WO2021214610A1

WO2021214610A1 - Method for controlling a powertrain system of a hybrid electric vehicle and related hybrid electric vehicle

Info

Publication number: WO2021214610A1
Application number: PCT/IB2021/053149
Authority: WO
Inventors: Pier Giuseppe ANSELMA; Claudio MAINO; Alessia MUSA; Ezio Spessa; Giovanni BELINGARDI; Daniela Anna MISUL
Original assignee: Politecnico Di Torino
Priority date: 2020-04-22
Filing date: 2021-04-16
Publication date: 2021-10-28

Abstract

The present invention relates to a method for controlling a powertrain system of a hybrid electric vehicle (100) by means of an onboard-control -unit (200) of the hybrid electric vehicle (100), said method comprising: - a recognition phase, wherein the onboard-control-unit (200) identifies by means of interface means (220) a driver of the hybrid electric vehicle (100) and retrieves by means of the interface means (220) optimized driving data specifically related to said driver from an external device (190); - an input phase, wherein said driver inputs navigation information of at least one journey to the onboard-control -unit (200) by means of input means (210); - a control phase, wherein the onboard-control-unit (200) collects navigation data (ND) from sensor means of the hybrid electric vehicle (100), wherein the onboard-control -unit (200) generates control data for controlling said powertrain system, and wherein the control data are based on said navigation data (ND) and said optimized driving data.

Description

METHOD FOR CONTROLLING A POWERTRAIN SYSTEM OF A HYBRID ELECTRIC VEHICLE AND RELATED HYBRID ELECTRIC VEHICLE

DESCRIPTION

The present invention relates to a method for controlling a powertrain system of a hybrid electric vehicle, according to the preamble of clam 1. In particular, the present invention describes a hybrid electric vehicle comprising a powertrain system adapted to be controlled by an onboard-control-unit of the hybrid electric vehicle, in accordance to said method thereof. The present invention can be implemented for managing the energy of a powertrain system of hybrid electric vehicles (HEVs) with the capability of maximizing the fuel economy performance.

The knowledge of the personal driving style of a driver is fundamental in order to better minimize the overall fuel consumption, and so to better minimize tailpipe emissions, while complying with the energy constraints imposed by the battery of a HEV. Most of the driving style tailored energy management systems for HEVs, known in the art, require an on-board driving system that operates in real-time while driving. A heuristic driving- style-recognition algorithm is implemented in the onboard-control-unit of a HEV that identifies the driving style of a driver according to certain predefined input data. Therefore, the heuristic driving-style-recognition algorithm would considerably increase the complexity of the onboard-control-unit of the HEV, thus remarkably increasing the cost associated to this type of technical solutions. Furthermore, the heuristic driving-style- recognition algorithm must operate efficiently in real-time, such that to select the correct driving behaviour shortly after the start of the journey, therefore an extra on-board system may be required to involve complexities both in terms of maintenance and potential failure.

In order to overcome these limitations some solutions known in the art replace said heuristic driving-style-recognition algorithm with an artificial intelligent (AI) agent implemented by using neural networks. As example, in the United States patent US7954579 is described a control strategy for a hybrid electric vehicle having an electric motor (MG), a battery and an internal combustion engine (ICE). The control strategy is implemented over a varying set of driving conditions through an adaptive control unit which employs an artificial neural network. The artificial neural network is trained on a pre-processed training set based on the highest fuel economies of multiple control strategies and multiple driving profiles. More precisely, one specific vehicle is chosen and programmed to operate under different control strategies. For each control strategy, the specific vehicle is operated over city, highway and other city-highway combined drive cycles. For each of these drive cycles, the particular control strategy that resulted in the highest fuel economy is chosen and its operating points are taken as part of the training set.

The technical solution described in the above-mentioned US patent has the following drawbacks.

A first drawback of such method is that the artificial neural network is trained considering one specific vehicle which operates under multiple control strategies related to a small set of driving profiles such as city driving profiles, highway driving profiles or their combinations. Therefore, the cited solution does not take into account the complexity related to a personal journey of a specific driver.

Another drawback of such method is that it only considers a small set of generic driving styles such as aggressive, passive, speeder and/or slowpoke. Therefore, the cited solution does not take into account the complexity related to a personal driving style of a specific driver.

A further drawback of such method is that it can consider only a small set of testing HEVs without taking into account the technical features, such as the battery efficiency, that might change during the operating time of the HEV owned by each specific driver.

