Tao LEI, Zhou YANG , Zicun LIN , Xiaobin ZHANG

a School of Automation, Northwestern Polytechnic University, Xi'an 710129, China

b Key Laboratory of Aircraft Electric Propulsion Technology, Ministry of Industry and Information Technology of China,Xi'an 710129, China

KEYWORDS Energy management strategy(EMS);Hybrid power system (HPS);New energy sources;Optimization algorithms;UAV

Abstract New energy sources such as solar energy and hydrogen energy have been applied to the Unmanned Aerial Vehicle (UAV), which could be formed as the hybrid power sources due to the requirement of miniaturization, lightweight, and environmental protection issue for UAV. Hybrid electrical propulsion technology has been used in UAV and it further enforces this trend for the evolution to the Hybrid-Powered System (HPS). In order to realize long endurance f light mission and improve the energy eff iciency of UAV,many researching works are focused on the Energy Management Strategy (EMS) of the HPS with digital simulation, ground demonstration platforms and a few f light tests for the UAV in recent years. energy management strategy, in which off-line or on-line control algorithms acted as the core part,could optimize dynamic electrical power distribution further and directly affect the eff iciency and fuel economy of hybrid-powered system onboard.In order to give the guideline for this emerging technology for UAV,this paper presents a review of the topic highlighting energy optimal management strategies of UAV.The characteristics of typical new energy sources applied in UAV are summarized f irstly,and then the classif ication and analysis of the architecture for hybrid power systems in UAV are presented. In the context of new energy sources and conf iguration of energy system,a comprehensive comparison and analysis for the state of art of EMS are presented,and the various levels of complexity and accuracy of EMS are considered in terms of real time,computational burden and optimization performance based on the optimal control and operational modes of UAV. Finally, the tendency and challenges of energy management strategy applied to the UAV have been forecasted.

1. Introduction

Compared with the unmanned aerial vehicle powered by an Internal Combustion Engine (ICE) which uses fossil fuel, the UAV driven by an electrical motor, which uses new energy sources, takes many advantages in terms of emission, eff iciency, stealth and noise.1For small rotor UAVs, only a suitable battery pack can meet all power requirements and support a f light mission of tens of minutes,and a series of civilian small UAVs have been successfully launched by some famous manufactures such as DJI-Innovations (China), 3D Robotics (USA) and Parrot (French). However, considering longer endurance and higher f light speed, solar cells and fuel cells, which have higher energy density, are used as main propulsion energy in UAV.

Solar cells, in which power's performance is signif icantly inf luenced by temperature and sunlight, are usually used in high-altitude UAV with Maximum Power Point Tracking(MPPT)control.2,3The UAV powered by a solar cell has light weight but more larger size and aspect ratio with the limitation of wind loading and aeroelasticity.On the other hand,in lower altitude and lower speed situation, fuel cells have more applications. This paper is more concerned with the hybrid power system based on fuel cells. The f irst documented fuel cell UAV f light was the Hornet UAV from Aero Vironment in 2003, with a wingspan of only 38 cm and a prof ile of 0.25 h.4In 2006, Georgia Tech University built a UAV powered by a 500 W fuel cell.5It was one of the f irst projects undertaken by a university to study the benef its and suitability of fuel cells for UAV. In 2013, the Off ice of Naval Research (ONR)achieved a world record endurance with its fuel cell powered UAV--Ion Tiger, with f light duration of 48 h.6In 2017, an innovative tilting-rotor UAV based on fuel cells from the Korea Aerospace Research Institute (KARI) was exhibited in TM Forum Smart City Submit, which can achieve hover monitoring, vertical taking-off and landing.7

However, one limitation of fuel cells is their slow dynamic response that leads to voltage f luctuations even system instability when an abrupt power change is encountered. This is due to the complex dynamics associated with mass and heat balances of the stack.8Moreover, fuel cells cannot absorb regenerative and surplus energy, and have diff iculty in cold start. Conclusively, neither solar cells nor fuel cells could supply total power demanding of UAV alone with satisfactory performance. As compensation, the new energy power system combined with energy storage systems such as batteries and Ultra-Capacitors (UC), which have high power density, is indispensable for improvement of dynamics response and stability of power system.

