Control device of internal combustion engine

Patent metadata

Field	Value
Publication number	US-10634081-B2
Application number	US-201815922151-A
Country	US
Kind code	B2
Filing date	Mar 15, 2018
Priority date	Feb 5, 2018
Publication date	Apr 28, 2020
Grant date	Apr 28, 2020

Abstract

A control device of an internal combustion engine using a neural network. When a value of an operating parameter of the engine is outside a preset range, the number of nodes of a hidden layer one layer before an output layer of the neural network is increased and training data obtained by actual measurement with respect to a newly acquired value of an operating parameter of the engine is used to learn a weight of the neural network so that a difference between the output value changing corresponding to the value of the operating parameter of the engine and training data corresponding to the value of the operating parameter of the engine becomes smaller.

First claim

The invention claimed is:

1. A control device of an internal combustion engine having an electronic control unit, said electronic control unit comprises:
an interface configured to acquire values of operating parameters of the engine,
a memory configured to store:
a neural network comprised of an input layer, hidden layers, and an output layer; and
training data obtained by actually measuring values of the operating parameters of the engine in preset ranges, wherein the preset ranges of the values of the operating parameters of the engine are set in advance and a number of nodes of the hidden layers of the neural network corresponding to the preset ranges of values of the operating parameters of the engine are set in advance, and

a processor configured to:
input the values of the operating parameters of the engine to the input layer;
when a newly acquired value of an operating parameter of the engine is inside a preset range, use the stored training data to learn a weight of the neural network so that a difference between an output value changing according to the newly acquired value of the operating parameter of the engine and the training data corresponding to the newly acquired value of the operating parameter of the engine becomes smaller;
when the newly acquired value of the operating parameter of the engine is outside the preset range:
increase a number of nodes of a hidden layer one layer before the output layer of the neural network to corresponding to a number or data density of training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine, and
use the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine to learn the weight of the neural network so that the difference between the output value changing according to the newly acquired value of the operating parameter of the engine and the training data corresponding to the newly acquired value of the operating parameter of the engine becomes smaller; and

output an output value corresponding to the value of the operating parameter of the engine using the neural network with the learned weight.

2. The control device of an internal combustion engine according to claim 1 , wherein the processor is configured to, when the newly acquired value of the operating parameter of the engine is outside the preset range;
increase the number of nodes of the hidden layer one layer before the output layer, and
use the stored training data and the training data obtained by actual measurement corresponding to the newly acquired value of the operating parameter of the engine to learn the weight of the neural network so that the difference between the output value changing according to the newly acquired value of the operating parameter of the engine and the training data corresponding to the newly acquired value of the operating parameter of the engine becomes smaller.

3. The control device of an internal combustion engine according to claim 1 , wherein the processor is configured to, when the newly acquired value of the operating parameter of the engine is outside the preset range and the data density of the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine is higher than a preset density;
increase the number of nodes of the hidden layer one layer before the output layer of the neural network, and
use the stored training data and the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine to learn the weight of the neural network so that the difference between the output value changing according to the newly acquired value of the operating parameter of the engine and the training data corresponding to the newly acquired value of the operating parameter of the engine becomes smaller.

4. The control device of an internal combustion engine according to claim 3 , wherein the processor is configured to increase the number of nodes of the hidden layer one layer before the output layer of the neural network such that the greater a number of newly acquired values of the operating parameter of the engine falling in a range, the more the number of nodes of the hidden layer one layer before the output layer of the neural network is increased.

5. The control device of an internal combustion engine according to claim 1 , wherein the processor is configured to, when the newly acquired value of the operating parameter of the engine is outside the preset range and the data density of the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine is higher than a preset density;
increase the number of nodes of the hidden layer one layer before the output layer of the neural network by preparing a new neural network, and
use the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine to learn a weight of the new neural network so that the difference between the output value changing according to the newly acquired value of the operating parameter of the engine and the training data corresponding to the newly acquired value of the operating parameter of the engine becomes smaller.

6. The control device of an internal combustion engine according to claim 5 , wherein
the operating parameters of the engine include a plurality of different engine operating parameters,
the preset ranges include learned areas that are divided corresponding to the values of the engine operating parameters and are finished being learned,
a neural network is prepared for each of the learned areas, and
the processor is configured to, when the newly acquired value of the operating parameter of the engine is inside a new area that is outside the learned areas and a data density in the new area of the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine is higher than the preset density:
prepare the new neural network for the new area, and
use the training data obtained by actual measurement with respect to the newly acquired value of the operating parameter of the engine to learn a weight of the new neural network prepared for the new area so that the difference between the output value changing according to the newly acquired value of the operating parameter of the engine and the training data corresponding to the newly acquired value of the operating parameter of the engine becomes smaller.

7. The control device of an internal combustion engine according to claim 6 , wherein the processor is configured to set the number of nodes of the hidden layer one layer before the output layer of the new neural network prepared for the new area based on a mean value of numbers of nodes of hidden layers of the neural networks prepared for the learned areas that are positioned around the new area.

8. The control device of an internal combustion engine according to claim 7 , wherein the processor is configured to increase the number of nodes of the hidden layer one layer before the output layer such that the greater a number of input data corresponding to the new area, the more the number of nodes of the hidden layer one layer before the output layer of the new neural network prepared for the new area is increased.

9. A control device of an internal combustion engine using a neural network, wherein preset ranges of values of operating parameters of the engine are set in advance and a number of nodes of hidden layers of the neural network correspondin

Assignees

• Toyota Motor Co Ltd

Inventors

• Kitagawa Eiki
• Ehara Masato

Classifications

• F02D2200/101

  Engine speed · CPC title

• G06N20/00

  Machine learning · CPC title

• F02D2200/0414

  Air temperature · CPC title

• F02D41/1461

  of the exhaust gases emitted by the engine · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title