Fuel cell anode line control device

Abstract

The invention relates to a fuel cell (1) comprising a stack (E) of electrochemical cells (T1, T2) equipped with at least one anodic feed collector (C1, C2) and one anodic discharge collector (C3), the fuel cell (1) being intended to supply an electric current to a load and comprising at least one anodic line passing through the stack (E) of electrochemical cells. The cell includes a control device (K) for developing the control of a fuel fluid injection group in order to regulate the pressure of this fuel fluid in the line to conform to a given setpoint pressure. The control device (K) is configured to evaluate a limited number of injection control profiles over a predetermined time horizon. Figure 1

Description

Title of the invention: Control device for a fuel cell anodic line. FIELD OF THE INVENTION

[0001] The present invention relates to a control device for an anodic line of a fuel cell, in particular a fuel cell of the "proton exchange membrane" type. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0002] In a fuel cell intended to supply an electric current to a load, and more particularly in a fuel cell of the "proton exchange membrane" type, the "anodic line" designates the fluidic network passing through the electrochemical cells composing the cell and allowing the contact of the fuel fluid (typically hydrogen) with one of the electrodes (the anode) of the cells.

[0003] This line extends from a first controllable valve, associated with a source of fuel fluid, and fluidly connected to a fuel cell supply manifold, to a second controllable discharge valve, fluidly connected to a fuel cell discharge manifold.

[0004] The electrochemical reaction occurring in the cells of the battery generates the electric current drawn by the load.

[0005] It is necessary to supply these cells with reactive fluids in quantities adjusted to this electric current, which can naturally vary. For obvious reasons of mechanical strength, the pressure of the reactive fluids, fuel and oxidizer, is balanced within a cell, so that the fluid quantities are not necessarily the stoichiometric quantities of the electrochemical reaction. It is also necessary to remove unconsumed reactive fluids and residues of the chemical reaction occurring in the cells, in particular the water produced on the anode side or migrating from that side.

[0006] To this end, the fuel cell is equipped with control means connected to the first and second valves, these control means being configured to regulate the fuel fluid pressure in the cells of the cell to a setpoint pressure and to control the removal of residues from these cells. The setpoint pressure is that which allows the reactive fluids to be delivered to the cells in quantities adjusted to the electrical current to be produced.

[0007] Document US20230129936 illustrates a conventional approach to regulation in which the first valve is of the "on or off" type, this type of valve being referred to as an "injector" in the remainder of this description.

[0008] In this conventional control approach, known as "bang-bang" or "hysteresis control," a band is defined around a setpoint pressure within which the fluid pressure in the anodic line is allowed to vary. When the pressure reaches the upper or lower limit of this band, the fuel fluid injector is triggered (open or closed) to bring the pressure back into the desired range around the setpoint pressure.

[0009] This conventional approach, however, has limitations. Indeed, due to the bandwidth within which the fluid pressure is allowed to vary, the inertia inherent in the electrochemical process, and the delays introduced by the injector and any other actuators and sensors in the control loop, it is difficult to stabilize the pressure in the line around the setpoint pressure, as the pressure tends to oscillate. This can be partially compensated for by operating the control device, and therefore the injector, at a very high frequency. This imposes a high computational load on the control device and excessive stress on the injector, which is obviously undesirable for reliability reasons.

[0010] Alternative control approaches have been considered in the prior art, but these approaches are highly dependent on the fuel cell architecture. They are not always suitable for an architecture without a recirculation pump, as such a pump allows the fuel fluid not consumed by the electrochemical reaction to be reinjected into the anodic line. SUBJECT OF THE INVENTION

[0011] One object of the invention is to offer an alternative to prior art solutions. More specifically, one object of the invention is to provide a fuel cell anode line control device that is simple to implement, suitable for a wide range of fuel cell architectures, and improves the quality of the regulation. BRIEF DESCRIPTION OF THE INVENTION

