Method for managing the energy of a hybrid vehicle equipped with an on-board navigation system and an electric powertrain comprising at least one electric machine

Abstract

The invention relates to a method for energy management of a vehicle having a navigation system and an electric machine (MEL) powered by a fuel-to-electricity conversion system and an electrical energy storage system a) providing a destination in a navigation system b) calculating the route by the navigation system c) applying a predictive model of the vehicle to said route, to deduce the predicted required power trace (Preq) d) managing the electricity supply of the electric machine (MEL) with d0) measuring the state of charge (SoC) of the electrical energy storage system (BAT), d1) calculating the control of the electrical energy sources (BAT,FC), from the predicted power trace (Preq) by the Pontryagin Maximum Principle (PMP) to determine the optimal trace of the state of charge (SoC) d2) determining in real time the electricity control to power the electric machine (MEL).Figure for abstract: Fig 1.

Description

Title of the invention: Method for managing the energy of a hybrid vehicle equipped with an on-board navigation system and an electric drive train comprising at least one electric machine Technical field

[0001] The invention relates to the management of the energy supply, in particular electrical energy, of vehicles, in particular land vehicles, equipped with an electric drive train comprising at least one electric machine. It relates to hybrid vehicles, comprising several energy sources, and in particular to all-electric hybrid vehicles, which may comprise one or more electricity storage systems (such as batteries) and one or more gas conversion systems, in particular hydrogen, into electricity (such as fuel cells).

[0002] The energy control of this type of hybrid vehicle is generally controlled in the vehicle by a so-called power management system (also known as Power Management System or by its acronym PMS). This system is the control unit responsible for distributing the power required by the electric machine between the different available energy sources.

[0003] Generally speaking, this management system seeks to minimize the energy consumption of the electrical machine, and / or to minimize the aging of the energy sources, by taking into account a certain number of constraints. Prior art

[0004] The so-called optimal control theory is known, which makes it possible to determine the control of a system which minimizes or maximizes a given performance criterion, possibly under constraints which may relate to the control or the state of the system considered.

[0005] Offline, that is to say when the driving cycle is known in advance, mathematical tools such as Dynamic Programming (DP) or the Pontryagin Maximum Principle (PMP) make it possible to find the optimal solution to this type of problem.

[0006] Online, that is to say in real conditions, with a management system embedded in the vehicle, the cycle is not known in advance, many methods have been studied to get closer to the optimal solution.

[0007] Among them, we can cite the equivalent consumption minimization strategy (the Anglo-Saxon expression of which is “Equivalent Consumption Minimization Strategy”, with its acronym ECMS), which is the most commonly used strategy. Its principle is to define, for a hybrid vehicle with an internal combustion engine and a battery, an equivalence factor between the use of the battery (electric energy) and the fuel consumption of the internal combustion engine. The challenge, in this strategy, is then to find the equivalence factor which makes it possible to achieve a minimum consumption on a given cycle while respecting the constraints of the different power sources. Details on this strategy can be found, for example, in the article "Equivalent Consumption Minimization Strategy, Energy management controller for P0-P4 hybrid electric vehicles", which can be found by following the following link: https: / / fr.mathworks.com / help / autoblks / ref / equivalentconsumptionminimizationstrateg y.html citing the following publications: [1] Balazs, A., Morra, E., and Pischinger, S., Optimization of Electrified Powertrains for City Cars. SAE Technical Paper 2011-01-2451. Warrendale, PA: SAE International Journal of Alternative Powertrains, 2012. [2] Onori, S., Serrao, L., and Rizzoni, G., Hybrid Electric Vehicles Energy Management Systems. New York: Springer, 2016.

[0008] This strategy, based on the heuristic principle of an equivalence between electrical energy and fuel consumed, is in fact derived from the Pontryagin Maximum Principle (PMP) mentioned above.

[0009] In the PMP theory, the factor that is called in the ECMS "equivalence factor" corresponds to what is called an adjoint state associated with the state variable of the system (in this case the state of charge of the battery, SoC (for "State of Charge" in English), in the case of the energy management of hybrid vehicles). This adjoint state is a priori variable over time. In particular, when the constraints on the state of charge of the battery are active (that is to say that the state of charge at a given instant approaches a minimum state of charge SoCmin or maximum state of charge SoCmax), it is supposed to see its value vary to avoid exceeding these constraints.

[0010] Until now, online PMS algorithms based on the PMP / ECMS principle assume, to simplify the problem, a constant adjoint state / equivalence factor. In these cases, the optimization performed is therefore blind to the constraints on the state of charge of the SoC battery: the optimal control is calculated without "knowing" that the state of charge of the SoC battery must not exceed certain values.

[0011] However, on certain cycles (journeys), in particular with large variations in slope, this hypothesis of a constant adjoint state / equivalence factor can however prove to be limiting, and lead to distributions between the different energy sources which are not optimal.

[0012] The aim of the invention is to improve these energy supply management methods for hybrid vehicles, in particular for all-electric hybrid vehicles with an electricity storage system and a fuel conversion system, for example gas, into electricity. This involves in particular developing such a power supply management method which is more efficient, in particular which makes it possible to size the electricity sources (batteries in particular) as accurately as possible, and / or which is more robust to hazards, and / or which does not require - or requires less - adjustment upstream of the journey to be traveled, and / or which takes greater account of the limitations of the overall system, whether these are constraints on the controls or on the charge states. Summary of the invention

[0013] The invention firstly relates to a method for managing the energy of a hybrid vehicle equipped with a navigation system and an electric drive train comprising at least one electric machine (MEL) which is supplied with electricity by electrical energy sources (BAT, FC) comprising at least one fuel-to-electricity conversion system, and at least one electrical energy storage system, such that said method comprises: (a) a step of providing information about a destination in a navigation system which retrieves and uses satellite navigation data in real time, b) a step of calculation by the navigation system of the route to be covered from navigation data, c) a step of applying a predictive model of the vehicle to said journey, from at least some of said navigation data and from vehicle parameters, to predict the vehicle speed trace (Vvh) on the journey and to deduce therefrom the predicted required power trace (Preq) for the electric machine (MEL) during said journey, d) a step of managing the electricity supply of the electric machine (MEL) by the electrical energy sources, which comprises dO) a sub-step of measuring the state of charge (SoC) of the electrical energy storage system(s) (BAT), dl) a sub-step of calculating the optimal control of the electrical energy sources (BAT, FC) in a loop, in particular called a slow loop, from the predicted power trace (Preq) in step c), said calculation being based on the Pontryagin Maximum Principle (PMP) and taking into account - the state of charge (SoC) of the electrical energy storage system(s) (BAT) measured in step dO) as the initial state of charge, - and the constraints on the limitations of the controls of the energy sources (BAT, FC) and on the limitations of the state of charge (SoC) of the electrical energy storage system(s) (BAT), to determine at each position of the vehicle on the route predicted in step b) the optimal trace of the state of charge (SoC) and the optimal adjoint states (p;), d2) a sub-step of determining in real time the control of the share of electricity coming from the electrical energy storage system(s) (BAT) and the share of electricity coming from the fuel conversion system(s) (FC) to power the electric machine (MEL) during said route, by minimizing the Hamiltonian of the optimal control problem which is defined by the Pontryagin Maximum Principle (PMP) and which depends on the optimal adjoint states (p;) calculated in step dl).

