Design and Implementation of a Hybrid Energy Supply System for Railway Vehicles

Design and Implementation of a Hybrid Energy Supply System for Railway Vehicles R. Li, A. Pottharst, N. Fröhleke, J. Böcker K. Witting, M. Dellnitz ...
Author: Audrey Manning
1 downloads 0 Views 471KB Size
Design and Implementation of a Hybrid Energy Supply System for Railway Vehicles R. Li, A. Pottharst, N. Fröhleke, J. Böcker

K. Witting, M. Dellnitz

O. Znamenshchykov, R. Feldmann

Institute for Power Electronics and Electrical Drives

Institute for Applied Mathematics

Institute for Efficient Use of Parallel Systems

Faculty of Electrical Engineering, Computer Science and Mathematics (EIM) University of Paderborn, Germany

Abstract- An energy supply system based on a hybrid energy storage unit combined of batteries and ultracapacitors for a railway vehicle is studied. In order to optimize the energy supply system architecture and to manage the energy distribution the power electronic converters connecting ultracapacitors, batteries and the DC-link are investigated together with control strategies comprising a multiobjective continuous and a discrete optimization. The prospective goal of applying latter optimization techniques via a so called Operator Controller Module is a self optimizing energy supply system. Simulated and measured results are presented, revealing that the separation of the dynamic load from batteries yield improvements in lifecycle, availability and long term costs, while the novel control facilitates prospectively large energy savings and obeying set constraints. I INTRODUCTION

Recently the advances of ultracapacitor technologies make these devices effectively to be employed as energy storage element in electric vehicles. The main advantages of ultracapacitors are their high power density and cycling capability, which release the costly battery from peak power, help in smoothing strong and transient power demand of a distribution network [2][6][8]. In most applications, however, ultracapacitor modules work at high voltage level which needs the series connection of a lot of ultracapacitor cells, increasing cost and resulting in large volume and heavy weight, whereby the application ranges of ultracapacitor are limited. In this contribution, a hybrid energy supply system based on battery and ultracapacitor is designed and implemented. The system structure is shown in Fig. 1.

Fig. 1 Hybrid energy supply system based on battery and ultracapacitor

High energy density batteries are usually applied as the primary energy storage component for providing the continuous power. Ultracapacitors are responsible to take over the transient demand of power. Two types of

bidirectional converters controlling the energy flow between ultracapacitors, battery and DC-link are investigated. By increasing the voltage boost ratio of the DC-DC converter, coupling the ultracaps with the DC-link the rated voltage of the ultracapacitor module is reduced [3][4][7]. In order to develop the control strategy we combine intelligent lookahead methods from the field of artificial intelligence [12] with multi-objective optimization methods [9][10]. The multi-objective approach is used to optimize the trade off between two contradicting goals, namely energy efficiency and life-cycle of the system. The lookahead method is used to make decisions based on future environmental and/or system changes. The latter method has often proven to be beneficial in planning and game playing problems. [13]. With the estimated power required from the DC-link the control signals of the DC-DC converters are generated to distribute energy in a way the performance of the energy supply system is promoted. Prospective objective of a self optimizing energy management, which is one subproject for validation of self optimizing schemes (see [11] for detailed information), is the power flow control between the DC-link of the electric grid, the battery and the ultracaps. Self optimization we define, if the following 3 actions are involved iteratively: 1) Analysis of the current situation; the latter includes not only the state of the system but also the observation of inputs and influences on this system. 2) Determination of system objectives: at this step new objectives of the system may be derived out of a given, discrete, closed set of possible targets by means of selection or adaptation. 3) Adaptation of the system behavior; the latter is determined by the aspects structure and parameter (of actuators, sensors, controllers and computing cores). Step 2 discriminates self optimizing systems from adaptive systems. In this contribution the test bed available via project NBP [1], the hybrid energy supply, incl. converters and control structure is briefed in chapter II. The underlying optimization algorithms for adaptation of the system behavior are explained in chapter III, which could be a reaction on altered influences and subsequent objective selection and weighing strategy. Simulation results for


different control strategies are discussed in chapter IV, while first experimental results gained on hardware-in-theloop testbeds are given in chapter V. II CONSTRUCTION OF HYBRID ENERGY SUPPLY SYSTEM

A. Railway Vehicle A railway vehicle shown in Fig. 2(a), developed within project NBP (Neue Bahntechnik Paderborn), is designed to drive with a maximum speed of about 36 km/h using the doubly-fed linear motor. It has a length of 3.2m and a mass of about 1200kg. A test track shown in Fig. 2(b) in scale 1:2.5 with a total length of about 530m was established at the University of Paderborn for various investigations [1].

