Load-Sharing Management for Fuel Cell Hybrid Electric Vehicle (FCHEV) Based on Intelligent Controllers and Optimization Algorithms

Abstract

This study proposes an intelligent controller for a hydrogen-powered plug-in fuel cell hybrid electric vehicle (FCHEV) that integrates a fuel cell (FC) with two energy storage systems, which are ultracapacitor (UC) and battery (BAT), which results in a high dynamic response along with maintaining efficient use of resources for energy storage. Moreover, this controller works on effectively managing the system power flow to reduce the amount of power needed for the FCHEV. An effective energy management system (EMS) has been developed that utilizes the fuzzy logic controller (FLC) and artificial neural networks (ANNs) to achieve the EMS requirements. Also, the proposed system operates these three power sources at high efficiency with their mechanism performance, meets load power demands efficiently, and uses less hydrogen. Furthermore, the Crow Search Algorithm (CSA) and Particle Swarm Optimization (PSO) methods are utilized to adjust the parameters of the wavelet neural network that is connected to the PI controller, called WNN-PI. The DC-DC converters control the output voltage of the FC and BAT for maintaining the DC-bus voltage constant at 300 volts. The state-of-charge (SOC) for the BAT and UC is also considered in this study. The proposed system is analyzed and evaluated using the MATLAB/Simulink environment, and two vehicle driving cycles were implemented using the ADVISOR Simulator.

1. Introduction

Nowadays, demand for electric vehicles (EV), as well as hybrid electric vehicles (HEVs), have grown in popularity for environmental reasons. Moreover, there are also several other benefits, such as a decrease in pollution also a cessation of dangerous gas emissions. The internal combustion engine (ICE) is regarded as one of the most widely utilized in transportation services as the primary power for driving automobiles. The ICE is a remarkable achievement in contemporary technology since it offers a high level of power and has the ability to drive a vehicle over extended distances. Due to the fact that ICE burns petrol, it emits CO₂ and has a number of other drawbacks, including a high degree of complexity, loud noise operation, and entire reliance on a single fuel source.

In recent years, scientists and developers have concentrated on enhancing the performance of EVs [1, 2, 3]. HEVs especially are more functional and environmentally friendly than traditional cars [4]. The fuel cell electric vehicle (FCEV) uses hydrogen as fuel and produces only water as emissions. The FC engine has certain disadvantages as well, including a sluggish dynamic response, a high hydrogen cost, and a failure to use the car’s regenerative braking power. To address these deficiencies, an FCHEV should be developed by integrating the FC engine with energy-storage devices such as UCs and BATs. The ultracapacitor (UC) has a very rapid dynamic reaction in order to adapt to the vehicle’s unexpected load fluctuations. Therefore, the FC output power must be regulated. In addition, throughout the charging and discharging phases, the power flow between the battery, UC, and the load also must be efficiently managed. In order to actively optimize the control system, an energy management system (EMS) is necessary; along with the central control system for each converter, power electronics converters are also necessary to connect the power sources with the loads. Consequently, a lot of publications and investigation papers [5, 6, 7, 8, 9, 10, 11], as well as comprehensive review papers [12, 13, 14] for studying the importance of energy management for FCHEV.

The objective of this work is to create an intelligent energy management system (EMS) for the FCHEV that utilizes less hydrogen while effectively meeting the load power needs. The main power source of our proposed model plug-in hybrid FCEV is that the FC is meant to give power for steady-state loads only, whereas the UC is responsible for providing the transient power and the BAT is responsible for providing the medium frequency power. Also controlling the DC bus voltage; additionally, the power of the BAT supports the power output of the fuel cell in feeding the load if needed. This possibly be accomplished by providing the FC and BAT switching converters with the proper signals to guarantee that the hybrid FC system is working properly to handle the load dynamics. Also, this work presents two optimization algorithms, which are PSO and CSA.

2. Fuel cell hybrid electric vehicle configurations

Developing EVs has many factors, the most important one is the energy source cost sector. Therefore, researchers are being conducted using different designs of storage energy as well as a control system that aims to reduce the cost of energy storage with consideration of keeping a high level of improving efficiency. Thus, different configurations of EVs have been introduced [15], for example, some classifications of hybrid electric vehicle (HEV) such as series, parallel, and series–parallel. Meanwhile, the plug-in hybrid electric vehicle (PHEV) has the same configuration as HEV but with an extra plug-in port connection to charge from an external source at home or public charge stations. As shown in Figure 1 the PHEV block diagram.