The strategy of energy management for HPS is considered as a broad optimization concept. Macroscopically speaking,EMS is implemented as a power regulator to control the power distribution of power sources so that each source will operate in a more energy-eff icient and reliable way,9the strategies specif ically used for power control are collectively called Power Management Strategy (PMS); and microscopically speaking,EMS aims to optimize the transient characteristics of HPS including the overshoot and voltage ripple when a power inf luence occurred. Energy management strategy, as the core part of energy management controller, determines energy sharing under some optimization criterions. Moreover, EMS has a direct bearing on the reliability, stability, power quality and fuel economy of the HPS in UAV. At present, the research on HPS is mainly concentrated in the f ields of Hybrid Electric Vehicle (HEV) and Micro Grid (MG), and a large amount of optimization algorithms, which are employed online or off line in the energy management system, have been verif ied or demonstrated to different extents. Compared with HEV and MG, the demanding of energy management system in aircraft is stricter because of the rigorous requirement of reliability and safety. In the same power size, the higher power density and faster response based on f light velocity are main challenges in designing EMS of UAV.

A comprehensive analysis based on the state of art and highlights about fuel cell hybrid-powered UAV is provided in this paper.The organization is as following:Section 2 makes a comparison of characteristics of main energy sources used in UAV. Section 3 discusses the architecture of electrical propulsion system,and the f low chart of designing the EMS of UAV is given. The different energy management strategies are presented in Section 4.Section 5 introduces trends and challenges related to strategy of energy management of UAV in the future. In the end, conclusions and perspectives are discussed in Section 6.

2. Characteristic of different power sources

The power system of UAV is expected to have both high energy density and power density, namely plenty energy storage capacity and fast power response. Unfortunately, no any single new energy source can meet these two abilities without any compensation in current technical condition. A comparison of power and energy density of different power sources is shown in Fig. 1,10-12For achieving similar overall performance to ICE, a hybridization of fuel cell, battery and UC is indispensable.

Proton Exchange Membrane Fuel Cell (PEMFC) is the most popular fuel cell in experimental UAV platform with high eff iciency, long life and low operating temperature relatively. In contrast, other types of fuel cells such as Phosphoric Acid Fuel Cell (PAFC) with a high operating temperature,Alkaline Fuel Cell (AFC) with a rigorous concentration for oxygen, and Molten Carbonate Fuel Cell (MCFC), Solid Oxide Fuel Cell (SOFC) with high cost are all improper for propulsion system of UAV. A comparison of characteristics of different types of fuel cells is shown in Table 1. Taking a PEMFC (FCS-C1000) as an example and the corresponding U-I curve (also called the polarization curve) is shown in Fig. 2. It can be concluded that PEMFC possesses approximate linearity in ohmic polarization region,and obvious attenuation in electrochemical region and concentration difference polarization region. Hence, the maximum and minimum output power of PEMFC must be limited.

Fig. 1 A comparison of power and energy density of fuel cell,batteries and UC.

Table1 Characteristic of different types of fuel cells12.

Fig. 2 FCS-C1000 PEMFC (made in Horizon).

Accumulators such as Lithium-ion battery, Ni battery and Lead-acid battery cannot only store surplus power but also act as the emergency power source with enough capacity, and Lithium-ion battery is a better choice for UAV due to its high eff iciency and light weight. Ultra-capacitor is another typical bidirectional power source, in which power density and cycle life are much greater than batteries,and it is especially applicable to compensate peak power demanding from the load. The discharge characteristic (The voltage with respect to time and discharged capacity) of Lithium-ion battery and UC is shown in Fig. 3. As a practical matter, these two kinds of energy sources need to be formed as a pack in serial or parallel ways due to the performance constraints of single unit and cooling demand.

3. Electrical propulsion system of UAV

3.1. Analysis of architecture for hybrid power system

A reliable architecture of HPS in UAV is the guarantee and precondition for the optimal operation of energy system. Several hybridization topologies have been studied especially in HEV and MG.13-16Unfortunately, very little researching works are focused on this topic in UAV so far. On the other hand, the architecture applied to UAV can be evolved from the experience of designing in HEV and MG. The design of architecture for HPS is based on some signif icant factors such as reliable operation,electrical isolation and reasonable energy distribution, and the design of the aircraft must pay much attention to the security, weight and layout of the HPS.

In a macro view of conf igurations for More-Electric Aircraft(MEA),it can be divided into centralized and distributed layout.In general,an integrated Multi-port Inputs and Multiport Outputs (MIMO) architecture will be adopted in centralized ones. This architecture is suitable for a complex hybridpowered system with different energy sources and high-level bus voltage due to its relatively lighter weight and smaller overall size.17Some verifying tests have been applied for the HPS of middle-sized or large-sized MEA.Buticchi et al.18proposed a quadruple active bridge converter to interface a fuel cell, a battery, and a super-capacitor bank to the dc bus of the electrical power distribution system, as shown in Fig. 4.Karanayil et al.19also presents an isolated multi-port power converter including a high frequency transformer, which can enhance the fault tolerance of HPS.