The present invention aims at solving these and other problems by providing an improved method for controlling a powertrain system of a hybrid electric vehicle such that to achieve an optimal HEV powertrain control policy to a wide range of different driving scenarios and different personal driving styles. The present invention can be easily implementable in an onboard-control -unit of a hybrid electric vehicle for selecting, in real-time, control actions related to the powertrain, such as for example the repartition of power between the internal combustion engine and the electrical part, related to the battery and the electric motors, of the HEV.

The invention will be described in detail hereinafter through non-limiting embodiments with reference to the attached figures, in which:

- Figure 1 schematically represents a diagram illustrating a hybrid electric vehicle, according to an embodiment of the present invention;

- Figure 2 schematically represents a block diagram illustrating an onboard-control -unit shown in Figure 1, according to an embodiment of the present invention;

- Figure 3 shows a flow chart exemplifying a method for controlling a powertrain system of a hybrid electric vehicle shown in Figure 1, according to an embodiment of the present invention;

- Figure 4 schematically represents a neural network model involved in the method shown in Figure 3, according to an embodiment of the present invention;

- Figure 5 exemplifies a training procedure of the neural network model shown in Figure 4, according to an embodiment of the present invention;

- Figure 6 summarizes some results provided by the Applicant according to an embodiment of the present invention.

In this description, any reference to “ an embodiment ’ will indicate that a particular configuration, structure or feature described in regard to the implementation of the invention is comprised in at least one embodiment. Therefore, the phrase “ in an embodiment ’ and other similar phrases, which may be present in different parts of this description, will not necessarily be all related to the same embodiment. Furthermore, any particular configuration, structure or feature may be combined in one or more embodiments in any way deemed appropriate. The references below are therefore used only for simplicity’s sake, and do not limit the protection scope or extension of the various embodiments.

With reference to Figure 1, a hybrid electric vehicle 100 comprises a powertrain system adapted to be controlled by an onboard-control-unit 200 of said hybrid electric vehicle 100. The powertrain system comprises a fuel reservoir 110, an internal combustion engine 120, clutch means 130, an electric motor 140, a battery pack 150, a gearbox 160 and transmission means 170 which are operatively connected.

The fuel reservoir 110, such as a tank, is adapted to contain a fuel, such as gasoline, which is delivered to the internal combustion engine 120 for example by means of a pipe. The electric motor 140 is electrically connected to the battery pack 150 which can comprise one or more electric batteries or electric accumulators, such as for example a lithium-ion battery. In an embodiment of the present invention, the hybrid electric vehicle 100 can comprise a rectifier circuit that converts alternating current (AC), for example from an electric power provider, to a direct current (DC) in order to charge the battery pack 150. The clutch means 130, the gearbox 160 and transmission means 170 are operationally connected to at least one wheel 171 of said vehicle 100, and are adapted to transmit mechanical power from the internal combustion engine 120 and/or the electric motor 140 to at least one wheel 171. In an embodiment of the present invention, during a braking phase of said vehicle 100, the electric motor 140 can operate as electric generator providing electric energy to the battery pack 150.

The hybrid electric vehicle 100 comprises sensor means adapted to measure navigation data ND during a journey. The sensor means can comprise for example at least one of the following devices: a speedometer, an accelerometer, a gyroscope, an altimeter, a location system such as GPS. The sensor means can also comprise transductor devices of said powertrain system in order to obtain navigation data ND from said fuel reservoir 110, internal combustion engine 120, clutch means 130, electric motor 140, battery pack 150, gearbox 160 and transmission means 170. The navigation data D can comprise temporal sequences including values related to at least one of the following parameters: a speed, an acceleration, a status-flag, a battery-current, a battery-voltage, a fuel level, a position, a travel -time interval, a gear-number GN of said hybrid electric vehicle 100. The navigation data ND from the sensor means can be collected by the onboard-control-unit 200 as example by means of a CANBUS network deployed among the sensor means and the onboard-control -unit 200.

Figure 2 illustrates a block diagram exemplifying the onboard-control-unit 200 adapted to perform the method according to the present invention, which will be described in detail with reference to Figure 3. The onboard-control -unit 200 can comprise input means 210, interface means 220, output means 230, sensor-interface means 240, memory means 250 and processing means 260, which can be operatively connected as example through a communication bus 201 which allows the exchange information among said input means 210, interface means 220, output means 230, memory means 240 and processing means 250. Alternatively, the input means 210, interface means 220, output means 230, sensor- interface means 240, memory means 250 and processing means 260 can be operatively connected by means of a star architecture, without said communication bus 201.

