A Generic Control-Oriented Model Order Reduction Approach for High Step-Up DC / DC Converters Based on Voltage Multiplier

A generic control-oriented model order reduction approach for high step-up DC/DC converters based on voltage multiplier. Abstract: The modeling and control system design of high step-up DC / DC converters based on voltage multipliers (VMs) are di ﬃ cult, due to the various circuit topologies and the presence of large number of capacitors in VMs. This paper proposes a generic approach to reduce the model order of such converters by replacing the VM capacitors with voltage sources controlled by the output voltage of the converter. Theoretical analysis and simulation results show that the derived models can accurately represent the low frequency response of the converter which is valuable for obtaining a small-signal AC model for control system design. The detailed modeling and controller design process are demonstrated for the converter, and the obtained simulation results are veriﬁed experimentally on a 400 W prototype.


Introduction
The desire for clean environment, low carbon emission and the fact that fossil fuel reserves are depleting have led to the rapid development of renewable energy generation systems. Renewable energy generating units, such as photovoltaic arrays and wind turbine generators, present lower output voltage compared to the interconnected grid systems, and thus high voltage step-up methods are required for grid integration. It is also important to provide active control to these energy conversion devices in order to maximize the energy harvested, which calls for high-efficiency power electronic converters. High step-up voltage conversion technology based on voltage multiplier (VM) is one of the solutions [1] and has recently become increasingly popular in renewable applications such as solar photovoltaic [2,3], wind power generation [4], and fuel cell power generation [5,6].
In recent years, various VM circuits have been proposed to achieve a family of high step-up non-isolated DC/DC converter topologies for the applications in renewables [7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22]. A generalized circuit structure of a family of non-isolated converters with VM is shown in Figure 1, where the switch/inductor circuit is connected to the diode-capacitor VM to supply the load R L . The output voltage u o of the converter can be stepped up to much higher level compared to the input voltage u in . For the diode-capacitor VM circuit, voltage doubler circuit [7], voltage quadrupler circuit [8], Dickson charge pump [9], Cockcroft-Walton VM [10], and amongst others [17] have been used, and based on which various other different VM circuits have been proposed for two-input-phase interleaved booster inductor has been used to replace the inductors to provide further performance improvement [6,23,24].  However, most of the existing literature on this type of converter focus on derivation of the circuit topology and/or design optimization, while the determination of the control structure and the tuning of the control parameters are simply based on empirical methods, or even not discussed. The lack of systematic approach to modeling and control system design will lead to cumbersome parameter tuning process and/or poor performance of the converter, especially under highly-varying conditions of the renewable generations, which can offset the benefits of this type of promising converter. A large number of capacitors are used in VM circuits, which can effectively increase the voltage conversion gain and reduce the voltage stress of the switching devices, but this also leads to a high-order system and makes the classical model-based design of high-performance control system difficult, unless a suitable control-oriented model order reduction approach is developed for practical applications.
Indeed, modeling and model order reduction approaches have been developed for many VMbased converters, such as those with three-state switching cell [25], switch capacitor [26], coupledinductor [27]. However, the existing approaches are specifically designed for relatively low-order system and they cannot be readily applied to other topologies, especially when the number of VM cells increases. To reduce the effort of control system design for a family of high step-up DC/DC converters based on diode-capacitor VM, for the first time, this paper proposes a novel general method to acquire their averaged-value reduced-order models. The approach uses the concept to replace all the capacitors in the VM with voltage sources controlled by the output voltage of the converter, and the system order is greatly reduced without sacrificing much model accuracy.
The rest of the paper is organized as follows: Detailed procedure of the proposed model-order reduction method is presented in Section 2. Using this proposed method, general small-signal AC models and relevant transfer functions for the converters based on VM are derived in Section 3. Controller design is included in Section 4, where the low frequency response of the proposed reduced-order small signal AC model is found to be in agreement with the simulation results from the circuit model with detailed switching dynamics. The proposed control design method based on the reduced-order model is verified through simulation and experiment in Section 5. The main findings of the work are concluded in Section 6.

