System and method for operating a powertrain system

Abstract

A method and apparatus for operating a vehicle including an internal combustion engine, an electric motor, and an electrically heatable catalyst is disclosed. According to the disclosed embodiments, it is advantageous to simultaneously evaluate energy consumption and emissions due to increasing or decreasing catalyst heating action and due to increasing or decreasing motor torque based on an operating model, and use the operating model to determine an operating mode for each of the internal combustion engine, the electric motor, and the electrically heatable catalyst.

Description

Technical Field

[0001] The invention relates to a way of operating a powertrain system comprising an internal combustion engine and in particular to a way of implementing a strategy for engine and emission management which is advantageous for vehicles with electrically heatable catalysts. The invention improves a system with a combination of a drive source (electric motor EM, internal combustion engine ICE) and an emission control device, in particular an electrically heatable catalyst (EHC). The optimization of different degrees of freedom of such a system can reduce fuel consumption or improve fuel efficiency while meeting emission limits. Background Art

[0002] The electrification of the drivetrain is important to reduce fuel consumption and meet increasingly stringent limits on pollutant emissions. These goals must also be achieved under real driving conditions. Summary of the invention

[0003] Improved control strategies for hybrid electric vehicles (HEVs) must take into account parameters related to the energy required by the internal combustion engine (ICE), the electric motor (EM), and the electrically heated catalyst (EHC). Such strategies should control the torque distribution between the ICE and the electric motor, the power allocated to the electrically heated catalyst, etc. By doing so, the energy consumption of the hybrid vehicle can be significantly reduced compared to conventional drivetrains.

[0004] A hybrid electric vehicle (HEV) typically includes a traction (or high voltage) battery that acts as an electrical energy storage and provides power to an electric drive or traction motor or machine for propulsion. This high voltage battery can be at 800v, ​​400v or 48v. Electrical energy storage such as a battery together with the electric motor enables kinetic energy recovery, load point adaptation of the internal combustion engine, torque assistance and supercharging.

[0005] The hybrid configuration can also achieve robust emissions management that limits emissions to within regulatory limits, independent of driving conditions. For example, during low load and short distance trips, where little heat is supplied by the internal combustion engine, the exhaust gas temperature can be increased or enhanced due to heat from the electrically heatable catalyst. In the alternative, the load on the internal combustion engine can be increased using the disconnect torque of the electric motor. This in turn reduces the time to reach the ignition temperature of the catalytic converter, thereby improving the pollutant conversion efficiency of the catalytic converter. Therefore, when the temperature of the vehicle's catalyst is expected to drop below a threshold, electrical power is supplied to the electrically heated catalyst. Alternatively, or simultaneously, the braking torque of the electric motor can be increased in the event that the temperature of the vehicle's catalyst is expected to drop below a threshold or when the temperature of the vehicle's catalyst is expected to drop below a threshold.

[0006] During high load phases, or when exhaust gas temperatures are high, the catalyst may exceed its optimal temperature range. This results in low conversion efficiency. In this case, the load of the internal combustion engine may be reduced due to torque support from the electric motor, which reduces the raw exhaust mass flow and acts to reduce the temperature of the catalytic converter. The load may be a current load, or an expected load based on forecast information. Therefore, in the event that the temperature of the vehicle's catalyst is expected to rise above a threshold or when the temperature of the vehicle's catalyst is expected to rise above a threshold, the boost torque of the electric motor may be increased.

[0007] At any time, the objectives or constraints are to provide driver demanded torque, maintain battery state of charge (SoC) within regulatory limits, and maintain regulated emissions and projected regulated emissions (such as NOx) within regulatory limits. The vehicle's operating model can be used to optimize the operating mode of the components according to the optimization objectives.

[0008] The necessary control strategy can be presented as a control strategy for multiple degrees of freedom that interact to influence fuel consumption and emissions:

[0009] a) Torque distribution between the internal combustion engine and the electric motor;

[0010] b) electrical power for the electrically heatable catalyst;

[0011] c) combustion mode of internal combustion engine;

[0012] d) gear selection and shifting; and

[0013] e) Comfort functions such as heating and air conditioning.

[0014] The control strategy can be implemented using different artificial intelligence techniques. One such technique is reinforcement learning (RL). The control strategy can be developed through a learning or training phase, followed by an optional testing phase. A testing phase may be necessary to ensure that the trained and implemented control strategy meets the mandatory emission requirements. Learning or adjusting parameters during normal operation may or may not be possible.

