Controller with a learning capability and automatic exploration function of an operating parameter space

Abstract

The invention relates to a motor controller for an internal combustion engine of a vehicle, comprising a control unit for setting one or more control variables on the basis of one or more measured variables according to a stored control scheme; wherein the control unit is designed to modify the stored control scheme when the control unit is used as intended with the operational internal combustion engine, which is being controlled by the motor controller, according to a specified learning algorithm, namely using at least one feedback parameter which is associated with an optimization criterion and is provided to the control unit, in order to provide an improved control of the internal combustion engine.

Claims

1. An engine controller for an internal combustion engine of a vehicle, comprising a control unit for setting one or more controlled variables based on one or more measured variables according to a stored control scheme, wherein the control unit is configured to vary the stored control scheme according to a predefined learning algorithm during the intended use with the operating internal combustion engine controlled by the engine controller, wherein the predefined learning algorithm is based on at least one feedback parameter that is provided to the control unit and associated with an optimization criterion, and wherein the at least one feedback parameter is a pressure or encompasses a pressure in the combustion chamber or an exhaust gas-side gas mixture pressure.

2. The engine controller according to claim 1, wherein the at least one measured variable encompasses an engine speed, a throttle valve position, an injected fuel volume, a combustion residual gas quantity, an ignition timing, a valve opening and valve closing time, an engine temperature, an intake-side gas mixture pressure, a pressure in the combustion chamber, an exhaust gas-side gas mixture pressure, an engine torque, or an engine mileage, and/or the at least one controlled variable encompasses a throttle valve position, an injected fuel volume, an ignition timing, or a valve opening and closing time.

3. The engine controller according to claim 1, wherein the predefined learning algorithm is an algorithm for reinforcement learning or encompasses such an algorithm.

4. The engine controller according to claim 3, wherein the control unit is configured to deliberately adapt one or more of the controlled variables so as to generate random samples having a higher informational content for the learning process.

5. The engine controller according to claim 1, wherein the at least one feedback parameter encompasses a torque of the internal combustion engine, a fuel consumption of the internal combustion engine, or an exhaust gas composition of the internal combustion engine.

6. The engine controller according to claim 1, wherein the optimization criterion encompasses a respective target value or a respective trend specification for each feedback parameter.

7. The engine controller according to claim 6, wherein the optimization criterion encompasses also a respective weighting factor and/or a respective prioritization over one or more other feedback parameters.

8. The engine controller according to claim 1, wherein the control unit is configured to ensure that the at least one feedback parameter is provided continuously.

9. The engine controller according to claim 8, wherein the at least one feedback parameter is provided continuously once per ignition of the internal combustion engine during the operation of the internal combustion engine, and is configured to continuously vary the control scheme.

10. The engine controller according to claim 9, wherein the at least one feedback parameter is provided to vary the control scheme once in response to each provision of the at least one feedback parameter as long as the variation is useful according to the optimization criterion.

11. The engine controller according to claim 1, wherein the control unit is a pretrained control unit in which the stored control scheme is varied according to the predefined learning algorithm prior to the intended use with the operating internal combustion engine within the scope of a simulation.

12. The engine controller according to claim 11, wherein the predefined algorithm is based on at least one feedback parameter that is provided to the control unit and associated with the optimization criterion, in conjunction with one or more simulated measured variables.

13. The engine controller according to claim 1, wherein the respective prohibited value ranges are predefined for the controlled variable or controlled variables in the control unit so that values from the prohibited value ranges cannot be set.

14. An internal combustion engine or a vehicle, comprising an engine controller according to claim 11.

15. A method for operating an engine controller for an internal combustion engine of a vehicle comprising a control unit for setting one or more controlled variables based on one or more measured variables according to a stored control scheme, the method comprising varying the stored control scheme according to a predefined learning algorithm during the intended use of the engine controller with the operating internal combustion engine, wherein the predefined learning algorithm is based on at least one feedback parameter that is provided to the control unit and associated with an optimization criterion, and wherein the at least one feedback parameter is a pressure or encompasses a pressure in the combustion chamber or an exhaust gas-side gas mixture pressure.

Description

(1) The subject matter according to the invention shall be described in more detail based on the schematic drawings shown in the following FIGURE, without limiting the subject matter to the specific embodiments shown here.

(2) The FIGURE shows a schematic illustration of an exemplary embodiment of a learning engine controller including independent exploration of an operating parameter space. The engine controller 1 is coupled to an internal combustion engine 2 of a vehicle, which is not shown. The engine controller 1 comprises a control unit 3 for setting one or more controlled variables 4 based on one or more measured variables 5 according to a stored control scheme. The control unit 3 is designed to vary the stored control scheme according to a predefined learning algorithm during the intended use with the operating internal combustion engine 2, in the present case operating in the vehicle, and controlled by the engine controller 1, and more particularly based on at least one feedback parameter 6 that is provided to the control unit 3 and associated with an optimization criterion. As in the shown example, the feedback parameter can be fed back by the internal combustion engine 2, and alternatively or additionally by a further unit, such as a sensor unit 7, for example.

(3) Since the feedback parameter or parameters 6 in the present example is or are collected continuously, that is repeatedly, for an undetermined number of times, in particular once per ignition of the internal combustion engine, and provided to the control unit 3, the learning engine controller 1 is able to adapt quickly and independently to changing environment conditions, such as, for example, a changing humidity or a changing air pressure, and can thus enhance the control of the internal combustion engine 2 in keeping with the optimization criterion.