The input means 210 are adapted to read input information, such as data and/or instructions, from a driver. Said input information can comprise navigation information of at least one journey that the driver intends to perform. Said navigation information can comprise as example the location information such as geographic coordinates of a destination point where the driver intends to go, the location information of a starting point of the journey, the direction information to reach the destination point from the starting point. The navigation information can be provided by a navigation system of the hybrid electric vehicle 100 which can operate with a location system such as for example the GPS system. Such input means 210 can comprise for example a keyboard, a touchscreen, a memory device, an interface device which can operate according to the USB, Bluetooth, Firewire, SATA, SCSI standards and the like.

The interface means 220 allows to identify the driver of the hybrid electric vehicle 100 and allows to retrieve optimized driving data specifically related to said driver from an external device 190. The optimized driving data comprise information according to the method of the present invention which will be described in detail with reference to Figure 3. The interface means 220 can comprise at least one of the following interfaces: an USB interface, a NFC interface, a Bluetooth interface, a Wi-Fi interface, a mobile-network interface, such as for example a GSM, a UMTS, a LTE and a 5G mobile-network interfaces. The external device 190 is adapted to be a USB memory device or a smartphone or a remote server.

As example, in an embodiment of the present invention, the external device 190 can be an USB memory device which stores said optimized driving data. When the driver connects the USB memory device to said interface means 220, comprising an USB interface, the onboard-control-unit 200 can identify the driver for example by an ID-code stored in the USB memory and then the onboard-control-unit 200 can retrieve the optimized driving data from the USB memory.

In another embodiment of the present invention, the external device 190 can be a smartphone, as example owned by the driver, which stores said optimized driving data. When the driver connects the smartphone to said interface means 220, comprising a Bluetooth interface, the onboard-control-unit 200 can identify the driver, for example by using the Bluetooth pairing procedure with the smartphone, then the onboard-control-unit 200 can retrieve the optimized driving data from the smartphone.

In a further embodiment of the present invention, the external device 190 can be a remote server which stores said optimized driving data. In this case, the interface means 220 can comprise a mobile-network interface such as an LTE interface. When the driver inputs his credentials into the onboard-control -unit 200, by means of said input means 210, the onboard-control-unit 200 can identify the driver in cooperation with the remote server, then the onboard-control-unit 200 would lead to retrieve the optimized driving data from the remote server. In this case, the optimized driving data can be translated into a suitable coding language before being loaded into the onboard-control-unit 200.

Further embodiments of the present invention can be realized considering as example an NFC interface instead of the above-mentioned Bluetooth interface or as combinations of the embodiments described above. The output means 230 are adapted to provide output information, such as processed data, to said user and/or to said powertrain system. Said processed data can comprise control data for controlling the powertrain system according to the method of the present invention. In a preferred embodiment of the present invention, the onboard-control-unit 200 is adapted to generate the control data for controlling said powertrain system. As example, the control data can comprise a power-flow parameter PF, i.e. a fraction of power partitioned between the internal combustion engine 120 and the electric motor 140, of the hybrid electric vehicle 100 and/or can comprise a gear-number GN, as example from 1 to 6, of the gearbox 160 of said vehicle 100. The control data can be generated in real-time in order to effectively control the powertrain system of the hybrid electric vehicle 100. As example, the following table shows the power-flow parameter PF related to a set of operating modes of said vehicle 100.

Such output means 230 can comprise for example a screen, a touchscreen, a memory device and an interface according to the CANBUS, USB, Wi-Fi, Bluetooth, Firewire standards.

The sensor-interface means 240 are adapted to receive the navigation data ND from said sensor means. As mentioned above, the navigation data ND can comprise temporal sequences including values related to at least one of the following parameters: a speed, an acceleration, a status-flag, a battery-current, a battery -voltage, a fuel level, a position, a travel -time interval, a gear-number GN of said hybrid electric vehicle 100. Such sensor- interface means 240 can comprise for example an interface according to the CANBUS, USB, Wi-Fi standards.