[0015] By appropriate adjustment of the different degrees of freedom, the control strategy can minimize both fuel consumption and emissions, with the advantages shown below. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 shows the layout of the HEV architecture including the exhaust aftertreatment system;

[0017] Figure 2 A reinforcement learning configuration is shown;

[0018] Figure 3 The steps of training, testing, and operation are shown;

[0019] Figure 4 The steps of controlling the degrees of freedom are shown; and

[0020] Figure 5 A decision curve related to SoC is shown. DETAILED DESCRIPTION

[0021] exist Figure 1 1 , the main elements of one embodiment of a hybrid vehicle are shown as 100. The connections are shown as mechanical 101, electrical 102, fuel flow 103, and exhaust 104. In the exhaust flow, an electrically heatable catalyst (EHC) 110 precedes a diesel oxidation catalyst (DOC) 111. The exhaust then passes to a selective catalytic reduction 112. An internal combustion engine 120 receives fuel from a fuel supply 121. In this embodiment, the internal combustion engine and electric motor 130 are mechanically connected via a belt 135. Electric power to and from the electric motor 130 can be passed to a battery 135, the electrically heatable catalyst 110, and other auxiliary loads shown as 136. Mechanical energy from the internal combustion engine and / or electric motor is passed to wheels 150 via a clutch 140 and a gearbox 145.

[0022] Figure 2 2 shows the principle elements of a reinforcement learning system 200. A reinforcement learning (RL) agent 230 provides an action vector a to an environment 240. t 210. The environment can be a real physical environment (such as a hybrid vehicle), or it can be a simulated environment in which the main elements of the hybrid vehicle are modeled in software. The environment takes the action vector as input and generates the resulting state vector s t 220 and return vector r t 225. The action vector contains the values ​​or elements corresponding to the degrees of freedom, as well as any additional actions or control elements required to operate the vehicle. For example, an action vector a t May contain values ​​that determine how much fuel is to be supplied to the internal combustion engine, or how much current is to be supplied to an electrically heatable catalyst, or how much current is to be supplied by an electric motor to, for example, a battery. Other settings or controls for the operating mode that may be set in or manipulated by the action vector include vehicle speed or target speed, target state of charge (SoC) for the battery, selection of a hybrid mode (e.g., recuperation, coasting), urea or AdBlue injection timing and amount, and timing for filter regeneration (e.g., diesel DPF regeneration), or gear shift and / or gear selection.

[0023] Return vector r t225 contains information corresponding to the environmental aspects to be optimized. For example, the reward vector may contain environmental values ​​for CO2, NOx, fuel consumption, and other values ​​related to environmental or emission considerations. The state vector s t and the return vector r t is returned as input to the RL agent.

[0024] The values ​​in the action vector will determine how the degrees of freedom are used, and the RL agent will use the reward vector and the state vector to optimize the action vector. The next action vector specified by the RL agent will determine the torque distribution between the ICE and EM, the electric power to the EHC (input or output), and the combustion mode of the ICE. In this way, the operating model will also predict future fuel consumption and emissions. Therefore, the operating model of the vehicle is used to optimize the operating mode of the components according to the selected optimization goal, such as minimizing fuel consumption while always complying with emission limits.

[0025] Other factors may also be considered in the action vector and / or state vector. For example, additional degrees of freedom may include gear shifting and gear selection, AddBlue injection, heating and cooling, etc.

[0026] A control strategy may be implemented using a cost-based comparison of different modes of a hybrid electric vehicle. Based on the cost comparison, the strategy may decide which of multiple modes is best suited for the current operating point and SoC. In one embodiment, these modes may be defined as battery charging, battery discharging, and zero battery current. For each mode and operating point, the cost of satisfying the mechanical power requested by the driver and the thermal constraints required by the aftertreatment system is calculated. The cost term is defined as the ratio of the increase or decrease in fuel power caused by the load point shift to the increment in battery power. The discharge cost may be expressed as the saved fuel power compared to the depleted battery power. On the other hand, the charging cost may be the additional fuel power used to restore the battery power. Therefore, the highest cost is optimal in the discharge mode and the lowest cost is optimal in the charge mode. By finding the lowest or highest cost, respectively, the torque set point and power for the EHC may be found. During online application, the hybrid mode may be selected based on a cost comparison of each mode to a cost criterion. The criterion maps the SoC to a maximum limit for the charge mode and a minimum limit for the discharge mode, such as Figure 5 For a discharge cost higher than the minimum cost, select Discharge. For a charge cost lower than the maximum cost, select Charge.