They are constructed by 280 Ni/Cd battery cells connected in series, which have a total energy of 2.4kWh with an energy density of 21 Wh/kg. The working voltage of this batteries is from 260V to 440V. In order to employ ultracapacitors as peak power buffers, it is necessary to know the usable energy storage (Wh) and the maximum power requirements, because the primary disadvantage of ultracapacitors is their relatively low energy density (Wh/kg and Wh/l) compared to batteries. The energy requirement is calculated based on a maximum power of 6 kW for 15 seconds, and assuming that 75% of the energy stored in the ultracapacitors can be used. A commercial 42V ultracapacitor module was selected and designed to operate in a voltage range of 20 to 42 V. Such a module is able to store 35.6Wh of energy, which satisfies the total energy storage requirement of 33.3Wh for the shuttle. B. Bidirectional DC-DC Converter Due to the low voltage conversion ratio a simple nonisolated buck-boost DC-DC converter is selected to couple batteries and DC-link shown in Fig. 3, serving as one part of the validation base for the controlling concept.

(a) Railway Vehicle

Fig. 3 Non-isolation buck-boost DC-DC converter between batteries and DC-link

(b) Test Track Fig. 2 the tested railway vehicle and test track

The on-board power supply system was designed especially for the demands of the doubly-fed linear motor drive structure. The energy consumers, which are the linear motor secondary side driving module, a hydraulic unit and a 24V auxiliary power supply for the control system, are coupled via the DC-link. Two types of energy storage devices are employed to satisfy the required energy and power density. The main energy required to supply the vehicle is stored in batteries.

Moreover the large voltage conversion ratio between ultracaps and DC-link necessitates an isolated converter type and the bidirectional power flow. A full-bridge bidirectional DC-DC converter shown in Fig. 4 is employed to control the peak power flow of the ultracapacitors. The high boost ratio of this converter enables using an ultracapacitor module of relative low voltage range.





Voltage Range


Energy Density



2.4 kWh



Voltage Range


Maximum Current


Stored Energy Maximum Series Resistance Weight

35.6Wh 10 mΩ

Life Cycle


16 kg

Fig. 4 Bidirectional DC-DC converter between ultracapacitors and DC-link

By means of a simple auxiliary circuit, consisting of Paux and two diodes at the voltage-fed side, and an active clamping circuit, consisting of Saux and Ccl on the currentfed side, ZVZCS is achieved for the main switches, if they are controlled by phase shifted PWM. Moreover, the current commutation between the clamping capacitor and


leakage inductance is accelerated, thereby improving the converter dynamic performance significantly. A 3kW converter prototype used to couple ultracapacitors and DC-link was developed and built to verify the operation principle of the proposed full bridge bidirectional DC-DC converter. In order to reduce the converter volume, the switching frequence is set to 50kHz. The current control loop of boost and buck mode are implemented by using analog current compensator and PWM modulator, while the voltage control loop is realized using a DSP controller. Converter modules will be switched in parallel in order to match the power requirement, if one is not sufficient. C. Control Structure The complete block diagram of an enhanced control scheme is shown in Fig. 5. Step 1 of the self-optimization process (see chapter 1) called analysis of the current situation contains the predictor of the prospective power requirement of the shuttle and power, which could be generated by the linear motor. Input quantities for the predictor are: track and driven profile, motor structure, headwind, convoy type etc. More details about the simulation model of this predictor are to find in [5]. With these estimated parameters the control strategies are generated to optimize the energy distribution, in a way the performance of the energy supply system is improved. Adaption of the system behaviour is performed via the reference quantities, which are calculated by the discrete lookahead algorithms and multiobjective optimization outlined in chapter III. The voltage and current control are used to control the

energy flow and improve the converter dynamic behavior and stability. The DC-link voltage and the voltage of energy storage devices are sampled by the DSP controller as voltage feedback. According to the working mode and the working situation of the load and energy storage devices, the operation modes of DC-DC converters are configurated. One of the bidirectional DC-DC converters is instructed to work in voltage control mode, namely to control the DC-link voltage, the other one should be set in current control mode and control the current of the storage devices. The individual control of the bidirectional converters provides the flexibility of the energy distribution. While batteries supply the continuous power, ultracapacitors are designed to absorb high regenerative braking energy and provide the demand of peak power during the acceleration of the shuttle, thus limiting the otherwise high charging and discharging currents of the batteries. In this way the negative effects of the characteristic limits of battery, such as limited life-cycle, cold intolerance and critical charging rates are reduced. Therefore the energy storage system provides an efficient means to optimize the energy management using a combination of different power sources, moreover provides the possibility to optimize the volume and weight of the energy storage system. III OPTIMIZATION ALGORITHMS

