Trained Neural network air/fuel control system

• US 5,781,700 A
• Filed: 02/05/1996
• Issued: 07/14/1998
• Est. Priority Date: 02/05/1996
• Status: Expired due to Fees

• Current Assignee: Ford Global Technologies, LLC (Ford Motor Company)
• Original Assignee: Ford Global Technologies, LLC (Ford Motor Company)
• Inventors: Feldkamp, Lee Albert, Puskorius, Gintaras Vincent, Davis, Leighton Ira
• Primary Examiner(s): Davis, George B.
• Application Number:
  US08/597,080
• Time in Patent Office: 890 Days
• Field of Search: 395/23, 395/22, 395/20, 395/905
• US Class Current: 706/14
• CPC Class Codes: F02D 2041/1423 Identification of model or controller parameters
• CPC Class Codes: F02D 2041/1429 Linearisation, i.e. using a feedback law such that the system evolves as a linear one
• CPC Class Codes: F02D 2041/1433 using a model or simulation of the system
• CPC Class Codes: F02D 41/1405 Neural network control
• CPC Class Codes: F02D 41/1473 characterised by the regulation method
• CPC Class Codes: F02D 41/2432 Methods of calibration
• CPC Class Codes: F02D 41/2454 Learning of the air-fuel ratio control
• CPC Class Codes: F02D 41/2477 characterised by the method used for learning
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

An electronic engine control (EEC) module executes both open loop and closed loop neural network processes to control the air/fuel mixture ratio of a vehicle engine to hold the fuel mixture at stoichiometry. The open loop neural network provides transient air/fuel control to provide a base stoichiometric air/fuel mixture ratio signal in response to throttle position under current engine speed and load conditions. The base air/fuel mixture ratio signal from the open loop network is additively combined with a closed loop trimming signal which varies the air/fuel mixture ratio in response to variations in the sensed exhaust gas oxygen level. Each neural network function is defined by a unitary data structure which defines the network architecture, including the number of node layers, the number of nodes per layer, and the interconnections between nodes. In addition, the data structure holds weight values which determine the manner in which network signals are combined. The network definition data structures are created by a network training system which utilizes an external training processor which employs gradient methods to derive network weight values in accordance with a cost function which quantitatively defines system objectives and an identification network which is pretrained to provide gradient signals representative of the behavior of the physical plant. The training processor executes training cycles asynchronously with the operation of the EEC module in a representative test vehicle.

Claims

1. Apparatus for controlling the air/fuel mixture ratio of the combustibles delivered to an internal combustion vehicle engine, said apparatus comprising, in combination:

a plurality of sensors coupled to said engine for producing a plurality of input signal values each indicative of a predetermined variable engine operating condition, a first one of said input signal values being indicative of the oxygen level in the exhaust gases produced by said engine;

a electronic engine control processor coupled to said sensors and programmed to execute at least one neural network signal translation algorithm for converting said input signal values into a plurality of intermediate node output values and an output signal value indicating a controlled fuel command in accordance with network weight values,

fuel delivery control means responsive to said output signal value for adjusting the rate at which fuel is delivered to said engine to achieve said controlled air/fuel mixture ratio;

data storage means coupled to said control processor for storing said input signal values, said intermediate node output values, said output signal value, and said weight values; and

a training processor external to said control processor and coupled to said data storage means for modifying said weight values, said training processor including;

monitoring means responsive to said input signal values for comparing said controlled air/fuel mixture ratio with a predetermined desired air/fuel mixture ratio which provides substantially stoichiometric engine operation, and

means responsive to said monitoring means for varying at least selected ones of said weight values such that said controlled air/fuel mixture ratio more nearly matches said predetermined desired air/fuel mixture ratio.

2. Apparatus as set forth in claim 1 wherein said means responsive to said control signal for adjusting the rate at which fuel is delivered to said engine includes closed loop control means responsive to a first one of said input signal values for varying said fuel delivery rate to reduce deviations in said controlled air/fuel mixture ratio from said desired air/fuel mixture ratio.

3. Apparatus for controlling a vehicle engine as set forth in claim 2 wherein said neural network signal translation algorithm includes recurrent processing means for combining said input signals with delayed ones of said intermediate output values.

4. Apparatus for controlling said vehicle engine as set forth in claim 3 wherein said means for varying at least selected ones of said weight values comprises:

dynamic gradient processing means for indicating the rates of change of said controlled air/fuel mixture ratio with respect to selected ones of said input signal values, and

means for varying said weights in accordance with said rates of change such that said controlled air/fuel mixture ratio more nearly matches said desired air/fuel mixture ratio.

