Battery state prediction apparatus and method of operating same

Abstract

A battery state prediction apparatus according to one embodiment disclosed in the present document may include: a generation unit that is trained based on battery data and features included in the battery data and generates a plurality of machine learning models to predict a gas generation amount of a battery; and a controller that obtains a plurality of pieces of prediction data predicting a gas generation amount of the battery by applying the battery data to the plurality of machine learning models, and predicts the gas generation amount of the battery based on the plurality of pieces of prediction data.

Description

Technical Field

[0001] Cross - Reference to Related Applications

[0002] This application claims the priority and benefits of Korean Patent Application No. 10-2022-0178733, filed with the Korean Intellectual Property Office on December 19, 2022, and Korean Patent Application No. 10-2023-0180528, filed with the Korean Intellectual Property Office on December 13, 2023, the entire contents of which are incorporated herein by reference. Technical Field

[0004] Embodiments disclosed herein relate to a battery state prediction device and an operation method thereof. Background Art

[0005] Electric vehicles are powered from the outside to charge the battery, and then the motor is driven by the voltage charged in the battery to obtain power. The battery of an electric vehicle may have heat generated therein through a chemical reaction occurring during the charging and discharging processes, and the heat may damage the performance and lifespan of the battery, leading to a phenomenon of generating internal gas in the battery.

[0006] The gas generated inside the battery may act as a resistor or may be a cause of deformation of the battery module and battery pack, thereby increasing the product defect rate. Accordingly, a battery management system (BMS) that monitors the temperature, voltage, and current of the battery is driven to predict whether gas is generated inside the battery and the amount of gas generation.

[0007] The battery management system may predict the amount of gas generation in the battery by training battery data in an artificial intelligence model that analyzes the state of the battery. However, when the same battery data is input as input data Input into the artificial intelligence model, the battery management device may always obtain the same output data Output, without being able to reflect the actual distribution of different amounts of gas generation actually generated by the battery in the same environment. Summary of the Invention

[0008] Technical Problem

[0009] Embodiments disclosed herein aim to provide a battery state prediction device and an operation method thereof in which prediction data of the amount of gas generation in a probability distribution form can be obtained by using multiple artificial intelligence models that predict the amount of gas generation in the battery.

[0010] The technical problems of the embodiments disclosed herein are not limited to the above technical problems, and other unmentioned technical problems will be clearly understood by those of ordinary skill in the art from the following description.

[0011] Technical Solution

[0012] A battery state prediction device according to an embodiment disclosed herein includes: a generation unit configured to generate a plurality of machine learning models trained based on battery data and features included in the battery data to predict a gas generation amount of a battery; and a controller configured to obtain a plurality of pieces of prediction data for predicting the gas generation amount of the battery by applying the battery data to the plurality of machine learning models, and predict the gas generation amount of the battery based on the plurality of pieces of prediction data.

[0013] According to an embodiment, the plurality of machine learning models may include a plurality of deep neural network (DNN) models, and the controller may further be configured to obtain a plurality of pieces of prediction data by applying the battery data to the plurality of DNN models.

[0014] According to an embodiment, the battery data may include the temperature, state of charge (SOC), and state of health (SOH) of the battery measured cumulatively, and the features of the battery data may include at least one or more of the electrode type, assembly process, and separator type of the battery.

[0015] According to an embodiment, the controller may further be configured to apply weights to each of the plurality of pieces of prediction data based on the weights applied based on the features of the battery data during the training of each DNN model that generates each of the plurality of pieces of prediction data, and input the plurality of pieces of prediction data to which the weights are applied into an ensemble learning model to generate prediction data of the gas generation amount in the form of a probability distribution.

[0016] According to an embodiment, the controller may further be configured to calculate an average value of the prediction data of the gas generation amount in the form of a probability distribution, and determine the accuracy of the plurality of machine learning models by comparing the average value with a pre-stored measured value of the gas generation amount of the battery.

[0017] An operation method of a battery state prediction device according to an embodiment disclosed herein includes the steps of: generating a plurality of machine learning models trained based on battery data and features included in the battery data to predict a gas generation amount of a battery; obtaining a plurality of pieces of prediction data for predicting the gas generation amount of the battery by applying the battery data to the plurality of machine learning models; and predicting the gas generation amount of the battery based on the plurality of pieces of prediction data.

[0018] According to an embodiment, the plurality of machine learning models may include a plurality of deep neural network (DNN) models, and the step of obtaining a plurality of pieces of prediction data for predicting the gas generation amount of the battery by applying the battery data to the plurality of machine learning models may include obtaining a plurality of pieces of prediction data by applying the battery data to the plurality of DNN models.

[0019] According to one embodiment, the step of generating multiple machine learning models for predicting the gas generation amount of a battery based on battery data may include: generating multiple machine learning models based on battery data including the temperature, state of charge (SOC), and state of health (SOH) of the battery measured cumulatively, and the features of the battery data may include at least one or more of the electrode type, assembly process, and separator type of the battery.

[0020] According to one embodiment, the step of predicting the gas generation amount of a battery based on multiple prediction data may include: applying weights to each of the multiple prediction data based on the weights applied based on the features of the battery data during the training of the DNN model for generating each of the multiple prediction data; and inputting the multiple prediction data to which the weights are applied into an ensemble learning model to generate prediction data of the gas generation amount in the form of a probability distribution.