Nonetheless,it should not be denied that a system failure is likely to occur in the integrated conf iguration when a certain port failed due to its strong coupling and correlation among each energy sources.20In order to achieve higher reliability and stability, the distributed architecture has been popularly used in HPS of UAV. The following introduction also will be organized around it.

Considering the above-mentioned factors,the load requirements of UAV and combination of different sources, one of the most common architecture for the PEMFC hybrid systems including a single DC/DC converter with the Auxiliary Power Unit(APU)is shown in Fig.5.The PEMFC and APU are connected to the DC bus by a unidirectional DC/DC converter and an electrical switch respectively.The converter can control output power of PEMFC and maintain DC bus voltage stable.Common components of the APU are UC, battery and UC/-battery,which are good compensation to high-frequency transient power demanding and peak power demanding. This architecture allows three kinds of working modes by controlling electrical switch. The f irst mode, denoted as Parallel Mode, the switch is on and the load power is shared by all power sources. The second mode, named as Charging Mode,the switch is still on and the battery is charged by PEMFC.Finally, in a third mode, the Alone Mode, only PEMFC supplies all load power demand alone and the switch is off.Energy management system is the core controller of the whole system and it is used to control the power output quantity of each power source, charging and discharging equably. Furthermore, the controller can also monitor the working state of HPS in real-time and prevent from overshoot, overheating and over discharging.

Fig. 3 Characteristic curve of discharge for the energy storage system.

Fig. 4 Architecture of quadruple active bridge converter.

Fig. 5 Single converter based architecture.

Another common architecture mentioned in the literature is based on multiple converters, as shown in Fig. 6, in which the auxiliary unit switch is replaced by a bidirectional DC/DC converter. The converter connected with PEMFC is still employed as a regulator of DC bus voltage and output power.The difference is that the bidirectional converter,instead of an electric switch,can make the battery and UC charged and discharged in a rate current, which can extend battery lifetime and improve fuel economy. However, this architecture results in a corresponding increase of the overall weight, size, and design complexity.

In recent years, considering diversity of power sources and further improvement of endurance and capacity, solar cells and fuel cells are all employed as main propulsion energy together in a hybrid power system. The modif ied architecture is shown in Fig.7.Solar cells are connected to the DC bus with another DC/DC converter. During f light mission, the controller will check sunlight intensity in real time f irst.If the sunlight is suff icient,solar cells will employ a role same as PEMFC in architecture of Figs. 5 and 6 and PEMFC here is in nonoperating state.Otherwise,EMS controller will give a warning and make solar cells disconnected, and the operation of the hybrid system will be completely identical to the architecture in Fig. 6, but the whole system will be more complicated and expensive inevitably.

To validate the performance of designed HPS before the f light experiment of UAV,a ground-based validation platform should be built f irstly.21,22A typical structure of PEMFC/UC/-battery hybrid-powered system based on a Hard-In-Loop(HIL)platform is shown in Fig.8.Energy management system for HPS is implemented as a two-stage structure which includes the master controller of HIL platform and local slave controller of each power converter. The HIL platform would send the power reference signals which are calculated through the EMS to the slave controller,and then the power signals are decoupled into the given reference voltage signals and reference current signals in slave controller. This hierarchical twostage management structure is explicit and reliable, and online adjustable power split between different power sources is realized.

3.2. Description of EMS in UAV

A typical f light prof ile for a f ixed-wing UAV is given as shown in Fig. 9,with f ive stages including taking off, climbing, cruising,descending and landing,where hmaxis the maximum height of total f lighting time.23Different from HEV and MG, the power demand of UAV during each stage is relatively stable,sequential and always uninterrupted. Taking a FC/battery hybrid system into account, the corresponding power prof ile is shown in Fig. 10, where Ploadis the load power, Pfcand Pbatare the output power of fuel cell and battery,respectively.During the takeoff stage, the power level is about 1 KW, and the rated power is 600 W during the landing stage.

It can be seen that,during the stage of taking off and climbing, the load power is the largest and the peak power of load exceeds the maximum power of each source. Thus the load power of these stages must be shared by both of them at the same time.The next-level power demand occurs in descending and landing with signif icant f luctuation, and the load power during the stage of cruising is at a relatively low and stable level. In traditional passive control methods, power output of each source depends on their own power characteristics,which results in serious energy loss and low security, but the power distribution will be managed through EMS controller in active methods.The fuel economy,eff iciency and f light time could be optimized further.

The main criterions commonly considered in hybrid power system are as following:

(1) Low energy consumption.

(2) High eff iciency of overall system.

