Method for determining measurement location groups, device for determining measurement location groups, and method for generating machine learning models.

Abstract

A method for determining a set of measurement locations, a device for determining such locations, and a method for generating a machine learning model are provided, which search for an appropriate set of measurement locations necessary for estimating the physical quantities of a fluid quickly and comprehensively. [Solution] The method for determining the measurement location group includes: a first step (S1) of creating a dataset of feature quantities and physical quantity distributions used for training the estimator; a second step (S2) of generating an estimator for estimating the distribution of physical quantities using the created dataset; a third step (S3) of calculating the importance of feature quantities measured at the measurement location group using the trained estimator; a fourth step (S4) of selecting a portion of the measurement location group that contributes highly to the distribution of the estimated physical quantity using the importance, and making the selected portion of the measurement location group a new measurement location group; and a determination step (S5) of repeating the first step, the second step, the third step, and the fourth step if the estimation accuracy of the estimator is not worse than a predetermined accuracy.

Claims

1. A method for determining a group of measurement locations for measuring characteristic quantities of a fluid, in order to estimate the distribution of physical quantities of the fluid, A first step is to create a dataset which is a set of paired data of the distribution of the feature quantities and the physical quantities used to train the estimator, A second step is to generate an estimator for estimating the distribution of the physical quantity using the created dataset, A third step involves using a trained estimator to calculate the importance of the features measured in the measurement location group, A fourth step involves using the aforementioned importance to select a group of measurement locations that have a high contribution to the estimated distribution of the physical quantity, and designating the selected group of measurement locations as a new group of measurement locations. A method for determining a group of measurement positions, comprising a determination step of determining whether the first step, the second step, the third step, and the fourth step are repeated if the estimation accuracy of the estimator is not worse than a predetermined accuracy.

2. The method for determining a group of measurement positions according to claim 1, wherein the fourth step involves extracting a group of measurement positions by multiplying by a reduction rate in a geometric progression when the number of measurement positions is greater than or equal to a predetermined number, and extracting a group of measurement positions by subtracting the reduction amount in an arithmetic progression when the number of measurement positions is less than a predetermined number.

3. A device for determining a group of measurement locations for measuring characteristic quantities of a fluid, in order to estimate the distribution of physical quantities of the fluid, A creation unit creates a dataset which is a set of paired data of the distribution of the feature quantities and physical quantities used for training the estimator, A learning unit that generates an estimator for estimating the distribution of the physical quantity using the created dataset, A calculation unit that uses a pre-trained estimator to calculate the importance of features measured in the measurement location group, An extraction unit selects a portion of measurement locations that have a high contribution to the estimated distribution of the physical quantity using the aforementioned importance, and sets the selected portion of measurement locations as a new measurement location group. A device for determining a group of measurement positions, comprising a determination unit that causes the processing of the creation unit, the processing of the learning unit, the processing of the calculation unit, and the processing of the extraction unit to be repeated if the estimation accuracy of the estimator is not worse than a predetermined accuracy.

4. A method for generating a machine learning model, comprising generating a machine learning model as the estimator by machine learning using training data that includes fluid feature quantities measured at the measurement position group determined by the method for determining the measurement position group according to claim 1 or 2.

Images