[0014] The calculation loop carried out in step dl) is preferably carried out according to a certain frequency, and preferably not in real time.

[0015] According to the invention, the or at least one of the (or all of the) electrical energy storage systems is preferably of the battery (BAT) and / or supercapacitor type.

[0016] According to the invention, the or at least one of the (or all of the) fuel-to-electricity conversion systems is preferably of the fuel cell (FC) type, or of the turbogenerator type (generally gas) or of the heat engine type associated with an alternator.

[0017] The term "fuel" is therefore understood as a fluid, in particular a gas such as hydrogen (for a fuel cell in particular) or a liquid such as gasoline (for a thermal engine coupled to an alternator in particular, which may also, in addition to or instead of gasoline, be supplied with hydrogen-type gas).

[0018] The invention thus defined is based on the theory of optimal control with use of the Pontryagin Maximum Principle (PMP).

[0019] It takes into account the constraints on the state of charge of the electricity storage system (battery type) in the calculation of the optimal control, and thus makes it possible to size the vehicle architecture as accurately as possible: it is thus possible to provide smaller batteries for equivalent applications. It has the same utility in applications with hybrid storage combining one or more supercapacitors and one or more batteries for example.

[0020] The invention takes into account the limitations of the overall energy management system, in particular the constraints on the controls or on the load states.

[0021] The invention can be implemented online, and it is robust to hazards (for example a traffic hazard or a driving behavior hazard).

[0022] Another advantageous point of the invention is that it only requires data from a navigation system with geolocation, and that it does not require special focus before implementing it.

[0023] Advantageously, steps c), d0) and dl) can be repeated to determine the control of sub-step d2) according to a given time step At, constant or variable, depending on the duration of the journey.

[0024] Restarting the calculation in this way makes it possible to obtain better precision, insofar as it may turn out that the predicted power trace only has the desired precision over a given time horizon, which may be a fairly short horizon (for example from 1 minute to a few tens of minutes, in particular a few minutes or at most 10 minutes).

[0025] It is thus possible to take into account an unforeseen power demand, a state of charge, for example due to a change of route, one or more stops of the vehicle during the journey, etc., by redoing the calculation of step dl) at a given frequency (constant frequency or changing in a predetermined manner or not throughout the journey, in particular according to its predicted duration and / or according to the type of journey or portions of journey). Preferably, this frequency can be fixed, but it is also possible to divide the cycle (the journey) according to the predicted power trace: to give an example, if the cycle provides for a portion of stabilized journey, in particular on a motorway, it is possible to avoid restarting the calculation during this stabilized portion.

[0026] Alternatively, steps c), d0) and dl) can advantageously be repeated to determine the control of sub-step d2) when a given deviation ASoC is reached between the measured state of charge (SoC) and the optimal state of charge predicted in sub-step dl).

[0027] Here again, we take into account an unforeseen power demand and / or state of charge, but only when a difference is observed between the predicted optimal charge rate and the actual charge rate: we can thus limit the calculation iterations to only carry them out when it is deemed necessary, only when there is a deviation deemed too significant compared to the predictions.

[0028] Advantageously, the navigation data may include traffic information, in particular the traffic conditions, speed limits and gradients of the possible traffic lanes for completing the journey.

[0029] Preferably, in step d), the fuel consumption of the fuel-to-electricity conversion system(s) (FC) (in particular hydrogen-type gas when it is a fuel cell) can be minimized.

[0030] Cumulatively or alternatively to this minimization of fuel consumption, it is also possible to minimize the aging of the storage system (battery(ies)) or that of conversion of fuel into electricity (fuel cell). This aging can be evaluated / quantified, in particular by measuring the voltage at the terminals of the systems in question.

[0031] In step d), the constraints on the limitations of the controls of the energy sources (BAT, FC) and on the limitations of the state of charge can be chosen from at least one of the following constraints, in particular all of the following constraints: - Preferably, in step d), a state of charge (SoC) of the or at least one of the electrical energy storage system(s) (BAT) is imposed between a minimum state of charge (SoCmin) and a maximum state of charge (SoCmax). - Preferably, in step d), a state of charge (SoC) of the or at least one of the electrical energy storage system(s) (BAT) is imposed at the end of the journey equal to said state of charge at the start of the journey. - Preferably, in step d), a given state of charge (SoCfin) of the or at least one of the electrical energy storage system(s) (BAT) is imposed at the end of the journey. - Preferably, in step d), an IFC current generated by the gas to electricity conversion system (FC) is imposed between a minimum current IFCmin and a maximum current IFC max. - Preferably, in step d), a current IBAT generated by the electricity storage system (BAT) is imposed between a minimum current IBATmin and a maximum current IBATmax. - Preferably, in step d), a gamma current gradient FC generated by the fuel-to-electricity conversion system (FC) is imposed, between a minimum gradient gamma FCmin and a maximum gradient gamma FCmax.

[0032] The vehicle parameters used in step c) of applying a predictive model may include at least the architecture of the vehicle's powertrain, and the characteristics of the electric machine (MEL). This may include, in particular, taking into account the weight and dimensioning of the vehicle, the auxiliary equipment consuming electricity, or the dimensioning and design of the electric machine. This may also include the dimensioning / number of power converters which are generally provided to adapt the electrical voltage of the various electrical components to the network voltage.

[0033] Preferably, in sub-step dl) and / or in sub-step d2), a so-called interior point method (MPI) can be used, in particular to manage the constraints of maximum voltage state of the electrical network Umax and the state of charge (SoC) of the electricity storage system(s) (BAT). This is indeed a very effective method for solving functional optimization problems under constraints.

[0034] This method of interior points preferably used both in sub-steps dl) and d2) corresponds to the method of solving the problem defined by the Pontryaguin Maximum Principle mentioned above.

[0035] Any primal method can also be used for sub-steps dl) and / or d2) dual: For real-time balancing and disturbance rejection (during sub-step d2), the proposed algorithm requires calculating the dual variables of the problem constraints. Then, to calculate the optimal trajectory, it is also possible to use so-called primal-dual methods, either in a discretization then optimization context, or in an optimization then discretization context. The choice of primal-dual methods in the optimization then discretization context is favored for reasons of speed of calculation time.

[0036] Preferably, in step dl) of loop calculation, called slow, the calculation is carried out over the entire remaining path.

[0037] The invention also relates to a computer, or calculator configured to implement the method as described above. Any computer means, on-board or not, can also be considered.

[0038] The invention also relates to a vehicle of the hybrid type and equipped with an on-board navigation system and an electric powertrain comprising at least one electric machine (MEL) supplied with electricity by electrical energy sources (BAT, FC) comprising at least one fuel-to-electricity conversion system, in particular of the fuel cell (FC) type, and at least one electrical energy storage system, in particular of the battery (BAT) and / or supercapacitor type. The vehicle further comprises means for controlling the electrical systems in electrical power supply of the electric machine (MEL) controlled according to the method described above or with the computer or the calculator mentioned above and implementing said method.

[0039] The vehicle according to the invention can be land-based (light vehicle, heavy goods vehicle, bus), but also maritime or air-based.

[0040] Preferably, the navigation system is integrated into the vehicle, and in particular has a screen integrated into the dashboard. Alternatively, the navigation system is autonomous, and connected to the computer or calculator implementing the method according to the invention mentioned above.

[0041] The invention also relates to any computer program product downloadable from a communications network and / or recorded on a medium readable by a computer or a calculator and / or executable by a processor, comprising program code instructions for implementing the method described above, when said program is executed on a computer or a calculator.

