the master and slave units was demonstrated in order to verify the validity of the proposed communication-less master-slave control method.
Figure 8 shows steady-state comparisons between a conventional droop control method and the proposed master-slave control method.In both cases, the total load is 1 kW, and the GIU generates 90% power in order to identify maximum bus voltage ripple.At the same controller bandwidth and hardware design, the bus voltage ripple of the proposed method is approximately 4.7 V, while that of the droop control method is approximately 8.3 V.Because the 120 Hz ripple from the utility side can be considered an external disturbance in the SIU side, this impedance clamping problem completely divides the regulation property of each unit.By contrast, the proposed method has no impedance clamping problem in any subsystem, and thus the equivalent bus impedance of the microgrid system can be compared as shown in Figure 10a.The This difference arises from an impedance clamping problem in the droop control method.As shown in the small-signal block diagram in Figure 9a, droop gain is added in the control loop and clamps the low-and high-frequency output impedance of the converter, as shown in Figure 9b.The droop gain R d clamps the bus impedance of the SIU, and interrupts its dynamic property in all regions.
Energies 2016, 9, 880 8 of 14 of the system was illustrated and compared with a conventional droop control system in order to verify the effectiveness of the proposed system in terms of bus regulation and dynamic properties.As a long-term simulation, the power sharing of the master and slave units was demonstrated in order to verify the validity of the proposed communication-less master-slave control method.Figure 8 shows steady-state comparisons between a conventional droop control method and the proposed master-slave control method.In both cases, the total load is 1 kW, and the GIU generates 90% power in order to identify maximum bus voltage ripple.At the same controller bandwidth and hardware design, the bus voltage ripple of the proposed method is approximately 4.7 V, while that of the droop control method is approximately 8.3 V.This difference arises from an impedance clamping problem in the droop control method.As shown in the small-signal block diagram in Figure 9a, droop gain is added in the control loop and clamps the low-and high-frequency output impedance of the converter, as shown in Figure 9b.The droop gain Rd clamps the bus impedance of the SIU, and interrupts its dynamic property in all regions.Because the 120 Hz ripple from the utility side can be considered an external disturbance in the SIU side, this impedance clamping problem completely divides the regulation property of each unit.By contrast, the proposed method has no impedance clamping problem in any subsystem, and thus the equivalent bus impedance of the microgrid system can be compared as shown in Figure 10a.The Because the 120 Hz ripple from the utility side can be considered an external disturbance in the SIU side, this impedance clamping problem completely divides the regulation property of each unit.By contrast, the proposed method has no impedance clamping problem in any subsystem, and thus the equivalent bus impedance of the microgrid system can be compared as shown in Figure 10a.The calculated bus impedance of the proposed control method at 120 Hz is −2.7 dB, which means that the bus voltage ripple of the DC microgrid system is compensated into 0.73 times the 120 Hz current disturbance, thereby satisfying the bus voltage ripple in Figure 9 of 4.7 V, with a 6.5 A peak current in the inductor current of the GIU.calculated bus impedance of the proposed control method at 120 Hz is −2.7 dB, which means that the bus voltage ripple of the DC microgrid system is compensated into 0.73 times the 120 Hz current disturbance, thereby satisfying the bus voltage ripple in Figure 9 of 4.7 V, with a 6.5 A peak current in the inductor current of the GIU.Similarly, the simulation results of the step-load changing comparison shown in Figure 10b illustrate the advantages of the proposed control method.The figure shows a comparison of the transient characteristics of the bus voltage at load changes from 100 W to 1 kW at time 0.3 s.The transient response is much smaller and faster in the proposed method, as the master unit, the SIU, can be considered an additional shunt regulator [14].
The results of the long-term simulations are shown in Figure 11 as the bus voltage, battery SoC, and output current of the proposed method with a downscaled battery capacity (the battery SoC level changes too slowly with an original battery status to identify the characteristics of the proposed method).Similarly, the simulation results of the step-load changing comparison shown in Figure 10b illustrate the advantages of the proposed control method.The figure shows a comparison of the transient characteristics of the bus voltage at load changes from 100 W to 1 kW at time 0.3 s.The transient response is much smaller and faster in the proposed method, as the master unit, the SIU, can be considered an additional shunt regulator [14].
The results of the long-term simulations are shown in Figure 11 as the bus voltage, battery SoC, and output current of the proposed method with a downscaled battery capacity (the battery SoC level changes too slowly with an original battery status to identify the characteristics of the proposed method).
Energies 2016, 9, 880 9 of 14 calculated bus impedance of the proposed control method at 120 Hz is −2.7 dB, which means that the bus voltage ripple of the DC microgrid system is compensated into 0.73 times the 120 Hz current disturbance, thereby satisfying the bus voltage ripple in Figure 9 of 4.7 V, with a 6.5 A peak current in the inductor current of the GIU.Similarly, the simulation results of the step-load changing comparison shown in Figure 10b illustrate the advantages of the proposed control method.The figure shows a comparison of the transient characteristics of the bus voltage at load changes from 100 W to 1 kW at time 0.3 s.The transient response is much smaller and faster in the proposed method, as the master unit, the SIU, can be considered an additional shunt regulator [14].
The results of the long-term simulations are shown in Figure 11 as the bus voltage, battery SoC, and output current of the proposed method with a downscaled battery capacity (the battery SoC level changes too slowly with an original battery status to identify the characteristics of the proposed method).Figure 11a shows the long-term simulation results with battery SoC variation.The battery capacity is changed at a 1/14,400 scale and the load changes from 100 W to 900 W at 2 s.The master and slave curve is designed as shown in Figure 11b and Table 2.In the simulation results, the GIU does not respond to the load changing and, therefore, the SIU immediately and solely covers the entire load.Thus, the battery SoC decreases through battery discharge and V Bus changes following the master DBS curve design.In the same manner, the generation power of the GIU changes slowly and becomes saturated at a 29% SoC level.Saturating the battery SoC level takes approximately 8 s, as opposed to a calculated value of approximately 32 h at the original battery capacity.