Proposed Model Order Reduction Method
For the type of converters as shown in Figure 1, it has been commonly found that by neglecting the effects of ripples, the voltage across each VM capacitor is in linear relationship to the output voltage uo of the converter [7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22]. Hence, each VM capacitor can be replaced by a controlled voltage source associated with uo. As the system order is determined by the number of the energy storage However, most of the existing literature on this type of converter focus on derivation of the circuit topology and/or design optimization, while the determination of the control structure and the tuning of the control parameters are simply based on empirical methods, or even not discussed. The lack of systematic approach to modeling and control system design will lead to cumbersome parameter tuning process and/or poor performance of the converter, especially under highly-varying conditions of the renewable generations, which can offset the benefits of this type of promising converter. A large number of capacitors are used in VM circuits, which can effectively increase the voltage conversion gain and reduce the voltage stress of the switching devices, but this also leads to a high-order system and makes the classical model-based design of high-performance control system difficult, unless a suitable control-oriented model order reduction approach is developed for practical applications.
Indeed, modeling and model order reduction approaches have been developed for many VM-based converters, such as those with three-state switching cell [25], switch capacitor [26], coupled-inductor [27]. However, the existing approaches are specifically designed for relatively low-order system and they cannot be readily applied to other topologies, especially when the number of VM cells increases. To reduce the effort of control system design for a family of high step-up DC/DC converters based on diode-capacitor VM, for the first time, this paper proposes a novel general method to acquire their averaged-value reduced-order models. The approach uses the concept to replace all the capacitors in the VM with voltage sources controlled by the output voltage of the converter, and the system order is greatly reduced without sacrificing much model accuracy.
The rest of the paper is organized as follows: Detailed procedure of the proposed model-order reduction method is presented in Section 2. Using this proposed method, general small-signal AC models and relevant transfer functions for the converters based on VM are derived in Section 3. Controller design is included in Section 4, where the low frequency response of the proposed reduced-order small signal AC model is found to be in agreement with the simulation results from the circuit model with detailed switching dynamics. The proposed control design method based on the reduced-order model is verified through simulation and experiment in Section 5. The main findings of the work are concluded in Section 6.