[0012] To achieve one of these objectives, the object of the invention proposes a fuel cell comprising a stack of electrochemical cells equipped with at least one anodic supply collector and at least one anodic discharge collector, the fuel cell being intended to supply an electric current to a load and comprising: • at least one anodic line passing through the stack of electrochemical cells and in which a combustible fluid is capable of circulating, the anodic line extending: i. of a fuel fluid injection group fluidly connected to the anodic supply manifold, the injection group comprising a plurality of selectively controllable on / off injectors, an injection group control being defined by the number of selectively controlled open injectors among the plurality of injectors. ii. to a controllable purge valve fluidically connected to the anodic discharge manifold; • a pressure sensor capable of providing a representative value of the fuel fluid pressure in the anodic line; • a control device intended to develop, during successive calculation periods, the control of the injection group to be applied at the end of the calculation period in question in order to regulate the pressure of the fuel fluid in the anodic line so that it conforms to a given setpoint pressure, the control device being configured to evaluate a predetermined number of injection control profiles, a profile being composed of the injection group controls over a determined prediction horizon.

[0013] The control device is further configured to, during each calculation period: i. apply each injection control profile to a predictive anodic line pressure model and calculate a predictive performance indicator associated with that injection control profile; ii. select the injection control profile that optimizes the predictive performance indicator; iii. extract from the selected injection command profile the injection group command to be applied to the injection group at the end of the calculation period.

[0014] According to other advantageous and non-limiting features of the invention, taken alone or in any technically feasible combination: • the injection group commands that make up an injection command profile are linked together; • the injection group commands composing an injection command profile are all equal to each other; • the predictive model of anodic line pressure links the anodic line pressure to the electrical current, a purge valve control and the injection group control; • the predictive performance indicator incorporates constraints on the pressure of the anodic line, so that it tends to remain between a minimum pressure and a maximum pressure; • the predictive performance indicator includes a time weighting term; • the fuel cell further includes a supervisory unit configured to control the purge valve and apply the command provided by the control device to the injection group; • the fuel cell comprises two injection groups respectively associated with two anodic feed manifolds of the electrochemical cell stack, defining two anodic lines extending to the purge valve, the supervisory unit being configured to apply the injection group control provided by the control device to one and / or the other of the two injection groups; • both injection groups have the same number of injectors; • the injectors of one injection group have the same characteristics as the injectors of the other injection group; • the pressure sensor is located at the level of the anodic drain manifold; • the fuel cell is of the proton exchange membrane type; • the fuel cell has no recirculation between the valve of purge and the anodic supply manifold. Brief description of the drawings

[0015] Other features and advantages of the invention will become apparent from the detailed description of the invention which follows with reference to the accompanying figures in which:

[0016] [Fig.1]

[0017] Fig. 1 represents a fuel cell according to one embodiment;

[0018] [Fig.2a]

[0019] [Fig.2b]

[0020] [Fig.2c]

[0021] Figures 2a, 2b and 2c illustrate the operation of the battery in [Fig.1];

[0022] [Fig.3a]

[0023] [Fig.3b]

[0024] Figures 3a and 3b compare, at two different operating points, the regulation of a battery according to the prior art and of a battery according to the invention. DETAILED DESCRIPTION OF THE INVENTION

[0025] Figure 1 represents a fuel cell according to one embodiment. In this embodiment, the fuel cell retains the general architecture and operating principles of the cell described in document EP2707921. The invention is, of course, in no way limited to such an architecture, which is used solely to illustrate the principles of the invention.

[0026] As is well established in the field, the fuel cell 1 is designed to supply an electric current to a load by electrochemically reacting reactive fluids, fuel and oxidizer, within a plurality of cells. The reactive fluids are usually in the form of gases and can be composed of hydrogen and air or of hydrogen and an oxygen-rich gas.

[0027] The fuel cell of [Fig. 1] comprises a stack E of electrochemical cells T1, T2 connected electrically and fluidically. Each cell of the stack E includes an anode, towards which the fuel fluid flows, a cathode, towards which the oxidizing fluid flows, and an electrolyte between the two, here formed by a proton exchange membrane. At the ends of the stack E of cells T1, T2 are, on one side, an anode connected to a negative terminal to discharge electrons, and on the other side, a cathode connected to a positive terminal. The load is connected to these terminals, and the electrons forming the electric current produced by the fuel cell flow from the anode to the cathode to power the load.