[0027] Go to Figure 3 , showing the steps of training, testing and operation. In the first step 310, training is performed to find the optimal operating model. In this embodiment, using Figure 2The loop and simulated environment shown determine the operation model. The RL agent is provided with different operation states and the resulting rewards given by the reward vector. Various action vectors are generated, which in turn change the state of the simulated environment. The RL agent then uses the state vector as a reference to evaluate the resulting reward vector.

[0028] Step 310 includes an operating model that has been prepared through the steps of simulating driving conditions, and optimizing the use of the internal combustion engine 120, the electrically heatable catalyst 110, and the electric motor 130 during the simulation to minimize both fuel consumption and emissions.

[0029] When the optimal operating model has been found, this can be passed to an optional testing step 320. In an embodiment with a testing step, different simulation environments are used to verify that the operating model always complies with regulations regarding emission conditions. For example, the simulation environment used to determine the operating model can consist of a large number of simulated training trajectories, such as 500 trajectories (car trips), and the test simulation environment can consist of a similar or smaller number of different verification trajectories, such as 400 trajectories (car trips). In this way, the learned behavior can be verified before being used in the product. Likewise, if there are weaknesses in the training data, incorrect learned behavior can be identified and corrected if necessary.

[0030] The RL agent can learn to adjust the emission profile in a signal-dependent manner to stay within regulatory limits. In particular, the EHC can be activated based on the signal. If the signal is missing in the real environment, the vehicle using the operating model may no longer meet regulatory requirements because the EHC is not operating correctly.

[0031] Once the operational model has been found, and in some embodiments tested and validated, then at step 330, the operational model is provided and used in the vehicle for use in a real operational environment. At step 330, the operational model is used to provide the motion vector a t 210, which optimizes the degrees of freedom and provides the control signals or operating modes required to operate, for example, the internal combustion engine ICE 120, the electrically heatable catalyst 110 and the electric motor EM 130. In a preferred embodiment, from the action vector a t The derived operating mode is operable to operate the electrically heatable catalyst and / or the electric motor and / or the internal combustion engine. The operating mode will be set to achieve the optimization goal.

[0032] In some embodiments, a further step 340 is possible. In step 340, the operating model is adapted to further optimize the operation, for example in terms of fuel efficiency or emissions. The operating model can then be used in step 330. In other embodiments, the set points can be taken from a cost comparison method.

[0033] The exhaust aftertreatment system (ATS) used in the example vehicle may consist of an electrically heatable catalyst (EHC) 110, a diesel oxidation catalyst (DOC) 111, and a selective catalytic reduction catalyst (SCR) 112. The main parameters of such an example HEV are given in Table 1.

[0034] Vehicle weight M 1523kg ICE max. mechanical power <![CDATA[P ice，max ]]> 134kW EM max. mechanical power <![CDATA[P em，max ]]> 30kW EHC max, electric power <![CDATA[P ehc，max ]]> 4kW 48V battery capacity <![CDATA[Q0]]> 40Ah

[0035] Table 1. Example vehicle parameters.

[0036] The same inventive concept can be used in various vehicles having different power levels.

[0037] One embodiment of reinforcement learning (RL) is to Figure 2 The agent-environment interface shown uses Figure 4 In step 410, the agent observes the state s of the environment at time t t and return r t , and then generate the action vector a by t In step 420, the environment receives the action vector a t At a later time t+1, the environment has reacted in step 430 and generates a state vector s as the environment transitions to a new state t and the return vector r t Then, at 410, the RL agent reads the newly generated state vector s from the simulated environment t and the return vector r t , and feeds the action vector back into the model to compute the resulting new state. The goal of the RL agent is to search for a policy that maximizes the cumulative reward before the learning process ends.

[0038] The agent weights decisions based on current rewards compared to future rewards: for a discount factor γ = 0, the agent chooses greedy decisions to get immediate rewards; as γ approaches 1, the agent prefers future rewards.

[0039] There are different ways to develop a trial-and-error model for RL agents. One embodiment is based on Proximal Policy Optimization (PPO), which shows good performance across various types of tasks. PPO is a policy gradient method where the policy is stochastic and modeled as a parameterized probability distribution based on the current state, from which actions are sampled.