[0021] According to one embodiment, the step of predicting the gas generation amount of a battery based on multiple prediction data may include: calculating the average value of the prediction data of the gas generation amount in the form of a probability distribution; and determining the accuracy of the multiple machine learning models by comparing the average value with the measured value of the gas generation amount of the pre-stored battery.

[0022] Advantageous Effects

[0023] The battery state prediction device and its operation method according to the embodiments disclosed herein can obtain prediction data of the gas generation amount in the form of a probability distribution by using multiple artificial intelligence models. Brief Description of the Drawings

[0024] Figure 1 Illustrates a battery pack according to one embodiment disclosed herein.

[0025] Figure 2 Is a diagram for generally describing a battery state prediction device according to one embodiment disclosed herein.

[0026] Figure 3 Is a block diagram showing the configuration of a battery state prediction device according to one embodiment disclosed herein.

[0027] Figure 4 Is a diagram for describing the operation of a generation unit according to one embodiment disclosed herein.

[0028] Figure 5 Is a diagram for describing the operation method of a controller according to one embodiment disclosed herein.

[0029] Figure 6 Is a flowchart of an operation method of a battery state prediction device according to one embodiment disclosed herein.

[0030] Figure 7 is a block diagram showing the hardware configuration of a computing system for implementing a battery state prediction device according to an embodiment disclosed herein. Detailed Embodiment

[0031] Hereinafter, some embodiments disclosed in this document will be described in detail with reference to the exemplary drawings. When adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals, even if they are shown in different drawings. In addition, when determining that a detailed description of a related known configuration or function interferes with the understanding of the embodiments disclosed in this document, its detailed description will be omitted.

[0032] To describe the components of the embodiments disclosed herein, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only used to distinguish one component from another component, and do not limit the component to the essence, order, sequence, etc. of the component. The terms used herein (including technical and scientific terms) have the same meaning as those commonly understood by those skilled in the art, as long as these terms are not defined differently. Generally, terms defined in a general dictionary should be interpreted as having the same meaning as the context of the related art, and should not be interpreted as having an ideal or exaggerated meaning, unless they are clearly defined in this document.

[0033] Figure 1 Illustrates a battery pack according to an embodiment disclosed herein.

[0034] Refer to Figure 1 , a battery pack 1000 according to an embodiment disclosed herein may include a battery module 100, a battery state prediction device 200, and a relay 300.

[0035] The battery module 100 may include a plurality of battery cells 110, 120, 130, and 140. Although a plurality of battery cells are illustrated as four in Figure 1 , the present disclosure is not limited thereto, and the battery module 100 may include n battery cells (n is a natural number equal to or greater than 2).

[0036] The battery module 100 may supply power to a target device (not shown). To this end, the battery module 100 may be electrically connected to the target device. Herein, the target device may include an electrical device, an electronic device, or a mechanical device that operates by receiving power from the battery pack 1000 including a plurality of battery cells 110, 120, 130, and 140, and the target device may be, for example, an electric vehicle (EV) or an energy storage system (ESS), but is not limited thereto.

[0037] Each of the plurality of battery cells 110, 120, 130, and 140 is a basic unit of a battery that can be used by charging and discharging electric energy. The plurality of battery cells 110, 120, 130, and 140 can be lithium-ion (Li-ion) batteries, lithium-ion polymer batteries, nickel-cadmium (Ni-Cd) batteries, nickel-metal hydride (Ni-MH) batteries, etc., and are not limited thereto. In addition, although one battery module 100 is illustrated in Figure 1 , according to one embodiment, the battery module 100 can be configured in multiple numbers.

[0038] The battery state prediction device 200 can predict the gas generation amount of the plurality of battery cells 110, 120, 130, and 140 based on the temperature, current, voltage, state of charge (SOC), and state of health (SOH) data of the plurality of battery cells 110, 120, 130, and 140. The battery state prediction device 200 can predict the gas generation amount of the plurality of battery cells 110, 120, 130, and 140 based on the battery data A of the plurality of battery cells 110, 120, 130, and 140.

[0039] According to one embodiment, the battery state prediction device 200 can be implemented in the form of a battery management system (BMS). In addition, according to one embodiment, the battery state prediction device 200 can be installed on the battery management system.

[0040] Herein, the battery management system can manage and / or control the state and / or operation of the battery module 100. For example, the battery management system can manage and / or control the state and / or operation of the plurality of battery cells 110, 120, 130, and 140 included in the battery module 100. The battery management system can manage the charging and / or discharging of the battery module 100.

[0041] In addition, the battery management system can monitor the voltage, current, temperature, etc. of the battery module 100 and / or each of the plurality of battery cells 110, 120, 130, and 140 included in the battery module 100. Sensors or various measurement modules (not shown) for monitoring performed by the battery management system can be additionally installed in the battery module 100, the charge / discharge path, any position of the battery module 100, etc. The battery management device can calculate parameters (such as SOC or SOH, etc.) indicating the state of the battery module 100 based on measurement values such as the monitored voltage, current, temperature, etc.

[0042] The battery management system can control the operation of the relay 300. For example, the battery management system can short-circuit the relay 300 to supply power to the target device. When the charging device is connected to the battery pack 1000, the battery management system can short-circuit the relay 300.