Experimental Results
To verify the suitability of the proposed control method, short-term experimental hardware was built and tested.Because the battery capacity would otherwise overly prolong the experiment, the battery SoC level was changed artificially using an external signal.
Figure 12 shows the experimental waveforms of the step-load changing condition corresponding to those in Figure 10b.To check the bidirectional operation of the SIU, the load was changed from 1 kW to 0 W, and the GIU generated nearly 1 kW of power.
Energies 2016, 9, 880 10 of 14 Figure 11a shows the long-term simulation results with battery SoC variation.The battery capacity is changed at a 1/14,400 scale and the load changes from 100 W to 900 W at 2 s.The master and slave curve is designed as shown in Figure 11b and Table 2.In the simulation results, the GIU does not respond to the load changing and, therefore, the SIU immediately and solely covers the entire load.Thus, the battery SoC decreases through battery discharge and VBus changes following the master DBS curve design.In the same manner, the generation power of the GIU changes slowly and becomes saturated at a 29% SoC level.Saturating the battery SoC level takes approximately 8 s, as opposed to a calculated value of approximately 32 h at the original battery capacity.

Experimental Results
To verify the suitability of the proposed control method, short-term experimental hardware was built and tested.Because the battery capacity would otherwise overly prolong the experiment, the battery SoC level was changed artificially using an external signal.
Figure 12 shows the experimental waveforms of the step-load changing condition corresponding to those in Figure 10b.To check the bidirectional operation of the SIU, the load was changed from 1 kW to 0 W, and the GIU generated nearly 1 kW of power.As a result, an inductor current of the SIU changed to a negative value and charged the battery with 1 kW of instantaneous power.A measured bus voltage ripple of 5.5 V well-matched with the 7.5 A peak-to-peak inductor current of the GIU with a designed bus impedance of −2.7 dB.
Figure 13a shows the experimental waveforms of the proposed system when the battery SoC level is changed from 10% to 90% artificially.As shown by the red line, VBus changes from 375 V to 385 V as a result of the external battery SoC signal, and PGIU changes from 1 kW to −1 kW, accordingly.The load condition does not change in this case, and IL_SIU charges and discharges the battery according to IGIU based on the dotted standard point.Figure 13b shows the same experimental result with DC voltage restoration; the bus voltage does not change, but IGIU changes by the same amount as in Figure 13a.As a result, an inductor current of the SIU changed to a negative value and charged the battery with 1 kW of instantaneous power.A measured bus voltage ripple of 5.5 V well-matched with the 7.5 A peak-to-peak inductor current of the GIU with a designed bus impedance of −2.7 dB.
Figure 13a shows the experimental waveforms of the proposed system when the battery SoC level is changed from 10% to 90% artificially.As shown by the red line, V Bus changes from 375 V to 385 V as a result of the external battery SoC signal, and P GIU changes from 1 kW to −1 kW, accordingly.The load condition does not change in this case, and I L_SIU charges and discharges the battery according to I GIU based on the dotted standard point.Figure 13b shows the same experimental result with DC voltage restoration; the bus voltage does not change, but I GIU changes by the same amount as in Figure 13a.

Conclusions
A DC distribution system controlled by a communication-free master-slave-based hierarchical control method is presented in this paper.Although droop control has many drawbacks, including low frequency impedance clamping problems and a comparatively slow transient response, there has not been any previous widespread use of master-slave control in microgrid systems, owing to the communication dependency and low expandability of such systems.The proposed master-slave technique utilizes the DBS method to operate freely without communication in order to overcome these drawbacks.By replacing the droop control technique with the proposed master-slave technique, a DC distributed microgrid system can operate autonomously without distinction between grid connecting and islanding conditions.Moreover, the transient response of bus voltage is accelerated because of the presence of a high voltage bandwidth master unit.The concept and control strategies for the proposed control method were presented and analyzed in this paper, along with a voltage restoration method for more efficient and effective system design, and design considerations were discussed.The results obtained using 1 kW simulation and hardware modules operating from a 380 V DC bus demonstrate the feasibility of the proposed method.