[0040] The input features of the agent and the so-called "critic" are calculated from the observations of the vehicle state. In one embodiment related to operating distance based limits, features are derived from the vehicle speed v, which depends on whether the traveled distance x(t) is greater or less than a distance such as 5 km. At the beginning of the trajectory, the emission limit is higher, and after a certain distance (e.g. 5 km), the emissions must be below the defined emission limit.

[0041] Another feature is calculated as the accumulated NOx emissions, which are compared to the distance travelled and multiplied by the NOx limit (e.g. 60 mg / km). Additional inputs are the state of charge of the battery SoC, the exhaust gas temperatures Texh and Tscr. The reward is defined as proportional to the (negative) fuel mass, which is proportional to the CO2 emitted. If the NOx emissions exceed the limit, a penalty is added.

[0042] In one embodiment, the agent consists of a single linear layer neural network for P(ehc) and tq(em) control, with only Tscr and SoC as input. For combustion mode i(ice), the linear layer output is added to a fully connected network with leaky-relu activation and 30 neurons in the hidden layer. Tanh activation is used to calculate tq(em). The positive output can be scaled from 0 to the current maximum torque of the EM as tq(em,max), and the negative output can be scaled from 0 to tq(em,min). Both tq(em,max) and tq(em,min) depend on the SoC and are subject to derating of the EM.

[0043] In one embodiment, the output of the electrically heated agent is scaled to range from zero to the maximum possible heating power P(ehc,max), limited by the SoC and the physical limitations of 4kW.

[0044] The linear part of the model is initialized with reasonable values ​​that make SoC and T scr Keep it within controllable range, as it is known that SCR efficiency drops significantly towards low and high temperatures.

[0045] During training, the models are repeatedly evaluated on the training data. The model that meets the NOx limits on all training traces and has the lowest fuel consumption among these is selected as the final model for testing.

[0046] exist Figure 5 A decision curve for optimizing costs based on the state of charge SoC (500) is shown in FIG. 5. The lowest discharge cost is shown as 510 and the highest charge cost is shown as 520.

Claims

1. A method of operating a vehicle comprising an internal combustion engine (120), an electric motor (130) and an electrically heatable catalyst (110), comprising: Simultaneously evaluating energy consumption and emissions due to increasing or decreasing catalyst electrical heating action and due to increasing or decreasing electric motor torque based on an operating model; as well as determining an operating mode for each of the internal combustion engine, the electric motor, and the electrically heatable catalyst using the operating model so that operation is optimized according to an optimization objective, The method further includes: if deceleration is expected, simultaneously evaluating, based on the operating model, predicted energy consumption and emissions due to increasing or decreasing catalyst electrical heating activity and due to increasing or decreasing motor torque; and determining an operating mode for each of the internal combustion engine, the electric motor, and the electrically heatable catalyst using the operating model.

2. The method according to claim 1, further comprising: Projected energy consumption and emissions due to increasing or decreasing engine torque are simultaneously evaluated based on an operating model to determine an operating mode for each of the engine, the electric motor, and the electrically heatable catalyst using the operating model.

3. The method of claim 1 or 2, wherein the evaluation uses previously learned or trained values ​​as an operating model.

4. The method according to claim 1 or 2, wherein the operating mode is operable to operate an electrically heatable catalyst and / or an electric motor and / or an internal combustion engine.

5. The method according to claim 1 or 2, wherein: When the temperature of the catalyst of the vehicle is expected to drop below a threshold, the braking torque of the electric motor (130) is increased.

6. The method according to claim 1 or 2, wherein: When the temperature of a catalyst of the vehicle is expected to drop below a threshold, current to the electrically heatable catalyst (110) is increased.

7. The method of claim 1 or 2, wherein the operating model is adapted (440) during vehicle operation.

8. The method according to claim 1 or 2, wherein the operating mode comprises vehicle speed or target speed, target state of charge (SoC) for the battery, selection of hybrid mode, urea or AdBlue injection time and amount, and timing for filter regeneration, or selection of gear shift and / or gear.

9. A control system adapted to perform the method of any one of claims 1 to 8, comprising an operation model prepared by the following steps: simulated driving conditions; and Use of the internal combustion engine (120), the electrically heatable catalyst (110), and the electric motor (130) is optimized during the simulation to minimize both fuel consumption and emissions.

10. A hybrid vehicle comprising an internal combustion engine (120), an electrically heatable catalyst (110), an electric drive or traction motor (130) and a battery (135), wherein the vehicle is suitable and adapted to perform the method of any one of claims 1 to 8.

Images