[0043] The battery management system can calculate the cell balancing time for each of the multiple battery cells 110, 120, 130, and 140. Herein, the cell balancing time can be defined as the time required for balancing the battery cells. For example, the battery management device can calculate the cell balancing time based on the SOC, battery capacity, and balancing efficiency of each of the multiple battery cells 110, 120, 130, and 140.

[0044] Figure 2 is a diagram for generally describing a battery state prediction device according to an embodiment disclosed herein.

[0045] Referring to Figure 2 , the battery state prediction device 200 can extract a part of the battery data A and input it into the multiple machine learning models 211, 212, 213, and 214.

[0046] Although the multiple machine learning models are illustrated as four in Figure 2 , the present disclosure is not limited thereto, and the multiple machine learning models 211, 212, 213, and 214 can include n machine learning models (n is a natural number equal to or greater than 2).

[0047] The battery state prediction device 200 can obtain the battery data A of the multiple battery cells 110, 120, 130, and 140. In order to identify the gas generation amount of the multiple battery cells 110, 120, 130, and 140, the battery management system can obtain the battery data A, which includes the measured values of the battery from the voltage value when the SOC of the battery is 0% to the voltage value when the SOC of the battery is 100%. Accordingly, the battery state prediction device 200 can obtain the battery data A including the voltage, current, temperature, SOC, and SOH of the multiple battery cells 110, 120, 130, and 140 measured cumulatively during the charge / discharge period.

[0048] Herein, the battery data A can also include at least one of the characteristics of the battery data A including the separator type of the battery, the assembly process, the electrode type, etc. That is, the battery data A can also include at least one of the data related to what type of separator type the battery has, how the battery is assembled by which assembly process, and / or what type of electrode the battery has.

[0049] The battery state prediction device 200 can predict the gas generation amount of the battery by inputting battery data A into multiple machine learning models 211, 212, 213, and 214. In this context, machine learning refers to the technology used to predict a certain result by training a computer. Generally, the results of using machine learning include: the process of preparing training data for training the machine and training the computer in a manner suitable for the problem, the process of validating the model using test data, and the process of predicting results using the model with the validated model.

[0050] For machine learning, it is important that the training data well represents the features to be generalized by machine learning, and thus training data is generated using training data selectively limited according to certain criteria. For cases where the correlation between the features to be generalized by machine learning and the features of the training data is low, sampling noise may occur, and it is difficult to find the pattern in which the machine problem analysis model is embedded. As a result, regardless of the accuracy of the machine problem analysis model, the error of the model increases, and thus the reliability of the model is reduced. Therefore, machine learning techniques require time investment in evaluating training data and processing data to select the training data set.

[0051] The battery state prediction device 200 can extract at least a part of the battery data A and generate a training data set to generate and train multiple machine learning models 211, 212, 213, and 214. In this context, the multiple machine learning models 211, 212, 213, and 214 can refer to learning models capable of predicting the state of the battery including the gas generation amount of the battery based on the input battery data.

[0052] That is, the battery state prediction device 200 can generate multiple machine learning models 211, 212, 213, and 214 with the same structure based on one training data set generated by extracting at least a part of the battery data A. The battery state prediction device 200 can obtain the prediction data for predicting the battery gas generation amount of each of the multiple machine learning models 211, 212, 213, and 214 through the multiple machine learning models 211, 212, 213, and 214 generated based on one training data set generated by extracting at least a part of the battery data A.

[0053] The battery state prediction device 200 can finally obtain single gas generation amount prediction data C in the form of a probability distribution by combining prediction data for predicting the battery gas generation amount of each of the multiple machine learning models 211, 212, 213, and 214. Therefore, the battery state prediction device 200 can obtain output data Output of each of the machine learning models 211, 212, 213, and 214 with the same structure based on a single input data Input, and combine each output data to finally obtain single gas generation amount prediction data C in the form of a probability distribution, thereby reducing the learning error and improving the reliability.

[0054] The output data generated by the artificial intelligence model can include random errors and main effects. When the test is repeated multiple times based on a single input data, the artificial intelligence model can obtain random errors or white noise as different output data. In this article, when the artificial intelligence model repeats the same test enough times, the mean value of the random errors can converge to 0 and the artificial intelligence model can only obtain main effect data.

[0055] The battery state prediction device 200 can simultaneously obtain multiple output data by inputting the same input data into the multiple machine learning models 211, 212, 213, and 214 to establish the same test environment, thereby obtaining the effect of repeated testing. That is to say, the battery state prediction device 200 can input at least a part of the same battery data A into the multiple machine learning models 211, 212, 213, and 214 with the same structure, thereby predicting the battery gas generation amount with high precision corresponding to the main effect data.

[0056] Figure 3 is a block diagram showing the structure of a battery state prediction device according to an embodiment disclosed herein, and Figure 4 is a diagram for describing the operation of a generation unit according to an embodiment disclosed herein.

[0057] Hereinafter, with reference to Figure 3 and Figure 4 , the configuration and operation of the battery state prediction device 200 will be described in detail.

[0058] With reference to Figure 3 , the battery state prediction device 200 can include a generation unit 210 and a controller 220.

[0059] The generation unit 210 may collect battery data A. For example, the battery data A may be defined as a value recording the change in the battery state from the discharge state to the fully charged state or from the fully charged state to the discharge state of a plurality of battery cells 110, 120, 130, and 140. For example, the battery data A may include the voltage, current, temperature, SOC, and SOH of the battery measured cumulatively. Herein, the SOH may include the capacity degradation and resistance degradation of the battery. In addition, the battery data A may further include at least one of data related to what type of separator the battery has, by which assembly process the battery is assembled, and / or what type of electrode the battery has.

