Method for labeling training data of deep neural network training model for battery state estimation, and training method using same

Abstract

The present invention relates to, with respect to the generation a deep neural network training model by training a deep neural network in order to estimate a battery state, a training data labeling method for training the deep neural network, a method for training the deep neural network through labeling of training data, a training system therefor, a method for estimating a battery state through the training model, and a system therefor.

Description

A method for labeling training data for a deep neural network learning model for battery condition estimation and a learning method using the same.

[0001] The present invention relates to a method for labeling learning data for training a deep neural network to create a deep neural network learning model by training a deep neural network to estimate the state (SOX) of a battery, a method for training the deep neural network through labeling of the learning data, a learning system thereof, and a method for estimating the state of a battery through the learning model and a system thereof.

[0002] Secondary batteries (hereinafter referred to as batteries) have various uses, such as for starting automobiles, emergencies, electric vehicles, and UPS, and can be used for a long time until their lifespan ends by repeating charging and discharging.

[0003] These batteries can be recharged and reused after discharge, making them an essential power source for a wide range of applications, from home appliances to electric vehicles. Therefore, accurately estimating the battery's condition is crucial for efficient battery management and improving operational efficiency and stability in diverse applications.

[0004] Key indicators that help us estimate the health of a battery include its state of charge (SOC), state of health (SOH), and state of power (SOP). SOC represents how long a battery can run (how long it will run), SOH represents how healthy a battery is (how healthy it is), and SOP represents how powerful a battery is (how powerful it is).

[0005] For example, in the case of batteries in electric vehicles, estimating and providing accurate status data (e.g., State of Charge (SOC), State of Health (SOH), State of Operation (SOP)) can help determine when to replace the battery. This, in turn, allows for timely battery replacement, ensuring stable and continuous operation of the electric vehicle. Furthermore, accurate battery status estimation can accelerate the design of battery cells and advances in high-speed charging methodologies, thereby shortening battery testing periods.

[0006] Recently, various deep neural networks (DNNs) such as artificial neural networks (ANNs) and temporal convolutional neural networks (TCNNs) have been attempted to estimate the lifespan of batteries.

[0007] The advantage of modeling using deep neural networks over directly measuring the battery's condition through modeling based on physical characteristics is that the more training data there is in various operating environments, the more precise the prediction performance of the deep neural network learning model becomes.

[0008] For this reason, it's crucial to develop methods for labeling and training data from real-world environments, rather than laboratory data. However, because it's virtually impossible to accurately predict the state of an electric vehicle, mobile phone, or laptop during runtime, there's currently no way to obtain labeled data for deep neural network training.

[0009] To solve this problem, the present invention proposes a method of recognizing a battery relaxation period at run time, extracting measurement data using an indirect measurement method after the battery stabilizes during the relaxation period, performing data interpolation between relaxation periods to construct overall learning data, and then automatically labeling the learning data with a target value.

[0010] In addition, the present invention proposes a method for generating a deep neural network learning model for battery state estimation by training a deep neural network using the labeled learning data.

[0011] In addition, the present invention proposes an estimation method that extracts measurement data from the idle period during run-time in an electric vehicle using an indirect measurement method and then inputs the data into a deep neural network learning model for estimating the battery state, thereby enabling accurate estimation of the battery state.

[0012] Next, we will briefly explain the prior art existing in the technical field of the present invention, and then describe the technical details that the present invention seeks to achieve differently from the prior art.

[0013] First, Korean Patent Publication No. 2021-0024962 (March 8, 2021) relates to a device and method for diagnosing the status and estimating the life of an ESS battery, which generates battery status data by using signal analysis and statistical analysis of a battery status signal including any one of voltage, current, and temperature collected from a battery included in an ESS (energy storage system), and inputs the generated battery status data into a pre-trained deep learning-based diagnosis estimation model to estimate the life.

[0014] The above Korean Patent Publication No. 2021-0024962 diagnoses the state of the battery and estimates its lifespan by inputting battery state data into a pre-generated diagnosis estimation model. However, there is no description at all of what data is learned and generated when learning battery state data to create a diagnosis estimation model.

[0015] On the other hand, the present invention proposes a technical feature of collecting measurement data during the idle period of the runtime of various battery-powered devices in operation or use, converting the collected measurement data into training data, automatically adding labels as target values ​​for battery status to the converted training data to train a deep neural network, and thereby generating a deep neural network learning model for battery status estimation. Therefore, the two inventions have significant differences in their technical configurations and effects.

[0016] In addition, Korean Patent No. 2485303 (January 2, 2023) relates to a data labeling device and method, which labels data according to labeling conditions input by a user to generate first learning data, and provides the first learning data to the user, and generates second learning data by modifying or supplementing the first learning data according to a processing command input by the user.

[0017] The above Korean Patent No. 2485303 creates learning data by labeling data, but has the disadvantage of requiring user intervention.

[0018] On the other hand, the present invention collects measurement data on the state of a battery during a rest period in the run-time when various devices using a battery are operated or used, converts the collected measurement data into learning data, and automatically adds a label as a target value for the state of the battery to the converted learning data. Korean Patent No. 2485303 does not describe, suggest, or imply any technical features of the present invention.