[0042] The invention also relates to any storage medium readable by a computer or calculator and storing instructions, which, when executed by a computer or a calculator, imply that the computer or the calculator implements the method described above. List of figures

[0043] [Fig.l] [Fig.l] represents the general energy architecture of a vehicle equipped with an electric machine powered by a battery and two fuel cells. [Fig.2] [Fig.2] represents an example of an energy control algorithm for a vehicle according to the invention. [Fig.3] [Fig.3] represents the energy architecture of a vehicle using an algorithm according to the invention. [Fig.4] [Fig.4] is a zoom on the fuel cell part of the vehicle. [Fig.5] [Fig.5] is a zoom on the battery part of the vehicle. [Fig.6a] [Fig.6a] is a plot of two identical journeys (made at different times) by a GPS-type navigation system, with longitude on the abscissa and latitude on the ordinate. [Fig.6b] [Fig.6b] is a corresponding power trace generated from plotting the two paths according to [Fig.6a]. [Fig.7a] [Fig.7a] is a plot of the calculated “off-line” SoC state of charge of the vehicle battery during a journey as a function of the distance traveled (in km). [Fig.7b] [Fig.7b] is a plot of hydrogen consumption (in kg) calculated “off line” by the vehicle’s fuel cell as a function of distance traveled (in km). [Fig. 8a] [Fig.8a] represents as a function of time the traces of the SoC state of charge of the optimal battery predicted at some of the iterations of the slow loop (1st and 37th iterations represented respectively by the solid line curve and the dotted line curve). [Fig.8b] [Fig.8b] represents as a function of time the sum of the predicted SoC state of charge trace segments (bottom graph) between each update of the slow loop as well as the state of charge trace actually produced (top graph). [Fig.9a] [Fig.9a] represents the trace of the optimal SoC state of charge calculated offline in function of the distance (in km). It serves as a reference (it is mathematically the best solution that can be found to minimize the criterion set by the invention (here the consumption of H2 type fuel)). [Fig.9b] [Fig.9b] represents the trace of the state of charge SoC actually made during the cycle (trip) as a function of the distance (km). [Fig.9c] [Fig.9c] gives an example of a SoC state of charge trace result (actually made during the cycle) as a function of distance (in km) if we had an algorithm that does not take into account the limitations of the SoC state of charge. In other words, this curve shows what we would have as a trace with a prior state-of-the-art algorithm. [Fig.10] [Fig. 10] is a fuel consumption graph for batteries of different sizes on the same journey, with the % reduction in battery capacity on the abscissa, and the % excess fuel consumption on the ordinate, the solid line curve corresponding to taking into account the state of charge constraints (according to the invention therefore) and the dotted line curve corresponding to the state of the art.

[0044] The figures, in particular those which are not graphs, are very schematic, not to scale and concern the illustration of embodiments as non-limiting examples of the implementation of the invention. Each reference or acronym designates the same component, the same calculation from one figure to another and throughout this text. Description of the embodiments

[0045] The invention will be described below using embodiments and examples in a non-limiting manner and by way of illustration, with the aid of the aforementioned figures.

[0046] By definition, a hybrid vehicle has several energy sources. In this case, on a vehicle equipped with one or at least one fuel cell and one or at least one battery, a power request from the vehicle's engine can be made either by the or at least one of the batteries, denoted BAT in this text, or by the at least one of the fuel cells, denoted FC in this text, or by both the battery and the fuel cell.

[0047] For the sake of brevity, unless otherwise indicated, when the term battery, or, respectively, fuel cell, is mentioned, it is a question of the only battery, or respectively the only fuel cell equipping the vehicle, or of at least one of them (or all of them) when the vehicle has several (mounted in parallel).

[0048] The vehicle's energy management system ("Power Management System" in English with its acronym PMS), is the control unit making it possible to arbitrate between these different sources, with the aim of maximizing the energy efficiency of the vehicle. (Note that throughout this text and in a known manner, this system can also be designated, in an equivalent manner, by the English term Energy Power Management, or by its acronym EMS).

[0049] [Fig.l] thus schematically represents the general energy architecture of a vehicle equipped with an electric machine powered by electricity from a battery and two fuel cells: the energy management system EMS controls the operation of the battery BAT and the fuel cell system FC comprising here n stacks SI to Sn. All the auxiliary equipment Aux allowing the fuel cell to operate, this notably concerns at least one compressor and at least one radiator, these auxiliaries Aux require an electrical power P Aux to operate. The battery BAT can supply an electrical power PBat to the electric motor MEL of the vehicle, and the fuel cell system FC can supply it with an electrical power PFC-

[0050] The power required by the electric motor to complete a journey Preq is therefore the sum of these two powers Pbat.Pfc supplied, less the power Paux spent on auxiliary equipment.

[0051] The sum of the two powers Pbat.Pfc provided makes it possible to provide the power required by the auxiliaries Paux and the power Preq requested by the electrical machine MEL.

[0052] The role of the EMS management system is to determine the optimal Pbat.Pfc values ​​to provide the required power Preq while minimizing the fuel consumption (hydrogen) of the fuel cell and / or the aging of the cell and / or the battery over a given cycle (i.e. over a given journey of the vehicle).

[0053] Note that the term “fuel” within the meaning of the invention, as already seen, designates the fluid, in particular gas, for example hydrogen, or gasoline to be supplied to the gas-to-electricity conversion system. In the examples described below, the conversion system is a fuel cell whose fuel is therefore gaseous hydrogen.

[0054] The objective, for a given driving cycle, is to control the two energy sources (FC and Battery) so as to minimize a criterion: here, in a non-limiting embodiment of the invention, we choose to minimize the consumption of H2, while respecting the following constraints: - The power balance must be satisfied, that is to say that the power Preq(t} requested by the electric motor must be satisfied at each instant of the cycle. - The limits of the actuators must be respected in this embodiment SoCmin < SoC(t) < SoCmax (minimum / maximum limits of the battery SoC state of charge)

[0055] ipcmin < 1fcO < ^FCmax Minimum / maximum current from the FC stack)

[0056] v <j (t< v (limites minimale / maximale du gradient de courant ^FCmin FC\ ' F Cmax from the FC stack) - The expected final state of charge must be reached: for example, we can impose a state of charge SoCfin of the battery at the end of the journey / cycle which is equal to its initial state of charge SoCfin = SoCjnit in order to directly see the gain in hydrogen consumption without having to convert the energy of the battery used into hydrogen equivalent. But in another embodiment, we can impose a charge level at the end of the journey SoCfindetermined (and therefore different from the initial charge state).

[0057] This amounts to an optimal control problem whose mathematical formulation is (over a cycle of T seconds): Minimization of a criterion J [T / . / wi J(soc, ifc)=^soc(T)) +}0mH2(iFC(t))dt

[0058] With ¢( soc( T ) ) the constraint on the final state of charge of the battery, and (T , , , the hydrogen consumption.