[0028] The stack E of [Fig. 1] is formed of two types of electrochemical cells T1, T2, which may be identical to each other but are supplied with fuel separately. For ease of understanding, the two types of cells T1, T2 are shown side by side in [Fig. 1], but in practice the two types of cells are nested in the stack E, according to a regular alternation of cells of the first type T1 and cells of the second type T2. A stack is generally composed of several dozen or several hundred T1, T2 cells.

[0029] The stack E is traversed by two anodic supply manifolds C1,C2 constituting means for distributing the fuel fluid on the anode side, respectively, of the first type T1 cells and the second type T2 cells. Similarly, the stack E is traversed by an anodic discharge manifold C3 constituting means for evacuating the residues of the electrochemical reaction which have accumulated on the anodic side of the cell (water and nitrogen in particular which do not participate in the reaction and which, on the contrary, tend to inhibit it).

[0030] It would be possible to equip the stack E with two anodic evacuation collectors, each collector allowing the collection of reaction products from each type of cell T1, T2, but such an arrangement is not mandatory. Similarly, for the sake of simplicity, the supply and evacuation means on the cathodic side of the cells T1, T2 of the stack E are not shown, although these means are of course present.

[0031] Continuing the description of the fuel cell 1 shown in [Fig. 1], it is connected to a fuel fluid reservoir R, here hydrogen, via conduits. Other elements can be provided between the reservoir R and the fuel cell 1 to condition the fuel fluid in such a way as to promote the electrochemical reaction occurring in the cells T1, T2 (filter, heat exchanger, compressor, etc.). In any case, the fuel fluid is delivered to two injection groups G1, G2 respectively connected, fluidically, to the anodic feed manifolds C1, C2 of the stack E.

[0032] Each injection group G1,G2 comprises a plurality of injectors arranged in a parallel configuration and selectively controllable in an on / off fashion, i.e., between an "open" mode and a "closed" mode. An injection group control Cdl,Cd2 (more simply "injection control" in the remainder of this description) is defined by the number of injectors selectively controlled in open mode among the plurality of injectors, the other injectors in the group being controlled in closed mode. Thus, depending on the number of injectors controlled open by the injection control, it is possible to vary the hydrogen flow rate injected into a type of cell from a zero flow rate, for which the injection control positions all injectors in closed mode, to a maximum flow rate, for which the injection control positions all injectors in open mode.

[0033] The number of injectors in each group G1,G2 can be arbitrary, typically between 2 and 10 injectors per group. For simplicity of implementation, in the embodiment illustrated in [Fig. 1], the first injection group G1 and the second injection group G2 comprise the same number Ninj of injectors, all having identical fluidic characteristics (such as, for example, the fluid flow characteristic for a pressure difference of 1 bar).

[0034] This should not, however, be seen as a limitation and, in other implementations, the number of injectors in each group and the fluidic characteristics of the injectors may be different, insofar as it is possible to control the two injection groups G1,G2 to allow the distribution of an identical flow rate in both types of cells T1,T2.

[0035] The stack 1 also includes a controllable purge valve Vp fluidically connected to the anodic drain manifold C3 (or to the plurality of These collectors (if the stack E is equipped with such a plurality) are used. When this purge valve Vp is controlled to be open, the residues of the electrochemical reaction collected from the anodic side of cells T1 and T2 by the discharge collector C3 can be discharged from the stack E. This purge valve Vp is advantageously of the "on / off" type, and the purge valve control Cdp (more simply "purge control" in the remainder of this description) allows it to be controlled open or closed. However, in other implementations, the purge valve can be of the "proportional" type, and the purge control allows setting an opening ratio for this valve Vp between its fully open and fully closed positions.

[0036] The stack 1 can also be equipped with a cooling circuit allowing a heat transfer fluid to circulate in the stack E in order to control its temperature.

[0037] The injection groups G1, G2 and the purge valve Vp which have just been described, as well as all the other undescribed elements of stack 1, are controlled by a supervisory device S. This device is implemented by elementary processing components (CPU, microcontroller, FPGA, memory, converters, input / output interfaces...) combined together.

[0038] The supervisory device S operates the fuel cell 1. Specifically, it applies a first injection command Cdl, a second injection command Cd2, and a purge command Cdp to, respectively, the first injection group Gl, the second injection group G2, and the purge valve Vp. The supervisory device S collects and records in memory all measurements provided by the sensors with which fuel cell 1 is equipped, including the current I produced by fuel cell 1, the pressure measured by a pressure sensor P, and an injection command Cd provided by a control device K, which will be detailed in a later section of this description.