Figure 1.

PHEV block diagram.

Overall, electric vehicles can be classified into three basic types, which are battery electric vehicles (BEVs) that use only batteries, fuel cell electric vehicles (FCEVs) that use only fuel cells stack, and hybrid electric vehicles (HEVs) that use a hybrid system. For example, HEVs such as batteries and other energies are called BHEVs or fuel cells stack and other energies are named FCHEVs. Moreover, all HEV types can be plug-in or not.

The working principle of FCHEV can be summarized and simplified in three steps:

• Two or three powers in FCHEV, which are a fuel cell, battery, and/or ultracapacitor combined; and controlled by an (ex: intelligent control method) for the energy management system (EMS). First starting of the vehicle at very low load driving (low speed) of the FCHEV run by one of the power sources ultracapacitor or battery if the ultracapacitor/battery has sufficient energy to ensure soft starting.

• The FCHEV is run by fuel cells stack individually at normal load driving (normal speed) to drive the vehicle and charge the ultracapacitor/battery; but at accelerating (high speed), the fuel cells stack and ultracapacitor/battery are combined together to get powerfully driving.

• When deceleration (Brake), the vehicle recovers the kinetic energy of tire friction and converts it into electricity, in order to charge the ultracapacitor/battery.

Generally, there are three main classification topologies of FCHEVs configuration, which are passive connection, full active connection, and semi-active connection [16]. In the passive configuration, the FC stack, BAT, and UC are directly connected to the DC bus, without employing any electrical power converters, as illustrated in Figure 2. The design of the passive connection structure is flexible and simple construction as well as high efficiency, low power losses, and low costs [16]. Meanwhile, this topology structure has disadvantages [16, 17] such as the output voltage of FC, BAT, and UC must be the same. Also, unable to be controlled power sources under the energy management strategy. Moreover, the vehicle weight is high and requires a large FC generator to meet the DC bus voltage.

Figure 2.

The fundamental block diagram of the passive connection topology.

The fully active configuration uses three DC-DC converters where each one is connected to the DC bus through independent power electronic converters as shown in Figure 3. The fully active configuration structure can use the FC generator with a voltage much less than the DC voltage bus due to combined to boost DC-DC converter; then, this is the main advantage of this topology. Also, the weight of the vehicle in terms of the power sources (FC, BAT, and UC) is smaller than the passive connection. However, the fully active configuration has some drawbacks such as the extra weight of adding three converters, high cost, and high power losses [17].

Figure 3.

The fundamental block diagram of the fully active connection topology.

In order to find the most efficient configuration, the semi-active topology is proposed, which combines and uses the features of the active and the passive connections. Moreover, there are two types of semi-active topology. The first is that the fuel cell is connected to the DC-bus voltage through a unidirectional DC-DC converter, whereas the BAT has a direct link to the DC-bus voltage, and the UC is attached to the DC-bus via a bidirectional DC-DC converter, as illustrated in Figure 4. The FC is connected to the DC-bus via a unidirectional DC-DC converter, although the UC is attached straight to the DC-bus, and the BAT is tied to the DC-bus voltage via a bidirectional DC-DC converter, as illustrated in Figure 5.

Figure 4.

Semi-active connection topology: BAT links directly to the DC-bus.

Figure 5.

Semi-active connection topology: UC links directly to the DC-bus.

This work aimed at improving the topology of the second type of semi-active topology (see Figure 5) via the proposed EMS that is designed to achieve that the BAT responds to load power that has low-frequency orders and the UC responds to power demand that has high-frequency orders since it is connected directly to DC bus. While the FC gives steady-state power, its efficiency is increasing.

3. Power unit converters of FCHEV

Regulation of the DC-bus voltage required DC-DC converters to regulate the voltage as well as adjust the power flow between the power sources of the developed FCHEV system. This work uses two topologies DC-DC converters, as shown in Figure 6, for the unidirectional boost converter that is used for the FC generator, and Figure 7 shows the converter of the BAT, which is a bidirectional buck-boost type.