(3) Long lifetime of power sources.

(4) Robustness of HPS.

(5) Power quality of DC bus.

Hydrogen cost is still at a high level even through industrial production,and the hydrogen consumption has become one of the most important optimization criterions. There are two main methods to estimate the hydrogen consumption by measuring output current or power24of fuel cell in project, and corresponding functions are

where LHVH2is the low calorif ic value of hydrogen; ηfcis the fuel cell eff iciency; F is Faraday constant, N is the number of fuel cell in stack, ifcis the fuel cell current.

Fig. 6 Multi-converter based architecture.

Fig. 7 Modif ied multi-converter based architecture.

Fig. 8 Structure of hybrid power system in UAV from the view of supervisor control.

Fig. 9 Typical f light prof ile of UAV.

Moreover, working condition of batteries also needs to be monitored in real time.State of Charge(SOC)is an important parameter to determine the operation mode of the battery.The SOC of battery is hard to estimate accurately due to its nonlinearity and time variation. The common methods used to estimate SOC are full discharge test method, open circuit voltage method, impedance method, Kalman f iltering-based method, neutral network and observer-based method. And now there has been a tendency that combining two or three methods together to compensate for each other. An approach combined neural network and Kalman f iltering-based method is proposed by He et al.25and Sun et al.26Recently, State of Health (SOH) of battery has been put forward, which can describe the aging of the battery with the degree of capacity fading.27On the other hand, SOH can modify the measurement of SOC. Cacciato et al.28established the equivalent circuit model of the Lithium-ion battery f irst, and def ined SOH as the ratio of the actual capacity of the battery to the rated capacity.Simultaneously,SOC and SOH are monitored in real time with PI-based observers as follows

Fig. 10 A power prof ile of UAV.

where i t( ) is the battery current, Ccapis the actual battery capacity, Xais the actual current rate used for the test, Xris the rated one, and Kcis the temperature correction factor.

It's widely accepted that the transient power change with high frequency and high peak will increase the chemical and mechanical stresses inside the FC stack or battery cell, and decrease the operation lifetime of power sources. Therefore,the limitation of output power and instantaneous rate of change for all power units must be considered.29We can conclude that the research on energy management in a hybrid system is an optimization problem with multi-objective and nonlinearity, which can be expressed as

where u means the control variables applied by the EMS,J is a series of objective functions with respect to u, The required load power is represented by Preq(t), Pfc(t), Pbat(t) and Puc(t)correspond to output power of fuel cell, battery and ultracapacitor respectively. The subscripts such as ‘min', ‘max'and‘scope' are constrains of maximum value, minimum value and instantaneous rate of charge for corresponding variables.

Based on above-mentioned descriptions, for designing an optimal energy management strategy, a proper architecture should be determined f irstly according to the power f low and standard requirements of UAV, and the optimization criteria need to be achieved by an objective function of algorithms.The modeling and parameters of algorithm will be modif ied and regulated under repeated tests within the corresponding constrains until obtaining the optimal performance as expected. The detail f low chart for designing EMS is shown in Fig. 11.

4. EMS for hybrid power system

Various energy management strategies for HPS have been reported in the public literature. Erdinc and Uzunoglu13presented a comprehensive review of recent trends in PEMFC hybrid systems including a detail explanation of application areas, architecture design and the energy management methods. Motapon and Dessaint24presented a comparative analysis of different energy management strategies for a fuel cellbased emergency power system to a more-electric aircraft with analysis of SOC, hydrogen consumption, system eff iciency and stresses. Sun et al.30and Martinez et al.31has proposed a review on the EMS and the Power-Split Strategy(PSS) which are suited for the Hybrid Energy Storage System (HESS) of HEV, and a thorough survey of the latest progress in optimization-based algorithms is provided. Summarized from the above-mentioned contents and vast amount of literatures, four classes of EMS of UAV are reviewed in this section i.e., rule-based, intelligent-based,optimization-based and others. Detailed classif ication is shown in Fig. 12.

Fig. 11 Flow chart of designing the EMS.

4.1. Rule-based energy management strategy

Rule-based control is a well-known and simple strategy for the management of HPS,which is described as a set of rules based on some pre-established thresholds over the control variables.The power system will operate in different modes by comparing these actual control variables with corresponding thresholds.32All rules established are deterministic and easy to realize with very low computational burden.These advantages make an on-line and real-time power distribution possible.The RB control framework in which converters are driven by PI controllers is shown in Fig.13.However,one drawback of this strategy is that the threshold setting is on the basis of heuristic and past experience. In other words, the optimization effect completely depends on how well designers know about the characters of different power sources and power demand33Furthermore, the dynamic adaptability to power is restricted by f inite states.Therefore,it's hard to say the optimization performance under RB strategy is satisfactory in practical use.