[0039] The supervision device S is configured in particular by hardware and / or software to operate stack 1 according to three different configurations, respectively represented in figures 2a, 2b, 2c.

[0040] In a first configuration shown in [Fig. 2a], the first injection group Gl is controlled to supply a flow of fuel fluid to the cell Tl of a first type. Conversely, the second injection group G2 is controlled to be fully closed. The purge valve Vp is also controlled to be closed. The fuel fluid therefore flows along a first anodic line extending from the first injection group Gl to the purge valve Vp, through the cell Tl of the first type, and also along a second anodic line extending from the purge valve Vp to the second injection group G2 through the second type T2 cell.

[0041] In a second configuration shown in [Fig. 2b], the first injection group G1 is controlled to be fully closed, and the second injection group G2 is controlled to supply a flow of fuel fluid to the second-type cell T2. The purge valve Vp is also controlled to be closed. The fuel fluid thus flows along a first anodic line extending from the second injection group G2 to the purge valve Vp, through the second-type cell T2, and also along a second anodic line extending from the purge valve Vp to the first injection group G1, through the first-type cell T1.

[0042] In a third configuration shown in [Fig. 2c], the first injection group Gl and the second injection group G2 are both controlled to supply the same fuel fluid flow rate to cells T1, T2. The purge valve Vp is controlled to be open. The fuel fluid thus flows along a first anodic line extending from the first injection group Gl, through the first-type cell T1 and the purge valve Vp. It also flows along a second anodic line extending from the second injection group G2, through the second-type cell T2 and the purge valve Vp. This configuration allows the reaction residues to be ejected from the anodic portion of the cells and from the fuel cell 1.

[0043] The monitoring device S is configured to repeatedly alternate the first two configurations a certain number of times, then follow this alternation with the third configuration. The previously cited document EP2707921 provides additional implementation methods for this architecture, which we do not reproduce in this description to keep it concise, and details all the advantages of this operating principle. In particular, the fuel cell in the implementation method of [Fig. 1] does not have any recirculation between the purge valve Vp and the anodic feed manifolds.

[0044] The present description is primarily concerned with the anodic line of the fuel cell 1, that is to say, the fuel flow extending, in one direction or the other, from an injection group to the purge valve Vp, through a cell T1,T2 of one of two types. As has just been seen, a fuel cell compatible with the present invention may have several anodic lines, which in this case intersect at the purge valve Vp, and therefore have identical or very similar pressures. But the invention can also be exploited in a simpler architecture, having only a single anodic line, and / or having possibly a recirculation line, between the purge valve Vp and the anodic feed manifolds, without crossing one of the cells of the stack E.

[0045] In all cases, and as stated in the introduction to this application, the aim is to control the fuel fluid pressure in the anodic lines, and more specifically in cells T1 and T2 of fuel cell 1, so that it conforms to a setpoint pressure Pref. This setpoint pressure Pref is the pressure required to deliver the reactive fluids to the cells in quantities adjusted to the electrical current to be produced. The setpoint pressure Pref can be determined by the supervisory control system S, for example, using a function or a table that associates the setpoint pressure Pref with a given current for fuel cell 1.

[0046] To enable this pressure regulation of the anodic line, the cell 1 includes a pressure sensor KP capable of providing a value P representative of the fuel fluid pressure in this anodic line. In the case of the implementation shown in [Fig. 1], this sensor KP is located in the section of conduit that separates the drain manifold C3 from the purge valve Vp.

[0047] Stack 1 also includes a control device K for generating, over successive calculation periods, an injection command Cd. These calculation periods can follow one another according to a determined sampling frequency. In the case of stack 1 of the chosen implementation mode, this sampling frequency can be on the order of 100 Hz, or more generally between 10 Hz and 1 kHz, defining a calculation period of between 1 millisecond and 100 milliseconds.

[0048] Just like the supervisory device S, the control device K is implemented by combining elementary processing components (CPU, microcontroller, FPGA, memory, converters, input / output interfaces, etc.). Although, for clarity, the control device K and the supervisory device S are shown in [Fig. 1] as two separate devices, it is entirely possible for their respective functions to be implemented by a single device; for example, for the control device K to be implemented by the elementary processing components constituting the supervisory device S.