[0019] The present invention was created to solve the above problems, and the purpose of the present invention is to provide a method for labeling learning data for training a deep neural network to create a deep neural network learning model by training a deep neural network to estimate the state of a battery, and a method for training the deep neural network through labeling the learning data.

[0020] In addition, another object of the present invention is to provide a method of recognizing a sleep period in a run-time in which a battery is being used, collecting measurement data on the characteristic values ​​of the battery during the sleep period, converting the data into learning data, and labeling the converted learning data as a target value for the battery status, and a learning method using the same.

[0021] In addition, another object of the present invention is to provide a method for converting the measurement data into learning data by interpolating data extracted through linear, nonlinear, and padding between the above rest periods.

[0022] In addition, another object of the present invention is to provide a method for accurately estimating the state of a battery by converting measurement data measured at run time into input data and inputting the converted input data into the learning model for estimating the state of the battery, thereby estimating the SOC (or SOE), SOH, SOP, or SOX including a combination thereof.

[0023] A method for labeling learning data of a deep neural network for estimating the state of a battery according to one embodiment of the present invention includes a learning data configuration step of configuring learning data by collecting measurement data on the state of the battery and converting it into learning data; and a learning data labeling step of automatically labeling the configured learning data with SOC (state of charge), SOH (state of health), SOP (state of power), or SOX including a combination thereof according to the state of the battery; wherein the learning data is characterized in that it includes feature values ​​collected during a relaxation period in the run-time of the battery.

[0024] Here, the feature values ​​include an open circuit voltage (OCV), voltage (V), current (I), charge (Q), temperature (T), or a combination thereof of the battery measured during a relaxation period in the run-time of the battery, and the learning data is characterized in that it further includes statistical values ​​including an average, variance, median, standard deviation, difference, or a combination thereof extracted from the collected feature values.

[0025] The above learning data configuration step is characterized by including an idle period recognition step of recognizing a stabilized section in which the measurement data changes within a predetermined range for a predetermined period of time or longer in the run-time of the battery as an idle period; a measurement data extraction step of extracting the measurement data from the recognized idle period; an SOX calculation step of calculating SOX including SOC, SOH, SOP, or a combination thereof of the battery from the extracted measurement data; and a learning data conversion step of converting the extracted measurement data into learning data in a format for training the deep neural network.

[0026] In the above learning data labeling step, the run-time SOH is extracted by dividing the OCV-SOC measured at the first point of the previous rest period by the OCV-SOC measured at the second point of the later rest period from the OCV-SOC between two adjacent rest periods, and the first point and the second point are set by detecting a flat area in the OCV-SOC curve of each rest period in which the OCV change amount is gentle or almost non-existent.

[0027] In the above learning data labeling step, the output power (P) is calculated as voltage (V) / current (I) through the OCV, voltage (V), and current (I) measured in the flat area with the smallest OCV change in the OCV-SOC curve of each resting period, and when the maximum power (Pmax) of the battery is known, the SOP is labeled by calculating it from the output power (P) / maximum power (Pmax).

[0028] If you want to calculate the maximum power (Pmax) and derive SOP from the calculated value, you can calculate the resistance (R = (OCV - V) / I)) through the OCV, voltage (V), and current (I) measured in the flat area with the smallest OCV change in the OCV-SOC curve of each rest period, and calculate the maximum current (Imax = (OCV - Vterm) / R) using the calculated resistance, and calculate the maximum power (Pmax = maximum current (Imax) * measured voltage (V)) from the maximum current, which can be used to derive SOP. Here, Vterm is the termination voltage of the battery. Of course, it is also possible to calculate the maximum current including the cutoff voltage.

[0029] In addition, the learning data labeling method further includes a learning data expansion step of expanding the learning data so that the deep neural network estimates the SOX at all data points other than the idle period; and the expansion is characterized in that the learning data is constructed by interpolating between run-time feature values ​​extracted in each of two consecutive idle periods on the time axis with data extracted through a predetermined statistical operation using the feature values, or by interpolating by padding with predetermined fixed data.

[0030] In addition, it is characterized in that the learning data is composed of only the feature values ​​of each rest period without interpolation of the learning data of the deep neural network.

[0031] Meanwhile, a learning method according to another embodiment of the present invention is characterized by generating a deep neural network learning model that estimates the state of the battery through the learning data labeling method of the deep neural network.

[0032] Meanwhile, a learning system according to another embodiment of the present invention is characterized by including: a memory storing a program configured to generate a deep neural network learning model for estimating the state of the battery through the learning data labeling method of the deep neural network; and a processor configured to execute the program stored in the memory.

[0033] Meanwhile, a battery state estimation method according to another embodiment of the present invention is characterized in that the state of the battery is estimated by measuring a feature value of the battery and inputting it into a deep neural network learning model that estimates the state of the battery generated through the learning data labeling method of the deep neural network.

[0034] Meanwhile, a battery state estimation system according to another embodiment of the present invention is characterized by including: a memory storing a program configured to estimate the state of the battery by measuring a feature value of the battery and inputting it into the deep neural network learning model, which estimates the state of the battery by using the deep neural network learning data labeling method; and a processor configured to execute the program stored in the memory.