Fig. 12 Classif ication of EMS of UAV.

Fig. 13 Control framework based PI.

In spite of these inherent defects, RB strategy is applied in UAV successfully due to its reliability and simplicity.Lee et al34designed an active power management system with six rules for a UAV powered by solar cells/PEMFC/batteries.Then a f light test of 3.6 h was carried out. Jin et al.35proposed a nine-rulebased energy management for PEMFC/batteries hybrid system and cold start is realized by a hysteresis voltage controller.Savvaris et al36validated the operation performance of an FC/battery hybrid powered system under different initial SOC of batteries in the hardware-in-the-loop testing. Our researching group also established a ground demonstration platform of PEMFC/Lithium-ion battery pack with a powerlevel of 1200 W, as shown in Fig. 14. The main parameters of this platform are also summarized in Table 2.The Speedgoat HIL platform is the energy management system, and the DC electronic load emulates the characteristics of onboard load of UAV.Furthermore,a f light prof ile of 15 minutes under RB method was validated successfully. The corresponding rule-based management strategy and power response curve are respectively shown in Figs. 15 and 16. The consumption of hydrogen in storage cylinder is compared based on different initial SOC of lithium battery under the proposed strategy of energy management.Where Ploadis the total power demanded,Pfcis power supplied by the fuel cell,Pbatis power supplied by the battery, Pfcminis minimum output power of the fuel cell,ibatis current to the battery for charging.

4.2. Intelligent-based energy management strategy

The intelligent algorithm, which is constructed by some imitative behaviors of human-being or animals, can achieve automatic control to a specif ic object. Last decades, these algorithms have made a great development and successful application in hybrid power system. Fuzzy Logic(FL)control is one of the major and typical intelligent-based energy management techniques with high robustness and fast response,in which the power distribution of the HPS is accomplished via membership function and a set of IF-THEN rules.37The remarkable difference between RB and FL method is that fuzzy rules are described as ‘probability' rather than accurate equation or precise numerical threshold. The overall system is usually abstracted in a multi-input and signal-output or multi-input and multi-output system. Complexity and nonlinearity of modeling could be ignored in this abstract model.Compared with other intelligent algorithm, the historical data is unnecessary in FL. As a result, FL has become a good way for on-line control with low computational burden. A simple FL scheme and trapezoidal membership function are shown in Fig.17,the SOC and load power are utilized as two inputs,and reference power of PEMFC is to be the single output.The membership function about load power consists of 4 fuzzy linguistic variables:Zero(ZE),Positive Small(PS),Positive Medium (PM), Positove Big (PB); the function about on SOC includes 3 variables:Low(L),Middle (M),High(H).In order to improve the stability of output power of fuel cell,7 variables are set in the corresponding membership function:OFF,Hold Off Average (HOA), Average (AVE), Hold Average Middle(HAM), Medium (MED), Hold Maxium Medium (HMM),Maxium(Max).Based on this scheme,Zhang et al.38proposes an on-line fuzzy logic strategy for a fuel cell/battery UAV,then demonstrated by a ground test-bench, the experiment results show that the hydrogen consumption under FL control is lower than typical RB control.Unfortunately,it is the pity that the designing of IF-THEN rules also needs to refer to the knowledge and experience of an expert.For further increasing performance, other advanced algorithms such as Genetic Algorithm (GA),39Particle Swarm Optimization (PSO)40and evolutionary algorithm41are combined with FL. It should be noted that these methods maybe ineff icient and timeconsuming if in improper use.

Fig. 14 A ground demonstration platform of PEMFC/battery.

Table 2 Main parameters of HPS.

Fig. 15 Scheme of a RB strategy.

Another intelligent-based strategy widely used is Neural Network (NN) energy management strategy. NN control can make a nonlinear mapping implementation through mathematical proof. It's a good choice for the optimal solution due to its ability to self-learning, associative memory, generalization and high-speed seeking.42In theory, the optimization performance of Neural network is better than fuzzy logic control. Nevertheless, there exists an incidental problem, named‘Paralysis', which is caused by the complex objective function and results in algorithmfailure.43Representative training samples play an important role in the process of learning and it means another extra burden in storage and calculation.Khayyam et al.44designed a neural network structure with 4-layer back-propagation neural network for FC/battery system, and the scheme of NN is shown in Fig. 18, the power demand and SOC are def ined as the two input layer neurons f irst. Then the historical data is trained with Leven-Verg-Marquardt method in the hidden layer, and f inally the reference output power of FC is obtained in the output layer

4.3. Optimization-based energy management strategy

A cost function is usually set up combined with constraints in advance in optimization-based energy management strategies.All methods can be classif ied as two types,non-casual controlling methods and casual controlling methods.The typical noncasual controlling algorithms include Dynamic Programming(DP), Pontryagin's Minimum Principle (PMP)45,46and Linear Programming (LP).47All of them belong to optimal control theory and they are only implemented off line because of their computational burden. Equivalent Consumption Minimization Strategy (ECMS) and Model Predictive Control (MPC)are two of typical casual algorithms. ECMS can be used both on-line and off-line control and MPC is only suitable for online and real-time situations.