[0049] The injection command Cd is provided by the control device K to the supervisory device S. The latter uses this injection command Cd to generate the first injection command Cd1 and the second injection command Cd2, respectively applied to the first and second injection groups G1, G2 at the end of the calculation period. The injection command Cd provided by the control device K defines the number of injectors controlled in open mode in one and / or the other of the injection groups G1, G2, it being understood that the non-injectors controlled in open mode in an injection group, are controlled in closed mode.

[0050] Thus, when the monitoring device S operates stack 1 in the first configuration, the monitoring device S: applies to the first injection group G1 the first injection command Cdl fixed to the injection command Cd supplied by the control device K; applies a second injection command Cd2 to the second injection group G2, which commands this group G2 to be fully closed. In other words, the second injection command is set to 0.

[0051] Similarly, when the supervisory device S operates stack 1 in the second configuration, the supervisory device S: applies to the first injection group G1 the first injection command Cdl which commands this group G1 to be fully closed. The second injection command is therefore set to 0; applies to the second injection group G2 the second injection command Cd2 fixed to the injection command Cd supplied by the control device K.

[0052] Finally, when the monitoring device S operates stack 1 in the third configuration, the monitoring device S: applies to the first injection group G1 the first injection command Cdl fixed to the injection command Cd supplied by the control device K; applies to the second injection group G2 the second injection command Cd2 fixed to the injection command Cd supplied by the control device K.

[0053] We now present the principles implemented by the control device K to develop the injection control Cde, selectively applied to one and / or the other of the injection groups G1,G2 by the supervision device S.

[0054] First of all, a model of the pressure P prevailing in the anodic lines of the pile 1 is developed, the model being able to be expressed by an equation of state linking this pressure to the electric current I, to the purge control Cdp and to the injection control Cd.

[0055]

[0056] P((k+1)-Ts) = AP(k-Ts) + B-^k-nd)-Ts) With ik uk = Cdp{k)

[0057] where * Ae€P€€Be Plx3 are the terms of the state representation; • Ts is the calculation period; • e € N is a delay; • P e € P is the pressure of the anodic line; • the vector ue P3xi concatenates the system inputs, namely the current Z e € P, the on / off purge control Cdp G {0,1} and the injection control, i.e., the number of open injectors w ¢= fri 1 V 1 in the first two configurations of stack 1 and Hinj ly, K ..., ïV jnjj n . c-lno A 7*V. .1 in the stack purge configuration 1. Fljnj t Z- zy ifijj

[0058] It is noted that when the purge command takes the value 1, stack 1 is then configured according to the third (purge) configuration presented above, in which both of the two injection groups G1,G2 are activated. Therefore, the number of open injectors ninj is necessarily even and can take a value equal to the total number of injectors 2*Ninj composing these two groups G1,G2, in the chosen implementation mode.

[0059] The numerical parameters of the state equation (A, B, and nd) can be obtained by identification from pre-existing measurements, as is common practice in this field. The delay nd is typically between 0 and 4, and it can notably depend on the chosen sampling frequency (defining the calculation period Ts). The term A is equal to or close to 1.

[0060] With this system model, the development of the injection control Cd can be formulated as an optimization problem for a predictive performance indicator, over a given prediction horizon defined by a number N of computation periods. This indicator can be based on a difference between a pressure estimated by the model over this horizon and the setpoint pressure Pref. The prediction horizon N is typically chosen between 2 and 20, for example N=10, although other values ​​outside the range of 2 to 20 are also possible.

[0061] By way of illustration, this approach can be formulated by a constrained quadratic optimization equation, in which we find the difference between the pressure estimated at an instant i of the prediction horizon P(k+i) and the setpoint pressure Pref that we seek to minimize:

[0062] = argminE^[ (P(k + i) -P^) Q (P(k+i) -) + ninj■ R• ] ninja

[0063] Complying with the constraints:

[0064] ^((£ + 0 'D = AP(kTs) + Bu((k-nd) Ts) for ie{Q .... A?-l} [°065] + ..., N^' ^Cdp=O [°°66] ninj(k + i^{o,2,4, ..., 2*^^^ ^Cdp=l

[0067] p(k) e [P,P],}>for i& {0, .... N -1}

[0068] Where • NE € N is the prediction horizon; • ninj is the injection command profile, consisting of the N injection commands ninj(k+i) used successively in the equation of state at the end of each of the N periods of the prediction horizon nmj=[ninj(k) ninj(k+l) ... ninj(k + Nl)]r; • the matrices Q, R are respectively an output weighting matrix and a control input weighting matrix u. • P and P are the upper and lower pressure limits to be respected.