[0035] As described above, the present invention recognizes the idle period of the battery at run time, measures the run-time characteristic value of the battery to generate learning data, and labels the generated learning data with SOX indicating the state of the battery to train a deep neural network, thereby creating a deep neural network learning model capable of more accurately estimating the state of the battery.

[0036] In addition, the present invention has the effect of creating a more sophisticated deep neural network learning model with learning data that accurately reflects actual operating conditions without using complex calculation procedures or statistical data.

[0037] In addition, the present invention estimates and provides the exact status of the battery, thereby enabling the timing of battery replacement to be determined, and thereby enabling stable and continuous use of the application-side device by replacing the battery at the right time.

[0038] In addition, the present invention has the effect of shortening the battery test period by accelerating the development of the design of cells constituting the battery and the fast charging methodology through accurate estimation of the battery state.

[0039] FIG. 1 is a conceptual diagram illustrating a concept of collecting measurement data for labeling learning data of a deep neural network learning model for estimating battery status according to one embodiment of the present invention.

[0040] FIG. 2 is a diagram illustrating the physical concept of SOX for labeling learning data to create a deep neural network learning model for battery state estimation according to one embodiment of the present invention.

[0041] FIG. 3 is a diagram for explaining the concept of configuring learning data during an idle period and labeling it with SOX to create a deep neural network learning model for estimating a battery state according to one embodiment of the present invention.

[0042] FIG. 4 is a diagram illustrating a method of labeling learning data with SOX to create a deep neural network learning model for battery state estimation according to one embodiment of the present invention.

[0043] FIG. 5 is a conceptual diagram illustrating a concept of estimating the state of a battery using a deep neural network learning model for estimating the state of a battery according to one embodiment of the present invention.

[0044] FIG. 6 is a block diagram showing the configuration of a learning system that generates a deep neural network learning model that estimates the state of the battery through a learning data labeling method of a deep neural network according to one embodiment of the present invention.

[0045] FIG. 7 is a block diagram showing the configuration of a battery state estimation system that estimates the state of the battery by measuring the characteristic values ​​of the battery and inputting them into a deep neural network learning model that estimates the state of the battery generated through a learning data labeling method of a deep neural network according to one embodiment of the present invention.

[0046] FIG. 8 is a flowchart illustrating a procedure for learning by labeling learning data according to one embodiment of the present invention and a procedure for estimating the battery state of an actual battery using a learning model for battery state estimation generated through the learning.

[0047] [Explanation of symbols]

[0048] 100: Learning system, 100a: Learning engine, 300: Battery state estimation system, 300a: Battery state estimation engine, 200: Database, 110: Learning data composition unit, 111: Idle period recognition unit, 112: Measurement data extraction unit, 113: SOX calculation unit, 114: Learning data conversion unit, 120: Learning data labeling unit, 130: Learning data expansion unit, 140: Learning model creation unit, 310: Measurement data reception unit.

[0049] Hereinafter, with reference to the attached drawings, a preferred embodiment of a training data labeling method of a deep neural network learning model for battery state estimation and a training method therethrough according to the present invention will be described in detail. The same reference numerals in each drawing represent the same elements. In addition, specific structural and functional descriptions of the embodiments of the present invention are merely illustrative for the purpose of explaining the embodiments according to the present invention, and unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by a person of ordinary skill in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and are preferably not interpreted in an ideal or excessively formal meaning unless explicitly defined in this specification.

[0050] FIG. 1 is a conceptual diagram illustrating a concept of collecting measurement data for labeling learning data of a deep neural network learning model for estimating battery status according to one embodiment of the present invention.

[0051] As illustrated in FIG. 1, the learning system (100) for generating a deep neural network learning model for estimating the battery state of the present invention cannot use the measurement data as learning data when the application side is an electric vehicle while driving, stopped for less than a predetermined time, or charging, and can only use the measurement data measured in a stabilized state so that the battery is recognized as an idle period when parked for a predetermined time or longer as learning data.

[0052] The above learning system (100) receives run-time measurement data from an application during the battery life cycle in various environments for use of multiple batteries, recognizes a relaxation period sufficient for the battery to stabilize, converts the measurement data received during the relaxation period into learning data, labels the converted learning data, and then trains a deep neural network. After testing whether sufficient accuracy is guaranteed through the learning, a learning model is created and stored in a database (200).

[0053] The above database (200) stores measurement data measured during the idle period and learning data converted into learning data, and stores not only the deep neural network to be trained but also the deep neural network for which training has been completed.

[0054] Here, the above learning data includes run-time features including OCV-SOC, voltage (V), current (I), charge (Q), temperature (T), or a combination thereof measured at run-time, and the database stores SOX, which is a label assigned to the above learning data.

[0055] Please note that when the application is running, operating, or in use, the battery measurement value is unstable and SOX cannot be accurately calculated, making it inappropriate to use it as learning data.

[0056] The above-mentioned idle period is defined as a period of time during which measured data (e.g., voltage) remains stable with little or no change within a given range. In other words, the idle period refers to a stable time period during which there is little or no charging or discharging of the battery.