The memory means 250 are adapted to store information and the set of instructions for carrying out the method according to an embodiment of the present invention. Said method will be described in detail with reference to Figure 3. The stored information can comprise the navigation information, the navigation data ND, the optimized driving data and the output information of the method according to an embodiment the present invention. Such memory means 250 can comprise for example volatile and/or non-volatile memory units based on semiconductor-electronic and/or opto-electronic and/or magnetic technologies.

The processing means 260 are adapted to process the data and to execute the set of instructions stored by the memory means 240. Such processing means 260 can comprise for example a Central Processing Unit (CPU) based on ARM architecture or X64 architecture and/or a Graphical Processing Unit (GPU). Such processing means 260 can be for example implemented by a microcontroller like Arduino or can be implemented by dedicated hardware components such as CPLD, FPGA, or can be implemented by purpose-built chipsets. The processing means 260 can control the operations performed by the input means 210, the interface means 220, the output means 230, the sensor- interface means 240 and the memory means 250.

Besides, the block diagram shown in Figure 2 is of exemplificative nature only; it allows to understand how the invention works and how it can be realized by the person skilled in the art. The person skilled in the art understands that these charts have no limitative meaning in the sense that functions, interrelations and information shown therein can be arranged in many equivalents ways; for example, operations appearing to be performed by different logical blocks can be performed by any combination of hardware and software resources, being also the same resources for realizing different or all blocks. With reference to Figure 3, a method for controlling a powertrain system of a hybrid electric vehicle 100 by means of the onboard-control -unit 200 is described.

At step 300 an initialization stage is performed by said processing means 260. During this stage, the processing means 260 fetch the information and the set of instructions for carrying out the method according to an embodiment of the present invention.

At step 310 a recognition phase is performed by said processing means 260. During this stage, the onboard-control-unit 200 identifies by means of the interface means 220 the driver of the hybrid electric vehicle 100 and retrieves by means of the interface means 220 said optimized driving data OD specifically related to said driver from the external device 190, for example, in accordance to the above described embodiments. The optimized driving data OD comprise at least one Neural -Network-Model which is previously trained by using previous navigation data from a plurality of personal journeys of said driver.

Figure 4 schematically represents an example of a Neural -Network-Model 400 in accordance of an embodiment of the present invention. The Neural -Network-Model 400 comprises a first neural network 420 and a second neural network 430 which are operatively connected in a series configuration. The first neural network 420 takes as input the navigation data ND and is adapted to control the gearbox 160 by predicting an optimal gear-number. The second neural network 430 takes as input both the navigation data ND and the optimal gear-number and is adapted to control both the internal combustion engine 120 and the electric motor 140 by predicting an optimal power-flow parameter. Since the navigation data ND involve relatively long sequences of values, for example from 500 to 2000 values for each retained journey, in a preferred embodiment of the present invention, the Neural -Network-Model 400 is implemented by at least one Long-Short Term Memory (LSTM) Neural-Network. The first neural network 420 can comprise a plurality of neurons 422 and can be represented by a weighted graph in which each neuron 421 is represented by a node of the graph and a connection 423, between two of said neurons 421, can be represented by an edge of the graph. The connection 421 can be characterized by a weight, i.e. a parameter of the first neural network 420 that can be represented for example by a real number encoded as four or eight bytes according to the IEEE754 standard. The neurons 422 are organized in layers 425, and the topology of the graph characterizes the neural network 420, for example the neurons 422 belonging to two adjacent layers 425 can be fully connected, i.e. each neuron 422 of a layer 425 has a connection 423 to each neuron 422 of its adjacent layer 425. Each neuron 422 has its own activation function to be applied after some affine function which can be a convolution, dot product, or any combination of them. According to the LSTM NNs architecture, the first neural network 420 comprises a given number of additional layers 421 which are featured by cells 424 instead of neurons 422. Each cell 424 of the additional hidden layers 421 includes two hidden states and four interacting gates which are responsible for allowing the information to optionally pass from one step of the network to the next one. Thanks to this unique network layout, compared with other NNs, a LSTM NN can achieve improved performance when dealing with time-related sequences of data. In the present embodiment the second neural network 430 has the same architecture of the first neural network 420 and the Neural -Network-Model 400 is characterized by the parameters, such as the weights and the bias, the activation functions, the affine functions, the topology of the graph and the configurations related to both the first neural network 420 and the second neural network 430. In another embodiment of the present invention, more than two neural networks can be employed having different architecture among each other and/or different configurations done by a plurality of combinations comprising series and parallel configurations. In further embodiment of the present invention, only one neural network can be employed.