[0069] This additional (and optional) constraint on the upper and lower pressure limits is advantageous in that it prevents the anodic pressure of cell 1 from reaching, even transiently, values ​​incompatible with the proper functioning of this cell. The parameters P and P can be configurable values, for example, stored in a memory of the monitoring device S. Within the scope of the invention, any type of additional constraint may be added in addition to or in place of the one given as an example, to take into account operational requirements.

[0070] The optimal solution to the constrained optimization problem described above is the n*nj profile of the injection commands applied over the prediction horizon N, denoted by:

[0071] ninj=[n-nj(k) n*„j(k+l) ... n-nj(k + Nl)]T

[0072] It should be noted that the delay nd is generally much smaller than the prediction horizon N. Therefore, the evaluation of the predictive performance indicator relies, through the pressure estimates P(k+i) provided by the state equation, on the current value I(k+i) and the purge command Cdp(k+i) that make up the input vector u(k+i) of this equation. Some of these values ​​are not necessarily known at time k+i, particularly when this time i is greater than the delay nd. In this case, the value of this current and the purge command can be set to their last known value in the input vector u(kl). As for the purge command Cdp, it can be assumed that this value is known in advance by the monitoring device S, and that this device S makes the value of this purge command available to the control device K over the entire prediction horizon or over a portion thereof.

[0073] Upon obtaining the optimal injection command profile n^j, which defines the time series of injection commands to be applied successively during the prediction horizon, the control device K extracts the injection command n^nj(k) to be applied to the injection group at the end of the current calculation period k. This injection command n-nj(k) extracted from the optimal profile corresponds to the injection command Cd communicated by the control device K to the supervisory device S, which applies it to one or both of the injection blocks G1, G2, depending on the current configuration of the stack, as previously described.

[0074] The optimization problem formulated above is a constrained linear quadratic problem that can be solved using a dedicated optimization solver. However, this formulation and the requirement to use a dedicated solver impose a strong constraint on the performance of the control device K necessary for its real-time implementation.

[0075] Due to the discrete nature of an injection command, chosen from the set {0,...,Ninj}, the space of possible injection command profiles, over a prediction horizon N, consists of NinjAN elements. Alternatively, instead of using a solver, one could explore the entire space of possible injection commands and search for the optimal n*nj injection command profile that optimizes the predictive performance indicator within this space. However, this space is far too large to be fully explored within the available computation time. For example, for injection groups composed of 4 injectors (Ninj taken as 4) and a time horizon of 10 computation periods, there are more than 10A6 possible injection command profiles.

[0076] To facilitate the implementation of the algorithm on the control device K, and according to a highly advantageous feature that reduces the computational load, the injection control profile is parameterized over the prediction horizon N. By parameterizing the injection control profile, we mean that the inputs of this profile, the injection controls flinj(k + i) over the prediction horizon, are linked together. This parameterization leads to a reduction in the number of injection controls ninj(k + i) that can be applied at each calculation period of the prediction horizon to a subset of injection controls with reduced cardinality.

[0077] Therefore, the number of ninj injection control profiles to be evaluated during a computation period is limited. The number of ninj profiles is chosen to allow the implementation of the algorithm on the control device K; this number can be 100, 50, 10 or less depending on the characteristics of this device and, in particular, its computing capacity.

[0078] In a preferred embodiment, and to illustrate this approach, the parameterization is carried out by fixing the injection commands n^j(k + i) composing the injection command profile ninj so that these commands are constant and equal to the value nfnj(k) throughout the prediction horizon:

[0079] n.nj: = [n^jik) n^k) ... ninj(k)]T

[0080] In this embodiment, the number of possible Ninj injection control profiles is limited to the number of Ninj injectors in each injection group G1,G2. Given this limited number (the number of Ninj injectors typically being on the order of 10 or less), it is possible to evaluate, during a calculation period, the predictive performance indicator for all possible Ninj injection control profiles.