Proposed Model Order Reduction Method
For the type of converters as shown in Figure 1, it has been commonly found that by neglecting the effects of ripples, the voltage across each VM capacitor is in linear relationship to the output voltage u o of the converter [7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22]. Hence, each VM capacitor can be replaced by a controlled voltage source associated with u o . As the system order is determined by the number of the energy storage Energies 2019, 12,1971 3 of 17 components in the circuit, the corresponding model order can be greatly reduced. Detailed procedure for the proposed model order reduction approach is described as follows: Step 1: Obtain the steady-state relationship between all VM capacitor voltages and the output voltage u o (t) according to the results from circuit analysis. The general relationship can be expressed as where u cn (t) is the voltage of n-th VM capacitor and k n is corresponding proportional gain.
Step 2: Replace all VM capacitors, except the ones at the output, with controlled voltage sources according to (1). This step is shown graphically in Figure 2.
Step 3: Reformulate and simplify the resulting model to construct an equivalent reduced-order circuit based on the state-space averaging method [28].
Energies 2019, 12, x FOR PEER REVIEW 3 of 18 components in the circuit, the corresponding model order can be greatly reduced. Detailed procedure for the proposed model order reduction approach is described as follows: Step 1: Obtain the steady-state relationship between all VM capacitor voltages and the output voltage uo(t) according to the results from circuit analysis. The general relationship can be expressed as where ucn(t) is the voltage of n-th VM capacitor and kn is corresponding proportional gain.
Step 2: Replace all VM capacitors, except the ones at the output, with controlled voltage sources according to (1). This step is shown graphically in Figure 2.
Step 3: Reformulate and simplify the resulting model to construct an equivalent reduced-order circuit based on the state-space averaging method [28]. In this connection, the converter proposed in [12] will be used as an example to demonstrate the proposed approach. Figure 3a shows the topology of this converter with n number of basic VM cells. It consists of two switches S1 and S2, two inductors L1 and L2, and the VM circuits with n VMs. Each VM consists of two diodes and two capacitors with the equal capacitance Cia = Cib = Cik (i = 1, 2, …, n). The voltage conversion gain is 2n times of the traditional boost converter. The driving signals of switches S1 and S2 are interleaved and both duty cycles d shall be larger than 0.5. The conversion gain of this converter can be easily adjusted by changing the number of VMs at the design stage. Voltage stress of all semiconductor devices are reduced significantly compared with conventional boost converter, and the two inductor currents can be automatically self-balancing. All capacitors have the same current stress, except the two which are connected with the load, that is, conducive to the thermal design. We denote this converter as the VM converter hereafter in this paper.
The steady-state relationships between the capacitor voltage in VMs and the output voltage are given in (2) based on the analytical results provided in Reference [12]. According to Step 2 of the proposed model order reduction method, the equivalent circuit of this converter by replacing all capacitors in VMs with corresponding controlled voltage sources is shown in Figure 3b. In this connection, the converter proposed in [12] will be used as an example to demonstrate the proposed approach. Figure 3a shows the topology of this converter with n number of basic VM cells. It consists of two switches S 1 and S 2 , two inductors L 1 and L 2 , and the VM circuits with n VMs. Each VM consists of two diodes and two capacitors with the equal capacitance C ia = C ib = C ik (i = 1, 2, . . . , n). The voltage conversion gain is 2n times of the traditional boost converter. The driving signals of switches S 1 and S 2 are interleaved and both duty cycles d shall be larger than 0.5. The conversion gain of this converter can be easily adjusted by changing the number of VMs at the design stage. Voltage stress of all semiconductor devices are reduced significantly compared with conventional boost converter, and the two inductor currents can be automatically self-balancing. All capacitors have the same current stress, except the two which are connected with the load, that is, conducive to the thermal design. We denote this converter as the VM converter hereafter in this paper. components in the circuit, the corresponding model order can be greatly reduced. Detailed procedure for the proposed model order reduction approach is described as follows: Step 1: Obtain the steady-state relationship between all VM capacitor voltages and the output voltage uo(t) according to the results from circuit analysis. The general relationship can be expressed as where ucn(t) is the voltage of n-th VM capacitor and kn is corresponding proportional gain.
Step 2: Replace all VM capacitors, except the ones at the output, with controlled voltage sources according to (1). This step is shown graphically in Figure 2.
Step 3: Reformulate and simplify the resulting model to construct an equivalent reduced-order circuit based on the state-space averaging method [28]. In this connection, the converter proposed in [12] will be used as an example to demonstrate the proposed approach. Figure 3a shows the topology of this converter with n number of basic VM cells. It consists of two switches S1 and S2, two inductors L1 and L2, and the VM circuits with n VMs. Each VM consists of two diodes and two capacitors with the equal capacitance Cia = Cib = Cik (i = 1, 2, …, n). The voltage conversion gain is 2n times of the traditional boost converter. The driving signals of switches S1 and S2 are interleaved and both duty cycles d shall be larger than 0.5. The conversion gain of this converter can be easily adjusted by changing the number of VMs at the design stage. Voltage stress of all semiconductor devices are reduced significantly compared with conventional boost converter, and the two inductor currents can be automatically self-balancing. All capacitors have the same current stress, except the two which are connected with the load, that is, conducive to the thermal design. We denote this converter as the VM converter hereafter in this paper.
The steady-state relationships between the capacitor voltage in VMs and the output voltage are given in (2) based on the analytical results provided in Reference [12]. According to Step 2 of the proposed model order reduction method, the equivalent circuit of this converter by replacing all capacitors in VMs with corresponding controlled voltage sources is shown in Figure 3b.
According to the working state of the switches, the operation mode of this converter can be separated into three regions during one switching period TS and the equivalent circuits for each region are shown in Figure 4.  Figure 4. Three working state equivalent circuits of the VM converter: (a) S1, S2 are both on; (b) S1 is on and S2 is off; (c) S2 is on and S1 is off.
Equation (3) can be obtained while S1 and S2 are both in on-state. Similarly, (4) is obtained when S1 is in on-state and S2 is in off-state. When S2 is turned-on and S1 is turned-off, (5) can be obtained. The steady-state relationships between the capacitor voltage in VMs and the output voltage are given in (2) based on the analytical results provided in Reference [12]. According to Step 2 of the proposed model order reduction method, the equivalent circuit of this converter by replacing all capacitors in VMs with corresponding controlled voltage sources is shown in Figure 3b.
According to the working state of the switches, the operation mode of this converter can be separated into three regions during one switching period T S and the equivalent circuits for each region are shown in Figure 4.
According to the working state of the switches, the operation mode of this converter can be separated into three regions during one switching period TS and the equivalent circuits for each region are shown in Figure 4. Figure 4. Three working state equivalent circuits of the VM converter: (a) S1, S2 are both on; (b) S1 is on and S2 is off; (c) S2 is on and S1 is off.
Equation (3) can be obtained while S1 and S2 are both in on-state. Similarly, (4) is obtained when S1 is in on-state and S2 is in off-state. When S2 is turned-on and S1 is turned-off, (5) can be obtained.  Equation (3) can be obtained while S 1 and S 2 are both in on-state. Similarly, (4) is obtained when S 1 is in on-state and S 2 is in off-state. When S 2 is turned-on and S 1 is turned-off, (5) can be obtained.
Furthermore, according to Reference [12], when the duty cycles of S 1 and S 2 are identical, the average inductor currents of L 1 and L 2 over one switch period T s shall be equal, i.e., Next, by replacing the state variables with their low-frequency averaged values over one switch period T s , (3)-(6) can be simplified to where d = 1 − d. Combining (2) and (7) yields the new governing equation of state of the capacitors, i.e., Hence, the reduced-order model for the VM converter is obtained as (7) and (8). Based on (7) and (8), its equivalent circuit can be obtained as shown in Figure 5a. It can be clearly seen that the voltage is stepped up through n stages by using n ideal transformers. In the first stage the voltage increases by a factor of 2/d (t), and later the voltage increases linearly. Indeed, by combining the Energies 2019, 12,1971 6 of 17 inductors, the capacitors and the ideal transformers respectively, Figure 5a can be readily simplified to Figure 5b, where the equivalent parameters A and L eq can be readily obtained as Considering that C ia = C ib = C ik , the equivalent capacitance C eq,i of the i-th stage in Figure 5a referring to the final (n-th) stage is Hence, the equivalent C eqo in Figure 5b can be calculated by If C 1k = C 2k = . . . C nk = C k , and considering, Equation (11) can be further reduced to

