Identification device, identification method, and identification program

Abstract

The challenge is to further improve the accuracy of individual battery identification using the magnetic flux density generated by the battery. [Solution] The identification device includes a feature extraction processing unit that inputs the measured magnetic flux density to a machine learning model that has been trained using the generated pseudo-data, removes disturbances which are noise dependent on the relative positions between each magnetic sensor included in the plurality of magnetic sensors, and extracts feature quantities which are information that represents individual differences in batteries, and an identification unit that identifies individual batteries based on the extracted feature quantities.

Claims

1. A generation unit generates pseudo-data by superimposing disturbances caused by physical phenomena occurring inside and around the battery onto data representing the magnetic flux density generated from the battery. A measuring unit that measures the magnetic flux density generated from the battery to be identified using multiple magnetic sensors, A feature extraction processing unit inputs the measured magnetic flux density into a machine learning model that has been trained using the generated pseudo-data, removes disturbances which are noise dependent on the relative positions between each of the multiple magnetic sensors, and extracts feature quantities which are information that represents individual differences in the battery. Based on the extracted feature quantities, an identification unit identifies the individual battery. An identification device equipped with the following features.

2. The generating unit is The data is generated using a cell model that calculates the magnetic flux density at any position in the battery. The disturbance is generated using a disturbance model that calculates disturbances at any location in the battery. The identification device according to claim 1.

3. The generating unit is The pseudo-data is generated using the disturbance model, which is created by combining multiple types of pseudo-disturbances representing noise generated inside the battery and noise generated around the battery. The identification device according to claim 2.

4. The aforementioned disturbance is a combination of multiple types of disturbances. The identification device according to claim 1.

5. A differential processing unit that generates a first magnetic flux density by removing components attributable to individual batteries from the aforementioned magnetic flux density, The feature extraction processing unit extracts feature quantities that represent individual differences in the batteries from the first magnetic flux density. The identification device according to claim 1.

6. Extracting the aforementioned features means From the first magnetic flux density, disturbances and noise are removed, and the feature quantities are extracted. The identification device according to claim 5, which includes the following:

7. Identifying individual batteries is A second feature quantity extracted from a second magnetic flux density previously measured for the battery, and the individual battery are identified based on the aforementioned feature quantity. The identification device according to claim 5 or 6, which includes the following:

8. Identifying individual batteries is A third feature quantity extracted from a third magnetic flux density previously measured, including a second battery other than the aforementioned battery, and the individual battery are identified based on the aforementioned feature quantity. The identification device according to claim 5 or 6, which includes the following:

9. Identifying individual batteries is An individual determination device identifies each battery based on the feature quantity, using an individual determination device that identifies each battery based on a second magnetic flux density previously measured for the battery, or a common determination device that identifies each battery based on a third magnetic flux density previously measured for a second battery other than the aforementioned battery. This includes, The identification device according to claim 5 or 6, further comprising a determination switching unit that switches the determination device used to identify an individual battery based on the battery identifier or a user designation to the individual determination device or the common determination device.

10. Extracting the aforementioned features means The feature quantities are extracted from the first magnetic flux density using a machine learning model. This includes, The identification device according to claim 5 or 6, further comprising an update unit that trains the machine learning model using a fourth magnetic flux density, generated by removing a component attributable to the individual battery from a predetermined magnetic flux density by the difference processing unit, as training data.

11. Training the aforementioned machine learning model is From the predetermined magnetic flux densities, select the predetermined magnetic flux density where the disturbance is less than or equal to a predetermined threshold, The machine learning model is trained using the fourth magnetic flux density, which is generated by removing components attributable to individual batteries from the predetermined magnetic flux density with minimal disturbances, as training data. The identification device according to claim 10, which includes the following:

12. Training the aforementioned machine learning model is The machine learning model is trained using the fourth magnetic flux density, which is generated by removing the component attributable to the individual battery from the pseudo-generated magnetic flux density by the difference processing unit, as training data. The identification device according to claim 10, which includes the following:

13. Identifying individual batteries is Using a machine learning model, identify individual batteries based on the aforementioned features. This includes, The identification device according to claim 5 or 6, further comprising an update unit that trains the machine learning model using a fourth feature representing individual differences in the battery, extracted by the feature extraction unit from a fifth magnetic flux density generated by removing a component attributable to the individual battery from a predetermined magnetic flux density by the difference processing unit, as training data.

14. The common classifier includes a machine learning model that identifies individual batteries from the feature quantities. The identification device according to claim 9, further comprising an update unit that trains the machine learning model using a fourth feature representing individual differences in the battery, extracted by the feature extraction unit from a fifth magnetic flux density generated by removing a component attributable to the individual battery from a predetermined magnetic flux density by the difference processing unit, as training data.

15. The individual classifier includes a machine learning model that identifies individual batteries from the feature quantities, The identification device according to claim 9, further comprising an update unit that trains the machine learning model using a fifth feature representing individual differences in the battery, extracted by the feature extraction unit from a sixth magnetic flux density generated by removing components attributable to individual batteries from the magnetic flux density by the difference processing unit, as training data.

16. The sixth magnetic flux density is generated from the previously measured magnetic flux density of the battery associated with the battery identifier, Training the aforementioned machine learning model is A predetermined threshold is set for the distance of the feature vectors that represent individual differences in the batteries in the machine learning model. The fifth feature quantity is divided into two in a predetermined ratio, One of the divided fifth features is input to the machine learning model to identify the individual battery, From the seventh magnetic flux density generated by removing components attributable to individual batteries by the difference processing unit from the magnetic flux density previously measured for the battery, which is associated with an identifier different from the aforementioned battery identifier, the sixth feature quantity representing the individual difference of the battery, extracted by the feature quantity extraction processing unit, is input into the machine learning model to identify the individual battery. The first proportion of the fifth feature quantity that was misclassified as a different individual, and the second proportion of the sixth feature quantity that was classified as the same individual are calculated. If the first and second percentages do not meet the predetermined conditions, the threshold is updated. The identification device according to claim 15, which includes the following:

17. Switching to the individual determination device or the common determination device means Based on the identification result obtained by identifying the individual battery from the magnetic flux density for evaluation using the individual and common determination devices respectively, the system switches to either the individual or common determination device. The identification device according to claim 9, which includes the following:

18. Pseudo-data is generated by superimposing disturbances caused by physical phenomena occurring inside and around the battery onto data representing the magnetic flux density generated from the battery. The magnetic flux density emanating from the battery to be identified is measured using multiple magnetic sensors. The measured magnetic flux density is input to a machine learning model that has been trained using the generated pseudo-data, and disturbances, which are noise dependent on the relative position between each of the multiple magnetic sensors, are removed, and feature quantities, which are information representing individual differences in the battery, are extracted. Based on the extracted features, identify the individual battery. An identification method in which an identification device performs the processing.

19. Pseudo-data is generated by superimposing disturbances caused by physical phenomena occurring inside and around the battery onto data representing the magnetic flux density generated from the battery. The magnetic flux density emanating from the battery to be identified is measured using multiple magnetic sensors. The measured magnetic flux density is input to a machine learning model that has been trained using the generated pseudo-data, and disturbances, which are noise dependent on the relative position between each of the multiple magnetic sensors, are removed, and feature quantities, which are information representing individual differences in the battery, are extracted. Based on the extracted features, identify the individual battery. An information processing program that causes an identification device to perform a process.

Images