[0081] The injection control profile could naturally be parameterized differently over the prediction horizon N than the one presented above as an illustration. More generally than parameterizing the inputs of an injection control profile, any heuristic approach can be used to define a limited and predetermined number of Ninj injection control profiles that will be evaluated during the optimization phase. This limited and predetermined number is strictly less than the total number of possible injection control profiles (the possible NinjAN profiles, mentioned above).

[0082] Advantageously, and to avoid the emergence of bias during regulation, the predictive performance indicator is adjusted by introducing a time weighting term (N4\". This weighting term tends to favor the first N / instants of the prediction horizon in the predictive performance indicator. We can also introduce certain constraints into the performance indicator itself, which, according to the preferred embodiment described above, allows us to express the optimization problem of the predictive performance indicator as:

[0083] = argmin^0'() [ (l'(k + i) Q- (P(k + i) - P"^) R .n^f+Qcn-rnax(Pik + i) -P, 0)" + 0cs-max(P- + / ).0)2]

[0084] Respecting the constraints:

[0085] P((k+i)T,)=AP(kT,)+Bu((kn^^^ for ie {0, .... TV -1}

[0086] mnj(k+e {o, ..., Mm J' for ie{0, ..., Nl} if Cdp =0 100871 . - Cdp = I

[0088] where • n is a priority coefficient of the time weighting factor / Ni \n ; \ N / * Qcs is 'a constraint violation weighting matrix.

[0089] In this expression, the search for the optimal profile of the n*n / injection commands is reduced to the search for a scalar quantity rifinj{k}.

[0090] Also, the optimization problem can be easily solved by applying the following calculation sequence: • Apply each ninj injection control profile to the anodic line pressure model and calculate the predictive performance indicator associated with this injection control profile; • select the injection control profile ri^j that optimizes the predictive performance indicator.

[0091] We can thus calculate the value of the indicator for each value Hjn j ( k ) of the injection command in its entirety admissible (({gj or 4 2*N- •] sc'on the value of the purge command Cdp. We retain the nmj(k) value of the injection command that optimizes this indicator.

[0092] The control of the injection group Cd provided by the control device is extracted from the selected injection control profile n^nj. In the preferred embodiment in which an injection control profile consists of injection controls all equal to each other, this extraction step then consists of providing the scalar value n^j ( k ) .

[0093] Figures 3a and 3b compare, at two different operating points (corresponding to two different currents produced by the battery), the regulation of a battery according to the prior art and of a battery according to the invention.

[0094] The graphs represent the evolution of the pressure (in arbitrary units), as measured by the pressure sensor P, over a period of time (approximately 30 seconds on the graph in [Fig.3a], and 20 seconds on the graph in [Fig.3b]).

[0095] Each of these graphs presents 3 time sections, 2 sections marked SI during which the stack was operated using a control device configured according to the invention, and another section marked S2 during which this control device was configured according to a "bang bang" approach of the prior art.

[0096] We also marked the time periods (Cp=l) during which the fuel cell 1 was configured in purge mode, with the purge valve open.

[0097] It is observed that in both operating points, the pressure excursion around the setpoint pressure is much smaller for the configured stack in accordance with the invention, which shows the good performance of the regulation and the full benefit of the invention.

[0098] Of course the invention is not limited to the modes of implementation described and alternative embodiments can be made without departing from the scope of the invention as defined by the claims.

[0099] Thus, the predictive pressure model used by the control device was chosen to linearly combine the pressure and the input vector u, composed of the injection command, the current, and the purge command. It could, of course, be expected that the predictive pressure model will differ from the one given as an example. In particular, it could use other measurements, for example, a temperature measurement of the fuel cell or a measurement of the purge valve flow rate (or an estimate of this flow rate). This model could also be nonlinear or take the form of a device configured by learning (neural network, Bayesian network).

[0100] Similarly, the predictive performance indicator is not necessarily expressed in quadratic form as presented. It can be expressed in any suitable form, based on the difference between the estimated pressure and the setpoint pressure over the prediction horizon.