A. Discrete Lookahead-Algorithms All the operations of energy consumption from the components of the energy supply system and their recharging are accompanied with certain energy losses,

Fig. 5 Control structure of energy supply system


which depend both on the current environment of the system as given by the value of PZK and the internal state of the system as given by the loading levels of the battery and the ultracaps. In this section we present a heuristic based on lookahead search to compute a control strategy with the goal to minimize energy losses. As mentioned earlier, exactly one of either the battery or the ultracaps are used to control the DC-link by connecting the DC-DC converter to them. At the same time the other device may be loaded, discharged or simply be idle. This results in exactly the six possible control states (c,l),(c,d),(c,i), (l,c),(d,c),(i,c) of the system, where the first component stands for the battery, the second component for the ultracaps and c,l,d,i stand for “controlling”, “loading”, “discharging” or “idling”, respectively. In a first approach the current for loading and discharging any of the devices is kept constant. Note that being in a certain control state for a certain time interval does not only create some energy loss, but also changes the internal state of the system and, thus, influences future energy losses. Therefore we use a lookahead approach to compute a strategy with small energy losses: Starting from the presence, the lookahead algorithm simulates future changes of the environment and the internal systems state. We model this problem by defining a tree T=(V,E), where the nodes V of T correspond to states of the system at a certain time t in the future. Edges represent possible transitions by switching from one control state into another one. Switching may be done at certain predefined time steps, or, if there are drastical changes in the input power PZK . Associated with each node v in the search tree is a measure M Z (v ) of the energy loss accumulated on the path from the root of the tree to v. The goal is to find a path P = (v0 ,..., vk = v) from the root v 0 of T to a leaf v such


that M Z (v) = min{M Z (u ) | u is leaf of T } . This path corresponds to an optimal switching strategy.

Fig. 6. Discrete system’s behavior model

The system under consideration is able to switch between the six control states every 42 ms. As a benchmark we are interested to find a switching strategy for a trip of 80 s. As a consequence the resulting lookahead tree has depth 1905 and width 6, resulting in a number of 61905 ≈ 2 ⋅101023 leaves. Clearly, a heuristic approach must

be used to reduce the size of the search tree. This is done by the introduction of a lookahead frontier, given by a depth parameter d. The tree is grown only until time d ⋅ 42 ms. This tree Td is grown implicitly by a depth-first-search routine that returns the energy loss of the best path P found in Td . The behavior of the system when d=1 is shown in Fig. 6. The pseudo-code of the algorithm is shown below: function dfs(System S, state s, time t, value L, int d) { m = infinity; if (d = 0) return(L); compute next time τ to switch control states; for all the six system states

vi , 1 ≤ i ≤ 6 {

simulate system S in state s from time τ to t;


= L + energyLoss(S,s, τ -t);

x = dfs((S, v i , τ ,d-1);

If (x < m) { best(d) = v_i; m = x; } } return(m); } /* end of dfs */ function control(System S, int d) { v = initial control state; forever { x = dfs(S,v,0,0,d); /* returns loss, sets best(d) */ switch to control state best(d) at next time to switch; v = best(d); } } /* end of control */

A sequence of search trees resulting from a depth 1 lookahead search is presented in Fig. 6. To reduce the size of the lookahead trees even further we introduce the use of an upper bound U. Paths of the lookahead tree T are then pruned, if their current loss already exceeds U. Further reductions in the size of the lookahead trees were obtained by using constraints for the efficiency of the battery or ultracapacitator. With the introduction of the lookahead approach we created a flexible, scalable control mechanism that allows to easily be extended to include further improvements. The first improvement that comes into one mind is to optimize the current used for charging or discharging, respectively, the battery and the ultracaps. Depending on the current state of our system the optimal current may vary. This is done by computing a current optimal for the next millisecond 42 times in each 42 ms interval between two different switching times. As a result the current used for charging or discharging may change 42 times on each edge of the tree T. A second, even more challenging extension is to use multiobjective optimization techniques to not only minimize the energy losses but in addition to maximize the life-cycle of the system. We will focus on the latter goal in the next section. B. Continuous Multiobjective Optimization The respective adjustment of the charging or discharging current is determined by global multiobjective optimization. In contrast to classical scalar optimization, where the