[0060] The generation unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 by extracting at least a part of the battery data A as a training data set. For example, the generation unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 by extracting 80% of the battery data A as a training data set.

[0061] According to one embodiment, the plurality of machine learning models 211, 212, 213, and 214 may include a plurality of deep neural network (DNN) models. The DNN model is an artificial neural network technology including a plurality of hidden layers between an input layer and an output layer. The DNN model can learn various complex non-linear relationships by including a plurality of hidden layers. The generation unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 capable of predicting the amount of battery gas generation by using at least a part of the battery data A as a training data set. The amount of internal gas generation of the battery may increase as the SOC and temperature of the battery increase, and the amount of internal gas generation of the battery may also vary depending on the separator type of the battery, the assembly process of the battery, and / or the electrode type of the battery. Therefore, the plurality of machine learning models 211, 212, 213, and 214 may predict whether internal gas of the battery is generated and the amount of gas generation based on the voltage, current, temperature, SOC, SOH, assembly process, electrode type, and separator type of the battery included in the battery data A.

[0062] Referring to Figure 4 , the generation unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 trained based on at least one of the features of the battery data A. For example, when considering the separator type and electrode type among the features of the battery data A, the generation unit 210 may extract the battery data A including the separator type and electrode type of the battery in addition to the voltage, current, temperature, SOC, and SOH of the battery as a training data set, thereby generating a plurality of machine learning models 211, 212, 213, and 214.

[0063] In this document, the generation unit 210 may use the battery data A to train multiple machine learning models 211, 212, 213, and 214 by applying various weights to the features of the battery data A, which include the diaphragm type, assembly process, and electrode type of the battery among the various features included in the battery data A.

[0064] For example, the generation unit 210 may use a regularization scheme to solve the overfitting of the multiple machine learning models 211, 212, 213, and 214 by minimizing the interference of relatively unimportant data among the diaphragm type, assembly process, and electrode type of the battery based on a drop-out scheme (a), thereby generating the multiple machine learning models 211, 212, 213, and 214. That is, the generation unit 210 may generate the multiple machine learning models 211, 212, 213, and 214 by training the machine learning models 211, 212, 213, and 214 using the battery data A with the connections of the nodes related to the features that are unwanted or whose interference is to be minimized among the diaphragm type, assembly process, and electrode type of the battery removed.

[0065] In another example, the generation unit 210 may generate multiple machine learning models 211, 212, 213, and 214 trained using the battery data A by fixing the weights (b) applied to the features of the battery data A, which include the diaphragm type, assembly process, and electrode type of the battery among the various features included in the battery data A. For example, to generate a machine learning model highly influenced by the diaphragm type of the battery, the generation unit 210 may train the machine learning model using the battery data A by applying a high weight (e.g., *a) to the diaphragm type among the features of the battery data A. To generate a machine learning model less influenced by the electrode type of the battery, the generation unit 210 may train the machine learning model using the battery data A by applying a low weight (e.g., *b) to the electrode type among the features of the battery data A.

[0066] In another example, the generation unit 210 may generate multiple machine learning models 211, 212, 213, and 214 trained using the battery data A by applying a bias (e.g., +a' or +b') (c) to the nodes related to the battery data A, which include the diaphragm type, assembly process, and electrode type of the battery among the various features included in the battery data A.

[0067] In addition, the generation unit 210 may generate multiple machine learning models 211, 212, 213, and 214 trained using the battery data A based on weight adjustment including L1 regularization Lasso and L2 regularization Ridge, and the embodiments disclosed in this document are not limited to this example.

[0068] The generation unit 210 may perform min-max scaling on at least a part of the battery data A. Herein, min-max scaling is a method of adjusting the range of all variables because when the magnitudes or units of numerical variables differ due to the variables, their effects on the dependent variable are not properly reflected. The generation unit 210 may convert at least a part of the battery data A into a value between 0 and 1 by performing min-max scaling on at least a part of the battery data A.

[0069] The generation unit 210 may determine the accuracies of the multiple machine learning models 211, 212, 213, and 214 through K-fold cross-validation with respect to the multiple machine learning models 211, 212, 213, and 214. K-fold cross-validation is a method of dividing a preprocessed data set into a training data set and a test set and dividing the training data set into 'K' folds to use one fold for validation and (K-1) folds for model training, thereby using each data for the training and validation processes. For example, the generation unit 210 may determine the accuracies of the multiple machine learning models 211, 212, 213, and 214 through 5-fold cross-validation with respect to the multiple machine learning models 211, 212, 213, and 214.

[0070] For example, the generation unit 210 may evaluate the performances of the multiple machine learning models 211, 212, 213, and 214 based on the mean absolute error (MAE), which is the average of the absolute values obtained by converting the errors between the actual values and the predicted values.

[0071] The controller 220 may input at least a part of the battery data A including the voltage, current, temperature, SOC, and SOH changes of the multiple battery cells 110, 120, 130, and 140 as a test data set into the multiple machine learning models 211, 212, 213, and 214.