A General Reduced-Order Model of the Converters Based on VM
Using the similar method presented in Section 2.1, the reduced-order averaged models of the existing VM-based converters proposed in References, [9][10][11]14,16], are investigated and derived. It was found that all the reduced-order models share an identical equivalent circuit structure, as shown in Figure 5b, while the corresponding expressions for parameters Leq, Ceqo and A are different; they are summarized in Table 1. Table 1. Parameters of the reduced-order average model for different converter circuits based on VM.

Ref.
Topology Expressions of Equivalent Parameters

A General Reduced-Order Model of the Converters Based on VM
Using the similar method presented in Section 2.1, the reduced-order averaged models of the existing VM-based converters proposed in References, [9][10][11]14,16], are investigated and derived. It was found that all the reduced-order models share an identical equivalent circuit structure, as shown in Figure 5b, while the corresponding expressions for parameters L eq , C eqo and A are different; they are summarized in Table 1.

Ref.
Topology Expressions of Equivalent Parameters [9] Using the similar method presented in Section 2.1, the reduced-order averaged models of the existing VM-based converters proposed in References, [9][10][11]14,16], are investigated and derived. It was found that all the reduced-order models share an identical equivalent circuit structure, as shown in Figure 5b, while the corresponding expressions for parameters Leq, Ceqo and A are different; they are summarized in Table 1. Table 1. Parameters of the reduced-order average model for different converter circuits based on VM.

Ref. Topology Expressions of Equivalent Parameters
[9] [10] existing VM-based converters proposed in References, [9][10][11]14,16], are investigated and derived. It was found that all the reduced-order models share an identical equivalent circuit structure, as shown in Figure 5b, while the corresponding expressions for parameters Leq, Ceqo and A are different; they are summarized in Table 1.