Figure 6.

Circuit diagrams of the unidirectional boost converter.

Figure 7.

Circuit diagrams of the bidirectional buck-boost converter.

The waveforms can be averaged over a period of time that is brief in comparison to the system’s inherent time constants without appreciably influencing the response [18]. When the basic condition is met, averaging the single-quadrature converter waveforms throughout the switching period is a good approximation. The converter’s low-frequency behavior is predicted by the averaged model, which ignores the high-frequency switching harmonics. The single-quadrature converter is modeled by using a conventional transformer-less single-quadrature architecture. Therefore, the proposed work uses the average model converter instead of the electronic circuit model, which in return speeds up the simulation time and gets the same results. Figure 8 illustrates the proposed single-quadrature averaged converter model.

Figure 8.

Equivalent averaged model of single-quadrature DC-DC converter.

The relationships between the average of currents for the input and output converter are represented as [18]:

$${\mathrm{I}}_{\mathrm{o}}=\mathrm{\eta }\left(1-\mathrm{D}\right){\mathrm{I}}_{\mathrm{i}}$$E1

Generally, the efficiency of the converter should be constant. Assumed to be 90% at full load. The relationship between VI and VO can be calculated via the voltage second balance across the inductor as shown in Ref. [19]:

$$V_O=\frac{V_i}{1-D}$$E2

The two-quadrature converter is directly controlled by the duty cycle signal. The two-quadrant bidirectional outputs can be utilized in either a buck or boost configuration. Two-quadrature DC-DC converter average switch model can be derived based on buck (Charging Mode) and boost (Discharging Mode) operations. Figure 9 illustrates the proposed two-quadrature averaged converter model.

Figure 9.

Average model of two-quadrature converter in the buck mode.

In the operation of buck mode, the relationships for V_i and I_o can be obtained as shown below:

$$\mathrm{Vi}={\mathrm{DV}}_{\mathrm{o}}$$E3

$${\mathrm{I}}_{\mathrm{o}}={\mathrm{\eta DI}}_{\mathrm{i}}$$E4

Where D consists of d1 for the boost switch and d2 for the buck switch.

4. Power load driving cycles and power sources parameters

In order to accomplish the design and sizing of the parameters in terms of the power sources for FC, UC, and BAT of the FCHEV; and evaluate the reliability of FC, UC, and BAT power sources when utilized by the proposed EMS. Load profiles are needed; therefore, two driving cycles have been proposed, which are Urban Dynamometer Driving Schedule (UDDS) and Federal Test Procedure (FTP) were acquired by the ADVISOR analysis program. Since UDDS has the worst peak power consumption variations of any of the driving cycle standards, it serves as the primary example for analyzing FCHEV performance. Figure 10a and b illustrate the power profile of UDDS and FTP, respectively.

Figure 10.

The power load profile of (a) UDDS and (b) FTP.

On the basis of UDDS driving cycle power profile characteristics, the maximum load power is 11 kW while the average power is 7 kW. As a result, the power requirement for this load has been modeled on a hybrid system that integrates FC, BAT, and UC, which are built by extracting the actual technical specifications of these resources from their datasheets and incorporating them into the Simulink model environment to make system performance outcomes as realistic as feasible in the real world. The following are the power sources:

1. Fuel cell: Hydrogenics 12.5 kW HyPM-HD12 PEMFC; Number of cells is 65; FC datasheet in [20].

2. Battery: Valence Technology U-Charge U1-12XP lithium-ion BAT. The BAT is represented by four series-connected cells. The nominal voltage of each string is 12.6 V. BAT datasheet in [21].

3. Ultracapacitor: Maxwell Boostcap®BCAP1200 UC; the Maxwell 1200F, 2.7 V/cell Boostcap®BCAP1200. 120 cells in series in order to produce 300 volts due to the DC link voltage of the FCHEV system being 300 V and the UC being directly connected to the DC link. UC datasheet in [22].