5. Apparatus for controlling said vehicle engine as set forth in claim 4 wherein said means for varying said weights employs a decoupled extended Kalman filter training algorithm for varying said weights in accordance with said rates of change.

6. Apparatus as set forth in claim 4 wherein said dynamic gradient processing means indicates the rates of change of said controlled air/fuel mixture ratio with respect to selected ones of said input signal values by processing said input signal values by means of a predetermined identification network algorithm previously trained by off-line processing to model the relationship between said input signals and said controlled air/fuel mixture ratio.

7. Apparatus for controlling the air/fuel mixture ratio supplied to an internal combustion vehicle engine, said apparatus comprising, in combination:

a plurality of sensors coupled to said engine for producing a plurality of input signal values each indicative of a predetermined variable engine operating condition including a first input signal value indicative of the oxygen level in the exhaust gases produced by said engine;

a electronic engine control processor coupled to said sensors and programmed to execute open loop and closed loop neural network signal translation algorithms for converting said input signal values into a plurality of intermediate node output values and an output signal value indicating a controlled fuel command in accordance with a plurality of network weight values,

means responsive to said output signal value for adjusting the rate at which fuel is delivered to said engine to achieve a controlled air/fuel mixture ratio;

data storage means coupled to said control processor for storing said input signal values, said intermediate node output values, said output signal value, and said weight values; and

a training processor external to said control processor and coupled to said data storage means for modifying said weight values, said training processor including;

monitoring means responsive to said input signal values for comparing said controlled air/fuel mixture ratio with a predetermined desired air/fuel mixture ratio which provides substantially stoichiometric engine operation, and

means responsive to said monitoring means for varying at least selected ones of said weight values such that said controlled air/fuel mixture ratio more nearly matches said predetermined desired air/fuel mixture ratio.

8. Apparatus for controlling a vehicle engine as set forth in claim 7 wherein said open loop neural network signal translation algorithm translates a first set of said input signals into said output signal indicating a base transient air fuel ratio, and wherein said closed loop neural network signal translation algorithm translates a second set of said input signals into a closed loop correction value, and wherein said control processor further includes means for forming said output value indicating said controlled air/fuel mixture ratio from the additive combination of said base transient fuel command and said closed loop correction value.

9. Apparatus as set forth in claim 8 wherein said second set of input signal values includes said first input signal value indicative of the oxygen level in the exhaust gases produced by said engine, wherein said vehicle is equipped with a throttle whose position is manipulatable by the drive, and wherein said first set of input signal values includes at least a second input signal value indicative of the position of said throttle.

10. Apparatus for controlling said vehicle engine as set forth in claim 9 wherein said means for varying at least selected ones of said weight values comprises:

dynamic gradient processing means for indicating the rates of change of said controlled air/fuel mixture ratio with respect to selected ones of said input signal values, and

means for varying said weights in accordance with said rates of change such that said controlled air/fuel mixture ratio more nearly matches said desired air/fuel mixture ratio.

11. Apparatus for controlling said vehicle engine as set forth in claim 10 wherein said means for varying said weights employs a decoupled extended Kalman filter training algorithm for varying said weights in accordance with said rates of change.

12. Apparatus as set forth in claim 11 wherein said dynamic gradient processing means indicates the rates of change of said controlled air/fuel mixture ratio with respect to selected ones of said input signal values by processing said input signal values by means of at least one predetermined identification neural network having weights predetermined by previous offline processing to model the relationship between said input signals and said controlled air/fuel mixture ratio.

13. The method of training a neural network to control the air/fuel mixture ratio of the cumbustibles supplied to the intake of an internal combustion engine, said neural network being implemented by an electronic engine control processor connected to receive input signal values indicative of the operating state of said engine and being further connected to supply output signals to control the operation of said engine, said method comprising the steps of:

interconnecting an external training processor to said electronic engine control processor such that said external training processor can access said input signal values,

generating and storing a data structure consisting of an initial set of neural network weight values,

operating a representative internal combustion engine and its connected electronic engine control processor over a range of operating conditions,

concurrently with the operation of said representative engine, executing a neural network control program on said external training processor to process said input signal values into an output fuel rate control value in accordance with the values stored in said data structures,

concurrently with the operation of said representative engine, utilizing said fuel rate control value to control the rate at which fuel is delivered to said representative engine,

concurrently with the operation of said representative engine, executing a neural network training program on said external training processor to progressively alter at least selected values in said data structure to modify the results produced during the execution of said neural network training program,

evaluating the operation of said representative engine to indicate when a desired operating behavior is achieved, and

utilizing the values in said data structure at the time said desired operating behavior is achieved to control the execution of said neural network control program on said electronic engine control processor to control production engines corresponding to said representative engine.