The smart control is class of control techniques which uses artificial intelligence computing approches like experts system,fuzzy logic,metaheuistic methods and machine learning for power electronic systems [87]-[88]. The application of smart control algorithms in three life cycle of power electronic systems including design,control and maintenance is as shown in fig. Smart control algorithms are used in terms of expert system,fuzzy logic,metaheuristic methods and machine learning with clsssical control methods like PID and SMC [91].

Expert system is earliest method and as of now it is used in the applications for 0.9% only.Due to its certain limitations, advanced algorithms like fuzzy and machine learning are used since they have superior capabilities.

Once the optimization of application is formulated, the optimal solution can be obtained with non deterministic method, i.e., metaheuristic method as shown with the direction in the Fig. 8(a) .Due to enormous advantages ,most of the optimization tasks are solved with metaheuristic methods like genetic algorithm and partial swarm optimization algorithm in power electronics. The usage statistics of metaheuristic methods in optimization for power electronic systems is as shown in Fig. 8(b).

Machine learning is designed with datasets of the system to be controlled. To use machine learning in power electronics,it is classified as supervised learning,unsupervised learning and reinforcement learning. The usage statistics of machine learning methods in power electronics system is as shown in Fig. 8(c). Supervised learning is used in power electronics where system models are difficult to formulate [89]. It has dataset having input-and output pairs. Supervised learning methods are classified as connection based method like Neural network (NN) method, probalistic graphical method and memory-based method. When it comes to unsupervised learning ,it has no output data for learning system during the learning process.

Data clustering and data compression are achieved with unsupervised learning in power electronics [90]. Unlike supervised learning and unsupervised learning, reinforecement learning does not require a training dataset, instead it obtains the experience from the interations between systems.It is prefered in the application where system is with less knowledge or difficult to formulate [92]-[94].

In real time applications with bidirectional converters decide to choose the right control technique. Thus this section projects the strategies connected with isolated and non-isolated bidirectional DC-DC converter topologies. Non-isolated topologies are preferred over isolated topologies for medium power applications since they are less costly and less complex as there is no transformer used. But in case of high power applications isolated topologies are preferred since isolation is required between high power source and low power load. In addition, just because of usage of transformers as they operate at high frequency in isolated converters, it offer advantages of ZVS and ZCS,electrical isolation and high reliability.

The task of controlling the power conveters like DC-DC converters in order to get high efficiency and better dynamic response is the major issue. To address such control issues in the view of getting high efficiency and better dynamic response ,there are couple of control algorithms like genetic algorithms(GA),improved GA,partial swarm optimization(PSO), evolutionary programming(EP), hybrid evolutionary strategy, seeker optimization algorithm(SOA),bacterial-foraging optimization(BFO),gravitational search algorithm(GSA),differential evolution(DE) and artificial bee colony algorithm(ABC). Recently, few more advanced algorithms such as whale optimization algorithm,enhanced red wolf optimization, improved social spider optimization(ISSO),antlion optimization algorithm(AOA), JAYA algorithms,PSO extended algorithms like R-PSO,L-PSO,PSO-CFA, improved PSO based on success rate (IPSO-SR), fruit fly optimization algorithm (FFOA),modified fruit fly optimization algorithm(MFFOA) also being used with basic control laws like PID and SMC for the control of power converters [78].

To find the solutions for the issues which exits in the control of BBBC systems ,the following control strategies are proposed. The comparative analysis is carried out and listed in table.3.

## 3.1 PID Control

The general strcture of proportional integral derivative(PID) conroller is as shown in Fig. 9.This control is combined with other control schemes in order to have hybrid control stratagies to obtain optimization of operation of systems w.r.t efficiency.It is used in different control problems which arise in different converter configurations.

To control is the common problem in bidirectional DC-DC converters,there are two main transitions of controllers due to the converters.In case of battery charging using conventional methods where VL and VH are used ,large transients occur during transition from VL to VH control.To avoid such large transitions, PID controller is used with pulse width modulation (PWM) scheme.This controller reduces the size of the capacitor used at DC bus side of the converter and also reduces the transition time[58].