[0057] Accordingly, the present invention extracts SOH through the ratio of the amount of charge measured in the previous idle period and the amount of charge measured in the next idle period during two adjacent idle periods in the life cycle of a battery, and labels learning data with SOX (SOC, SOH, SOP, etc.) including the extracted SOH to train the deep neural network.

[0058] For example, SOX is extracted from the feature values ​​measured during the idle period at run-time, the feature values ​​are converted into learning data, and then the learning data is labeled with SOX.

[0059] Therefore, it becomes possible to estimate the state of a battery by estimating SOX using the label estimated through the output of the learning model for battery state estimation. In other words, the state of the battery is calculated based on the SOX label, which is the output of the learning model for battery state estimation.

[0060] Accordingly, since the learning system (100) of the present invention creates a learning model by labeling learning data with a battery state (SOX, state of X), the battery state estimation system (300) estimates the labeled SOX from the learning model. Here, 'X' is a term representing each or all of charge, health, power (energy), etc.

[0061] Accordingly, the above battery state estimation system (300) configures measurement data measured at run-time from an actual battery as input data to be input into a learning model for battery state estimation, and inputs the configured input data into the learning model for battery state estimation to estimate the state of the battery according to the SOX label, which is the output of the learning model for battery state estimation.

[0062] In addition, the deep neural network in the present invention refers to an artificial neural network with multiple hidden layers, and is divided into various types depending on its structure and purpose. That is, it can be composed of various machine learning networks such as an artificial neural network (ANN), a deep convolutional neural network (DCNN), a transformer, a temporal convolutional neural network (TCNN), a recurrent neural network (RNN), and a long short-term memory (LSTM).

[0063] FIG. 2 is a diagram illustrating the physical concept of SOX for labeling learning data to create a deep neural network learning model for battery state estimation according to one embodiment of the present invention.

[0064] As illustrated in Fig. 2, OCV-SOC is usually measured while performing full charging and full discharging in a laboratory environment. For example, charging is performed using a constant current-constant voltage method, and the set temperature is set to various settings such as an upper limit of 25 degrees, a lower limit of 40 degrees, and a rest period of 3 hours or 6 hours, and it is obtained by indirectly measuring while charging or discharging at a low C-rate of 0.01C or 0.02C or a high C-rate of 0.1C, 0.3C, or 1C.

[0065] Here, the data from a laboratory environment measured using an indirect measurement method are (1) the OCV changing from the full charge voltage to the termination voltage or vice versa, which is called the OCV-SOC in the laboratory environment. Depending on the battery's usage status, a capacity gap with respect to SOC occurs (2) when the battery is new and (3) when the battery has deteriorated over time.

[0066] If the OCV-SOC is measured during two adjacent idle periods and the OCV-SOC measured during the previous idle period is the same as the OCV-SOC measured during the next idle period, then the SOH is 1 and the battery has not aged at all.

[0067] The above SOC is an indicator of the current charge level of the battery, SOH is an indicator of the health status of the battery, and SOP is an indicator of the current output capability of the battery.

[0068] In addition, the SOC can be calculated by calculating the amount of charged and discharged charge, by measuring the voltage or temperature of the battery, or by measuring the current and accumulating it over time to calculate the amount of current, the SOH can be measured by measuring the capacity of the battery and comparing it with the initial capacity, by measuring the internal resistance to evaluate the loss, or by tracking the change in performance through the charge and discharge cycle, and the SOP can be estimated by estimating the current output capability of the battery or by detecting the change in output capability according to the change in temperature.

[0069] In the present invention, a deep neural network learning model for battery state estimation is used to estimate SOX including SOC, SOH, and SOP.

[0070] In addition, the output power (P) is calculated as voltage (V) / current (I) through the OCV, voltage (V), and current (I) measured in the flat area with the smallest OCV change in the OCV-SOC curve of the above-mentioned rest period, and when the maximum power (Pmax) of the battery is known, the SOP can be calculated from the output power (P) / maximum power (Pmax).

[0071] If you want to calculate the maximum power (Pmax) and derive SOP from the calculated value, you can calculate the resistance (R = (OCV - V) / I)) through the OCV, voltage (V), and current (I) measured in the flat area with the smallest OCV change in the OCV-SOC curve of each rest period, and calculate the maximum current (Imax = (OCV - Vterm) / R) using the calculated resistance, and calculate the maximum power (Pmax = maximum current (Imax) * measured voltage (V)) from the maximum current, which can be used to derive SOP. Here, Vterm is the termination voltage of the battery. Of course, it is also possible to calculate the maximum current including the cutoff voltage.

[0072] Also, the battery's SOC (Ah) indicates charging as it moves to the left, and discharging as it moves to the right. SOH is derived by finding two points on the OCV curve where the amount of change changes sharply and calculating the charge ratio between the two points. SOP indicates the battery's output state, indicating the availability of power that the battery can charge or discharge over a certain period of time. SOE (Wh) is another unit of measurement for SOC (Ah).

[0073] In the present invention, the physical concept of the OCV-SOC curve in a laboratory environment is applied to compute the run-time OCV-SOX by constructing training data and labeling the constructed training data using feature values ​​including current, voltage, charge, C-rate, and temperature measured during the idle period at run-time. This will be described in detail with reference to Fig. 3.