global minimizer of one single function f : ℜ n → ℜ is to be computed, in multiobjective optimization several objective functions are taken into account at the same time. More precisely, a multiobjective optimization problem is given by min {F ( x) : x ∈ S }, (MOP) where F is defined as the vector of the objective functions f 1 , K , f k : IR n → IR, F : IR n → IR k , F ( x) = ( f 1 ( x), K , f k ( x)) and S denotes the feasible region for x (cp. e.g. [9]). Here the meaning of ‘min’ in MOP is as follows: Definition: Let u, v be two vectors in IR k . Then the vector u is less than v ( u i ≤ p vi ), if

ui ≤ vi ∀i = 1, K , k . Definition: A point x ∈ S is called (globally) Pareto optimal for MOP (or a (global) Pareto point), if there is no x ∈ S with F ( x) ≤ p F ( x ) and f j ( x) < f j ( x ) for at least

one j ∈ {1, K , k} . In case of the energy supply module two partly conflicting objectives are of interest. More precisely we consider the following problem: F : IR → IR 2 , f1 (iUC ) = LE (iUC ), f 2 (iUC ) = η ESM (iUC ), where the battery life time is defined as E T − T0 1 LE = n ⋅ BN ⋅ 7 K ⋅ ln iB I 5 S ⋅ N TB − T0 1+ TBN − T0 and the efficiency of the energy supply module is given by ( P + P )η + PLB + PLUC η ESM = B UC GS PB + PUC PLB = Ri iB2 + u B iO 2 +



arg max LE − arg max η ESM . 1 − q Bmin

If q B = q Bmin (lower bound for q B ), then the computed value for the current is exactly that one where η ESM is optimal. On the other hand, if q B = 1 , then the result is the maximum of LE . In between, the values are linearily interpolated by this formula. 2) Considering the Pareto set in image space, i.e. the Pareto-optimal values of LE versus η ESM , it is obvious that such solutions are not desirable where slight variations of iUC cause only little losses in one objective but great benefits in the other one. Therefore, in the second step the solution is corrected to a better one by comparing the ratio of the two objective functions in image space.

(a) Total efficiency

(b) Battery life-cycle

u B2 RV

2 PLUC = RUC iUC + PAE . To compute the entire Pareto set several numerical algorithms have been designed (for an overview we refer e.g. to [10]). The result is – in this special case – the set of optimal compromises for the values of the current iUC .


for each special application. For the case of the energy supply module, the decision heuristic consists of two steps : 1) First of all the load value of the battery q B is taken into account. If the load value is high, solutions with a high battery life time are preferred. If it is low, then the efficiency is emphasized. The following formula characterizes this effect: opt = arg max η ESM + (q B − q Bmin ) ⋅ iUC

Meaning Nominal energy flow rate


(c) Optimization step 1

(d) Optimization step 2

Fig. 7 Example of Multiobjective optimization

As an example the results of the multiobjective optimization are shown in Fig. 7 for the state (c, l) with q B = 0.6, qUC = 0.6 and a power transfer of 3kW. The feasible region for iUC is defined depending on the zero of

Application days pro year

iB (which directly corresponds to the points in LE and

Cycles Nominal operating temperature of the battery Temperature of the battery


Battery discharge current over five years Balance power for ultracaps cell




Within this set the best adjustment has to be chosen. This step is performed by a decision heuristic which is designed

η ESM where the function is not differentiable). The fourth plot shows the Pareto set in image space and in the next plot the correction step of the decision heuristic is illustrated. The multiobjective optimization returns a point of the pareto set as a good trade-off between energy loss and lifecycle of the system to the discrete optimization routine. The decision of which control state to choose next is then solely based on the energy loss. The tree search as


described in chapter III.A chooses as next control state the one that minimizes energy loss.

ultracapacitors and DC link was built and tested in the test bed.


A simulation model for energy distribution was built in MATLAB/SIMULINK to verify the control schemes . The different control strategies are tested and evaluated under the profile of NBP test track. The simulation result shown in Fig. 8 indicates, that the total energy losses are reduced and the life-cycles of battery are increased evidently compared to an ordinary method, in which the ultracapacitor is used only to cut off power peaks of the battery at 2.3kW. Fig. 9 Hybrid energy supply test bed

the power and the current of battery for start-up test are shown In Fig. 8. The voltage of DC-link is controlled to 650V by DC-DC converter connecting battery tank. The measured efficiencies of the DC-DC converter in boost mode and buck mode experiments are more than 96%. (a) Battery current restriction

(c) State of charge of Ultracaps

(b) Multiobject optimization

(d) State of charge of Battery

Fig.8 Battery power


(black) and current iBat (red)

