Extension Sliding Mode Controller for Maximum Power Point Tracking of Hydrogen Fuel Cells

Abstract

Fuel cells (FCs) are characterized by low pollution, low noise, and high efficiency. However, the voltage-current response of an FC is nonlinear, with the result that there exists just one operating point which maximizes the output power given a particular set of operating conditions. Accordingly, the present study proposes a maximum power point tracking (MPPT) control scheme based on extension theory to stabilize the output of an FC at the point of maximum power. The simulation results confirm the ability of the controller to stabilize the output power at the maximum power point despite sudden changes in the temperature, hydrogen pressure, and membrane water content. Moreover, the transient response time of the proposed controller is shown to be faster than that of existing sliding mode (SM) and extremum seeking (ES) controllers.

1. Introduction

With growing international concern regarding the depletion of the world’s natural resources and the effects of global warming, the need for renewable energy sources has emerged as a critical issue in recent years. Among the various clean energy technologies available, for example, wind power, solar energy, and fuel cells (FCs), FCs are particularly attractive due to their low pollution, virtually silent running, simple maintenance, long service lives, and so on. Moreover, FCs have a generating efficiency of approximately 40–60%; with an overall efficiency of more than 80% being possible through their integration with cogeneration technologies. Thus, as the cost of FC materials continues to decrease, the commercialization of FCs for stationary and standalone applications is becoming increasingly attractive [1, 2].

The output voltage of an FC depends on the cell temperature, the air and hydrogen pressure, the oxygen partial pressure, and the membrane water content [3]. Notably, FCs have a nonlinear voltage-current characteristic, and thus there exists just one unique operating point which maximizes the output voltage and power under a particular set of operating conditions. Accordingly, a requirement exists for maximum power point tracking (MPPT) algorithms capable of controlling the stack current and fuel flow in such a way as to maximize the cell efficiency and output power under variable operating conditions [3].

Many MPPT methods for photovoltaic systems have been proposed, including Hill-climbing/Perturb and Observe (P&O), incremental conductance, fractional open-circuit voltage, fractional short-circuit current, fuzzy logic control (FLC), neural networks (NNs), ripple correlation control, current sweep, DC-Link capacitor droop control, load current/load voltage maximization, and sliding mode control (SMC) [4]. Such methods provide an effective means of solving the MPPT problem, but vary widely in terms of their complexity, implementation hardware, convergence speed, and sensed parameters [4]. Various MPPT algorithms have been presented for maximizing the output power of FC modules, for example, P&O [5–9], adaptive MPPT control [10], motocompressor control [11], adaptive FLC control [12], resistance matching [13], voltage and current based MPPT [14], and adaptive extremum seeking (ES) control [15].

As discussed previously, the FC efficiency and output power are highly sensitive to changes in the operating conditions. Previous studies have proposed reducing the effects of system disturbances on the FC performance by means of sliding mode (SM) [16] or fuzzy sliding mode (FSM) [17] controllers. In the present study, the output voltage of an FC is stabilized at the point of maximum power by means of an SM controller based on extension theory. The performance of the proposed extension sliding mode (ESM) controller is evaluated and compared with that of existing MPPT control schemes by means of Matlab/Simulink simulations.

2. Mathematical Model of Fuel Cell

Figure 1 presents a schematic illustration of the FC power generation process. As shown, the FC is fed with hydrogen and oxygen, and an electrical current is generated as a result of the subsequent oxidation/reduction reactions at the anode and cathode electrodes, respectively. The relevant chemical reactions are as follows [18, 19]: Anode∶H₂⟶2H⁺+2e⁻,Cathode∶O₂+4H⁺+4e⁻⟶2H₂O. (1)(2)

Figure 1 

FC power generation process.

In developing the proposed MPPT control scheme, the present study considers the hydrogen FC developed by Ballard Power Systems (Canada). For a single FC, the output voltage is given as 𝑉_FC=𝐸_𝑛−𝑉_act−𝑉_ohmic−𝑉_con, (3) where 𝐸_𝑛 is the thermodynamic reversible voltage exported from the FC and has the form 𝐸_𝑛=1.229−8.5×10⁻⁴(𝑇−298.15)+4.31×10⁻⁵𝑇ln𝑃_H2𝑃_O2, (4) in which 𝑃_H2 and 𝑃_O2 are the hydrogen and oxygen pressures, respectively, and 𝑇 is the operating temperature.

