Method for generating neural network model and control device using neural network model

1 . A method for generating a neural network model, the method comprising:

wherein an engine control device includes:

a processor, and

an engine inference device constituted by the neural network model, wherein the processor executes a neural network program and causes the neural network model to input time series data of a manipulating variable, which is also input to an actual engine, and calculate time series data of a predicted controlled variable of the actual engine, and

the engine control device calculates the manipulating variable based on the predicted controlled variable of the actual engine that is calculated by the neural network model,

acquiring, by the processor, first time series data that includes time series data of the manipulating variable and the controlled variable being obtained from the actual engine during operation, the first time series data having a first period that is shorter than a calculation period of the neural network model when the processor executes the neural network program;

extracting, by the processor, from the first time series data, a plurality of sets of second time series data having a second period that is longer than the first period, wherein the plurality of sets of second time series data includes the time series data of the manipulating variable and the controlled variable of the actual engine; and

executing, by the processor, training of the neural network model using training data that include the plurality of sets of second time series data to update parameters of the neural network model so as to reduce difference between:

the time series data of the predicted controlled variable of the actual engine that is calculated, based on the time series data of the manipulating variable in the training data, by the neural network model, and

the time series data of the controlled variable of the actual engine in the training data.

2 . The method for generating a neural network model according to claim 1 , wherein, during the extraction process, a plurality of sets of second time series data having different phases are extracted from the first time series data.

3 . The method for generating a neural network model according to claim 1 , wherein, during the extraction process, data acquired at time points shifted forward or backward from a time point corresponding to the second period are extracted from the first time series data, such that a plurality of the second time series data having a third period that is shorter or longer than the second period and shorter than the first period are extracted.

4 . The method for generating a neural network model according to claim 1 , wherein the acquisition process includes:

acquiring third time series data having the calculation period of the neural network model from the actual engine during operation; and

generating the first time series data by performing interpolation between data that are adjacent on a time axis of the third time series data so as to add a plurality of time series data between the data that are adjacent to the third time series data.

5 . The method for generating a neural network model according to claim 1 , further comprising:

evaluating the trained neural network model following the training process; and

increasing the amount of the second time series data by executing the extraction process when an evaluation acquired during the evaluation process does not reach a reference level,

wherein, during the training process, training of the neural network model is executed using training data including the increased second time series data.

6 . The method for generating a neural network model according to claim 1 , wherein the second period is the calculation period of the neural network model.

7 . The method for generating a neural network model according to claim 6 , wherein the time series data of the input are constituted by a chirp signal having a varying frequency.

8 . The method for generating a neural network model according to claim 1 , wherein the neural network model is a recurrent neural network.

9 . An engine control device comprising:

a processor, and

an engine inference device constituted by the neural network model, wherein the processor executes a neural network program and causes the neural network model to input time series data of a manipulating variable, which is also input to an actual engine, and calculate time series data of a predicted controlled variable of the actual engine, and

the engine control device calculates the manipulating variable based on the predicted controlled variable of the actual engine that is calculated by the neural network model,

wherein the neural network model is generated by the steps including:

acquiring, by the processor, first time series data that includes time series data of the manipulating variable and the controlled variable being obtained from the actual engine during operation, the first time series data having a first period that is shorter than a calculation period of the neural network model when the processor executes the neural network program;

extracting, by the processor, from the first time series data, a plurality of sets of second time series data having a second period that is longer than the first period, wherein the plurality of sets of second time series data includes the time series data of the manipulating variable and the controlled variable of the actual engine; and

executing, by the processor, training of the neural network model using training data that include the plurality of sets of second time series data to update parameters of the neural network model so as to reduce difference between:

the time series data of the predicted controlled variable of the actual engine that is calculated, based on the time series data of the manipulating variable in the training data, by the neural network model, and

the time series data of the controlled variable of the actual engine in the training data.