5. Methodology

The proposed controllers (WNN-PI, and the developed (EMS by FLC and ANNs)) supervise the BAT and FC current boundaries to guarantee proper timing for the charging and discharging of the BAT/UC and no negative current supply to the FC. The BAT current is supplied to a two-quadrature DC-DC converter, whereas the current of the FC is sent to a single-quadrature converter and responds to steady-state load power. Due to the UC capabilities that allow it to react fast to sudden changes in power load and the UC is directly attached to the DC bus, therefore, it is recharged before the BAT recharges. BAT converter is bidirectional and thus capable of utilizing rapid brakes of the vehicle for recharging the BAT. Wavelet methods are used to regulate the BAT converter with wavelet and PI controller identified as WNN-PI, which are tuned online using optimization algorithms (PSO or CSA). The boost converter is used for the FC, which is managed by an EMS with either an FLC or an ANN. The overall objective of the EMS control procedures is to efficiently deliver the required power to the vehicle load power, to control hydrogen consumption as effectively as possible, to increase the responsiveness and run efficiency of the FC, to increase the lifespan of the FC, BAT, and UC, and to minimize the size of the FC stack system, which in return lowers the FC cost.

5.1 Fuzzy and ANN for EMS of FCHEV

First, the FC reference current is managed by the FLC control system. Given that the ANN requires input and output data to be trained. As a result, the FLC method has been used to build the EMS first in order to get the necessary input/output data. Then, the ANNs have been trained after obtaining the necessary data via FLC. Keeps the fuel cell power generator output under control and satisfies all the FCHEV requirements for safe and efficient operation. Therefore, to control and achieve an efficient run of FC, consider the output power of the fuel cell in three cases; which are the minimum FC power (Pfc_min) is from zero till reaches 0.49 kW, and the maximum FC power (Pfc_max) is from 1.59 kW till reaches 11 kW, then the optimal FC power output (Pfc_opt) is from 0.5 kW till reach 1.6 kW. The fuel cell reference current is calculated by using a set of Fuzzy If-Then Rules which take into account both the power needed from the FC and the SOC limits for the battery, as illustrated in Figure 11, the FLC model in the Simulink environment. Moreover, the developed FLC has been implemented using a trapezoidal memberships function (trapmf).

Figure 11.

The EMS model with fuzzy logic control.

The neural control system has two inputs, which are the load power and the SOC of the battery. The output of ANNs is the FC reference power. The parameters of the ANNs have been defined as the following:

1. The network used is “feedforwardnet” training algorithm that used the “Levenberg–Marquardt” method to train the neural network.

2. The number of epochs is 1000 epoch

3. Three hidden layers have been employed, and each layer’s neurons are 30, 20, 15, and 10, which are chosen by the trial-and-error theorem.

4. Training goad (MSE) is 10⁻³⁰

Whereby 15% are used for testing, 15% for validation, and 70% are for training. The codes used to create this network and the network diagram are shown in Figure 12.

Figure 12.

ANNs codes and network diagram.

5.2 Wavelet strategy and optimization algorithm for EMS of FCHEV

The actual DC link voltage is subtracted from the reference DC voltage 300 V to get the error signal, which is then fed to the WNN-PI controller in order to get the BAT current reference, which in return controls the BAT DC-DC converter. The feed-forward wavelet neural network is used in this work (WNN-PI). Figure 13 shows the Simulink model; the PSO and CSA are used to tune the WNN-PI parameters, which are WNN variables dilation (a’s), translation (b’s), and weights (w’s); PI controller parameters are Ki and Kp.

Figure 13.

Battery controller scheme by wavelet and PI in Simulink model.

The fitness function for an optimal value is obtained using the “Integral of Squared Error” (ISE), which is the same function used by both PSO and CSA optimization methods.

$$\text{fitness function}=min\left(\mathrm{ISE}\right)$$E5

$$\mathrm{ISE}=\int {\mathrm{e}}^2\left(\mathrm{t}\right)\mathrm{dt}$$E6

$$e\left(j\right)=\mathrm{D}\left(\mathrm{j}\right)-\mathrm{y}\left(\mathrm{j}\right)$$E7