Small-Signal AC Models
Based on the state-space averaging method and the proposed reduced-order averaged model as shown in Figure 5, the small-signal AC model of the VM converter can be obtained as Energies 2019, 12, x FOR PEER REVIEW 8 of 18 [11] 1 2 eq 1 2

Small-Signal AC Models
Based on the state-space averaging method and the proposed reduced-order averaged model as shown in Figure 5, the small-signal AC model of the VM converter can be obtained as Energies 2019, 12, x FOR PEER REVIEW 8 of 18 [11] 1 2 eq 1 2

Small-Signal AC Models
Based on the state-space averaging method and the proposed reduced-order averaged model as shown in Figure 5, the small-signal AC model of the VM converter can be obtained as

Small-Signal AC Models
Based on the state-space averaging method and the proposed reduced-order averaged model as shown in Figure 5, the small-signal AC model of the VM converter can be obtained as where I in , U o and D are the quiescent values of the input current i in (t), output voltage u o (t), and d (t), respectively.î in (t),û o (t),û in (t) andd(t) are the small AC values of the input current i in (t), output voltage u o (t), input voltage u in (t), and d (t), respectively. Using the same method, a general AC small-signal model of the converters based on VM can be obtained, as shown in Figure 6. The expressions of the model parameters including A, B and C, for different topologies, are given in Table 2.

Transfer Functions
Based on Figure 6, the general transfer functions related to the controller or filter design of the converters based on VM are derived and given in (13)-(17), where A, B and C for different converter circuits are listed in Table 2.

Transfer Functions
Based on Figure 6, the general transfer functions related to the controller or filter design of the converters based on VM are derived and given in (13)-(17), where A, B and C for different converter circuits are listed in Table 2.

Accuracy of the Small Signal AC Model
To verify the accuracy and examine the performance of the proposed general small signal AC models and relevant transfer functions of the converters based on VM, a comparison between the proposed models and practical circuits, and controller design of the VM converter, will be presented in this section. The specifications for the simulation are listed in Tables 3 and 4. The simulation results are shown in Figures 7-9, where the VM numbers are 2, 3, and 5, respectively. The blue line is the actual circuit simulation result using PSIM software, and the red line is the proposed model simulation result obtained from MATLAB R2016a/Simulink 8.7 (The MathWorks, Inc., Natick, MA, USA). When the VM number is 2, Figure 7a,c,e show the output voltage waveforms when the duty cycle changes around 0.6 with different voltage ripple on capacitors in VM due to different C na and C nb . Figure 7b,d,f show the output voltage waveforms when the input voltage changes around 40 V with different voltage ripple on capacitors in VM due to different C na and C nb . It can be seen from Figure 7 that the increase of voltage ripple on capacitor in VM will make the steady-state accuracy of the proposed model worse, but the dynamic response of the proposed model is not affected significantly. When the voltage ripple on capacitors in VM is 10%, Figures 7c, 8a and 9a show the output voltage waveforms when the duty cycle changes around 0.6 with different VM number at C na = C nb = 2 µF, while Figures 7d, 8b and 9b show the output voltage waveforms when the input voltage changes around 40 V with different VM number at the same capacitance. Clearly, as the VM number increases, the steady-state error between the proposed model and the actual circuit increases too. This steady-state error is causes by the ripple in the capacitor voltage which affects the converter voltage gain [12].