[0059] With the following constraints: (Qbat being the battery capacity) > Initial condition and dynamics of the system 5oC( / 0) = SoCüüt SoC(r)- with Cat the battery current depending on the FC current iF( / t) and the power required by the electric motor Pre^t) > State / order constraints: SoC mifl SoC ( / ) SoCmax ipCmin < < hcmax yrr, . <ïFrÛ)< * rCnun 1 e\ J f bCmax

[0060]

[0061] > Global constraints: SoC(T) = SoCÏJq) We therefore seek to minimize the criterion J ( soc, iFF ) — J q mH2 ( ^FC^ SOc(t) =

[0062] If we rely, as in the present invention, on the Principle of Maximum Pontryaguin to solve this problem, then:

[0063] This amounts to minimizing the Hamiltonian H associated with the problem: H(soc(t),p(j), iFCU)) = (iFC(t)) - pit\ ■ p(t) = -* ■ 7 dsoc

[0064] With p(t) the adjoint state associated with the state variable soc(t).

[0065] The prior art using the PMP would consider the following hypothesis to solve the problem:

[0066] Assuming that — q, dsoc

[0067] on ap( / ) = p = este and therefore:

[0068] tji - (t\\_ ; / .y] „ . XxJtf clOAcAd) ) ~ lnHAlFCV)) ~ Pa ô ' ^'bot

[0069] After changing variables: P' PCIP bat — Pbat ' Cat and p-PCI

[0070] We find that the optimal control is obtained by minimizing at each instant a weighting function between thermal and electrical power:

[0071] i*FC = œg min Ph2(.îFc) +^Pbat{ Ce) (7¾^ Cmit^FCmu. J

[0072] This weighting function depends on a parameter Xo which is generally called the "equivalence factor" and which in fact corresponds to the adjoint state p0 associated with the state variable soc. The challenge is then to find the Xo (or the p0) which allows, for a given cycle, to respect the different constraints which we have set, in particular the constraints at both ends (Soc(t=0) = Soc_init and Soc(t=T) = Soc_init).

[0073] Let's take the "off line" case, when the cycle is known in advance: The simplest thing is then to determine 20 iteratively so as to respect the final state of charge constraint (SoC ( Tj = &?C(f0))

[0074] Let's take the "on line" case, when the cycle is unknown in advance: The equivalence factor X will be very dependent on the driving cycle as well as on random factors (traffic, driving behavior, etc.). Consequently, finding an adequate X is the main challenge in on line algorithms if we want to have performances close to the global optimum. Many methods exist, including: - methods based on X regulation: an initial X is calculated upstream. During the cycle, the X is then regulated by a PI controller to ensure that the SoC remains close to the final target SoC. - methods based on pattern recognition: upstream, a network of neurons are trained to recognize driving cycle patterns (highway, mountain, city, etc.) and associate the appropriate X with them. - methods based on load prediction: using navigation data, we can estimate the power required over a more or less long time horizon. We then perform an optimization calculation on this time window, which will give a reference SoC charge state trace to follow, and we adapt the X accordingly.

[0075] These methods, which are based on the principle of ECMS or PMP and aim to adapt the X in real time to take into account the hazards of the cycle, are often called A-ECMS or A-PMP (for “Adaptive” ECMS / PMP). Model Predictive Control (MPC) methods are part of the same family of methods based on load prediction and optimization. The general principle is to use past information, real-time vehicle data, and navigation data to deduce how to control energy sources.

[0076] Among all the MPC methods, some are based on the principle of PMP. Their principle is to predict the power over a given time horizon, and then to carry out an optimization calculation based on the PMP to output a corresponding adjoint state p. This adjoint state can be updated during the cycle if the road conditions, and therefore the power prediction, change.

[0077] This approach is interesting because, being based on the theory of optimal control, it guarantees approaching the global optimum provided that the hazards of the cycle are correctly predicted, and this, for a relatively low computing power required.

[0078] The present invention is part of this type of approach (MPC - PMP), with - Prediction of power demand from GPS data - Carrying out an optimal control calculation under constraint using the PMP. - Calculation of the adjoint state p, calculation of the Hamiltonian from p, minimization of the Hamiltonian. - Refresh during the cycle of the predicted power trace and the adjoint state P-

[0079] Several publications have already proposed “on line” methods of the MPC-PMP or A-ECMS type, including: Musardo C., Staccia B., Bittanti S., Guezennec Y., Guzzella L.,Rizzoni G. (2004a) An adaptive algorithm for hybrid electric vehicles energy management, FISITA 2004 World - Automotive Congress, Barcelona, Spain R. Schmid, J. Bürger and N. Bajcinca, "Efficient Optimal Control of Plug-in-Hybrid Electric Vehicles including explicit Engine on / off Decisions," 2018 European Control Conférence (ECC), Limassol, Cyprus, 2018, pp. 596-601, doi: 10.23919 / ECC.2018.8550516 : J. Buerger and M. Cannon, "Nonlinear MPC for supervisory control of hybrid electric vehicles," 2016 European Control Conférence (ECC), Aalborg, Denmark, 2016, pp. 135-140, doi: 10.1109 / ECC.2016.7810276: Simona Onori, Laura Tribioli, Adaptive Pontryagin’s Minimum Principle supervisory controller design for the plug-in hybrid GM Chevrolet Volt, Applied Energy, Volume 147, 2015, Pages 224-234, ISSN 0306-2619, https: / / doi.Org / 10.1016 / j.apenergy.2015.01.021.

[0080] These works all promise satisfactory results and not far from the global optimum (generally around +1% on the cycles considered).

[0081] A limitation that they all have in common, however, concerns the consideration of the state of charge constraints of the SoC battery: SoCmin < SoC(t) < SoCmax

[0082] Indeed, in all these works, the assumption taken is that the Hamiltonian: 100831 H(soc(t),

[0084] does not depend on the SoC state of charge

[0085] And therefore that: n = — q and therefore n( t ) = n — este ' dsoc t \ ' 0

[0086] Even in prior art algorithms that update the adjoint state p, it is considered constant between each update and these updates do not predict variations in p. Algorithms that regulate the adjoint state to follow a reference of the SoC state of charge are necessarily far from the optimal solution. Most of the time, this hypothesis is justified as noted in the aforementioned publication by Simona Onori et al., because although the battery current depends on battery parameters which themselves depend on the state of charge SoC, in reality this evolution over time is small.

[0087] However, when the minimum or maximum limits are reached, the hypothesis of the constant adjoint state is no longer valid: in fact, if the optimal control took into account the state constraints linked to the state of charge SoC, the adjoint state linked to said state of charge SoC would evolve during the cycle to avoid going beyond the limits.

[0088] In the prior art, the optimization carried out is therefore blind to the constraints on the SoC state of charge: the optimal control is calculated without "knowing" that the SoC state of charge must not exceed certain values. In order for the SoC state of charge to still remain within the admissible limits, a penalty function is generally applied to the Hamiltonian, but this can only lead to suboptimal results when the SoC state of charge limits are reached.

[0089] Some studies, such as the aforementioned publication J Buerger et al., consider that In hybrid vehicles with internal combustion engines (known as "HEVs"), the SoC limits are rarely reached because the battery is sized accordingly, the assumption of constant p is then not a problem. This argument may be acceptable most of the time, but may become false on certain mountain-type cycles with large variations in slopes or if we want to size the battery as precisely as possible.

[0090] More generally, taking into account state constraints related to the SoC state of charge in optimal control can allow for more accurate sizing of the vehicle architecture, for example with smaller batteries for equivalent applications. This is also useful in applications with supercapacitor / battery hybrid storage.