Before the Neural -Network-Model 400 can be deployed, it needs to be trained. Its training can be performed by means of a training-set, representative of a task that the Neural- Network-Model 400 has to deal with, such as for example the prediction of the optimal gear-number and the prediction of the optimal power-flow. Usually, said training-set comprises a large number of examples, such as pairs (<¾½), where each pair comprises an input value d_k and its corresponding target value v¾. During the training procedure, the parameters of the Neural -Network-Model 400 evolve from a learning epoch / to a next learning epoch t+1 according to a predefined algorithm such as for example the well- known Backpropagation algorithm.

Figure 5 exemplifies a training procedure 500 of the Neural -Network-Model 400 according to a preferred embodiment of the present invention. The training procedure 500 is performed off-line by training means which can be, as example, a laptop, a remote server, a smartphone, a purpose-built device. The external device 190 is adapted to collect previous navigation data from a plurality of personal journeys of the driver such that the training means can perform the training procedure 500. The previous navigation data, related to a complete journey previously performed by the driver, comprise the same information as the navigation data ND.

According to the embodiments above described for example, if the external device 190 is a USB memory connected to the interface means 220, then the previous navigation data can be memorized in said USB memory at the end of each personal journeys of the driver. In this way the driver can perform the train procedure by the training means, such as its owned laptop for example, by using the information stored in the USB memory. Successively, the driver can save the resulting Neural -Network-Model 400 in the USB memory.

In another embodiment of the present invention, the training means can be implemented by the external device 190. As example, if the external device 190 is the remote server which can communicate with the onboard-control-unit 200 by means of the interface means 220, such as a LTE interface, the previous navigation data can be memorized in said remote server at the end of each personal journey of the driver. Next the remote server, as example managed by a service provider, can perform the training procedure in accordance to the present invention and save the resulting Neural -Network-Model 400 in cloud.

According to a preferred embodiment of the present invention, the Neural -Network- Model 400 is previously trained by taking as input values said previous navigation data and by taking as target values optimized data provided by an optimization algorithm which takes as input said previous navigation data. As example, see Figure 5, for each driver the training procedure 515, 516, 517 takes as input values (d_k) the previous navigation data 505, 506, 507 and takes as target values (v_k) optimized data provided by an optimization algorithm 510, 511, 512 which takes as input said previous navigation data 505, 506, 507. As the result of the training procedure 500, the Neural -Network- Models 400, 401, 402 are obtained. In this way, the Neural -Network-Models 400, 401, 402 are optimally trained on the basis of previous navigation data 505, 506, 507 which take advantageously into account the personal behaviour of each driver.

Said optimization algorithm 510, 511, 512 can be a dynamic programming algorithm or a Pontryagin’s minimum principle algorithm, known in the art. As example, the optimization algorithm 510, 511, 512 can operate by minimizing overall fuel consumption and tailpipe emissions while complying with constraints imposed to the battery state-of- charge (SoC). As example, the optimization algorithm 510, 511, 512 can be implemented by a dynamic programming algorithm by setting the final battery SoC equal to the initial one, such as for example equal to 60% of the overall battery charge. It should be noted that, if required, the training procedure 500 is capable of optimally managing plug-in HEVs as well in charge-depleting conditions. Knowing the percentage of completion of the cycle particularly helps the Neural-Network-Model 400 to understand how the optimization algorithm 510, 511, 512 achieves the desired management of the battery SoC over the analysed journeys.

In another embodiment of the present invention, the topology of said Neural -Network- Model 400 is obtained by selecting the best performing combination over a set of hyper parameters comprising at least one of the following parameters: a number of layers, a number of neurons, a number of cells, neurons activation functions, learning-rate parameters, backpropagation optimizer parameters and dropout percentage parameters. As example, said set of hyper-parameters can be fine-tuned thanks to a pipeline made by a random-search algorithm and a grid-search algorithm. Once the upper and lower thresholds are defined for each hyper-parameter domain, the random-search algorithm selects random values within the limits so to produce an overview of the best and worst prediction performance. The random-search algorithm operations are defined as rough tuning. Given the entire number of hyper-parameters exploited through the random-search algorithm, the 5% of the combinations producing the best prediction results is selected to build up a grid featured by any combined hyper-parameter value. Therefore, each grid node is explored through the grid-search algorithm so as to define the optimal algorithm topology by selecting the best performing hyper-parameters combination. It should be noted that as the combination of hyper-parameters changes, the Neural -Network-Model 400 could produce significantly different results.