[0072] For example, after the generation unit 210 generates the multiple machine learning models 211, 212, 213, and 214 by extracting 80% of the battery data A as a training data set, the controller 220 may extract the other 20% as a test data set and input it into the multiple machine learning models 211, 212, 213, and 214.

[0073] Figure 5 is a diagram for describing an operation method of a controller according to an embodiment disclosed herein.

[0074] Refer to Figure 5, the controller 220 may apply at least a part of the battery data A to multiple machine learning models 211, 212, 213, and 214 to obtain multiple prediction data B1, B2, B3, and B4 for predicting the gas generation amount of the battery from the multiple machine learning models 211, 212, 213, and 214 respectively. That is, the controller 220 may obtain multiple output data Output by respectively inputting at least a part of the battery data A, which is a single input data Input, into each machine learning model. The controller 220 may generate gas generation amount prediction data C for predicting the gas generation amount of the battery based on the multiple prediction data B1, B2, B3, and B4.

[0075] The controller 220 may predict the state of the battery based on the multiple prediction data B1, B2, B3, and B4. According to one embodiment, the controller 220 may apply weights x1, x2, x3, and x4 to the multiple prediction data B1, B2, B3, and B4 generated by the multiple machine learning models 211, 212, 213, and 214 respectively, based on at least one or more of the separator type, assembly process, and electrode type of the battery among the features of the battery data A. For example, when applying the weight x1 to the first prediction data B1 based on any one of the features of the battery data A, the size of the first prediction data B1 may be enlarged to increase the ratio of the first prediction data B1 to the entire prediction data B1, B2, B3, and B4. Herein, the weights x1, x2, x3, and x4 applied to the multiple prediction data B1, B2, B3, and B4 respectively may be related to the weights applied to the features of the battery data A in the training of each of the multiple machine learning models 211, 212, 213, and 214.

[0076] The controller 220 may apply weights to the multiple prediction data B1, B2, B3, and B4 according to the weights applied based on the features of the battery data A during the training of the DNN models for generating the multiple prediction data B1, B2, B3, and B4. That is, the weights applied to the features of the battery data during the training of the multiple machine learning models 211, 212, 213, and 214 are different from the weights applied to each of the multiple prediction data B1, B2, B3, and B4. For example, to determine the weight to be applied to the prediction data B1, the controller 220 may consider the weights applied based on the features of the battery data A during the training of the DNN model included in the machine learning model 211 that generates the prediction data B1.

[0077] Similarly, the controller 220 may consider the weights applied based on the characteristics of the battery data A in the training of the DNN model included in the machine learning model 212 that generates the prediction data B2 to determine the weights to be applied to the prediction data B2, consider the weights applied based on the characteristics of the battery data A in the training of the DNN model included in the machine learning model 213 that generates the prediction data B3, and consider the weights applied based on the characteristics of the battery data A in the training of the DNN model included in the machine learning model 214.

[0078] According to one embodiment, the weights applied to the plurality of prediction data B1, B2, B3, and B4 may be proportional to the weights applied to the characteristics of the battery data A in the training of each of the machine learning models 211, 212, 213, and 214. For example, when, during the training of the machine learning model 211 that generates the prediction data B1, characteristics related to the separator type and the assembly process among the characteristics of the battery data A are considered, the controller 220 may apply the weight corresponding to the separator type and the weight corresponding to the assembly process type to the prediction data B1, where the weight corresponding to the separator type and the weight corresponding to the assembly process may be proportional to the weight applied to the separator type-related characteristic of the battery data and the weight applied to the assembly process-related characteristic of the battery data during the training of the machine learning model 211. Here, the weight x1 may be the sum of the weight corresponding to the separator type and the weight corresponding to the assembly process type, but is not limited to this example.

[0079] That is, the controller 220 may calculate the final gas generation amount prediction data C by applying a fuzzy algorithm capable of reflecting the importance and characteristics of specific variables, rather than simply calculating the average of the plurality of prediction data B1, B2, B3, and B4 by applying the weights x1, x2, x3, and x4 to the plurality of prediction data B1, B2, B3, and B4, respectively.

[0080] For example, the controller 220 may input the plurality of prediction data B1, B2, B3, and B4 to which the weights x1, x2, x3, and x4 are applied into an ensemble learning model, thereby generating gas generation amount prediction data C in the form of a probability distribution. Here, the ensemble learning model may be a machine learning scheme that has better performance than a single learning model by combining two or more learning models. When each model has different reliabilities, the ensemble learning model may calculate a weighted sum by applying weights to the output data of each model, rather than calculating the average of the output data of each model. Here, the weighted sum may be defined as the average obtained by reflecting the weight value corresponding to the importance or influence of the data value when obtaining the average of the data.

[0081] The controller 220 may calculate the average value of the gas generation amount prediction data C in the form of a probability distribution. More specifically, the controller 220 may calculate a 95% prediction interval of the average value of the gas generation amount prediction data C in the form of a probability distribution. For example, the controller 220 may calculate the average value and standard deviation of the gas generation amount prediction data C in the form of a probability distribution, and then calculate "average value ± standard deviation * 1.96" as the 95% prediction interval of the average value of the gas generation amount prediction data C.

[0082] The controller 220 may compare the prediction interval of the gas generation amount prediction data C in the form of a probability distribution with the pre-stored battery gas generation amount measurement value to determine the accuracy of the multiple machine learning models 211, 212, 213, and 214.