[0091] The invention therefore proposes an algorithm for an EMS energy management system of a hybrid vehicle in order to control the distribution of the power coming from the battery and that coming from the combustion cell to power the electric motor of a vehicle, said system having the following properties: - based on the theory of optimal control using the Pontryagin Maximum Principle PMP - that can be implemented online, which is robust to hazards (traffic conditions, route changes, etc.) with the sole need for data from a GPS-type geolocation system. - taking into account all the limitations of the system whether on the commands (for example: ipcmm- ^FCmax, ^FCmin, ^FCmax, etc.) or on the states (here SoC, fnifp SoC^ax) - not requiring any upstream development (other than providing the vehicle parameters required by the models used by the algorithm)

[0092] An embodiment of the algorithm according to the invention is described below at a point from a methodological point of view, with the description of the vehicle architecture model, the definition of the optimal control problem, and the precise description of the resolution algorithms:

[0093] - Before departure, we enter a destination in the GPS which gives the route to travel, speed limits and altitude - A model taking into account vehicle parameters makes it possible to predict, from navigation data (traffic, speed limits, slopes, etc.), the speed trace and therefore the power trace required by the electric motor at each moment - The PMS takes as input the measured SoC state of charge and the predicted power to perform two types of calculation: > a so-called “slow” loop calculation: in this loop, an “off line” calculation is carried out from the predicted power trace, taking the initial SoC state of charge as the the state of charge SoC measured at the time the calculation is launched. This calculation is based on the PMP and takes into account the constraints on the commands and on the states (here SoC state of charge). It allows to determine the optimal SoC trace and the corresponding adjoint states P[ at each position in the cycle. > a so-called "fast (real-time)" loop calculation: at each time step, the power actually requested may be different from the predicted power. This loop ensures that the power requirement is met, while minimizing the chosen criterion (here the H2 consumption). We therefore write the Hamiltonian at time t from the adjoint states Pi calculated by the slow loop: H(t, iFC, iBAr) = COnsoB^iFC) -p^t)*SoC(iB^ )We then solve a numerical optimization problem: We can look for the commands iFcl ^bat which minimize H( t, iFC, Cat) at time t while respecting the constraints: ' 5oCraill < SoC ( l ) < SoCmax The Ce! ^bat solutions will be the commands hcmin < S hcmax <VFO„„ ~^FC^ “0 effective sent to the vehicle.

[0094] The fast loop occurs at each time step.

[0095] The so-called slow loop can be recalculated during the cycle when the actual SoC state of charge trace moves away from the calculated reference, or more simply at regular time intervals (e.g. every 10 minutes).

[0096] This is what [Fig.2] represents, with some of the references of [Fig.2] already described above in this text. The other references correspond respectively to: - 1: journey = origin and destination - 2: traffic information - 3: calculation of the route to be taken - 4: speed limit data, altitude - 5: calculation of achievable speed - 6: optimal control (calculation of the co-states or adjoint states p mentioned above) - 7: calculation of the power distribution between the battery and the two fuel cells - Pm: measured power - Ppred: predicted power over the entire cycle (the journey) - VH: vehicle - Dr: actual distance - SoCm: measured state of charge of the battery

[0097] The fast loop is described in more detail below: This is a constrained numerical optimization problem, meaning that the solutions sought belong to a numerical space. One of the most effective methods for solving this type of problem is the so-called interior point method.

[0098] This method consists of adding logarithmic penalties to the initial cost to relax the inequality constraints. In other words, the original problem, which is to minimize the Hamiltonian under inequality constraints (here the inequality constraints are the constraints on the SoC, VFC and ipc), is rewritten as a minimization without inequality constraints of the Hamiltonian to which penalties corresponding to each constraint have been added. We therefore transform a problem under constraints that is difficult to solve into a problem without constraints that we know how to solve and whose solution is the same as the solution to the original problem.

[0099] The slow loop is described in more detail below: This time, the problem we want to solve does not correspond to a numerical optimization problem (solution belonging to a numerical space) but to a functional optimization problem (solution belonging to a space of functions). In other words, we seek to minimize a function of a function. Here it is the H2 consumption which is among other things a function of the SoC (itself being a function of Ibat}-

[0100] One of the reasons why state constraints were not taken into account in "online" algorithms until now is that this greatly complicates the equations to be solved and that there were no resolution algorithms (and whose convergence towards a solution was proven) efficient enough to allow real-time implementation.

[0101] Recently P.Malisani, in the publications P. Malisani, “Interior point methods in optimal control of affine Systems: Convergence results and solving algorithms,” to appear in SIAM Journal on Control and Optimization, 2023, and P. Malisani. Interior point methods in optimal control. ESAIM COCV (in review), 2023, demonstrated that it was possible to apply interior point methods for functional optimization problems and that they provided an exact solution. Based on this work, we can therefore write the system of differential and algebraic equations defining the necessary conditions of optimality (given by the PMP) into a new system without constraints according to the interior point method. We can therefore quickly solve this system using an interior point algorithm.

[0102] To make the calculation over the entire cycle even faster, we define 3 variables: - the “Start_time_undersampling”: time on the predicted cycle from which we undersample - the “Under_sampling_steps”: new sampling step -1' "Actualization_horizon", which allows you to set the time after which the calculation is restarted

[0103] For the calculated values ​​of the adjoint states to be close to the optimal, we need to know the trend of the requested power Preq over the entire cycle. Indeed, to be optimal, the control needs to know if, later in the cycle, a very high demand will occur (e.g., mountain climbing) or, conversely, if there will be a high potential for energy recovery (mountain descending). This is why the calculation of the slow loop is advantageously done over the entire remaining cycle.

[0104] However, as the calculation is restarted after a certain time horizon (in order to take into account random events and correct prediction errors), we only need to have an accurate prediction (and therefore with a high sampling rate), over a relatively short time horizon. The rest of the cycle can be sampled more coarsely, the main thing being to keep enough information on the average Preq. We therefore preferably take a much lower sampling rate beyond the "actualization_horizon".

[0105] This allows the problem to be solved in just a few tens of seconds. Example 1 Vehicle energy architecture

[0106] As mentioned previously, we consider a hybrid architecture equipped with two identical FC fuel cells and a BAT battery supplying a DC bus from which an inverter draws power to operate the MEL electric motor.

[0107] This example corresponds to [Fig.3], with the following additional references compared to the previous figures:

[0108] - PFC1: 1a power supplied by the first fuel cell FC1 - PFC2: 1a power supplied by the second fuel cell FC1 - Pauxl: the power required by the auxiliaries of the 1st fuel cell - Paux2: the power required by the auxiliaries of the 2nd fuel cell - Ubus: the voltage at the terminals of the DC bus to which the MEL electrical machine is connected

[0109] The most recent studies tend to show that if, on light vehicles, non-hybridized full electric remains more profitable, on heavier vehicles on the other hand this type of architecture would become much more interesting. We consider so here for example a heavy goods vehicle of 44 tonnes.

[0110] This choice is all the more interesting for the implementation of the invention since this type of vehicle can require very high power and performs long but relatively predictable cycles (obligation to respect speed limits, GPS route necessarily known). [YES] Battery specifications: Cell capacity: ~ 113520 As Min current: Ibat»™ — - 378 A (charge rate up to 3 times) Max current: here assumed to be infinite for simplicity but a max constraint could be taken into account without problem No DC / DC bus voltage Uhus = Uhat (hereinafter we will denote U as the DC bus voltage) Nominal battery / bus voltage: 780 V

[0112] Characteristics of FC modules: Number of Sl-SN stacks: 4 Number of cells per stack = 200 Min current: ipcmin — OA Max current: ipcmax ~ 000 Min current gradient: . = - 120 A / s Max current gradient: yr„ — 120 A / s

[0113] In order to operate, the fuel cell FC consumes a power P Aux-The fuel cell modules

[0114] As shown in [Fig.4], which is a part of the architecture shown in the previous figure, each FC module is composed of 4 stacks with auxiliary equipment Aux including a compressor. The cooling system is common to both FC modules. Each FC module is connected to the bus by a DC / DC converter.