At step 320 an input phase is performed by said processing means 260. During this phase, the driver inputs navigation information of at least one journey to the onboard-control- unit 200 by means of input means 210. As mentioned above, the navigation information can comprise as example the location information, such as geographic coordinates, of a destination point where the driver intends to go, the location information of a starting point of the journey, the direction information to reach the destination point from the starting point. The navigation information can be provided by a navigation system of the hybrid electric vehicle 100 which can operate together with a location system such as for example the GPS system.

At step 330 a control phase is performed by said processing means 260. During this phase, the onboard-control -unit 200 collects, during the journey, navigation dataND from sensor means of the hybrid electric vehicle 100, then the onboard-control -unit 200 generates control data for controlling said powertrain system, wherein the control data are based on said navigation data ND and said optimized driving data. The control data are generated in real-time during the journey and can be outputted by the output means 230 which can comprise, for example, a CANBUS interface operatively connected with the gearbox 160, the internal combustion engine 120 and the electric motor 140. The control data can comprise the power-flow parameter PF and/or can comprise a gear-number GN resulting as output from said Neural -Network-Model 400 which receives as input the navigation data ND, see for example Figure 4.

At step 340 a check phase is performed by said processing means 260. During this phase, the processing means 260 evaluate if the journey is terminated. As example, the processing means 260 can verify if the current position of the hybrid electric vehicle 100, provided by a GPS system, is equal to the destination point inputted by the driver. In the affirmative case, the processing means 260 execute step 350, while they execute step 330 otherwise.

At step 350 a finalization stage is performed by said processing means 260. During this phase, the onboard-control-unit 200 sends said navigation data ND to the external device 190 by means of said interface means 220. In this way, the navigation data ND which are now related to a complete journey performed by the driver, i.e. they become previous navigation data, can be collected by the external device 190, in other to form the plurality of personal journeys of the driver, such that the training means can perform the training procedure 500 as described with reference to Figure 5.

The method according to the present invention can be advantageously performed by one or more drivers for the same hybrid electric vehicle 100. As example, considering the hybrid electric vehicle 100 to be included into a fleet of similar featured HEVs of a car sharing service, in this case the present invention can be performed by one or more drivers for the same hybrid electric vehicle 100. This allows advantageously to achieve an optimal driver-tailored powertrain control of each HEV of the fleet, having similar features of the hybrid electric vehicle 100, among a plurality of drivers.

With reference to Figure 6, the results of performance tests conducted by the Applicant are going to be discussed.

Personal driving data were experimentally collected by the Applicant for several real- world journeys, each of them performed by each user involved in the analysis. Nine specific driving missions were selected by means of Google Maps application and identified by specific geographical positions of starting and ending. Three different users were then asked to perform each of the 9 driving missions using the same vehicle, thus resulting in 27 personal driving mission data collected. Out of the 9 missions, 6 were particularly aimed at being used for the training process of the Neural -Network-Model, according to an embodiment of the present invention, here referred as AI agent, while the remaining 3 driving missions were used as testing of the AI agent. Training missions were selected in order to include both urban, extra-urban, hilly and highway driving scenarios. One of the training missions has been defined in such a way that a combination of urban, extra-urban and highway driving conditions was considered. The 3 testing missions are representative of all possible driving conditions as well: however, they differ from any of the training missions so that multiple testing simulations could realistically be carried out. All the selected driving missions are based both in the city centre and the neighbourhoods of Turin, Italy.