Fig. 16 Power response curves under RB.

Fig. 17 Example of FLC.

Fig. 18 Scheme of neutral network.

4.3.1. Dynamic programming

DP is a famous global optimization algorithm and it is considered as an ideal and recognized one for energy management for HPS. The core of DP is to decompose a multi-stage process into a series of single process interrelated and to solve them in order or reverse order. The optimal solution is achieved by minimizing an unwanted outcome considering present and future cost of control decisions. The cost function J for DP with deterministic implementation can be expressed as

where gNrepresents terminal cost;gKis additive cost in current time k; xk, uk, and wkdenote system states, control decision,and disturbances.

On the other hand,DP is suitable for a given operating condition, and large static and historical information must be known ahead of time. There exists the problem of ‘curse of dimensionality' for a complex system with multi-dimension of states, which is entrenched property of the Bellman's optimalprinciple.48Thus it is only used in off-line state due to its computational burden.The result obtained from DP is usually regarded as a criterion to assess other strategies. A Multi-Dimensional Dynamic Programming(MDDP)applied in FC/-battery/UC system was proposed by Ansarey et al.49In order to solve diff icult questions such as the different operation cycle and computational burden, the algorithm of Stochastic Dynamic Programming (SDP) and the Adaptive Dynamic Programming (ADP) were developed. Much work has been focused on this topic to solve the energy optimal distribution in UAV.

4.3.2.Equivalent Consumption Minimization Strategy(ECMS)

As a casual controlling algorithm, ECM is not restricted by specif ic conditions and aims to achieve minimum consumption of every sampling time. The total energy consumption is divided into hydrogen consumption of FC and equivalent hydrogen consumption of battery or UC. The cost function E can be described as50,51

where E is total energy consumption,ΔT is the sampling time,α is a penalty factor, μ is the SOC balance coeff icient.

The penalty factor α can modify the cost function and keep SOC of the battery within a reasonable range.Motapon et al.52presented an Equivalent Energy Minimum Strategy (EEMS),which aims to minimize total consumption by maximizing the output of the battery under SOC constraints. And the increasing performance of EEMS has been validated by analysis between ECMS and EEMS in terms of fuel economy and overall eff iciency.

4.3.3. Model predict control (MPC)

MPC has been widely used in process control industry because of its algorithm robustness,the key benef it of MPC over other optimal control methods is that its f inite horizon allows for a convenient trade-off between the online computational burden of solving the minimum principle, and the off-line burden of generating the penalty surfacefunction,and it is especially suitable for nonlinear or uncertain dynamic systems. MPC can achieve on-line control with lower computational burden than ECMS, and the hybrid power system of UAV in MPC framework is described in equations of state instead of complex or accurate models.53,54The typical MPC framework is shown in Fig. 19, which consists of three parts: prediction model,receding horizon optimization and on-line correction. The operation of MPC comprises four main steps: (A) predictive state over a f ixed horizon with length N, which depends on the historical data recorded and system model;(B)control policy calculation from t to t+N based on the previous prediction; (C) application of the control policy calculated for the current instant t, discarding the rest control quantities; (D)update with real measurements at t, and return to Step. The algorithm performance relies on model quality,sampling step,and the length of prediction horizon.55The horizon length can be tuned accordingly with control strategy used, computational effort, model accuracy, external conditions and disturbances.56Due to these characteristics, MPC has become a researching hot topic of energy management for HPS. On the other hand,key weakness of the model predictive approach remains its susceptibility to uncertainties in the model of hybrid power system.These disturbances would make the performance of EMS degradation.The MPC controller to predict SOC and mechanical power is designed and a conclusion that MPC can achieve similar optimization to ECMS with lower computation time is validated.57,58While a fairly welldeveloped body of theory has been developed within the framework of robust-MPC, reaching an acceptable balance between computational complexity and conservativeness of the control remains a serious problem. In the more general control literature, adaptive control has evolved as an alternative to a robust-control paradigm.However,the incorporation of adaptive techniques into the MPC framework has remained a relatively open problem. In the future, the adaptive and robust MPC must be applied in the strategy of energy management in UAV.