𝑉_act in (3) is the activation overvoltage, that is, the combined cathode and anode activation overvoltage, and is given as 𝑉_act=−𝛿₁+𝛿₂𝑇+𝛿₃𝑇lnC_O2+𝛿₄𝑇ln𝐼_FC, (5) where 𝛿_𝑛  (𝑛= 1~4) are parametric coefficients, and C_O2 is the concentration of dissolved oxygen at the gas/liquid interface and is computed as C_O2=𝑃_O25.08×10⁶×𝑒^{(−408/𝑇)}. (6)𝑉_ohmic in (3) is the ohmic overvoltage, that is, the voltage loss within the FC resulting from the resistance of the various cell components and interconnects; that is, 𝑉_ohmic=𝐼_FC𝑅_𝑀+𝑅_𝐶, (7) where 𝑅_𝐶 denotes that the resistance contacting and the electron stream is constant and 𝑅_𝑀 is the impedance of the proton exchange membrane; that is, 𝑅_𝑀=𝜌_𝑀×𝑙𝐴, (8) in which 𝐴 is the cell reaction area, 𝑙 is the membrane thickness, and 𝜌_𝑀 is the membrane resistance coefficient given as 𝜌_𝑀=181.61+0.03𝐽+0.062(𝑇/303)²(𝐽)^2.5[][(.𝐺−0.634−3𝐽exp4.181−303/𝑇)] (9)

Note that in (9), 𝐺 is the membrane water content and 𝐽 is the current density; that is, 𝐼𝐽=^FC𝐴. (10) Finally, 𝑉_con in (3) is the concentration overvoltage resulting from diffusion-limited mass transfer and is given as 𝑉_con=−𝑅𝑇𝐽𝑛𝐹ln1−𝐽_max𝐽=−𝐵⋅ln1−𝐽_max, (11) where 𝑅 is the universal gas constant, 𝐹 is Faraday’s constant, 𝐽_max is the maximum current density, and 𝐵=𝑅𝑇/𝑛𝐹.

For an FC cell stack comprising 𝑁 fuel cells, the output power is equal to 𝑃_stack=𝑁𝑉_FC×𝐼_FC. (12)

As shown in (3)–(12), the output power of an FC varies in accordance with changes in the operating temperature, hydrogen pressure, reaction area, and membrane water content, respectively. Figure 2 presents the characteristic curve of the Ballard FC considered in the present study. As shown, in the ideal state, the open-circuit voltage of the FC is equal to approximately 1.244 V for any load current condition. In practice, however, the voltage drops with an increasing load due to a polarization effect. Thus, as described in the previous section, the present study proposes an SM control scheme based on extension theory to stabilize the output power of the FC at the point of maximum power.

Figure 2 

Characteristic curve of Ballard FC.

3. Overall Design of Maximum Power Point Tracking Controller

Figure 3 illustrates the functional architecture of the proposed FC/control system. As shown, the main components include the FC, the extension sliding mode (ESM) controller, a DC-DC boost convertor, and a resistive load. In the proposed approach, the output voltage and current signals of the FC are taken as inputs to the ESM controller, which then regulates the DC-DC boost converter in such a way as to maintain the output power of the FC at the maximum power point by means of a pulse-width modulation (PWM) switching scheme [20].

Figure 3 

Schematic diagram of proposed FC system.

The details of the PWM switching scheme are as follows.

State 1.  When switch 𝑄 is on, the diode 𝐷 is cut off, and the input voltage flows through the inductor 𝐿 directly. Thus, the inductive current 𝐼_𝐿 increases and capacitor 𝐶 is required to supply energy to the load. As a result, current 𝐼_𝐿 flows through the switch to form a loop. The corresponding voltage and current equations are as follows: 𝐼̇_{𝐿(𝑄_on)}=𝑉_FC𝐿,𝑉̇_{𝑜(𝑄_on)}=−𝑉_𝑜.𝐶𝑅 (13)(14)