In case of inverters connected with bidirectional converters,active and reactive currents to be controlled in sequence to control the powers independently and then the inverters can control active and reactive powers at AC sides.Later, PWM with reference values can control the inverter[57].Anoher issue with bidirectional power flow with bidirectioal converters is the delay during mode transition.This can be resolved by using axillary switch with main switch which usally have fixed turn-on time[59]. Eventhough control problems are implemented using digital signal processing(DSP), the discrete current sampling creates some issues like large power loss. Then again conventional control method is unable to control switching time of auxillary switch as the cause of resonant current sensing problem therefore to obtaining higher outcome with PID controller is used by availing discrete voltage sampling. In bidirectional converters, dead time of the switch. This dead time is taken into consideration for the effects of non-linear dependancy of the current on duty cycle [60]. PID controller may not be the choice for this condition to regulate the current in an entire range. On the other hand control scheme is prepared based on either continuous current or discontinuous current. The current is discontinuous with positive and negative peaks along with DC value in between those two peaks of signals provided dead-time occurs in inductor current. There should be fast and stable transition between continuous conduction mode (CCM) and discontinuous conduction mode(DCM) provided if both mode are involved .Thus PID controller regulates the current in DCM while PID requires a preset with control algorithm in CCM.

To have optimization with respect to pwer conversion efficiency and low cost of the systems like multiple converters which will have multiple inputs and multiple outputs(MIMO),Designing proper control scheme is very important step in case of bidirection DC-DC converters [61]. The capacitor at DC bus side or battery source side will have voltage regulation with logic of volltage controller due to PI control algorithm while current control algorithm with PI controller regulates flow of electronic in magnetic device and duty cycle is set for switches in converters.Switching devices are protected from over current with proper regulation of inductor current and also load is protected with zero abnormal current into the load. In order to analyse performance of bidirectional DC-DC converters which are non-linear systems, closed loop control scheme to be linearized around its equilibrium point eventhough the stability analysis is same for both step-up and step-down mode.Bode plot is the one of the linear methods to analise the stability of transition between bidirectional power flow [62]. Therefore the development of mathematical model is needed. The proper control scheme is designed based on stability condition given by bode plot analysis.

## 3.2 Sliding Mode Control

There are some reasons why sliding mode control as shown in Fig. 10 is coming into picture when there is PID controller. The reasons are; In bidirectional DC-DC converters ,non-linear elements are existing which will make the dynamic equations of the converter non-linear and existing linear method used stabilize the system[63]. However these methods accomplish the linearization and characterize the right devices. Also these models neglects dynamic changes. Unlike these methods,sliding mode control which is non-linear control scheme capable to offer exact control action by considering presence of perturbation and disturbance.

Furthermore in a control loop of bidirectionl converter, if large signal enters the control loop,there is chance that external perturbances also can come along with it then the earlier method i.e state space-averaging model not be able to detect the behaviour of regulator.sliding mode control scheme can overcome this condition since it has property of dynamic change. But still this control method is more complicated as it requires the right message.

In bidirectional Cuk converter,three specific switching states are studied with three different sliding surfaces. The analysis show that the system is insensitive to the variation of output voltage under steady state.This happens when magnetic coupling between inductor is negative and incoherence surface is a linear set of power parameters[64].

When BBBC is connected to non-linear load since in this converters states of opeartion are indeterministic. Therfore conventional control method can not solve the control problem since conventional control method can design a linear controller by linearization of system at a region of interest[65]. In this case sliding mode controller is designed in terms of high pass filter which ensures roboustness to parameter variations and hence reduce the transient response and regulates the DC bus voltage under non-linear load variations. However this approach also finds difficulty in controlling the converters that have poles and zeros in right half planes and prediction of stability in large signal behaviour since this is designed with modeling of converter.Therefore to address such issues there are two configurations are proposed [66].

Based on this control method,controller is designed with numerical study for BBBC used in applications like power backup systems. In this case some assumption are made that dynamics of inductor current and capacitor voltages are faster than the supercapacitor dynamics. At the end this control technique shows that it is highly insensitive to strctural perturbations.In some applictions like micro-grid systems where BBBC offers non-linear property with state-space modeling of the load and converter and also in this its with time. Hence in this case its find good for stable voltage[68]. Sometimes two or more control techniques required to integrate to solve the problems which can not be possible by single control technique. For instance,two PI controllers are used in cascade control method.One PI controller stabilize all controls but then due to severe variations of load and line,PID alone can not accomplish required performance.

Thus PID is controller to have better performance[69]. In another instance,sliding mode is integrated with fuzzy logic controller to get rid of chattering process in the usage of sliding mode control alone.In an application like regenerating the energy of an ultra capacitor[70],fuzzy logic and sliding mode controller are combined since the combined controller accomplishes strong robust in changes and minimize the variation required to the expected point.