Accuracy of the Small Signal AC Model
To verify the accuracy and examine the performance of the proposed general small signal AC models and relevant transfer functions of the converters based on VM, a comparison between the proposed models and practical circuits, and controller design of the VM converter, will be presented in this section. The specifications for the simulation are listed in Tables 3 and 4. The simulation results are shown in Figures 7-9, where the VM numbers are 2, 3, and 5, respectively. The blue line is the actual circuit simulation result using PSIM software, and the red line is the proposed model simulation result obtained from MATLAB R2016a/Simulink 8.7 (The MathWorks, Inc., Natick, MA, USA). When the VM number is 2, Figure 7a,c,e show the output voltage waveforms when the duty cycle changes around 0.6 with different voltage ripple on capacitors in VM due to different Cna and Cnb. Figure 7b,d,f show the output voltage waveforms when the input voltage changes around 40 V with different voltage ripple on capacitors in VM due to different Cna and Cnb. It can be seen from Figure 7 that the increase of voltage ripple on capacitor in VM will make the steady-state accuracy of the proposed model worse, but the dynamic response of the proposed model is not affected significantly. When the voltage ripple on capacitors in VM is 10%, Figure 7c, Figure 8a and Figure 9a show the output voltage waveforms when the duty cycle changes around 0.6 with different VM number at Cna = Cnb = 2 μF, while Figure 7d, Figure 8b and Figure 9b show the output voltage waveforms when the input voltage changes around 40 V with different VM number at the same capacitance. Clearly, as the VM number increases, the steady-state error between the proposed model and the actual circuit increases too. This steady-state error is causes by the ripple in the capacitor voltage which affects the converter voltage gain [12].
From the above observation, it can be seen that the proposed model will become less accurate in steady-state as the capacitor voltage ripple in VM or the number of VM cells increases. Nevertheless, these negative effects are very limited and negligible for control system design, which can be observed from Figures 7-9. On the other hand, the proposed reduced-order model is capable of capturing the major dynamic characteristics of the detailed circuit model. This can be seen from Figures 7-9, as the dynamic responses are very close. Above characteristics of the proposed reduced order model make it valuable in the controller design of the converter.

Controller Design
A voltage and current double-loop control scheme is designed and shown in Figure 10, where Gv(s) and Gi(s) are the transfer functions of the voltage and current regulators respectively. One/VM is the equivalent gain of the PWM modulator and it is equal to 1.33 V based on the control chip (ISL6558) specifications. Furthermore, H1(s) and H2(s) are the voltage and current measurement gain respectively. The system output voltage is 400 V, and its reference voltage uref is set to 4.0 V, so the voltage sampling factor H1(s) is kept at 0.01. The current measurement gain H2(s) is kept at 0.1. The voltage loop gain is affected by the current loop gain, so the design of the current loop compensator needs to be completed first. The current loop gain without the current loop compensator Gi(s) can be obtained by (18) as Ti-e(s).
Based on Table 3 and (18) The current loop compensator Gi(s) with crossover frequency at 4.7 kHz, are designed as in (21). The Bode diagram for current loop gains with and without compensation are shown in Figure 11.  From the above observation, it can be seen that the proposed model will become less accurate in steady-state as the capacitor voltage ripple in VM or the number of VM cells increases. Nevertheless, these negative effects are very limited and negligible for control system design, which can be observed from Figures 7-9. On the other hand, the proposed reduced-order model is capable of capturing the major dynamic characteristics of the detailed circuit model. This can be seen from Figures 7-9, as the dynamic responses are very close. Above characteristics of the proposed reduced order model make it valuable in the controller design of the converter.

Controller Design
A voltage and current double-loop control scheme is designed and shown in Figure 10, where G v (s) and G i (s) are the transfer functions of the voltage and current regulators respectively. One/V M is the equivalent gain of the PWM modulator and it is equal to 1.33 V based on the control chip (ISL6558) specifications. Furthermore, H 1 (s) and H 2 (s) are the voltage and current measurement gain respectively. The system output voltage is 400 V, and its reference voltage u ref is set to 4.0 V, so the voltage sampling factor H 1 (s) is kept at 0.01. The current measurement gain H 2 (s) is kept at 0.1.