State 2.  When switch 𝑄 is off, the diode 𝐷 is switched on. Thus, the energy stored in the inductor 𝐿 is discharged to the output end and 𝐼_𝐿 reduces toward zero. The relevant current and voltage equations are as follows: 𝐼̇_{𝐿(𝑄_off)}=𝑉_FC𝐿−𝑉_𝑜𝐿,𝑉̇_{𝑜(𝑄_off)}=𝐼_𝐿𝐿−𝑉_𝑜.𝐶𝑅 (15)(16)

Combining (13)–(16), the current and voltage equations for the PWM controller are as follows: 𝐼̇_𝐿=(𝑈−1)×𝑉_𝑜𝐿+𝑉_FC𝐿,𝑉̇_𝑜=(1−𝑈)×𝐼_𝐿𝐶−𝑉_𝑜,𝐶𝑅 (17) where 𝑈 is the duty cycle, 𝑉_𝑜 is the output voltage, and 𝐼_𝐿 is the inductor current.

4. Proposed Extension Sliding Mode Controller

In the sliding mode control (SMC) method, the control rules are designed in such a way that the system slides continuously along a sliding surface comprising the boundaries of multiple control structures. In the MPPT control system developed in this study, the sliding surface is set as 𝑑𝑝/𝑑𝐼=0, and the maximum power point is tracked via the appropriate manipulation of the DC-DC converter by the ESM controller [21, 22].

4.1. Mathematical Derivation

For the proposed ESM control scheme, the maximum power point occurs when 𝑑𝑝/𝑑𝐼=0. In other words, the sliding surface to be controlled is defined as follows: 𝑆=𝑑𝑃𝑑𝐼_FC. (18)

Substituting 𝑃=𝐼𝑉 into (18) and simplifying, the sliding surface is obtained as 𝑆=𝑑𝐼_FC𝑉_FC𝑑𝐼_FC=𝑉_FC+𝐼_FC⋅𝜕𝑉_FC𝜕𝐼_FC, (19) where 𝜕𝑉_FC𝜕𝐼_FC=𝜕𝜕𝐼_FC𝐸_𝑛−𝜕𝜕𝐼_FC𝑉_act−𝜕𝜕𝐼_FC𝑉_ohmic−𝜕𝜕𝐼_FC𝑉_con,𝜕𝜕𝐼_FC𝐸_𝑛𝜕=0,𝜕𝐼_FC𝑉_act𝛿=−⁴𝑇𝐼_FC,𝜕𝜕𝐼_FC𝑉_ohmic=𝑅_𝑀+𝑅_𝐶+𝐼_FC×𝜕𝑅_𝑀𝜕𝐼_FC,𝜕𝑅_𝑀𝜕𝐼_FC=𝑙𝐴_𝐶×𝜕𝜌_𝑀𝜕𝐼_FC,𝜕𝜕𝐼_FC𝜌_𝑀=𝑀×0.03𝐴+𝑁×2.5×𝐽^1.5×1𝐴−𝑂,𝑀=181.6𝐺−0.634−3𝐼_FC[(,𝑇/𝐴exp4.181−303/𝑇)]𝑁=0.062×303²𝑂=−31+0.03𝐽+𝑁×𝐽^2.5(),𝐺−0.634−3𝐽𝜕𝑉_con𝜕𝐼_FC1=𝐵×𝐴𝐽_max−𝐼_FC. (20)(21)(22)(23)(24)(25)(26)(27)(28)(29)

As described previously, the aim of the ESM controller is to track the maximum power point, that is, to drive the system toward the following sliding surface: 𝑆=𝑉_FC+𝐼_FC⋅𝜕𝑉_FC𝜕𝐼_FC=0. (30)

From Figure 3, it is seen that 𝐼_𝐿=𝐼_FC. Thus, the PWM control rule required to satisfy (30) can be expressed as the following (31) for the system dynamics to remain on the switching surface 𝑆̇=𝜕𝑆𝜕𝐼_FC𝜕𝐼_FC=𝜕𝑡𝜕𝑆𝜕𝐼_FC×)×𝑉(𝑈−1^𝑜𝐿+𝑉_FC𝐿=0. (31)

In the ideal case, 𝐼̇_FC=0. Thus, the duty cycle in (31) has the form 𝑈_eq𝑉=1−^FC𝑉_𝑜. (32) The duty cycle 𝑈 must have a value in the range of 0 to 1. Thus, in implementing the proposed ESM control scheme, the duty cycle is derived as follows: 𝑈=1𝑈_eq𝑈+𝜂𝑆≥0_eq+𝜂𝑆0<𝑈_eq+𝜂𝑆<10,𝑈_eq+𝜂𝑆≤0, (33) where 𝜂>0 is the control gain which is to be determined. In previous studies, the control gain required to minimize steady-state chattering was determined using a fuzzy sliding mode scheme. By contrast, in the present study, the optimal value of 𝜂 is determined using extension theory.