[0074] FIG. 3 is a diagram for explaining the concept of configuring learning data during an idle period and labeling it with SOX to create a deep neural network learning model for estimating a battery state according to one embodiment of the present invention.

[0075] As illustrated in Fig. 3, run-time OCV-SOX is extracted through measurement data collected after the battery's state stabilizes during the battery's idle period at run-time.

[0076] Here, among the SOX, the SOH is extracted by dividing the OCV-SOC measured at the first point of the previous rest period by the OCV-SOC (charge) measured at the second point of the next rest period in the OCV-SOC between two adjacent rest periods. Here, the first point and the second point are set by detecting a flat area in the OCV-SOC curve of each rest period in which the OCV change amount is gentle or almost non-existent.

[0077] In addition, the output power (P) is calculated as voltage (V) / current (I) through the OCV, voltage (V), and current (I) measured in the flat area with the smallest OCV change in the OCV-SOC curve of the above rest period, and when the maximum power (Pmax) of the battery is known, the SOP is calculated from the output power (P) / maximum power (Pmax) and labeled as a target value.

[0078] Among the two rest periods above, learning data is generated using the feature values ​​measured during the next rest period, and the generated learning data is labeled with SOX including the SOH.

[0079] The feature values ​​in the process of transitioning from the previous rest period to the next rest period are interpolated, each target value (SOX) is calculated from the interpolated feature values, and each learning data in the process of transitioning from the previous rest period to the next rest period is labeled with each of the calculated target values.

[0080] Of course, it is possible to automatically label the training data with the run-time OCV-SOX calculated for each pause period. It is also possible to label the training data by calculating the run-time OCV-SOX for each interpolated value between two pause periods.

[0081] FIG. 4 is a diagram illustrating a method of labeling learning data with SOX to create a deep neural network learning model for battery state estimation according to one embodiment of the present invention.

[0082] As shown in Fig. 4, SOX is extracted from measurement data measured at run time, learning data is extracted from the measurement data, and the extracted learning data is automatically labeled with SOX.

[0083] The above run-time measurement data includes the characteristic values ​​of the battery, such as OCV (open circuit voltage), voltage (V, measured voltage), current (I, current), charge (Q, capacity), and temperature (T, temperature). The SOX extracted from the characteristic values ​​includes SOC, SOH, and SOP.

[0084] Here, the feature values ​​are converted into learning data, and the learning data is labeled with the SOX extracted from each learning data. Here, the learning data further includes statistical values ​​including the mean, variance, median, standard deviation, difference, or a combination thereof extracted from the collected feature values.

[0085] The above SOX is assigned to each learning data as a label, and the learning data can be arranged in various ways, such as 3x3 or 1x9 matrix data, as shown in Fig. 4, and the learning data arranged in this way is input to a deep neural network to perform learning.

[0086] These training data formats can be arranged in a wide variety of ways, depending on the deep neural network. Figure 4 is merely an example, and numerous variations are possible. Furthermore, new training data items can be added as needed, allowing the deep neural network to be trained in various sizes.

[0087] That is, the format of the above learning data has a very diverse structure depending on the arrangement of each feature value and the time series range of the feature values ​​measured continuously.

[0088] Here, Vs can provide statistical information (mean, variance, median, standard deviation, difference, etc.) about the measured voltage V, allowing for controlling the learning of the learning network through various combinations with the measured voltage. For example, when interpolating the measured voltage, the interpolated voltage value can be set to 0 or the average value to control the reflection of the learning results in the learning network.

[0089] FIG. 5 is a conceptual diagram illustrating a concept of estimating the state of a battery using a deep neural network learning model for estimating the state of a battery according to one embodiment of the present invention.

[0090] As illustrated in FIG. 5, the deep neural network learning model generated by the learning system (100) according to the present invention is stored and managed in a database (200). In this case, the deep neural network learning model may be utilized to independently estimate the state of a battery through input data provided from each device in an application area such as an electric vehicle, or may be provided to a battery state estimation system (300) on a network to remotely estimate the state of the battery.

[0091] The above learning system (100) can be implemented on a server on a network (or in the cloud) to also perform the role of an estimation system (300) that estimates the state of the battery.

[0092] In other words, the deep neural network learning model can be uploaded to each device or cloud server (which serves as the estimation system) in the application domain and used to estimate the battery condition. In application domains such as electric vehicles, self-collected measurement data can be converted into input data for the deep neural network learning model to locally estimate the battery condition.

[0093] Meanwhile, when computing power is insufficient in each device in the application area, the input data of the self-generated deep neural network learning model is transmitted to a remote cloud or network-based estimation system (300), and the input data is input into the deep neural network learning model uploaded to the estimation system (300) to remotely estimate the state of the battery.

[0094] In each case, the deep neural network learning model can be uploaded and used from a database (200) on a network (cloud).

[0095] In addition to the learning model for battery state estimation, the above database (200) can also store and manage run-time measurement data or labeled learning data.

[0096] Ultimately, by collecting measurement data from each device in each application area and configuring input data, and then inputting it into a deep neural network learning model, which is a learning model for estimating battery status, the SOX label is estimated as a result, and the status of the battery can be estimated based on the label.