[0083] As described above, the battery state prediction device and its operation method according to an embodiment disclosed herein may obtain gas generation amount prediction data in the form of a probability distribution by using multiple artificial intelligence models.

[0084] The battery state prediction device may input a small amount of input data into multiple artificial intelligence models, thereby calculating the final prediction data with high accuracy and reducing the time and cost required for data collection and management.

[0085] In addition, the battery state prediction device may set weights according to the characteristics of the data to reflect the unique characteristics of the actual battery data and the characteristics of the usage environment of the actual battery.

[0086] Figure 6 is a flowchart of an operation method of a battery state prediction device according to an embodiment disclosed herein.

[0087] The battery state prediction device 200 may be substantially the same as the battery state prediction device 200 described with reference to Figures 1 to 5 and will therefore be briefly described to avoid redundant description.

[0088] With reference to Figure 6 , the operation method of the battery state prediction device may include: an operation S101 of generating multiple machine learning models for predicting battery gas generation amount based on battery data; an operation S102 of obtaining multiple prediction data for predicting battery gas generation amount by applying the battery data to the multiple machine learning models; and an operation S103 of predicting the battery gas generation amount based on the multiple prediction data.

[0089] Hereinafter, operations S101 to S103 will be described in detail.

[0090] In operation S101, the generating unit 210 may collect battery data A. For example, the battery data A may be defined as values recording the change in the battery state from the discharging state to its fully charged state or from the fully charged state to the discharging state of a plurality of battery cells 110, 120, 130, and 140. For example, the battery data A may include the voltage, current, temperature, SOC, and SOH of the battery measured cumulatively. Herein, the SOH may include the capacity degradation and resistance degradation of the battery. Additionally, the battery data A may further include at least one of data related to what type of separator type the battery has, by which assembly process the battery is assembled, and / or what type of electrode type the battery has.

[0091] In operation S101, the generating unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 by extracting at least a part of the battery data A as a training data set. In operation S101, for example, the generating unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 by extracting 80% of the battery data A as a training data set.

[0092] In operation S101, the generating unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 trained based on at least one of the features of the battery data A. For example, when considering the separator type and electrode type among the features of the battery data A, the generating unit 210 may extract the battery data A including the separator type and electrode type of the battery in addition to the voltage, current, temperature, SOC, and SOH of the battery as a training data set, thereby generating a plurality of machine learning models 211, 212, 213, and 214.

[0093] Herein, the generating unit 210 may use the battery data A to train a plurality of machine learning models 211, 212, 213, and 214 by applying various weights to the features of the battery data A (which includes the separator type, assembly process, and electrode type among various features included in the battery data A).

[0094] For example, the generating unit 210 may use a regularization scheme to solve the overfitting of the plurality of machine learning models 211, 212, 213, and 214 by minimizing the interference of relatively unimportant data among the separator type, assembly process, and electrode type of the battery based on the dropout scheme (a), thereby generating a plurality of machine learning models 211, 212, 213, and 214. That is, the generating unit 210 may generate a plurality of machine learning models 211, 212, 213, and 214 by training the machine learning models 211, 212, 213, and 214 with the battery data A in which the connections of the nodes related to the features that are unwanted or whose interference is to be minimized among the separator type, assembly process, and electrode type of the battery are removed.

[0095] In another example, the generating unit 210 may generate multiple machine learning models 211, 212, 213, and 214 trained using battery data A by fixing (b) the weights applied to the features of battery data A (which includes the separator type, assembly process, and electrode type of the battery among various features included in battery data A). For example, in order to generate a machine learning model that is greatly affected by the separator type of the battery, the generating unit 210 may train a machine learning model using battery data A by applying a high weight (e.g., *a) to the separator type among the features of battery data A. In order to generate a machine learning model that is less affected by the electrode type of the battery, the generating unit 210 may train a machine learning model using battery data A by applying a low weight (e.g., *b) to the electrode type among the features of battery data A.

[0096] In another example, the generating unit 210 may generate multiple machine learning models 211, 212, 213, and 214 trained using battery data A by applying a bias (e.g., +a' or +b') (c) to the nodes associated with battery data A (which includes the separator type, assembly process, and electrode type of the battery among various features included in battery data A).

[0097] In addition, the generating unit 210 may generate multiple machine learning models 211, 212, 213, and 214 trained using battery data A based on weight adjustment including L1-regularized Lasso and L2-regularized Ridge, and the embodiments disclosed herein are not limited to this example.

[0098] In operation S101, according to one embodiment, the multiple machine learning models 211, 212, 213, and 214 may include multiple deep neural network (DNN) models. The DNN model is an artificial neural network technology that includes multiple hidden layers between the input layer and the output layer. The DNN model can learn various complex non-linear relationships by including multiple hidden layers. In operation S101, the generating unit 210 may generate multiple machine learning models 211, 212, 213, and 214 capable of predicting the battery gas generation amount by using at least a part of battery data A as a training data set.

[0099] In operation S101, the generating unit 210 may perform min-max scaling on at least a part of battery data A. Here, min-max scaling is a method of adjusting the range of all variables because when the size or unit of numerical variables varies due to the variables, its impact on the dependent variable is not properly reflected. The generating unit 210 may convert at least a part of battery data A into values between 0 and 1 by performing min-max scaling on at least a part of battery data A.