4.4. Other energy management strategy

In addition to above-mentioned strategies, Frequency Decoupling (FD) is another important energy management strategy including wavelet transform, low-pass f ilter and fast Fourier transform, which can decouple power demanding into different frequencies.There is the fact that frequent transients would increase mechanical and chemical stress inside FC and decrease its service life. For extending the lifetime of FC, FD is commonly combined with other management strategies as a primary optimization of power. Vural et al.59and Erdinc et al.60proposed a wavelet-fuzzy energy management strategy.The load power is f irstly decoupled into three frequency band components including low frequency, middle frequency and high frequency power demanding,and then all input variables are calculated by FLC. The corresponding schematic diagram is shown in Fig. 20. Besides, other control algorithms such as Game Theory (GT),61Adaptive Control (AC),62classic PI control,63,64and sliding mode control,65are also applied in energy management of HPS. These control methods are often used for the local controller of power converter in HPS.

Finally,a comparison of main EMS mentioned in the literature is summarized in Table 3.

Fig. 19 Framework of MPC.

Fig. 20 Schematic diagram of wavelet-fuzzy control.

5. Trends and challenges of energy management of UAV

There have been wealth efforts on EMS of UAV, including rule-based, intelligent-based and optimization-based ones as reviewed in Sections 3-4. As a prosperous area of research,various innovative controlling strategies are expected to emerge for enhancing the performance, maintenance, and functional diversity of UAV. Further research opportunities will def initely gain considerable achievements from the advancement of optimization algorithms, aerodynamics,f light control technology and smart-cyber-physical systems. We brief ly discuss future trends of UAV EMS from the following different perspectives,which could contribute to more eff icient,safer, and greener UAV.

5.1. UAV with more air duct fans

A summary of different architectures of fuel cell-based hybrid electrical propulsion system has been discussed in Section 2,but mainly concerned on the UAV with centralized electric propulsion. All above-mentioned strategies are only considered for energy management of traditional electrical propulsion system including a few propulsion motors, but for UAVs with distributed electric propulsion such as vertical take-off and landing unmanned aerial vehicle (VTOLUAV)which has many air duct fans to obtain the optimal aerodynamic performance, as shown in Fig. 21,66,67the energy coupling relationship between the f light thrust system and the energy management system must be integrated in the energy optimal management strategies further.68So the coupling mechanism of energy f low in the HPS must be analyzed, and the stochastic power demanding from the air duct fan must be considered in the modeling of electric load.

5.2. Bottlenecks of power sources

Hydrogen,especially liquid hydrogen is considered as an ideal fuel for electrical propulsion system and the performance of fuel cell has been closed to ICE.69,70There is thereby an urgent need for hydrogen energy but still a signif icant challenge to higher overall cost, easy aging of stack, technology of hydrogen storage, and lower power density for fuel cells.71Moreover, although a great progress has been made in accumulators technology last decades, and Lithium-ion battery such as 18650, in which power density is up to 200 w/kg, has been already applied in Tesla electric vehicle, but not all accumulators can match power requirement of UAV especially in high speed and heavy weight situations in terms of ratio of power to weight, cost, and capacity.72There is still a long way to enhance the energy density and capacity of power sources.

5.3. From UAV to MEA

It's no doubt that the promotion and popularization of electrical propulsion technology and new energy sources in MEA are an inevitable trend. UAV is a scaled-down MEA with much more simplif ied architecture and much lower power rating.Research on energy management of UAV would be the guidance to further application in MEA. As the f irst phase to thistrend, the whole FC hybrid power system will be applied in manned aircraft as APU with a parallel architecture which remains the ICE as the main propulsion, as shown in Fig.22.73,74This architecture has also been more mature applications in hybrid electrical vehicles. The architecture can improve the eff iciency and fuel economy further but the output power is restricted due to the coupling of different types of power in the gear box.

Table 3 A comparison of main EMS mentioned in the literature.

Fig. 21 Architecture of VTOLUAV with more fans.

In order to lift this restriction, another series hybrid architecture which can supply much higher rate power is proposed,as shown in Fig. 23.75,11,76Decoupling of engine and motor has been achieved in this architecture and all propulsion power is generated by a motor. Sizing these two hybrid power systems, which including fossil fuel, fuel cell, battery and UC,the realization of hybrid powered system optimization based on the multi-disciplinary design including the aerodynamics of UAV,f light control and electrical power system is very complex and diff icult. In addition, the design of a strategy for energy management integrated with the health management and thermal management in UAV is very challenging.