[0097] FIG. 6 is a block diagram showing the configuration of a learning system that generates a deep neural network learning model that estimates the state of the battery through a learning data labeling method of a deep neural network according to one embodiment of the present invention.

[0098] As illustrated in FIG. 6, a learning system (100) according to another embodiment of the present invention comprises, as its core components, a learning engine (100a) including a memory storing a program configured to generate a deep neural network learning model for estimating the state of the battery through the learning data labeling method of the deep neural network, and a processor configured to execute the program stored in the memory.

[0099] As illustrated in FIG. 6, a learning engine (100a) according to one embodiment of the present invention is configured to include a learning data configuration unit (110), a learning data labeling unit (120), a learning data expansion unit (130), and a learning model generation unit (140).

[0100] The above learning data configuration unit (110) includes a rest period recognition unit (111), a measurement data extraction unit (112), a SOX calculation unit (113), and a learning data conversion unit (114).

[0101] The above learning data configuration unit (110) collects characteristic values ​​for the battery status as measurement data, and converts the collected measurement data into learning data in a predetermined format according to a deep neural network to configure learning data.

[0102] In addition, the learning data labeling unit (120) is configured to automatically label the learning data configured in the learning data configuration unit (110) with SOX according to the state of the battery. Here, the label is SOX including the state of charge (SOC), state of health (SOH), state of power (SOP), or a combination thereof of the battery extracted from the feature values.

[0103] Here, the learning data includes feature values ​​including OCV-SOC, voltage (V), current (I), charge (Q), temperature (T), or a combination thereof, measured indirectly during a relaxation period in the run-time of the battery.

[0104] In addition, the above learning data can be configured to further include statistical values ​​including the mean, variance, median, standard deviation, difference, or a combination thereof extracted from the collected feature values.

[0105] The above-mentioned idle period recognition unit (111) is configured to recognize a stabilized section in which measurement data changes within a predetermined range for a predetermined period of time or longer in the battery run-time as an idle period. Measurement data is extracted from the recognized idle period.

[0106] The relaxation period refers to the period of time after a significant period of time has elapsed, during which the battery stabilizes after a vehicle, such as an electric vehicle, has stopped operating or using the vehicle. Beyond simply measuring time, such as 3 or 6 hours, it also measures whether the battery's OCV changes within a specified range over time.

[0107] The above measurement data extraction unit (112) receives feature values ​​including OCV-SOC, voltage (V), current (I), charge (Q), temperature (T), or a combination thereof measured by an indirect measurement method, and processes them into data suitable for use as learning data. For example, this may include unifying the format (resolution, integer, floating point, etc.) of each data.

[0108] The above SOX calculation unit (113) is configured to extract SOX through measurement data for the recognized run-time feature value at points in two consecutive rest periods.

[0109] The above learning data conversion unit (114) is configured to convert the run-time feature values ​​extracted from the feature value measurement unit (112) into learning data in a format usable by a deep neural network by arranging them. The learning data may also be configured to further include statistical values ​​including the mean, variance, median, standard deviation, difference, or a combination thereof extracted from the collected feature values. The format of the learning data is omitted here as it has been described with reference to FIG. 4.

[0110] The above-mentioned learning data labeling unit (120) is configured to automatically label the converted learning data with the calculated SOX. Here, the label may be recognized as the target value of SOX. In fact, estimating SOX using the learning model according to the present invention is synonymous with estimating the target value.

[0111] The above learning data expansion unit (130) is configured to expand the learning data and labels so that the deep neural network can estimate SOX at all data points. Here, the expansion is performed by interpolating between measurement data extracted from two consecutive rest periods on the time axis with data extracted through a predetermined statistical process or by interpolating with predetermined data to form the learning data.

[0112] In the above deep neural network, all data points of the learning data used for learning are data columns that list measurement data through indirect measurement regardless of time interval.

[0113] For example, in the learning data labeling method according to the present invention, the learning data of the deep neural network is set to a first window having a fixed sampling interval, and the first window is set to be larger than a second window having a predetermined sampling interval for the collected measurement data, and sampling points that are insufficient in the measurement data are interpolated and configured with predetermined data.

[0114] In addition, the learning data of the above deep neural network includes configuring the learning data only with the feature values ​​of each rest period without interpolation.

[0115] The above learning model generation unit (140) is configured to generate a learning model for battery status estimation by training a deep neural network using the labeled learning data.

[0116] The above learning system (100) labels the learning data with a target value according to the above learning data, and is configured to perform deep learning on the labeled learning data by minimizing the error by backpropagating the error for the output of the deep neural network.

[0117] The learning model for estimating the battery status thus generated is configured to output a value that is probabilistically closest to the target value of SOX according to the battery's idle period when input data composed of measurement data measured at run time is input.

[0118] FIG. 7 is a block diagram showing the configuration of a battery state estimation system that estimates the state of the battery by measuring the characteristic values ​​of the battery and inputting them into a deep neural network learning model that estimates the state of the battery generated through a learning data labeling method of a deep neural network according to one embodiment of the present invention.