[0100] In operation S101, the generation unit 210 may determine the accuracies of the multiple machine learning models 211, 212, 213, and 214 through K-fold cross-validation with respect to the multiple machine learning models 211, 212, 213, and 214. K-fold cross-validation is a method of dividing a preprocessed data set into a training data set and a test set and dividing the training data set into 'K' folds to use one fold for validation and (K - 1) folds for model training, thereby using each data for the training and validation processes. For example, the generation unit 210 may determine the accuracies of the multiple machine learning models 211, 212, 213, and 214 through 5-fold cross-validation with respect to the multiple machine learning models 211, 212, 213, and 214.

[0101] In operation S101, for example, the generation unit 210 may evaluate the performance of the multiple machine learning models 211, 212, 213, and 214 based on the mean absolute error (MAE), which is the average of the absolute values of the errors between the actual values and the predicted values.

[0102] In operation S102, the controller 220 may input at least a part of the battery data A including the voltage, current, temperature, SOC, and SOH changes of the multiple battery cells 110, 120, 130, and 140, as well as the separator type and electrode type of the battery, into the multiple machine learning models 211, 212, 213, and 214 as a test data set.

[0103] In operation S102, for example, after the generation unit 210 generates the multiple machine learning models 211, 212, 213, and 214 by extracting 80% of the battery data A as a training data set, the controller 220 may extract the other 20% as a test data set and input it into the multiple machine learning models 211, 212, 213, and 214.

[0104] In operation S102, the controller 220 may apply at least a part of the battery data A to the multiple machine learning models 211, 212, 213, and 214 to obtain multiple prediction data B1, B2, B3, and B4 for predicting the gas generation amount of the battery from the multiple machine learning models 211, 212, 213, and 214 respectively.

[0105] In operation S102, that is, the controller 220 may obtain multiple output data Output by respectively inputting at least a part of the battery data A as a single input data Input into each machine learning model. The controller 220 may generate gas generation amount prediction data C for predicting the gas generation amount of the battery based on the multiple prediction data B1, B2, B3, and B4.

[0106] In operation S102, the controller 220 may apply at least a portion of the battery data A to the multiple DNN models included in the multiple machine learning models 211, 212, 213, and 214.

[0107] In operation S103, the controller 220 may predict the state of the battery based on the plurality of prediction data B1, B2, B3, and B4.

[0108] In operation S103, according to one embodiment, the controller 220 may apply weights x1, x2, x3, and x4 to the plurality of prediction data B1, B2, B3, and B4 generated by the multiple machine learning models 211, 212, 213, and 214 based on the characteristics of the obtained battery data A. Herein, the weights x1, x2, x3, and x4 applied to the plurality of prediction data B1, B2, B3, and B4 may be related to the weights applied to the characteristics of the battery data A in the training of each of the multiple machine learning models 211, 212, 213, and 214.

[0109] According to one embodiment, the controller 220 may apply weights to the plurality of prediction data B1, B2, B3, and B4 according to the weights applied to the characteristics of the battery data A during the training of the DNN models for generating the plurality of prediction data B1, B2, B3, and B4. That is, the weights applied to the characteristics of the battery data in the training of the multiple machine learning models 211, 212, 213, and 214 are different from the weights applied to each of the plurality of prediction data B1, B2, B3, and B4. For example, to determine the weight to be applied to the prediction data B1, the controller 220 may consider the weights applied to the characteristics of the battery data A during the training of the DNN model included in the machine learning model 211 that generates the prediction data B1.

[0110] Similarly, the controller 220 may consider the weights applied to the characteristics of the battery data A during the training of the DNN model included in the machine learning model 212 that generates the prediction data B2 to determine the weight to be applied to the prediction data B2, consider the weights applied to the characteristics of the battery data A during the training of the DNN model included in the machine learning model 213 that generates the prediction data B3, and consider the weights applied to the characteristics of the battery data A during the training of the DNN model included in the machine learning model 214.

[0111] In operation S103, for example, the controller 220 may input multiple pieces of prediction data B1, B2, B3, and B4 to which weights x1, x2, x3, and x4 are applied into the ensemble learning model, thereby generating prediction data C of the gas generation amount in the form of a probability distribution. Herein, the ensemble learning model may be a machine learning scheme that has better performance than a single learning model by combining two or more learning models.

[0112] In operation S103, the controller 220 may calculate the average value of the prediction data C of the gas generation amount in the form of a probability distribution. More specifically, in operation S103, the controller 220 may calculate a 95% prediction interval of the average value of the prediction data C of the gas generation amount in the form of a probability distribution. In operation S103, for example, the controller 220 may calculate the average value and the standard deviation of the prediction data C of the gas generation amount in the form of a probability distribution, and then calculate "average value ± standard deviation * 1.96" as the 95% prediction interval of the average value of the prediction data C of the gas generation amount.

[0113] In operation S103, the controller 220 may compare the prediction interval of the prediction data C of the gas generation amount in the form of a probability distribution with the pre-stored battery gas generation amount measurement value to determine the accuracy of the multiple machine learning models 211, 212, 213, and 214.

[0114] Referring to Figure 7 , the computing system 2000 according to an embodiment disclosed herein may include an MCU 2100, a memory 2200, an input / output I / F 2300, and a communication I / F 2400.

[0115] The MCU 2100 may be a processor that executes various programs (such as the battery gas generation amount program, etc.) stored in the memory 2200, processes various data through these programs, and executes Figure 1 the above functions of the battery management device 200 shown.