5.4. Optimization algorithms

It should be pointed out that the accuracy and adaption of model building have a strong inf luence on the performance of EMS of UAV,and static-basic models of HPS have a prevalent adoption due to their simplicity and lower computation burden.However,the result error between simulation and real test is inevitable just because of it.To bridge the gap,dynamic models are welcome.77,78On the other hand, as elucidated in Section 4, every optimization algorithm has its own strengths and limitations. A mixture of different algorithms with complementary characteristics is a potential direction for the EMS of UAV. GA, PSO and FD can be combined with fuzzy logic strategy to optimize the membership function.The result of DP can be used as the reference tracking of MPC to improve the prediction accuracy.79,80More such combinations could be anticipated in the near future.

Fig. 22 Parallel hybrid architecture.

In parallel with the previous work, optimization itself represents a vast area of research. Novel optimization algorithms are emerging,some of which are expected to solve EMS problem of UAV with some advantages,such as dynamic programming,81,82Ideal Operating Line (IOL) control,83and hybrid optimal control methods (model-based optimization methods).84,85Machine Learning (data-based optimization methods) is a growing area and provides numerous advanced learning technologies including the Support Vector Machine(SVM), Bayesian inference, and reinforcement learning, and the uncertainty of parameters of HPS and stochastic characteristic of load are considered.86-88These methods can be integrated in the EMSs of UAV to strengthen their autonomy and environmental sensible.

5.5. Multiple-objectives optimization

Most of existing EMSs only concerned on a single optimization objective such as the fuel consumption minimization and caused hidden trouble in other unconsidered objectives.For achieving multi-objective optimization,the objective function can be set as a most important target combined with other subordinate targets via weighting eff icients.89So in some situations, these objective cost maybe contradicted with each other,such as the eff iciency of the power system and the weight of EMS of UAV.We can use optimization methods to get the credible Pareto solution. And this approach can return a suboptimal result but more reasonable and more systemic. Considering the EMS of UAV for a collaborative formation, the dimensionality and complexity of the optimized objectives will be more diff icult with vast computation burden. In fact, it should be pointed out that that very little optimization work has been carried out, but this is a more hopeful prospect.

5.6. Longer time scale and space scale

Fig. 23 Series hybrid architecture.

The mentioned EMSs of UAV were evaluated under a single f light stage or several f light cycles. So the timescale is for only onboard f light and very short.Nonetheless,there is a trend for the long endurance UAV operation. EMS must consider the longer time scale from the minutes to the hours, even to the days-time scale, which manages energy utilization in both f light and on-ground stage.90Traditionally, the UAV EMSs are considered at a single UAV level, and therefore, the space scale is relatively limited. With the development of Smart UAV, UAV-to-UAV communication technology, and the wireless power transfer,there will be inevitable increasing connection between the UAVs and UAVs formation to improve the f light performance and energy eff iciency. How to realize the coordination management of energy of UAVS formation is a challenging topic.91,92The EMS problem of such a formation of UAVS might be extraordinary different from the case of a single UAV, due to the spatial distribution, communication control and so on.This tendency may accelerate the development and application of multi-agent cooperative EMS,distributed MPC-based EMS and many other advanced networked EMSs to the UAV.

6. Conclusions

This paper presents a classif ication and analysis of architectures and energy management strategies for hybrid-powered systems in UAV. Energy optimal distribution of the HPS in UAV is a complex, nonlinear, strong coupling constraint and multi-objective optimization problem.

Compared with hybrid power tramways and vehicles based on the new energy sources, the architecture of HPS in UAV is more complex. So the energy optimal algorithm of UAV is more concerned with the aerodynamics and f light control law. There have been more complex operation modes and working constraints which must be integrated in strategies of energy management for hybrid-powered UAV. In addition,the performance of EMS of UAV such as the overall weight,stability, structural layout, and the ratio of power to weight can be achieved. It's noted that most EMS applied in UAV are still in the stage of theoretical simulation and HIL demonstration on the ground. Only a few simple algorithms have been applied in experimental UAV f light test successfully.

This review on UAV EMS highlights the advantages and disadvantages of virtually all existing approach in the existing literature.It is not concluded that any single algorithm prefers for the EMS of UAV, but it advocates mixing several ones to compensate for each other for each own disadvantages. With the development of artif icial intelligence and big data technology, a new era will be imaged to be advanced by means of machine learning as common framework of EMS in UAV.

Acknowledgements

This study was supported by the Open Innovative Fund of Xi'an Aisheng Technology Group Company of China (No.ASN-IF2015-0202).