## 3.3 Partial swarm Optimization(PSO) with Sliding Mode Control (SMC) [77].

The PSO can be applied for bidirectional DC-DC converter along with well established control laws like SMC as shown in Fig. 11. Here we have discussed w.r.t to electric vehicle to charge the battery with tight regulation and efficiency. The SMC control parameters are selected using PSO. The PSO is integrated with SMC to achiev optimization in the control actions. The PSO minimizes objection function which is the sum of errors by inductor current error,Dc bus error and battery voltage error. The PSO evaluate the objection function at each iteration and the best value of particle is saved and later compared to get the best value known as global values out of the group of best values for the SMC.

## 3.4 Dynamic Evolution Control(DEC)

The dynamic evolution controller is as shown in Fig. 12. This controller is suitable for nonlinear systems. It works with the principle of following evolution path irrespective of disturbance existence and hence minimizes the dynamic state error. It’s with respect to time. The advantage of this controller is that avails better performance of the system even though it does not require exact values of the parameters belong to the system.

For instance, in electric vehicle the frequent acceleration and deceleration take place and hence fast dynamic response is needed. When fuel cell based electric vehicle is considered, such fast response may not be offered. This issue can be resolved by connecting storing element with fuel cell [71]-[72] along with bidirectional power converter with above said control technique. The outcome of this controller used with such applications show that controller can meet dynamic loads and get battery charges when battery with fuel cell voltage is bigger than the load demands or during regenerative braking.

## 3.5 Model Predictive Controller(MPC)

The model predictive controller is as shown in Fig. 13. Its extended version of predictive control method. It takes functions to make the system variables follow the reference values. And hence it offers fast system with easy rig-up of system due to microprocessor technology [73]. This controller at first need to be designed by applying mathematical models for the future predicted value and previous values are taken into model of algorithm. The predicted values are forwarded to optimizer which will solve the optimization problem based on predefined cost function and predicted values at each time step. This process will obtain the optimal control actions for converters.

In the system, working of the system is divides as: idle, charge and discharge modes .This division is done based on available and desired values of voltage. Then control algorithm decides everything for the system. Since this algorithm is to any models like variable systems. This algorithm is tested the BBBC for any applications. The extended versions of these models are linear MPC, multi MPC and dynamic matrix control [74]. The converter model supposed to be linear model inside the control algorithm. This would be the limitation associated with Linear MPC algorithm. To overcome this limitation, Multi MPC is coming into picture and this will use multi models system to linearize the nonlinear models locally at different operating points. It will take nonlinearity effect at each stage of sampling. Further it will take the difference between linear and nonlinear models into account to get minimized. Thus this method will solve the constrains associated with multiple MPC algorithm.

## 3.6 Fuzzy Logic Control (FLC) [75].

The basic fuzzy logic control for power converters like DC-DC converters is as shown in Fig. 14.Since the DC-DC converters are non-linear system which can be controlled accurately without mathematical model. This is possible with fuzzy logic where human brain analysis is applied to determine system characteristics. Such analysis can be modelled to construct control logic so called rule base with uncertain inputs. For power converters like bidirectional buck –boost system, FLC receives two inputs like error signal e(k) and change in error signal de(k).Fuzzy rule is set for these two inputs based on dynamic behaviour of error signal. Different algorithms can be applied for fuzzification and de-fuzzification process. For instance, the fuzzy rule is Antecedent: IF X is Medium AND Y is Zero, Consequent: Then Z is Positive. For both the antecedent and consequent, the degree of fulfilment is determined by the membership functions. The type of fuzzy interference technic is classified as Mamdani-type [81]-[84] and Takagi-sugeno-Kang-type [80]. The output of de-fuzzification process represents control signal to generate switching signal for the switching device. This can replaced traditional PID control logic where complex mathematical modelling is carried out. Fuzzy base rule is set Hybrid model of fuzzy control is fuzzy PID control which can offer better transient response over load changes.

## 3.7 Artificial Neural Network (ANN) Control [76].

The computation burden with FLC for generating precise control signal can be reduced by using ANN. Also the better system performance can be achieved with ANN control. In case of DC-DC converter control, ANN plays very important role as it works with prediction control which is better than fuzzy control since intelligent control techniques involving ANN s are found to be simpler for implementation. ANN based PID control gives better system response than Fuzzy based PID control. Nevertheless, the expert experience can be copied from fuzzy logic and incorporated with conventional neural networks (NN) techniques like feed-forward neural network (FFNN) and radial basis function network (RBFN) form various control algorithms like fuzzy neural network (FNN) and adoptive neuro-fuzzy interference system (ANFIS) [85]-[86]. The architecture of whole system of DC-DC control with ANN is depicted as shown in Fig. 15. The traditional PID is not enough to better control of bidirectional buck-boost converters (BBBC) since BBBC load parameters changes with time. Such change cannot be controlled precisely by the PID control as its linear control model therefore for better control, model predictive control algorithm is used as expert and it provides the data to train the ANN which can be tuned finely to better control of BBBC so that to get highest efficiency and performance.