Controller Design
A voltage and current double-loop control scheme is designed and shown in Figure 10, where Gv(s) and Gi(s) are the transfer functions of the voltage and current regulators respectively. One/VM is the equivalent gain of the PWM modulator and it is equal to 1.33 V based on the control chip (ISL6558) specifications. Furthermore, H1(s) and H2(s) are the voltage and current measurement gain respectively. The system output voltage is 400 V, and its reference voltage uref is set to 4.0 V, so the voltage sampling factor H1(s) is kept at 0.01. The current measurement gain H2(s) is kept at 0.1. The voltage loop gain is affected by the current loop gain, so the design of the current loop compensator needs to be completed first. The current loop gain without the current loop compensator Gi(s) can be obtained by (18) as Ti-e(s).
Based on Table 3 and (18)   The voltage loop gain is affected by the current loop gain, so the design of the current loop compensator needs to be completed first. The current loop gain without the current loop compensator G i (s) can be obtained by (18) as T i-e (s).
Based on Table 3 and (18) and the above parameters of V M and H 2 (s), G id (s) and T i-e (s) can be obtained as follows.
The voltage loop gain without the voltage loop compensator G v (s) can be obtained by (22) as T v-e (s).
Energies 2019, 12, x FOR PEER REVIEW 13 of 18 The voltage loop gain without the voltage loop compensator Gv(s) can be obtained by (22) as Tv-e(s).
Based on Table 3, (9)-(11), (13) The voltage loop compensator Gv(s) with crossover frequency at 513 Hz, are designed as in (25). The Bode diagrams for voltage loop gains with and without compensation are shown in Figure 12.  Furthermore, in order to validate the derived transfer functions (13) and (15) in frequency domain, Bode diagrams for Gid(s) and Gud(s) are obtained using both the derived small-signal AC model transfer functions using MATLAB/Simulink, and the computer simulations based on PSIM circuit, as shown in Figure 13. Apparently, the frequency response of the transfer functions from the proposed reduced-order model are in close agreement with the PSIM simulation results within the frequency range below 3 kHz, which means the proposed reduced-order model and the transfer functions can be used in the controller design.

Simulation and Experimental Verification
This section examines the performance of the controller designed using the method presented in Section 4. The simulation circuit shown in Figure 10a is implemented in PSIM and the simulation results are shown in Figure 14. The specifications of the simulation is given in Table 3 as the rated parameters. Figure 14a shows the response of the output voltage uo to step changes in the input voltage uin. The input voltage uin increases from 30 V to 40 V, and then drops back to 30 V in 50 ms. Figure 14b shows the simulation waveforms of the output voltage uo and the output current io for step changes in load resistance. The load resistance increases from 400 Ω to 800 Ω first, and then reduces to 400 Ω in 50 ms. Clearly, when uin changes suddenly, uo is capable of tracking its reference Furthermore, in order to validate the derived transfer functions (13) and (15) in frequency domain, Bode diagrams for G id (s) and G ud (s) are obtained using both the derived small-signal AC model transfer functions using MATLAB/Simulink, and the computer simulations based on PSIM circuit, as shown in Figure 13. Apparently, the frequency response of the transfer functions from the proposed reduced-order model are in close agreement with the PSIM simulation results within the frequency range below 3 kHz, which means the proposed reduced-order model and the transfer functions can be used in the controller design. Furthermore, in order to validate the derived transfer functions (13) and (15) in frequency domain, Bode diagrams for Gid(s) and Gud(s) are obtained using both the derived small-signal AC model transfer functions using MATLAB/Simulink, and the computer simulations based on PSIM circuit, as shown in Figure 13. Apparently, the frequency response of the transfer functions from the proposed reduced-order model are in close agreement with the PSIM simulation results within the frequency range below 3 kHz, which means the proposed reduced-order model and the transfer functions can be used in the controller design.

Simulation and Experimental Verification
This section examines the performance of the controller designed using the method presented in Section 4. The simulation circuit shown in Figure 10a is implemented in PSIM and the simulation results are shown in Figure 14. The specifications of the simulation is given in Table 3 as the rated parameters. Figure 14a shows the response of the output voltage uo to step changes in the input voltage uin. The input voltage uin increases from 30 V to 40 V, and then drops back to 30 V in 50 ms. Figure 14b shows the simulation waveforms of the output voltage uo and the output current io for step changes in load resistance. The load resistance increases from 400 Ω to 800 Ω first, and then reduces to 400 Ω in 50 ms. Clearly, when uin changes suddenly, uo is capable of tracking its reference