[0115] - Calculation of the DC / DC current on the bus side Îdcdc'- The stack is defined by its polarization curve which gives the cell voltage VCeii as a function of the current density j: = / Ü)

[0116] From this polarization curve, and the DC / DC model (which takes into account conduction and switching losses), we define an analytical relationship between the current drawn on the FC and the current supplied on the DC bus: ^DCDC — a ï + a2 + a3 + a4 *u2 + a5 *u + a6

[0117] with (ab a2, a3, a4, a5, a6 ) constants

[0118] - Calculation of auxiliary power P Aux'- Phest-^ + £1 * The total auxiliary power (of the two fuel cells FC) is composed of the compressor power P plot for FCi and FC2 and the cooling system power for the two cells PCoolTot- Again an analytical expression gives the relation PAux = f(iFC): PgampTotf^ = * UFCl + * OfCI + te) + c2 SetToTry = le3 « ( / ¾ * + ^2) + ^2 * (te + te) + 2 * / ¾)

[0119] With (Pc? Pc* pCv q5 p.^ p^, pf^ SetNumber) constants

[0120] We finally obtain: PAuxkJjFCl' ipez) “ PcompTot^FCb hc?) + PmolTot^FCb hcè

[0121] With P heat the power dissipated in the form of heat by the batteries

[0122] Calculation of H2 consumption: The total consumption (of both batteries) in H2 is written:

[0123] (iFCÏ + iFC2) *Constcmso

[0124] With Constcons„ a constant depending on the stacking parameters (“stack”) and H2 stoichiometry The battery

[0125] As shown in [Fig.5], which is a part of the architecture shown in [Fig.2], the system uses a BAT battery.

[0126] - Calculation of battery current:

[0127] The battery current Îbat^expressed as a function of fcf fa -, ?req and U via the power balance: Preq = Ppc\ + PpC2 + P HA! " P Aux

[0128] In addition, a power Pf is introduced into this equation, which is called braking power. Indeed, it may happen that the power balance cannot be resolved only with fc and Îbat to be controlled. For example, if during a regenerative braking phase we have Preq < Pbatmin then the battery will only be able to absorb part of the energy. In reality, the rest of the energy would be dissipated during braking. To enable the control to find a solution in these cases, we add Pf as a command with the constraint: Pf <0.

[0129] We therefore finally have: Preq = P Fa + PF C2 + P BAT + Pf " P Aux

[0130] let Preq = ( iDCDCi + ÎdcDCZ + ^BAT ) + Pf “ P Aux

[0131] And so / ■ . TT nn \_^Te(l+PAii.kFCViFC^-Pi} . / . T j\ ' {' TJ lBAl\lFC^ lFC2' L' “fi ' req)— y ' lDCDCA\lFCY U ) ~ lDCDC^FC2> U

[0132] - Calculation of battery voltage:

[0133] When current is drawn from the battery, its voltage and therefore the DC bus voltage will be impacted. This is governed by the following differential equation:

[0134] 7\_ i *. / . ■ / appxu(t)-ocv MO- _ Cr lBAT\iFCt> lFC2? “VA “.fi “req) ~ R”cf

[0135] With Cf (filtering capacitance), OCV (Open Circuit voltage) and Rohm (Internal Resistance) as constants.

[0136] Definition of the system of equations to be solved: slow loop

[0137] The problem to be solved is as follows:

[0138] We consider x the vector of state variables: x = [soc iFC1 iFC? and P = [Pj P2P3P4 ] the vector of associated adjoint states.

[0139] And c the vector of control variables: yy Pf^

[0140] We note Preg the vector of the powers requested at each time step over the entire along the predicted cycle.

[0141]

[0142]

[0143]

[0144]

[0145] The state variables x are subject to the following differential equations: ( ( ipC'Afmi ) SOC{t^---- ZFC1(Z) -EFC1 zFC2(0 “ )'fc2 u(t) - lBAT\lFCV lFC2' Pf' ^req) - State variables are subject to the constraints: 'SOC(t) -SOCP SOCm-soc(t) <0 hm'^FClU) -0 ' hm~ - 0 The mixed state / command constraints are:

[0146]

[0147]

[0148]

[0149]

[0150]

[0151]

[0152]

[0153]

[0154]

[0155]

[0156]

[0157]

[0158]

[0159]

[0160]

[0161] ^FcS^ ~^2p “ -y (f)^0​ 2m r G 2X 7 pf ^BATmin " BAl{^FC^ Gc2> W' Preq) 0 2 i 2 t 2-. 2--. 2i 2 t 2 i 2 o 2 / 2^ are the variables y^ocp? ^socrm Æ?lp 'Mnr zt#lp *2 / ^ / vgbm ^^2^ 7itJ duals associated with each constraint. The Hamiltonian of the problem is written: H = consumption (iFCl, iFC2) +p, *sôc +t FC1 The PMP and the interior point method give us the Necessary Conditions of Optimality through the following system of equations to be solved: - ODE (for “Ordinary Differential Equation” in English or ordinary differential equation in French) FCÏ '^DCDCïGfC]? U) '^DCDCiGfCQ? M) ) lFC2 FC2 ; / P Aux ( ^FC^Fd ) +P- ■ P' ■socm - dconso(iF(2j.FC2) will say damso( iFFi said, I dim i 4 diFCl „ A dsdc „ du ■Pl TSF -Pi FF. ​ufi()soc FC^Ft — -P4 ^ +Ài - Equations algébriques diluA. iFC AFC2’U’ Pr^Pj ) dPf ■soc p '.wcm du 0 = 7ib, u(t)-OCV Pobn , 5 ! 1 ^BATX^FC^FC^ ‘ / ! + Alm+A’ --------------- j i ,i ^‘BAt(Pc]4fC2’'^ PmfPf) ’i2m _ ) O — n 4- n *-^4. g2p AgZmU — +p4 dp "soc p ? (O-hp) +2f0 = à. 2 / A socm + y SOCm

[0162]

[0163]

[0164]

[0165]

[0166]

[0167]

[0168]

[0169]

[0170]

[0171]

[0172]

[0173]

[0174]

[0175]

[0176]

[0177]