4.2. Proposed Gain Tuning Method

Extension theory was originally proposed by Cai in 1983 and contains two main disciplines, namely, matter-element theory and extension set theory [23, 24].

(1) Definition of Matter-Element.  Defining the name of the matter as 𝑁, the characteristics of the matter as 𝐶, and the value of 𝐶 as 𝑈, the matter-element can be expressed as follows: 𝑅=𝑁𝐶₁𝑈₁𝐶⋮⋮_𝑛𝑈_𝑛, (34) where 𝐶=[𝑐₁,𝑐₂,…,𝑐_𝑛] is a characteristic vector and 𝑈=[𝑢₁,𝑢₂,…,𝑢_𝑛] is the value vector of 𝐶.

(2)  Definition of Extension Set.  Let Ω be the universe of discourse. The extension set 𝐴 on Ω  is then defined as the following set of ordered pairs: 𝐴=𝑥,𝑦∣𝑥∈Ω,𝑦=𝐾(𝑥)∈(−∞,∞), (35) where 𝑦=𝐾(𝑥) is referred to as the correlation function for extension set 𝐴 and has the form shown in Figure 4.

Figure 4 

Extended correlation function.

As shown in Figure 4, 𝐾(𝑥) maps each element of Ω  to a membership grade between −∞ and ∞. The higher the degree of the membership grade, the more the number of elements which belong to the set. It is noted that the particular case of 0≤𝐾(𝑥)≤1 corresponds to a normal fuzzy set. Moreover, the condition 𝐾(𝑥)≤1 implies that element 𝑥 has no chance of belonging to the set. By contrast, the condition −1<𝐾(𝑥)<0 represents the extension domain; that is, element 𝑥 has a chance of belonging to the set.

The gain tuning method proposed in this study can be summarized as follows [10].

Step 1. Let the duty cycle of the PWM switch be redefined as 𝑈𝑈=𝜂𝑆,𝑆∈𝐸_eq+𝜂𝑆,𝑆∉𝐸, (36) where 𝐸 is the extension field (see Figure 4). Appropriate control laws are designed in the rational region and extension region, respectively, in order to ensure system stability. The extension correlation function and corresponding weights are then calculated in accordance with the defined characteristic values of 𝑆 and 𝑆̇, respectively.

Step 2. Partition the unknown matter element into various grade set characteristics, where the number of grade sets is equal to the number of sets in the classical domain and joint domain; that is, 𝑅=𝑁,𝐶_𝑖,𝑉_𝑖=𝑁𝐶₁𝑎₁,𝑏₁𝐶₂𝑎₂,𝑏₂𝐶⋮⋮_𝑛𝑎_𝑛,𝑏_𝑛, (37) where 𝑁 is the matter element which is to be divided, 𝐶_𝑖 are the characteristics of the matter element, and 𝑉_𝑖 is the characteristic value range, that is, the distribution range of the 𝑖th characteristic. Finally, 𝑎_𝑖 indicates the maximum value of the 𝑖th characteristic among the matter element grade sets, while 𝑏_𝑖 denotes the minimum value of the 𝑖th characteristic.