Simulation and Experimental Verification
This section examines the performance of the controller designed using the method presented in Section 4. The simulation circuit shown in Figure 10a is implemented in PSIM and the simulation results are shown in Figure 14. The specifications of the simulation is given in Table 3 as the rated parameters. Figure 14a shows the response of the output voltage u o to step changes in the input voltage u in . The input voltage u in increases from 30 V to 40 V, and then drops back to 30 V in 50 ms. Figure 14b shows the simulation waveforms of the output voltage u o and the output current i o for step changes in load resistance. The load resistance increases from 400 Ω to 800 Ω first, and then reduces to 400 Ω in Energies 2019, 12,1971 14 of 17 50 ms. Clearly, when u in changes suddenly, u o is capable of tracking its reference at a constant value, although a small overshoot about 18 V is observed during the transient. When the load resistance suddenly changes, u o can also follow its reference with a small overshoot of 28 V. It is worth noting that the output voltage will be rapidly re-stabilized to 400 V.  In order to further verify the efficacy of the proposed model, experimental prototype was set up as shown in Figure 15. The specifications of the experimental prototype are consistent with the rated parameters of the simulation model, as shown in Table 3. Two scenarios are tested and the experimental results are compared with the simulated results. First, the input voltage uin increases from 30 V to 40 V and then drops back to 30 V, and the voltage changes are limited by a maximum ramp rate of ±1 V/ms. The second scenario is similar to that for Figure 14b: the load increases suddenly from 400 Ω to 800 Ω and then drops back to 400 Ω. The experimental waveforms of the circuit variables are shown in Figure 16. It can be clearly observed that in both scenarios, the output voltages are well-regulated at the set-point. The overshoot of the output voltage is much smaller than that which is obtained from the simulation shown in Figure 14; such differences are caused by the limitation of the maximum change rate of the of the DC source and the load. In order to further verify the efficacy of the proposed model, experimental prototype was set up as shown in Figure 15. The specifications of the experimental prototype are consistent with the rated parameters of the simulation model, as shown in Table 3. at a constant value, although a small overshoot about 18 V is observed during the transient. When the load resistance suddenly changes, uo can also follow its reference with a small overshoot of 28V. It is worth noting that the output voltage will be rapidly re-stabilized to 400V.

uo[V] io[A]
100V/div 0.5A/div 10ms/div (a) (b) In order to further verify the efficacy of the proposed model, experimental prototype was set up as shown in Figure 15. The specifications of the experimental prototype are consistent with the rated parameters of the simulation model, as shown in Table 3. Two scenarios are tested and the experimental results are compared with the simulated results. First, the input voltage uin increases from 30 V to 40 V and then drops back to 30 V, and the voltage changes are limited by a maximum ramp rate of ±1 V/ms. The second scenario is similar to that for Figure 14b: the load increases suddenly from 400 Ω to 800 Ω and then drops back to 400 Ω. The experimental waveforms of the circuit variables are shown in Figure 16. It can be clearly observed that in both scenarios, the output voltages are well-regulated at the set-point. The overshoot of the output voltage is much smaller than that which is obtained from the simulation shown in Figure 14; such differences are caused by the limitation of the maximum change rate of the of the DC source and the load. Two scenarios are tested and the experimental results are compared with the simulated results. First, the input voltage u in increases from 30 V to 40 V and then drops back to 30 V, and the voltage changes are limited by a maximum ramp rate of ±1 V/ms. The second scenario is similar to that for Figure 14b: the load increases suddenly from 400 Ω to 800 Ω and then drops back to 400 Ω. The experimental waveforms of the circuit variables are shown in Figure 16. It can be clearly observed that in both scenarios, the output voltages are well-regulated at the set-point. The overshoot of the output voltage is much smaller than that which is obtained from the simulation shown in Figure 14; such differences are caused by the limitation of the maximum change rate of the of the DC source and the load.

Conclusions
This paper proposes a method to develop control-oriented reduced-order models for a family of high step-up DC/DC converters based on VM with diode-capacitor network. It was found that a general reduced-order averaged model can be derived by replacing the VM capacitors with output-voltage-controlled sources and performing state-space averaging, and that the simplified model is the same as that of the traditional booster converter with an ideal step-up transformer. The corresponding small-signal AC models and transfer functions have been derived. The proposed approach can be readily applied to other high-gain DC/DC converters circuit using VM, and the model difference only lies in the expressions of circuit parameters of the generic model. Theoretical analysis, simulation and experimental results are given, showing that: (1) the complexity of modeling the high step-up DC/DC converter based on VM can be greatly reduced; (2) the reduced-order model can represent the low-frequency performance of the converter well and it is suitable for control system design.