Where e(j) is the error that is produced when the desired value D(j) is subtracted from the actual value of the model y(j). Following that, PSO should use Eq. (6) to continuously update each swarm particle’s current position xj (m) and velocity vj at (m):

$${\mathrm{v}}_{\mathrm{j}}^{\mathrm{m}+1}=\mathrm{w}\ast {\mathrm{v}}_{\mathrm{j}}^{\mathrm{m}}+{\mathrm{c}}_1\ast {\mathrm{R}}_1\ast \left({\text{lbest}}_{\mathrm{j}}-{\mathrm{x}}_{\mathrm{j}}^{\mathrm{m}}\right)+{\mathrm{c}}_2\ast {\mathrm{R}}_2\ast \left({\text{gbest}}_{\mathrm{j}}-{\mathrm{x}}_{\mathrm{j}}^{\mathrm{m}}\right)$$E8

Applying Eq. (9) to modify the current position:

$${\mathrm{x}}_{\mathrm{j}}^{\mathrm{m}+1}={\mathrm{x}}_{\mathrm{j}}^{\mathrm{m}}+{\mathrm{v}}_{\mathrm{j}}^{\mathrm{m}+1}$$E9

A generalized wavelet is created by linearly combining a collection of daughter wavelets ${\mathrm{\psi }}_{\mathrm{a},\mathrm{b}}$:

$${\psi }_{a,b}=\psi \left(\frac{x-b}{a}\right)$$E10

The wavelet’s ultimate output is as follows:

$$y=\sum _{n=1}^N{w}_N{\psi }_{a_N,b_N}$$E11

For each crow, the CSA creates a new location in the search space. Assume that crow i follows a crow (for example, crow j) at random to find out where crow j is. Crow i’s position update is divided into two circumstances. First, crow j is unaware that crow i is following it. Using Eq. (12), evaluate the fitness function, and choose a random position. Second, crow j detects crow i and follows it, and crow j moves crow i to a random position. Based on Eq. (6) results, Eq. (13) follows [23]:

$${\mathrm{X}}^{\mathrm{i},\mathrm{itr}+1}=\begin{array}{c}{\mathrm{X}}^{\mathrm{i},\mathrm{itr}}+{\mathrm{r}}_{\mathrm{i}}\times \mathrm{f}{\mathrm{l}}^{\mathrm{i},\mathrm{itr}}\times \left({\mathrm{m}}^{\mathrm{j},\mathrm{itr}}-{\mathrm{X}}^{\mathrm{i},\mathrm{itr}}\right){\mathrm{r}}_{\mathrm{j}}\ge \mathrm{A}{\mathrm{P}}^{\mathrm{j},\mathrm{itr}}\\ \mathrm{a}\text{random position otherwise}\end{array}$$E12

Then, using Eq. (13) for updating the memory matrix of each crow based on Eq. (12) [24]:

$${\mathrm{m}}^{\mathrm{i},\mathrm{itr}+1}=\begin{array}{c}{\mathrm{X}}^{\mathrm{i},\mathrm{itr}+1},\mathrm{f}\left({\mathrm{x}}^{\mathrm{i},\mathrm{itr}+1}\right)\text{is better than}\mathrm{f}\left({\mathrm{m}}^{\mathrm{i},\mathrm{itr}}\right)\\ {\mathrm{x}}^{\mathrm{i},\mathrm{itr}+1}\text{otherwise}\end{array}$$E13

The algorithms of PSO and CSA are shown below, respectively:

PSO algorithm: Find optimum values for WNN-PI parameters “

Step 1:    Initialization PSO

       No. of birds n = 20, No. birds_steps = 10

         Dimension of the problem (dim)

          WNN-PI = 14

       PSO parameters c1 = 1.4 & c2 = 1.6

         Inertia w = 0.85

         >  fitness = 0*ones(n,bird_setp);

Step 2:    Initialize the parameter

         R1 & R2 = rand(n,dim);

         >  current_fitness = 0*ones(n,1);

Step 3:    Initializing swarm and velocities and position

       current_position = abs(10*(rand(n,dim)-.6));

         >  velocity = .25*randn(n,dim);

       local_best_position = current_position;

Step 4:    Evaluate initial population

        >   for i = 1:n

            PI = current_position(j,:);

               Set all (a’s, b’s, w’s) and (kp & ki)

               a_1–4 = PI(_1–4); and same for b’s, w’s and (kp & ki)

Step 5:    Initialize sim options (Simulink)

          >     Simout = sim(‘WNN_PI.slx’);

               Compute the error

           e = max(V_Actual)-300;

              m = abs(e);

               error = sum(m);

                Fitness Function is F = error

           current_fitness(j) = F;

           end

Step 6:    Velocity Update

         >  velocity = w *velocity + c1*(R1.*(local_best_position-

            current_position)) + c2*(R2.*(globl_best_position-

            current_position));

Step 7:   Swarm Update

Step 8:   Evaluate anew swarm: Back to Step.5:

Step 9:   Choose the optimum values and submit them to the Simulink model.

