Machine learning device of amount of unburned fuel, machine learning method, learned model, electronic control unit, method of production of electronic control unit, and machine learning system

Patent metadata

Field	Value
Publication number	US-10991174-B2
Application number	US-201816007446-A
Country	US
Kind code	B2
Filing date	Jun 13, 2018
Priority date	Apr 20, 2018
Publication date	Apr 27, 2021
Grant date	Apr 27, 2021

Abstract

A machine learning device of an amount of unburned fuel using a neural network in which the correlation functions showing correlations between parameters and an amount of unburned fuel are found for parameters relating to operation of an internal combustion engine, and the parameters with strong degrees of correlation with the amount of unburned fuel exhausted from the engine are selected from among the parameters based on the correlation functions. The amount of unburned fuel is learned by using the neural network from the selected parameters and amount of unburned fuel.

First claim

The invention claimed is:

1. A machine learning device configured for estimating an amount of unburned fuel by performing a machine learning function using training data, the machine learning device comprising an electronic control unit configured to receive signals from one or more sensors associated with an internal combustion engine, retrieve parameters relating to operation of the internal combustion engine, the training data, and control programs from a memory, and execute the control programs to perform the machine learning function, wherein the machine learning function comprises:
selecting first parameters having effects on the amount of unburned fuel exhausted from the internal combustion engine from among the parameters relating to operation of the internal combustion engine,
calculating the amount of unburned fuel based on the selected first parameters,
determining a correlation function showing a correlation between each of the selected first parameters and the amount of unburned fuel, wherein said correlation function comprises a linear function showing a relation between values of the selected first parameters and the amount of unburned fuel calculated using the least squares method,
calculating a variance of the amount of unburned fuel from an average value of the amount of unburned fuel and said linear function, and
selecting second parameters with strong degrees of correlation with the amount of unburned fuel exhausted from the internal combustion engine from among the selected first parameters based on said variance.

2. The machine learning device according to claim 1 , wherein the machine learning function further comprises:
calculating contribution ratios of the selected first parameters based on said variance, wherein the second parameters are selected based on said contribution ratios.

3. The machine learning device according to claim 2 , wherein the selected second parameters are a group of the parameters with the highest contribution ratios.

4. The machine learning device according to claim 2 , wherein the selected second parameters are the parameters with contribution ratios greater than a preset lower limit value.

5. The machine learning device according to claim 4 , wherein a degree of change of said variance caused by a fluctuation in a true amount of unburned fuel is set as the lower limit value.

6. The machine learning device according to claim 2 , wherein the machine learning function further comprises:
determining a cumulative contribution ratio by cumulatively adding the contribution ratios in order from the highest contribution ratio of a parameter that reaches a preset upper limit value, wherein the parameters contributing to the cumulative contribution ratio are selected as the second parameters.

7. The machine learning device according to claim 1 , wherein the selected second parameters are based on a mounting of any sensor for detecting the parameter value and a cost of the sensor.

Assignees

• Toyota Motor Co Ltd

Inventors

• Takasu Yusuke
• Oyagi Hiroshi
• Nagasaka Keisuke
• Kaneko Tomohiro

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• F02D41/1459

  the characteristics being a hydrocarbon content or concentration · CPC title

• F02D41/1405

  Neural network control · CPC title