[0119] As illustrated in FIG. 7, a battery state estimation system (300) according to another embodiment of the present invention comprises a battery state estimation engine (300a) including a memory (2000) storing a program configured to estimate the state of the battery by measuring a feature value of the battery and inputting the measured value into the deep neural network learning model for estimating the state of the battery generated through the learning data labeling method of the deep neural network, and a processor (1000) configured to execute the program stored in the memory.

[0120] Here, the user interface (3000), data interface (4000), and network interface (5000) are means for providing user, database, and network interfaces, which are typically provided on a user terminal or a server on a cloud or network. Furthermore, the web server (6000) is provided for easy access by user terminals to a server on a cloud or network.

[0121] As illustrated in FIG. 7, the battery state estimation engine (300a) is configured to estimate the state of an actual battery using actual run-time measurement data and a learning model for battery state estimation, and includes a measurement data receiving unit (310), an input data configuration unit (320), a SOX estimation unit (330), and a battery state estimation unit (340).

[0122] The above measurement data receiving unit (310) is configured to receive measurement data of the battery from an application provided by the user during actual runtime. The measurement data is a characteristic value including OCV-SOC, voltage (V), current (I), charge (Q), temperature (T), or a combination thereof.

[0123] The above input data configuration unit (320) converts the measurement data received from the measurement data reception unit (310) into data to be input into a learning model for battery state estimation to configure the input data.

[0124] The above input data can be configured for at least one idle period, and depending on the learning model, it is possible to configure the input data for not only one idle period but also multiple idle periods.

[0125] The above SOX estimation unit (330) is configured to estimate SOX by inputting the input data configured in the above input data configuration unit (320) into a learning model for battery state estimation.

[0126] That is, the SOX estimation unit (330) uploads a learning model from the database (200) and then estimates SOX at the corresponding data point using the input data configured in the input data configuration unit (320).

[0127] The above battery state estimation unit (340) analyzes SOX based on the label estimated and acquired by the SOX estimation unit (330) to estimate the current state of the battery. That is, the state of the battery is estimated through SOC, SOH, SOP, etc.

[0128] The life of the battery can also be predicted through the above-mentioned estimated SOC, SOH, and SOP. SOC is an indicator of the current charge level of the battery, SOH is an indicator of the health status of the battery, and SOP is an indicator of the current output capability of the battery.

[0129] The above SOC can be calculated by calculating the amount of charged and discharged charge, by measuring the voltage or temperature of the battery, or by measuring the current and accumulating it over time to calculate the amount of current; the SOH can be measured by measuring the capacity of the battery and comparing it with the initial capacity, by measuring the internal resistance to evaluate the loss, or by tracking the change in performance through the charge and discharge cycle; the SOP can be estimated by estimating the current output capability of the battery or detecting the change in output capability according to the change in temperature; in the present invention, a deep neural network learning model for estimating the battery state is used to estimate SOX including SOC, SOH, and SOP.

[0130] Meanwhile, the learning system (100) requires high-performance computing power because it learns labeled learning data to create a learning model for estimating battery status.

[0131] However, since the battery state estimation system (300) estimates the state of the battery using a learning model for battery state estimation generated by the learning system (100), it can be executed even with relatively low computing power. Accordingly, the battery state estimation engine (300a) according to the present invention can be provided to a user's application in the form of an application program.

[0132] Accordingly, users can integrate the battery status estimation unit (340) into their applications to locally check the battery status in real time. Of course, more accurate status estimation is possible by continuously estimating the battery status over multiple idle periods.

[0133] FIG. 8 is a flowchart illustrating a procedure for labeling and learning learning data according to one embodiment of the present invention and a procedure for estimating the state of an actual battery using a learning model for battery state estimation generated through learning.

[0134] As illustrated in FIG. 8, the procedure for creating a learning model for estimating a battery state by labeling learning data and training a deep neural network according to one embodiment of the present invention first performs a step of recognizing a rest period by having the learning system (100) recognize a rest period from learning measurement data stored in a database (200) (S110).

[0135] The above learning measurement data refers to learning data labeled with SOX, which is measurement data measured for multiple batteries under various environments.

[0136] The above-mentioned idle period recognition step recognizes the idle period by recognizing a section in which the measurement data (i.e., the learning measurement data) changes only slightly within a predetermined range or does not change at all for a predetermined period of time or longer.

[0137] Next, the learning system (100) performs a learning data configuration step of configuring the measurement data measured during the rest period into learning data based on the result of recognizing the rest period (S120).

[0138] At this time, the learning data configuration step may further include a learning data expansion step that expands the learning data by interpolating or padding data between the measurement data so that the time interval between the measurement data (i.e., the measurement time interval) is independent of the time interval of the deep neural network (i.e., the input time interval).

[0139] Next, the learning system (100) performs a learning data labeling step of labeling different SOXs in the learning data (S130).

[0140] The order of interpolation and labeling can be interchanged. That is, after labeling, the training data can be interpolated, and then SOX can be labeled on the interpolated training data.

[0141] In other words, the learning data labeling step labels the learning data with SOX so that the deep neural network can learn various battery states.

[0142] Next, the learning system (100) performs a learning model generation step of generating a learning model for battery state estimation by training a deep neural network using labeled learning data (S140).