         >  a_1–4 = globl_best_position(n,_1–4);

CSA algorithm: Find optimum values for WNN-PI parameters”

Step 1:    “Optimization Initialization and system definition

          No. of Variables pd = 14 and No. of Population size N = 30

             Awareness Probability AP = 0.6 and Flight Step FI = 2

Step 2:    Initialize Function

       [x l u] = init(N,pd); % Function for initialization, l = 0; u = 300; %

       Lower and upper bounds

Step 3:    Generate Random Position

           num = ceil(N*rand(1,N));

Step 4:  >  Compute the error

       function ft = fitness(FI,N,pd) % Function for fitness evaluation

         simout = sim(‘WNN_PI.slx’);

         m = abs(e); error = sum(m); F = error; ft(i) = F;

Step 5:  >  for i = 1:N

       tmax = 300; % Maximum number of iterations (itermax)

       ft = fitness; % fitness evaluation

Step 6:    Evaluate Memory Initialization

    >  fit_mem = ft; % Fitness of memory positions

    >   if rand > AP

    >  xnew(i,:) = x(i,:) + FI*rand*(mem(num(i),:)-x(i,:));

        % Generation of a new position for crow I (state 2)

       else

       > for j = 1:pd

   >   xnew(i,j) = l-(l-u)*rand; % Generation of a new position for crow i (state 2)

       end

       end

       end

Step 7:    Evaluate Fitness Function and Update new error

   >    F = xnew;

         ft = fitness;

Step 8:    Update new Position and the Memory

   >     for i = 1:N % Update position and memory

          if xnew(i,:) > =l & xnew(i,:) < =u

          ft = fitness; % Function for fitness evaluation

          mem = x; % Memory initialization

          fit_mem = ft; % Fitness of memory positions

          x(i,:) = xnew(i,:); % Update position

           if ft(i) < fit_mem(i)

           mem(i,:) = xnew(i,:); % Update memory

          fit_mem(i) = ft(i);

           end

          end

         end

      ffit(t) = min(fit_mem); % Best value until iteration t

        min(fit_mem);

       end

Step 9:   Evaluate fitness function: Back to Step.5:

      >  ngbest = find(fit_mem== min(fit_mem));

Step 10:   Select the optimum values and send them to Simulink.

      >  g_best = mem(ngbest(1),:);

      >  a_1–4 = g_best_position(_1–4);”

6. Simulation results and discussions

This section summarizes the simulation results for AI EMS, which was evaluated over two drive cycles (UDDS and FTP) with two alternative control approaches. The first is FC control by FLC, while the second is FC control by ANNs. WNN-PI controls the BAT, which is adjusted via PSO and CSA. The results reveal that FC controlling by ANNs is superior to FLC due to ANNs are a predictable system, resulting in more efficiency than FLC, particularly when utilized in the FTP drive cycle. Moreover, the ANNs offered optimal flow of power among power sources and FCHEV load. Furthermore, WNN-PI has superior tuning by PSO than CSA since CSA has sluggish speed convergence and readily falls into the local optimum [23, 24, 25]. Figure 14 illustrates the DC-bus voltage under UDDS driving cycle load profile.

Figure 14.

DC-bus voltage during USSD cycle.

It is observed that the fluctuation of the DC-Bus voltage by PSO is less than the voltage fluctuation of CSA due to that PSO tuned the parameters of the WNN-PI more precisely than CSA. By modifying the duty cycle for the buck and boost converters of the BAT and for the boost mode of the FC converter, the voltage of the DC-bus was preserved at an acceptable level. Also, observed that all of the converters’ duty cycle values by PSO are more stable, where all are below 90% which is an acceptable percent as illustrated in Figure 15 the duty cycle of the BAT in boost mode, and as per in Figure 16 the buck mode, while the FC converter as shown in Figure 17.