[0101] This setpoint pressure, provided by the monitoring device, is not necessarily constant.

[0102] It has been shown that the predictive performance indicator can incorporate constraints tending to maintain the pressure between a minimum and maximum pressure, and it can be predicted that this indicator incorporates other constraints (additional or as a replacement) or that it does not incorporate such constraints.

[0103] As already indicated in a previous passage, the control device can be applied to a fuel cell composed of several types of cells, each type being associated with its own injection group as illustrated in the present description, but this should not be considered exhaustive. The control device can also be used in a fuel cell composed of cells of a single type, these cells being associated with a single injection group. This fuel cell may include a recirculation line for unconsumed fuel, although the invention is more particularly suited to a fuel cell without any recirculation between the purge valve and the anodic feed manifold.

Claims

1. Demands Fuel cell (1) comprising a stack (E) of electrochemical cells (T1,T2) equipped with at least one anodic feed collector (C1,C2) and at least one anodic discharge collector (C3), the fuel cell (1) being intended to supply an electric current to a load and comprising: • at least one anodic line passing through the stack (E) of electrochemical cells and in which a combustible fluid is capable of circulating, the anodic line extending: i. of an injection group (G1,G2) of a fuel fluid fluid connected to the anodic supply manifold (C1,C2), the injection group (G1,G2) comprising a plurality of injectors selectively controllable on or off, an injection group control being defined by the number of selectively controlled injectors open among the plurality of injectors. ii. to a controllable purge valve (Vp) fluidically connected to the anodic evacuation manifold (C3); a pressure sensor (KP) capable of providing a representative value of the fuel fluid pressure in the anodic line; a control device (K) designed to develop, over successive calculation periods, the injection unit control (Cdl,Cd2) to be applied at the end of the calculation period in question in order to regulate the fuel fluid pressure in the anodic line so that it conforms to a given setpoint pressure, the control device being configured to evaluate a predetermined number of injection control profiles, a profile being composed of the injection unit control commands over a determined prediction horizon, the control device also being configured to, during each calculation period: i. Apply each injection control profile to a predictive model of anodic line pressure and calculate a predictive performance indicator associated with this injection control profile; ii. Select the injection control profile that optimizes the predictive performance indicator; iii. Extract from the selected injection control profile the injection group control (Cd) to be applied to the injection group (G1,G2) at the end of the calculation period.

2. Fuel cell (1) according to the preceding claim in which the injection group controls composing an injection control profile are linked together.

3. Fuel cell (1) according to the preceding claim in which the injection group commands composing an injection command profile are all equal to each other.

4. Fuel cell (1) according to any one of the preceding claims wherein the predictive model of anodic line pressure links the anodic line pressure to the electric current, to a purge valve control and to the injection group control.

5. Fuel cell (1) according to the preceding claim in which the predictive performance indicator incorporates constraints on the pressure of the anodic line, so that the latter tends to remain between a minimum pressure and a maximum pressure.

6. Fuel cell (1) according to any one of the preceding claims wherein the predictive performance indicator includes a time weighting term.

7. Fuel cell (1) according to any one of the preceding claims further comprising a supervisory unit (S) configured to control the purge valve (Vp) and apply the command (Cd) supplied by the control device (K) to the injection group (G1,G2).

8. Fuel cell (1) according to the preceding claim comprising two injection groups (G1,G2) respectively associated with two anodic feed collectors (C1,C2) of the stack (E) of electrochemical cells, defining two anodic lines extending to the purge valve (Vp), the supervisory unit (S) being configured to apply the injection group control (Cd) supplied by the control device (K) to one and / or the other of the two injection groups.

9. Fuel cell (1) according to the preceding claim in which the two injection groups (G1,G2) have the same number of injectors.

10. Fuel cell (1) according to the preceding claim in which the injectors of one injection group (G1,G2) have the same characteristics as the injectors of the other injection group.

11. Fuel cell (1) according to any one of the preceding claims wherein the pressure sensor (KP) is disposed at the anodic exhaust manifold (C3).

12. Fuel cell (1) according to any one of the preceding claims wherein the fuel cell (1) is of the proton exchange membrane type.

13. Fuel cell (1) according to any one of the preceding claims devoid of any recirculation between the purge valve and the anodic feed manifold.

Images