The experimental campaign was performed with a Fiat Scudo 2.0 TD, a citizen vehicle classified in the family of light duty vehicles. Vehicle characteristics, extracted from technical documents and considered in numerical simulations, are listed in the following: Tyres: 215/60 R16; Vehicle length: 5.143 m; Vehicle width: 1.895 m; Vehicle curb weight: 1911 kg; Engine 2.0 LTD, max power 94 kW, max torque 320 Nm @ 2000 rpm; Transmission: manual 6 gears. The 3 retained users drove the reported vehicle travelling the above-mentioned missions while adopting their own driving style. Vehicle velocity and altitude were recorded over time during each test (characterized by specific mission and driver) by means of the GPS -based MATLAB Mobile application. The selected representative HEV layout was defined as follows: ICE capacity 1.3 1; MG power 99 kW; Battery capacity 10 kWh; ICE-MG ratio 3.42; Final drive ratio 3.5. The results of the table pictured in Figure 6 show the simulated CO2 emission values from different driving missions, where RBC refers to a heuristic rule-based control. Even though the 3 retained driving missions are featured by equal starting and finishing geographical positions, different minimum values of CO2 emitted can be achieved by each driver due to the characteristics of the personal driving style, e.g. aggressive. As example, in DM07 and DM08 ideal CO2 emitted by Driver 1 is higher by 29.64% and by 28.55% compared with Driver 3, respectively, while in DM09 ideal CO2 emitted by Driver 2 is higher by 31.79% compared with Driver 1. In conclusion, the obtained results show that the method described in the present invention can outperform the state-of-the-art powertrain control schemes of a HEV in terms of emitted CO2 and therefore in terms of fuel consumption. The advantages of the present invention are therefore evident from the description provided above.

The method for controlling a powertrain system of a hybrid electric vehicle according to the present invention, and the hybrid electric vehicle thereof, allows advantageously to achieve an optimal HEV powertrain control policy to a wide range of different driving scenarios and different personal driving styles. The present invention can be easily implementable in an onboard-control-unit of a hybrid electric vehicle, without involving an extra on-board system and thus, without increase complexities in terms of both maintenance and potential failure.

A further benefit of the present invention is that it allows to control a powertrain system of a hybrid electric vehicle by taking into account the personal driving style and the complexity related to a plurality of personal journeys for each driver, because of an off line training procedure over a custom-made dataset of driving profiles. This allows advantageously to perform an improved control of a powertrain system of a hybrid electric vehicle which is more specifically tailored for each driver.

Another benefit of the present invention is that it allows an improved real-time control policy obtained by the onboard-control -unit also because the topology of the Neural- Network-Model is automatically adjusted for each driver according to a fine-tuning operation.

A further benefit of the present invention is that it allows to take into account the technical features, such as the battery efficiency, that might change during the operating time of the HEV owned by each specific driver by updating the custom-made dataset with the journeys recently performed by the driver.

The present description has tackled some of the possible variants, but it will be apparent to the man skilled in the art that other embodiments may also be implemented, wherein some elements may be replaced with other technically equivalent elements. The present invention is not therefore limited to the explanatory examples described herein, but may be subject to many modifications, improvements or replacements of equivalent parts and elements without departing from the basic inventive idea, as set out in the following claims.

Claims

1. A method for controlling a powertrain system of a hybrid electric vehicle (100) by means of an onboard-control -unit (200) of the hybrid electric vehicle (100), said method comprising:

- a recognition phase, wherein the onboard-control-unit (200) identifies by means of interface means (220) a driver of the hybrid electric vehicle (100) and retrieves by means of the interface means (220) optimized driving data specifically related to said driver from an external device (190);

- an input phase, wherein said driver inputs navigation information of at least one journey to the onboard-control -unit (200) by means of input means (210);

- a control phase, wherein the onboard-control-unit (200) collects navigation data (ND) from sensor means of the hybrid electric vehicle (100), wherein the onboard-control -unit (200) generates control data for controlling said powertrain system, and wherein the control data are based on said navigation data (ND) and said optimized driving data.

2. Method according to claim 1, wherein said optimized driving data comprise a Neural- Network-Model which is previously trained by using previous navigation data from a plurality of personal journeys of said driver.

3. Method according to claim 2, wherein said navigation data (ND) and said previous navigation data comprise temporal sequences including values related to at least one of the following parameters: a speed, an acceleration, a status-flag, a battery-current, a battery-voltage, a fuel level, a position, a travel-time interval, a gear-number of said hybrid electric vehicle (100).

4. Method according to claim 2 or claim 3, wherein said Neural -Network-Model is previously trained by taking as input values said previous navigation data and by taking as target values optimized data provided by an optimization algorithm which takes as input said previous navigation data.

5. Method according to claim 4, wherein said optimization algorithm is a dynamic programming algorithm or a Pontryagin’s minimum principle algorithm.

6. Method according to one or more of the claims from 2 to 5, wherein said Neural- Network-Model is implemented by at least one Long-Short Term Memory (LSTM) Neural-Network.