[0178] 0 = - (hatt) ' hp) - + ( Pc2(0 - 2 + 2s 0 = Ài2m- ( i2m-iFC2(» ) - fe.+OVWO )2 + 2s 0 = 41?-(ylm -yFC}^) I. ~2 0 “ " ( y P C2 ~^2p ) ' (y FC2^^ 0 = -U, - ( y^ - yFC1 ( t ) ) - ^2," + ( y,a - y ( t ) ) + 2s 0 = 2 / -(Pz)-^ + (P / )2+2s 0 = 2; - {J bai: mm ■ ^â / COci' 9c2' PF Pre^} - + ( IBATltlin ' ^BAl{h'CV Pc?? M' Pf' Preg) ) + 2t' - Conditions aux deux bouts 0 = SOC ( tQ ) - socsM (socxtrt = 0.5) 0 = ÎfC 1 ( ^>) " ÎFChfrt(ÎFClsM ~ 0 A) 0 - ÎfC2 ( ^0 ) " iFC2strt Q = u(t0)-uxtrî O = soc(tf)-socend ® = P2M 0 = P3M V = PÀ!j) With re tf] (ÎFC2M11 - 0 A) (^=77()19 = 0.5) The "solver" (English term which can be translated as resolver) uses the systems of equations defined previously to find the solution vectors x* (state vector), p* (adjoint state vector) and z* (constrained command vector) by interior point algorithm and damped Newton algorithm. Definition of the system of equations to be solved: the fast loop At each instant t, we want to find, for the real requested power Preq(t) and the real voltage li(t - 1), the vector x = [(g 47-( / ) ^fc~^) ^X^)] cIu' m'nim'se la H2 consumption, therefore according to the PMP, which minimizes the Hamiltonian of the problem. We recover the optimal vector p p* _ p* p* p*^ calculated by the slow loop. We can then write at each time step the Hamiltonian to be minimized: H^ = conso(iFCI, lFa)-p'^ ) As we are at a given time t iFCs(t) = y = 0 and iFC2(t) = VFC? = 0' it so it is enough to minimize: H\t) - consumption (lFC}, lFC2) - Pi -p4 — lBAT +)

[0179] This while respecting the constraints:

[0180] Equality:

[0181] »( t) *'(iBAT+ÎDCDCl( ipCb «(<-!))+ ÎdcDcA - 1) ) ) - P (t) ~ PauxUfCI' +Pf = 0

[0182] Inequality: 9 — ^BATmin ~ ^BAT 9 IPCI {) * S.... ~ 0 — lFC2 — hp 0 — ^FC2 0 > Pf «. 7 0:> SOC--* at — SOCy >]bat i) • • ( SOC - ' (f: 1 \ Qbat / 0 ■ iFCi(t) - - 1) - Zip 0 > y1JM - (iFCi(t) - iFcitt ~ 1)) 0 — lFC2 (0 — iFcztt — 1) — Yîp 0 — 72m ~ (^FC2^) ÎFC2(t 0)

[0183] By an interior point algorithm and a damped Newton, we find at each instant t the solution x — Ifca « Îfc~> t Pf ] ct 'cs commands ipc J ^fcz to send to the two stacks.

[0184] We consider two cycles for a Lyon-Bordeaux journey: - a “predicted” cycle: LB_TrajPred - a “real” cycle: LB_TrajReel

[0185] Both cycles were generated by the power trace generation tool from the GPS trace. In order to simulate a real and predicted cycle, two cycles were simulated with the same route, but different traffic times: the aim is to simulate traffic hazards.

[0186] We therefore consider: - the “trajPred” cycle as the “predicted” cycle and used by the slow loop. - the “trajReel” cycle as the “real” cycle used by the real-time loop.

[0187] [Fig.6a] is a plot of the route taken by “TrajPred” and “TrajReel” using a GPS-type navigation system (with longitude on the abscissa / latitude on the ordinate). [Fig.6b] (with the distance on the abscissa and the power required by the electric machine in kW on the ordinate) is a corresponding power trace generated from the plot of the two paths according to [Fig.6a].

[0188] Simulation assumptions: 2 fuel cells FC1 and FC2 with the characteristics indicated below (same characteristics for both cells: 4 stacks for each fuel cell ilm = i2m = 0 A (minimum battery current) ilp = i2p = 600 A (maximum battery current) gamma 1m = gamma 2m = - 120 A / S gamma Ip = gamma 2p = 120 A / S

[0189] “Start_Time_Undersampling” = 500s (time from which the predicted cycle is undersampled) “pUnder_sampling_steps” = 70s (the undersampling step) “Actualization_horizon” = 500s Battery charge rate BAT = 3 (which means that here the battery can deliver a maximum power equal to three times the power it could deliver for 1 hour if its charge was fully charged). The battery is limited by the minimum and maximum current intensities.

[0190] In order to calculate the global optimum and have a trace of the SoC state of charge and a reference H2 consumption, we carry out an “off line” calculation on the “real” trace (trajReel):

[0191] [Fig.7a] is a plot of the calculated “off line” SoC state of charge of the vehicle battery on the “real” journey trace (trajReel) (with the distance on the abscissa and the SoC state of charge on the ordinate). [Fig.7b] is a plot of the hydrogen consumption calculated “off line” by the vehicle’s fuel cell on the “real” travel trace (trajReel) (with the distance on the abscissa and the hydrogen consumption in kg on the ordinate).

[0192] A reference consumption of 29.27 kg of H2 is obtained.

[0193] We perform the calculation “on line”, with the slow loop which predicts the cycle (trajPred) and the real-time loop that follows the cycle (trajReel):

[0194] [Fig.8a] represents the traces of the optimal battery state of charge SoC predicted at several iterations of the slow loop (for the sake of readability, only two iterations are represented): each time the calculation starts from the current SoC value to recalculate a trace which satisfies the final constraint while minimizing the criterion. [Fig.8b] represents on the one hand the sum of the predicted SoC charge state trace segments between each update of the slow loop and on the other hand the SoC charge state trace actually produced.

[0195] The actual consumption is measured and compared to the reference as well as to an algorithm which does not take into account the constraints on the SoC state of charge and which uses a penalty function instead: [Fig.9a] represents the trace of the optimal state of charge SoC of the vehicle battery over the journey (with the distance in km on the abscissa). This is mathematically the global optimum which was calculated off line. The consumption is 29.27 kg of hydrogen and serves as a reference to judge the efficiency of the online algorithms. [Fig.9b] represents the trace of the actual SoC state of charge of the BAT battery during the journey (therefore online): it is 29.3 kg of hydrogen (i.e. + 0.1% compared to the reference consumption). [Fig.9c] represents the trace of the optimal state of charge SoC calculated over the cycle, with an algorithm that does not take into account the state of charge limits in its calculations and uses a Hamiltonian penalty instead: it is 29.65 kg of hydrogen (+ 1.3% compared to the reference consumption). We can thus see that the solution according to the invention ([Fig.9b]) is very close to the global optimum ([Fig.9a]) and produces better results than the solution usually used with penalization of the Hamiltonian ([Fig.9c]). Example 2

[0196] Taking into account the constraints linked to the SoC in the calculation of the optimal order makes it possible to better exploit the battery capacity and therefore potentially to optimize its dimensioning: We replay the Lyon-Bordeaux cycle in “off line” with or without taking into account the SoC limitations by taking increasingly smaller battery capacities.

[0197] We start from the nominal case where the battery cell capacity is 113,520 As. We then perform two offline calculations on the cycle considered: one with the algorithm of the present invention which takes into account the state constraints linked to the battery SoC, the other with an algorithm representative of the prior art, which does not take into account these constraints on the SoC in the optimal control and which instead uses a penalization of the Hamiltonian to avoid the limits on the SoC being exceeded. We note the final hydrogen consumption in both cases. This operation is then repeated on the same cycle but with an increasingly reduced battery capacity. Reducing the battery capacity necessarily leads to an increase in consumption on the same cycle and therefore overconsumption compared to the nominal case (battery cell capacity at 113,520 As).