Figure 15.

Duty cycle BAT converter in boost mode during UDDS cycle.

Figure 16.

Duty cycle of BAT converter in buck mode during UDDS cycle.

Figure 17.

Duty cycle of FC during UDDS cycle.

The duty cycles for the converters were successfully given by PSO and CSA. However, the findings indicate that PSO is more effective than CSA in terms of the DC-Bus voltage and duty cycles of the converters for the FC and the BAT during the UDDS even during the FTP driving cycle. Also, the ANN is more effective than FLC. Therefore, the remainder results of the proposed model have been carried out based on PSO and ANN. Whereby, the power of the FC, BAT, and UC during UDDS cycle are illustrated in Figures 18–20, respectively. Additionally, the power of the FC, BAT, and UC during FTP cycle are illustrated in Figure 21–23, respectively.

Figure 18.

BAT power during UDDS cycle.

Figure 19.

FC power during UDDS cycle.

Figure 20.

UC power during UDDS cycle.

Figure 21.

BAT power during FTP cycle.

Figure 22.

FC power during FTP cycle.

Figure 23.

UC power during FTP cycle.

The proposed control approach properly satisfies the AI EMS of FCHEV needs under the load power profile of UDDS and FTP driving cycles as well as the charging/recharging requirements of the BAT and UC, as illustrated in Figure 24. The FC power is also steady-state during the demand of load power and does not react to rapid fluctuations of the power load; whereas the BAT delivers a medium-frequency power to the power load, then it supports the FC to cover the power for the remaining load needed; and the UC power delivers high-frequency power to the power load to overcome abrupt load fluctuations. The power resources (FC, BAT, and UC) accomplished the optimal power flow for the FCHEV, which in turn makes the BAT and UC function safely and extend their lifespans as well as decreasing H2 (Hydrogen) use.

Figure 24.

BAT and UC state of charge during UDDS cycle.

The plug-in FC hybrid electric vehicle analysis and evaluation in this study were carried out under two vehicle driving cycles using the Advanced Vehicle Simulator (ADVISOR) and the MATLAB/Simulink 2022b (64bit) environment based on the proposed AI EMS structure and optimization algorithm. Figure 25 illustrates the proposed Simulink model for a compact car with a maximum load of 11 kW.

Figure 25.

The proposed Simulink model.

7. Conclusion

An energy management system that manages a hybrid plug-in FCEV is presented by this study as a combined intelligent controller system. The objective of this study was to develop the next generation of an EMS system employing a combination of fuel cells, ultracapacitors, and batteries in order to reduce the fuel consumption of the FCHEV powertrain and improve the system’s efficiency. In the design system, the created EMS scheme took into account the characteristics of FC, UC, and BAT. Also, the proposed EMS considers the dynamic responsiveness of the power sources. FLC and ANN have been used with WNN-PI that is tuned via two optimization algorithms, which are PSO and CSA. The most significant findings of this work are as follows:

• The National Renewable Energy Laboratory’s ADVISOR software was used to construct acceleration and grading test processes for analyzing hybrid vehicle dynamic properties and determining their electrical load power profile. The software was used to extract two varieties of load power profiles (UDDS and FTP), which were used to simulate FCHEV and evaluate the effectiveness of the proposed intelligent EMS.

• The FC, BAT, and UC are modeled using datasheets from their respective manufacturers. As a result, the characteristics of the power sources have been developed as close to their real-world characteristics.

• The power converters were modeled using the average modeling approach. Average models use less time to simulate a system than the switching model.

• The proposed model guarantees that the BAT provides specified energy with a safe run for the BAT, extends its lifespan, as well as not reposed to high load power variations. Moreover, the FC output power responds only to the steady-state power load, which in turn ensures that the FC works efficiently. Whereas UC has the first response to unexpected load fluctuations.

• WNN-PI has superior tuning by PSO than CSA. Whereby, the findings indicate that PSO is more effective than CSA in terms of the DC-Bus voltage and duty cycles of the converters for the FC and the BAT