Figure 1 .
Figure 1.Block diagram of the proposed (DC) distributed microgrid system.

Figure 1 .
Figure 1.Block diagram of the proposed (DC) distributed microgrid system.

Figure 2 .
Figure 2. (a) Operational flow chart and (b) design principle of the proposed communication-less master-slave control method.

Figure 2 .
Figure 2. (a) Operational flow chart and (b) design principle of the proposed communication-less master-slave control method.

Figure 3 .
Figure 3. Nonlinear Km and Ks curve designs of the proposed master-slave control method using (a) small Ks and (b) large Ks cases.

Figure 4 .
Figure 4. Control structure of the proposed method.Local Controller (inner control loop + primary control): bus voltage regulation, current and voltage control loop design, system stability issues, protection of each module, etc.

Figure 3 .
Figure 3. Nonlinear K m and K s curve designs of the proposed master-slave control method using (a) small K s and (b) large K s cases.

Figure 3 .
Figure 3. Nonlinear Km and Ks curve designs of the proposed master-slave control method using (a) small Ks and (b) large Ks cases.

Figure 4 .
Figure 4. Control structure of the proposed method.Local Controller (inner control loop + primary control): bus voltage regulation, current and voltage control loop design, system stability issues, protection of each module, etc.

Figure 4 .
Figure 4. Control structure of the proposed method.

Figure 5 .
Figure 5. Local controller of the proposed method.

Figure 5 .
Figure 5. Local controller of the proposed method.

Figure 6 .
Figure 6.Central controller block diagram and DC bus voltage restoration concept of the proposed master-slave control method.

Figure 6 .
Figure 6.Central controller block diagram and DC bus voltage restoration concept of the proposed master-slave control method.

Figure 7 .
Figure 7. System architecture of the proposed DC distribution system.

Figure 7 .
Figure 7. System architecture of the proposed DC distribution system.

Figure 8 .
Figure 8. Steady-state simulation comparisons between a conventional droop control method and the proposed master-slave control method.This difference arises from an impedance clamping problem in the droop control method.As shown in the small-signal block diagram in Figure9a, droop gain is added in the control loop and clamps the low-and high-frequency output impedance of the converter, as shown in Figure9b.The droop gain Rd clamps the bus impedance of the SIU, and interrupts its dynamic property in all regions.

Figure 9 .
Figure 9. (a) Small-signal block diagram of the storage interface unit (SIU) and (b) the output impedance bode diagram of the SIU in the droop control system.

Figure 8 .
Figure 8. Steady-state simulation comparisons between a conventional droop control method and the proposed master-slave control method.

Figure 8 .
Figure 8. Steady-state simulation comparisons between a conventional droop control method and the proposed master-slave control method.

Figure 9 .
Figure 9. (a) Small-signal block diagram of the storage interface unit (SIU) and (b) the output impedance bode diagram of the SIU in the droop control system.

Figure 9 .
Figure 9. (a) Small-signal block diagram of the storage interface unit (SIU) and (b) the output impedance bode diagram of the SIU in the droop control system.

Figure 10 .
Figure 10.Comparisons between (a) equivalent bus impedance bode diagram and (b) simulation waveforms at step load changing condition.

Figure 11 .
Figure 11.(a) Simulation waveforms at step-load changing condition with the battery state-of-charge (SoC) varying, and (b) master and slave curve design of the system.

Figure 10 .
Figure 10.Comparisons between (a) equivalent bus impedance bode diagram and (b) simulation waveforms at step load changing condition.

Figure 10 .
Figure 10.Comparisons between (a) equivalent bus impedance bode diagram and (b) simulation waveforms at step load changing condition.

Figure 11 .
Figure 11.(a) Simulation waveforms at step-load changing condition with the battery state-of-charge (SoC) varying, and (b) master and slave curve design of the system.

Figure 11 .
Figure 11.(a) Simulation waveforms at step-load changing condition with the battery state-of-charge (SoC) varying, and (b) master and slave curve design of the system.

Figure 12 .
Figure 12.Experimental waveforms of the step-load changing condition.

Figure 12 .
Figure 12.Experimental waveforms of the step-load changing condition.

Energies 2016, 9 , 880 11 of 14 Figure 13 .
Figure 13.Experimental waveforms of the proposed system when the battery SoC level changes from 10% to 90% (a) without voltage restoration and (b) with voltage restoration.

Table 2 .
Master and slave curve design values of the simulation.

Table 2 .
Master and slave curve design values of the simulation.
G viSmall signal transfer function of output current to output voltage G vd Small signal transfer function of duty to output voltage G idSmall signal transfer function of duty to inductor current G iiSmall signal transfer function of output current to inductor current H i