According to the a. m. discussion, while the battery unit is controlled to keep a const DC-link voltage, the ultracapacitor unit is controlled in current control mode. The current and voltage of the ultracapacitor unit are shown in Fig. 9 for normal operation. The measured efficiency of the bidirectional DC-DC converter in boost mode and buck mode experiments achieve 89% and 93%. (e) Energy losses (ELESM)

(f) Battery life-cycle

Fig. 8 Simulated results of different control strategy

In Fig. 8 PESM is the power of the DC-link supplied from both energy storage devices, i.e. power of battery PB and the power of Ultracaps PUC. The energy losses ELESM are calculated also from both stroage devices. V EXPERIMENTAL RESULTS

In order to verify the principle of operation and the optimization control strategies, a test bed of the energy supply system consisting of the bidirectional DC-DC converter, measuring facilities battery and ultracapacitor shown in Fig.7 was built and tested. The battery is constructed by 280 Ni/Cd battery cells connected in series and have a total energy of 2.4kWh. A commercial ultracapacitor module BMOD0115AV is selected and used at a voltage range of 20 to 42 V. A 3kW experimental converter prototype connecting

Fig. 9 Ultracaps current (Ch1, 20A/div) and voltage (Ch4).

A regneration of braking energy yields positive charging current of the ultracapacitor, while delivery of energy to the dc-link yields negative current.



Hybrid energy supply system using an ultracapacitor and battery as the energy storage elements and bidirectional DC-DC converters controlling the energy distribution of these storage devices are investigated and presented in this contribution. In order to control the energy stored in ultracapacitor and battery, the storage system structure and specifications are studied. Moreover, the control strategy as a result of the discrete and continuous multiobjective optimization procedures is used to decide how much electric energy for the operation of electric vehicle is taken from or charged to the ultracapacitor or battery. The operation of the energy distribution is verified by measurements.



[4] [5] [6]


The authors acknowledge gratefully the financial support granted under project SFB614 (Collaborative Research Center 614 – Self-optimizing Concepts and Structures in Mechanical Engineering) University of Paderborn, and the work was published on its behalf and funded by the Deutsche Forschungsgemeinschaft.


[9] [10] [11] [12] [13]

Web-Page of the “Neue Bahntechnik Paderborn”-Project A. Rufer and P. Barrade, “A supercapacitor-based energy-storage system for elevators with soft commutated interface”, IEEE Transactions on Industry Applications, Vol.38, pp 1151 -1159, Oct. 2002. K. Wang, F.C. Lee and Lai, J, “Operation principles of bidirectional full-bridge DC/DC converter with unified soft-switching scheme and soft-starting capability”, Proc. of IEEE APEC 2000, Vol.1, pp 111-118. R. Li, A. Pottharst, N. Fröhleke, J. Böcker, “Analysis and Design of Improved Isolated Full-Bridge Bi-Directional DC-DC Converter”, Record of IEEE PESC2004, Aachen, Germany R. Li, A. Pottharst, N. Fröhleke, J. Böcker, “Energy Storage Scheme for Rail-guided Shuttle using Ultracapacitor and Battery”, Record of EPE-PEMC 2004 A. Schneuwly, M. Bärtschi, V. Hermann, G. Sartorelli, R. Gallay, R. Koetz, “BOOSTCAP Double-Layer Capacitors for Peak Power Automotive Applications”, momentarily white paper. Ingo Voss, Rik W. De Doncker, Ho-Gi Kim, “Selection of a Bidirectional DC/DC Converter for Future Fuel Cell Hybrid Electric Vehicles”, Electric Vehicle Symposiom (EVS 20), USA, 2003. Erik J. Cegnar, Herb L. Hess, and Brian K. Johnson, “A Purely Ultracapacitor Energy Storage System for Hybrid Electric Vehicles Utilizing a Microcontroller- Based dc-dc Boost Converter”, Proc. of IEEE APEC2004, California, USA. K. Miettinen: Nonlinear Multiobjective Optimization, Kluwer Academic Publishers, 1998. Dellnitz, M.; Schütze, O.; Hestermeier, T.: Covering Pareto Sets by Multilevel Subdivision Techniques, to appear in the Journal of Optimization, Theory and Applications, 2005. Web-Page of the "Selbstoptimierende Systeme des Maschinenbaus"-Project R. E. Korf, “Depth-first iterative-deepening: An optimal admissible tree search”, Artificial Intelligence, Vol. 33, No 1, pp 65-88, 1987. R. Feldmann, “Computer chess: Algorithms and Heuristics for a Deep Look into the future”, SOFSEM 97, Theory and practice of informatics


Suggest Documents