[0198] In [Fig. 10], we plot for each percentage reduction in battery capacity (on the abscissa), the percentage of overconsumption of hydrogen (on the ordinate) compared to the nominal case: - The solid line curve is obtained with the algorithm of the present invention, which takes into account the state constraints on the SoC. - The dotted curve is obtained with a representative algorithm of the prior art which is also based on the PMP but without taking into account the state constraints linked to the SoC.

[0199] As the battery capacity decreases, the curve without taking into account the SoC limits (dotted lines) deviates more and more from the curve produced with the algorithm according to the invention (in solid line). This is due to the fact that the limits on the SoC states of charge are reached more and more frequently in the cycle. By taking them into account, the algorithm according to the invention manages to better manage the overconsumption due to the reduction in capacity. In this specific case study, we see that the consumption of the algorithm representative of the prior art with nominal battery can be obtained with a battery capacity reduced by 25% if we use the algorithm of the present invention.

[0200] It can be concluded that the algorithm according to the invention can also make it possible to dimension the energy architecture of the vehicle more efficiently: the invention offers greater room for maneuver to reduce in particular the size of the electricity storage system (battery).

[0201] Furthermore, the invention makes it possible to avoid reaching points where the battery can no longer supply current and where the entire power must be supplied by the fuel cells. It is therefore no longer necessary to have fuel cells sized to supply the entire power on their own: the power of the cells can also be adapted / adjusted downwards as needed thanks to the invention.

Claims

Claims

1. Method for energy management of a hybrid vehicle equipped with a navigation system and an electric powertrain comprising at least one electric machine (MEL) which is supplied with electricity by electrical energy sources (BAT, FC) comprising at least one fuel-to-electricity conversion system, in particular of the fuel cell (FC) or turbogenerator or heat engine type associated with an alternator, and at least one electrical energy storage system, in particular of the battery (BAT) and / or supercapacitor type, characterized in that said method comprises: (a) a step of providing information about a destination in a navigation system which retrieves and uses satellite navigation data in real time, (b) a step of calculation by the navigation system of the route to be covered for the journey from navigation data, c) a step of applying a predictive model of the vehicle to said journey, from at least some of said navigation data and from vehicle parameters, to predict the vehicle speed trace (Vvh) on the journey and to deduce therefrom the predicted required power trace (Preq) for the electric machine (MEL) during said journey, d) a step of managing the electricity supply to the electric machine (MEL) by the electric energy sources, which comprises dO) a sub-step of measuring the state of charge (SoC) of the electric energy storage system(s) (BAT), dl) a sub-step of calculating optimal control of the electric energy sources (BAT,FC) in particular in a loop, of the slow loop type, from the predicted power trace (Preq) in step c), said calculation being based on the Pontryagin Maximum Principle (PMP) and taking into account - the state of charge (SoC) of the electrical energy storage system(s) (BAT) measured in step dO) as the initial state of charge, - and the constraints on the limitations of the controls of the energy sources (BAT, FC) and on the limitations of the state of charge (SoC) of the electrical energy storage system(s) (BAT), to determine at each position of the vehicle on the path predicted in step b) the optimal trace of the state of charge (SoC) and the optimal adjoint states (Pi), d2) a sub-step of determining in real time the control of the share of electricity coming from the electrical energy storage system(s) (BAT) and the share of electricity coming from the fuel conversion system(s) (FC) to power the electric machine (MEL) during said journey, by minimizing the Hamiltonian of the optimal control problem which is defined by the Pontryagin Maximum Principle (PMP) and which depends on the optimal adjoint states (p;) calculated in step dl).

2. Method according to the preceding claim, characterized in that steps c) d0) and dl) are repeated to determine the control of sub-step d2) according to a given time step At, constant or variable, depending on the duration of the journey.

3. Method according to claim 1, characterized in that steps c), d0) and dl) are repeated to determine the control of sub-step d2) when a given difference ASoC is reached between the measured state of charge (SoC) and the state of charge predicted in sub-step dl).

4. Method according to one of the preceding claims, characterized in that the navigation data comprises traffic information, in particular the traffic situation, speed limits and gradients of the possible traffic lanes for carrying out the journey.

5. Method according to one of the preceding claims, characterized in that, in step d), the fuel consumption of the fuel-to-electricity conversion system(s) (FC) is minimized.

6. Method according to one of the preceding claims, characterized in that, in step d), a state of charge (SoC) of the or at least one of the electrical energy storage system(s) (BAT) is imposed between a minimum state of charge (SoCmin) and a maximum state of charge (SoCmax).

7. Method according to one of the preceding claims, characterized in that, in step d), a state of charge (SoC) of the or at least one of the electrical energy storage system(s) (BAT) is imposed at the end of the journey equal to said state of charge at the start of the journey.

8. Method according to one of claims 1 to 6, characterized in that, in step d), a given state of charge (SoCfin) of the or at least one of the electrical energy storage system(s) (BAT) is imposed at the end of the journey.

9. Method according to one of the preceding claims, characterized in that, in step d), an IFC current generated by the system is imposed conversion of gas into electricity (FC) between a minimum current IFCmin and a maximum current IFC max.

10. Method according to one of the preceding claims, characterized in that, in step d), a current IB AT generated by the electricity storage system (BAT) is imposed between a minimum current IBATmin and a maximum current IBATmax.

11. Method according to one of the preceding claims, characterized in that, in step d), a gamma current gradient FC generated by the fuel-to-electricity conversion system (FC) is imposed, between a minimum gradient gamma FCmin and a maximum gradient gamma FCmax.

12. Method according to one of the preceding claims, characterized in that the vehicle parameters used in step c) of applying a predictive model comprise at least the architecture of the vehicle's powertrain, and the characteristics of the electric machine (MEL).

13. Method according to one of the preceding claims, characterized in that, in sub-step dl) and / or in sub-step d2), a so-called interior point method (MPI) is used, or any primal-dual method, in particular to manage the constraints of maximum voltage state of the electrical network Umax and the state of charge (SoC) of the electricity storage system(s) (BAT).

14. Method according to one of the preceding claims, characterized in that, in step dl) of slow loop calculation, the calculation is carried out over the entire remaining path.

15. Computer, or calculator configured to implement the method according to one of the preceding claims.

16. Vehicle of hybrid type and equipped with an on-board navigation system and an electric powertrain comprising at least one electric machine (MEL) supplied with electricity by electrical energy sources (BAT, FC) comprising at least one fuel-to-electricity conversion system, in particular of the fuel cell (FC) type and at least one electrical energy storage system, in particular of the battery (BAT) and / or supercapacitor type, characterized in that it comprises means for controlling the electrical systems in electrical power supply of the electric machine (MEL) controlled according to the method according to one of claims 1 to 14 or with the computer, the calculator of claim 15.

17. Vehicle according to the preceding claim, characterized in that the navigation system is integrated into the vehicle, and in particular has a screen integrated into the dashboard, or in that said navigation system is autonomous, said navigation system being connected to the computer, or to the calculator implementing the method according to one of claims 1 to 14.

18. Computer program product downloadable from a communication network and / or recorded on a medium readable by a computer or a calculator and / or executable by a processor, comprising program code instructions for implementing the method according to one of claims 1 to 14, when said program is executed on a computer or a calculator.

19. A computer-readable storage medium, or computer, storing instructions, which, when executed by a computer or computer, involve the computer or computer implementing the method according to one of claims 1 to 14.

Images