Battery diagnosis method and battery diagnosis apparatus

1. A battery diagnosis method performed by a computing device that drives a battery prediction model trained to predict battery state information, the battery diagnosis method comprising:

receiving time series data including at least one of voltages, currents, and temperatures of a battery sequentially measured for a certain time period;

receiving non-time series data including battery impedance measured at a certain time point; and

predicting the battery state information by inputting the time series data and the non-time series data to the battery prediction model,

wherein, training data for the battery prediction model comprises time series training data of at least one of a voltage, a current, and a temperature measured for a certain time period for a plurality of battery samples, non-time series training data includes impedances of the plurality of battery samples, and validation data includes battery state information, and

the battery prediction model is a supervised learning model made by comparing the validation data with prediction data based on the time series training data and the non-time series training data,

wherein the battery prediction model comprises a recursive neural network and a fully connected layer,

wherein the predicting comprises:

generating a first feature value by inputting the time series data to the recursive neural network;

generating a second feature value by concatenating the first feature value with the non-time series data; and

inputting the second feature value to the fully connected layer.

2. The battery diagnosis method of claim 1 , wherein the non-time series data is impedance information measured by applying at least two sine wave signals of different frequencies to the battery.

3. The battery diagnosis method of claim 1 , further comprising optimizing a hyperparameter of the battery prediction model,

wherein the optimizing comprises:

training the battery prediction model by using pre-defined training data and a plurality of pre-defined candidate hyperparameters;

selecting at least two top candidate hyperparameters as parent hyperparameters, based on fitness evaluation of the battery prediction model trained based on each of the candidate hyperparameters;

generating at least one new candidate hyperparameter through intersection or transformation between the parent hyperparameters; and

repeating a process of selecting parent hyperparameters again based on fitness evaluation of the battery prediction model trained based on the new candidate hyperparameter and generating a new candidate hyperparameter, until a predefined condition is met or a predefined number of times is reached.

4. A non-transitory computer-readable recording medium having recorded thereon a computer program, which, when executed by a computer, performs the method of claim 1 .