## 3.8 Digital Control [76].

In this method of control, either voltage/current signals are taken by sensors from either source side(feed-forward control) or load side (feedback control).Such a sensor signals are converted into digital signals by A/D converter as sown in Fig. 16 .

Later it will be compared with desired output value. Based on the value of error signal(either + ve value,-ve value or zero value ),the control algorithms like PID,PIDN,PSO-PID,Fuzzy, Fuzzy-PID and adoptive-network-based fuzzy interface system(ANFIS) used to reduce the error signal equal to zero by adjusting the control parameters so that to obtain stable output signal. Finally the control signal will be fed to the digital pulse-width modulation (DPWM) unit which will generate control signal for the power converter. In high power converters, DSP/FPGA control boards are being used to implement such control algorithms since those control boards are having highly computational performance at low cost because of its high processing cores like cortex cores in DSPs and Spartan cores in Field programmable gate array(FPGA) boards. .also these board are having high immunity to electromagnetic interference. For the change of powerflow,the intelligent control algorithms like dead-band, switch and soft-start control are proposed in [45] to change the power flow directions in the converter at smooth –space in order to protect the converter from rush current at the rise time or fall time duration.

Table 3

Control algorithms in bidirectional converters

Control algorithms | Control issues | Advantages | Disadvantages | Recommended Applications |

**PID control** | -Power flow control issues -Analysis of mode of operation -Reducing the time delay and dead time during switching - Safeguarding the devices from over current. | -Minimum cost -Good dynamic response -High reliability | -Minimum efficiency -Lack of robustness in presence of disturbance and uncertainty -Poor in avoiding large transients between directions. | Smart grid systems[57] Electric vehicles[59], Fuel cells & satellite applications [60]-[70]. |

**Sliding mode control** | -Taking the account of external perturbations in large signal -Large variations with load & line | -Tracking the reference signal -Speed of response in defined time -Powerful against variation of parameters and external disturbances - Capable enough to estimate the system under both small and large signal conditions | Exact information of state variables and parameters are required. | Hybrid electric vehicle[61], DC motor control[63], Dc micro grid [64], Energy storage applications [67]. |

**Dynamic Evaluation control** | Voltage drop is avoided to the maximum extent for the change in load current | -Robust to the variation of load signal -Exact Knowledge of design model is not required -Good performance -Well tracking of reference | Division parameter appears in the estimation of duty cycle of control signal which results in complexity in the implementation of analog control form. | Ultra-capacitor based energy storage, fuel cell system [72]. |

**Model Predictive control** | -Power flow control issues -Establishing load & line regulation | -Tracking the reference signal -Fast dynamic response -Easy to implement, this credit goes to programmable devices like DSP | This model can allow the algorithm to use mathematical model of the type linear with some limitations | Hybrid power trains[72] DC distributed power system[73] Battery applications [74]. |

**Fuzzy control** | -Minimizing control time -Better dynamic response compare to PID control and reduces steady state error | -No mathematical knowledge of the system is required -Fast response -Flexible in setting design rules -East to design, understand and expressing control rules in human language. -It can reduce the impact of imprecise, noisy and distorted input information on control action | Vague approach to design control law Approachable only when the things are simple Needs heavy computation while converting the linguistic control rules into its respective control actions | Automotive systems [75], Consumer electronic goods [56], domestic goods, Environmental control [63]. |

**Artificial neural network control** | -Power flow control issues -Establishing load & line regulation -Minimizing control time - Superior system response | -Expert law like MPC can be removed once data sets are ready for tuning ANN -Less computation burden -Decreases the inaccuracy of the system model even with inaccurate parameters -The accuracy of the trained model is about 97% -Easy to implement on microprocessor based system -High computation rate | -Large complexity of network structure -It needs the training to operate -For large neural network, needs huge processing time. -It has to balance number of neurons used in the system and system performance in DC-DC conversion system | Battery applications [74]. Automotive systems [75], Consumer electronic goods [56], domestic goods, Environmental control [63]. |

**Digital control** | -Precise small signal modeling -With inrush current protection, changing the power flow directions smoothly | -Increase the efficiency with charge and discharge speed -Better EMI immunity -Use to use -Low cost -more flexible than analog design -More reliable | -Cause delay with ADC processing -Control loop gain crosses more than one | Hybrid electric vehicle[61], DC motor control[63], Dc micro grid [64], Energy storage applications [67]. Power management[72] |