7. Method according to one or more of the claims from 2 to 6, wherein a topology of said Neural-Network-Model is obtained by selecting the best performing combination over a set of hyper-parameters comprising at least one of the following parameters: a number of layers, a number of neurons, a number of cells, neurons activation functions, learning-rate parameters, backpropagation optimizer parameters and dropout percentage parameters.

8. Method according to one or more claims from 1 to 7, wherein said control data comprise a power-flow between an internal combustion engine (120) and an electric motor (140) of the hybrid electric vehicle (100) and/or comprise a gear-number of a gearbox (160) of the hybrid electric vehicle (100).

9. Method according to one or more claims from 1 to 8, wherein the onboard-control-unit (200) sends said navigation data (ND) to the external device (190) by means of said interface means (220).

10. Method according to one or more claims from 1 to 9, wherein said interface means (220) comprise at least one of the following interfaces: an USB interface, a NFC interface, a Bluetooth interface, a Wi-Fi interface, a mobile-network interface.

11. Method according to one or more claims from 1 to 10, wherein said external device (190) is a USB memory device or a smartphone or a remote server.

12. A hybrid electric vehicle (100) comprising a powertrain system adapted to be controlled by an onboard-control -unit (200) of said hybrid electric vehicle (100), the onboard-unit (200) being adapted to identify a driver of the hybrid electric vehicle (100) and being adapted to retrieve optimized driving data specifically related to said driver from an external device (190) by means of interface means (220), the onboard-control-unit (200) being adapted to acquire navigation information of at least one journey by means of input means (210) from the driver, the onboard-control-unit (200) being adapted to collect navigation data (ND) from sensor means of said hybrid electric vehicle (100) and the onboard-unit (200) is adapted to generate control data for controlling said powertrain system, wherein the control data are based on said navigation data (ND) and said optimized driving data.

13. Hybrid electric vehicle (100) according to claim 12, wherein said optimized driving data comprise a Neural -Network-Model which is adapted to be previously trained by using previous navigation data from a plurality of personal journeys of said driver.

14. Hybrid electric vehicle (lOO)accordingto claim 13, wherein said navigation data (ND) and said previous navigation data comprise temporal sequences including values related to at least one of the following parameters: a speed, an acceleration, a status-flag, a battery-current, a battery -voltage, a fuel level, a position, a travel-time interval, a gear- number of said hybrid electric vehicle (100).

15. Hybrid electric vehicle (100) according to claim 13 or claim 14, wherein said Neural- Network-Model is adapted to be previously trained by taking as input values said previous navigation data and by taking as target values optimized data provided by an optimization algorithm which takes as input said previous navigation data.

16. Hybrid electric vehicle (100) according to claim 15, wherein said optimization algorithm is a dynamic programming algorithm or a Pontryagin’s minimum principle algorithm.

17. Hybrid electric vehicle (100) according to one or more of the claims from 13 to 16, wherein said Neural -Network-Model is implemented by at least one Long-Short Term Memory (LSTM) Neural-Network.

18. Hybrid electric vehicle (100) according to one or more of the claims from 13 to 17, wherein a topology of said Neural-Network-Model is adapted to be obtained by selecting the best performing combination over a set of hyper-parameters comprising at least one of the following parameters: a number of layers, a number of neurons, a number of cells, neurons activation functions, learning-rate parameters, backpropagation optimizer parameters and dropout percentage parameters.

19. Hybrid electric vehicle (100) according to one or more of the claims from 12 to 18 wherein, said control data comprise a power-flow between an internal combustion engine (120) and an electric motor (140) of the hybrid electric vehicle (100) and/or comprise a gear-number of a gearbox (160) of the hybrid electric vehicle (100).

20. Hybrid electric vehicle (100) according to one or more of the claims from 12 to 19, wherein the onboard-control -unit (200) is adapted to send said navigation data (ND) to the external device (190) by means of said interface means (220).

21. Hybrid electric vehicle (100) according to one or more of the claims from 12 to 20, wherein said interface means (220) comprise at least one of the following interfaces: an USB interface, a NFC interface, a Bluetooth interface, a Wi-Fi interface, a mobile- network interface.

22. Hybrid electric vehicle (100) according to one or more of the claims from 12 to 21, wherein said external device (190) is adapted to be a USB memory device or a smartphone or a remote server.