[0116] The memory 2200 may store various programs regarding the operation of the battery state prediction device 200. In addition, the memory 2200 may store the operation data of the battery state prediction device 200.

[0117] As needed, multiple memory units 2200 may be provided. The memory unit 2200 may be a volatile memory or a non-volatile memory. For the memory unit 2200 as a volatile memory, random access memory (RAM), dynamic RAM (DRAM), static RAM (SRAM), etc. may be used. For the memory unit 2200 as a non-volatile memory, read only memory (ROM), programmable ROM (PROM), electrically variable ROM (EAROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), flash memory, etc. may be used. The examples of the memory unit 2200 listed above are only examples and are not limited thereto.

[0118] The input / output I / F 2300 may provide an interface for sending and receiving data by connecting input devices (not shown) such as a keyboard, a mouse, a touch panel, etc. and output devices (not shown) such as a display to the MCU 2100.

[0119] The communication I / F 2400, as a component capable of sending various data to and receiving various data from a server, may be various devices capable of supporting wired communication or wireless communication. For example, a program or various data for resistance measurement and abnormality diagnosis of a battery cell may be sent to and received from an externally provided separate server through the communication I / F 2400.

[0120] The above description only illustrates the technical concept of the present disclosure, and those of ordinary skill in the art to which the present disclosure pertains may make various modifications and variations without departing from the essential features of the present disclosure.

[0121] Therefore, the embodiments disclosed in the present disclosure are intended to describe rather than limit the technical spirit of the present disclosure, and the scope of the technical spirit of the present disclosure is not limited by these embodiments. The protection scope of the present disclosure should be interpreted by the appended claims, and all technical spirits within the same scope should be understood to be included within the scope of the present disclosure.

Claims

1. A battery state prediction device, the battery state prediction device comprising: A generation unit configured to generate a plurality of machine learning models trained based on battery data and features included in the battery data to predict the gas generation amount of the battery; And A controller configured to obtain a plurality of pieces of prediction data for predicting the gas generation amount of the battery by applying the battery data to the plurality of machine learning models, and predict the gas generation amount of the battery based on the plurality of pieces of prediction data.

2. The battery state prediction device according to claim 1, wherein, The plurality of machine learning models include a plurality of deep neural network (DNN) models, and The controller is further configured to obtain the plurality of pieces of prediction data by applying the battery data to the plurality of DNN models.

3. The battery state prediction device according to claim 2, wherein, The battery data includes the temperature, state of charge (SOC), and state of health (SOH) of the battery measured cumulatively, and The features of the battery data include at least one or more of the electrode type, assembly process, and separator type of the battery.

4. The battery state prediction device according to claim 2, wherein, The controller is further configured to: apply weights to each of the plurality of pieces of prediction data based on the weights applied based on the features of the battery data during the training of each DNN model that generates each of the plurality of pieces of prediction data, and input the plurality of pieces of prediction data to which the weights have been applied into an ensemble learning model to generate prediction data of the gas generation amount in the form of a probability distribution.

5. The battery state prediction device according to claim 4, wherein, The controller is further configured to calculate the average value of the prediction data of the gas generation amount in the form of the probability distribution, and determine the accuracy of the plurality of machine learning models by comparing the average value with the pre-stored gas generation amount measurement value of the battery.

6. An operation method of a battery state prediction device, the operation method comprising the following steps: Generating a plurality of machine learning models trained based on battery data and features included in the battery data to predict the gas generation amount of the battery; Obtaining a plurality of pieces of prediction data for predicting the gas generation amount of the battery by applying the battery data to the plurality of machine learning models; And Predicting the gas generation amount of the battery based on the plurality of pieces of prediction data.

7. The operating method of the battery state prediction device according to claim 6, wherein, The plurality of machine learning models include a plurality of deep neural network (DNN) models, and The step of obtaining the plurality of pieces of prediction data for predicting the gas generation amount of the battery by applying the battery data to the plurality of machine learning models includes the following steps: obtaining the plurality of pieces of prediction data by applying the battery data to the plurality of DNN models.

8. The operation method of the battery state prediction device according to claim 6, wherein, The step of generating the plurality of machine learning models for predicting the gas generation amount of the battery based on the battery data includes the following steps: generating the plurality of machine learning models based on the battery data including the temperature, state of charge (SOC), and state of health (SOH) of the battery measured cumulatively, and The features of the battery data include at least one or more of the electrode type, assembly process, and separator type of the battery.

9. The operating method of the battery state prediction device according to claim 6, wherein, The step of predicting the gas generation amount of the battery based on the multiple pieces of prediction data includes the following steps: Based on the weights applied based on the characteristics of the battery data during the training of each DNN model that generates each of the multiple pieces of prediction data, apply the weights to each of the multiple pieces of prediction data; and Input the multiple pieces of prediction data to which the weights have been applied into an ensemble learning model to generate prediction data of the gas generation amount in the form of a probability distribution.

10. The operating method of the battery state prediction device according to claim 6, wherein, The step of predicting the gas generation amount of the battery based on the multiple pieces of prediction data includes the following steps: Calculate the average value of the prediction data of the gas generation amount in the form of the probability distribution; and Determine the accuracy of the multiple machine learning models by comparing the average value with the pre-stored measured value of the gas generation amount of the battery.

Images