Step 3. Assign weights to the various characteristics such that ^𝑛_𝑖=1𝜆_𝑖=1. (38)

In the ESM control scheme proposed in the current study, the weightings are determined via the following entropy-based approach; that is, 𝜆_𝑖=^𝑛_𝑖=11−𝜍×𝑉_𝑖⁻¹,0<𝜍<1. (39)

Step 4. Calculate the degree of association between the data to be measured in accordance with 𝐾𝑥_𝑖=−𝜌𝑥_𝑖,𝑉_𝑖𝑉_𝑖,−1≤𝜌𝑥_𝑖,𝑉_𝑖𝑥≤0or𝜌_𝑖,𝑉_𝑖𝜌𝑥<−1_𝑖,𝑉_𝑖𝜌𝑥_𝑖,𝑉_𝑡−𝜌𝑥_𝑖,𝑉_𝑖,otherwise, (40) where 𝜌𝑥_𝑖,𝑉_𝑖=𝑥_𝑖−𝑎_𝑖+𝑏_𝑖2−𝑏_𝑖−𝑎_𝑖2,𝜌𝑥_𝑖,𝑉_𝑡=𝑥_𝑖−𝑐_𝑖+𝑑_𝑖2−𝑑_𝑖−𝑐_𝑖2. (41)

Step 5. Calculate the relative values of the degree of association of the various grade sets using the normalization equation shown in (42) such that the degree of association of each grade set falls within ⟨1,−1⟩. 𝐾𝑞^∗=2𝐾𝑞−𝐾𝑞_min−𝐾𝑞_max𝐾𝑞_max−𝐾𝑞_min. (42)

Step 6. Determine which grade set the matter element belongs to. If 𝐾(𝑞)^∗=1, the degree of correlation is of the evaluation result of type 𝑖. The probability of other set types is determined according to the degree of association. The outcome of this step is used to identify the gain value 𝜂 which ensures that the output power of the FC tracks the maximum power point.

5. Simulation Results and Discussion

The performance of the proposed extension sliding mode (ESM) controller was evaluated by means of a series of Matlab/Simulink simulations based on the integrated model shown in Figure 5. Note that in performing the simulations, the maximum output power of the FC was assumed to be 550 W and the simulation run time was specified as 10 s. In addition, the matter-element module was set as shown in (43).

Figure 5 

Integrated Simulink model of FC and ESM control system.

The characteristic parameters of the simulated fuel cell system and DC-DC converter are shown in Table 1.

Table 1 

Parameters of fuel cell system and DC-DC converter.

Figure 6 shows the simulation results obtained for the output power of the FC given an operating temperature of 25°C. Figures 7, 8, and 9 show the simulated response of the controlled FC given step changes in the temperature, hydrogen pressure, and membrane water content from 25°C to 60°C, 0.5 atm to 10 atm, and 10 to 20, respectively. The results confirm the ability of the proposed ESM control scheme to stabilize the output power of the FC at the point of maximum power despite perturbations in the FC operating conditions. Consider the following: ⟨⟩⟨⟩⟨⟩⟨⟩⟨⟩⟨⟩⟨⟩⟨⟩⟨⟩𝑅=sliding𝑆50,9020,5010,205,101,50.5,10.1,0.50.05,0.10,0.05. (43)

Figure 6 

Evolution of FC output power under normal operating conditions.

Figure 7 

FC output power response given step change in operating temperature.

Figure 8 

FC output power response given step change in hydrogen pressure.

Figure 9 

FC output response given step change in membrane water content.

Figure 10 and Table 2 compare the transient and steady-state performance of the FC when controlled by three different schemes, namely, the ESM scheme proposed in the present study, the conventional SM scheme [21], and an extremum seeking (ES) scheme [15]. It is seen in Figure 10 that the ES method yields a slow transient response and a poor steady-state stability. Furthermore, it is seen that while the ESM and SM methods both stabilize the output power at the maximum power point, the transient response of the proposed ESM method is faster than that of the traditional SM method.

Table 2 

Comparison of different control methods.

Figure 10 

Comparison of control performance of ESM, SM, and ES methods.

6. Conclusions

A sliding mode (SM) controller based on extension theory has been proposed for stabilizing the output power of a fuel cell (FC) at the maximum power point. In the proposed approach, the sliding mode is used to track the maximum power point, while extension theory is used to determine the appropriate value of the controller gain which ensures convergence of the controlled system to the sliding surface. The simulation results have shown that the proposed controller successfully stabilizes the output power of the FC at the maximum power point and has a faster transient response time than existing sliding mode (SM) or extremum seeking (ES) control schemes.

Acknowledgment

This paper was supported by the National Science Council of Taiwan, under Grant NSC-2628-E-167-002-MY3.