[0143] Also, as a procedure for estimating the state of a battery installed in an actual user's application using a learning model for battery state estimation, first, the battery state estimation system (300) receives measurement data for the battery from the user's application (S210).

[0144] The above measurement data includes characteristic values ​​of the battery, and may further include battery specifications as characteristic values.

[0145] Next, the battery state estimation system (300) converts the received actual measurement data into input data to configure input data for inputting a learning model for battery state estimation (S220).

[0146] Next, the battery state estimation system (300) inputs input data into a learning model for battery state estimation and performs a SOX estimation step (S230).

[0147] After this, the battery state estimation system (300) performs a battery state estimation step of estimating the state of the corresponding battery based on the estimated SOX label (S240).

[0148] As described above, estimating the battery status is performed by calculating the run-time current status of the battery based on the estimated SOX label.

[0149] As described above, the present invention recognizes the idle period of the battery at run time, collects measurement data of the battery to generate learning data, and labels the generated learning data with SOX indicating the state of the battery to train a deep neural network, thereby creating a deep neural network learning model capable of more accurately estimating the state of the battery.

[0150] In addition, the present invention has the effect of creating a more sophisticated deep neural network learning model with learning data that accurately reflects actual operating conditions without using complex calculation procedures or statistical data.

[0151] In addition, the present invention estimates and provides the exact status of the battery, thereby enabling the timing of battery replacement to be determined, and thereby enabling stable and continuous use of the application-side device by replacing the battery at the right time.

[0152] In addition, the present invention has the effect of shortening the battery test period by accelerating the development of the design of cells constituting the battery and the fast charging methodology through accurate estimation of the battery state.

[0153] In addition, although the preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and various modifications can be implemented by a person having ordinary skill in the art to which the invention pertains without departing from the gist of the present invention claimed in the claims, and such modifications should not be understood individually from the technical idea or prospect of the present invention.

[0154] As described above, the present invention recognizes the idle period of a battery at run time, measures the run-time characteristic value of the battery to generate learning data, and labels the generated learning data with SOX indicating the state of the battery to train a deep neural network, thereby generating a deep neural network learning model capable of more accurately estimating the state of the battery, and can generate a more sophisticated deep neural network learning model with learning data that accurately reflects the actual operating situation without using complex calculation procedures or statistical data, and by estimating and providing the accurate state of the battery, the timing of battery replacement can be determined, and thereby, if the battery is replaced at the right time, stable and continuous use of the application-side device is enabled, and furthermore, by estimating the accurate state of the battery, the design of cells constituting the battery and the development of a high-speed charging methodology can be accelerated, thereby shortening the test period of the battery, and thus has industrial applicability.

Claims

1. In a method for labeling learning data of a deep neural network for estimating the state of a battery, A learning data configuration step for configuring learning data by collecting measurement data on the status of the above battery and converting it into learning data; and A learning data labeling step for automatically labeling the learning data with SOC (state of charge), SOH (state of health), SOP (state of power), or SOX including a combination thereof according to the state of the battery in the above-described learning data is included; A learning data labeling method characterized in that the above learning data includes feature values collected during a relaxation period in the run-time of the battery.

2. In claim 1, The above characteristic values include the open circuit voltage (OCV), voltage (V), current (I), charge (Q), temperature (T), or a combination thereof of the battery measured during a relaxation period in the run-time of the battery. A learning data labeling method characterized in that the learning data further includes statistical values including the mean, variance, median, standard deviation, difference, or a combination thereof extracted from the collected feature values.

3. In claim 1, The above learning data configuration step is: A rest period recognition step for recognizing a stabilized section in which the measurement data changes within a predetermined range for a predetermined period of time or longer in the run-time of the battery as a rest period; A measurement data extraction step for extracting the measurement data from the above-described rest period; A SOX calculation step for calculating SOX including SOC, SOH, SOP, or a combination thereof of the battery from the extracted measurement data; and A learning data labeling method characterized by comprising a learning data conversion step of converting the extracted measurement data into learning data in a format for training the deep neural network.

4. In claim 1, In the above learning data labeling step, the run-time SOH is In the OCV-SOC between two adjacent rest periods, the OCV-SOC measured at the first point of the previous rest period is divided by the OCV-SOC measured at the second point of the later rest period, and A learning data labeling method characterized in that the first and second points are set by detecting a flat area in the OCV-SOC curve of each rest period in which the amount of OCV change is gentle or almost non-existent.

5. In claim 1, A learning data labeling method characterized in that, in the learning data labeling step, the output power (P) is calculated as voltage (V) / current (I) through the OCV, voltage (V), and current (I) measured in the flat area with the smallest OCV change in the OCV-SOC curve of each resting period, and when the maximum power (Pmax) of the battery is known, the SOP is calculated from the output power (P) / maximum power (Pmax) and labeled.

6. In claim 1, The above learning data labeling method is, The above deep neural network further includes a learning data expansion step for expanding the learning data so that the deep neural network estimates the SOX at all data points other than the rest period; The above extension is a learning data labeling method characterized in that the learning data is constructed by interpolating data extracted through a predetermined statistical operation using the feature values between run-time feature values extracted in each of two consecutive rest periods on the time axis or by padding with predetermined fixed data.

7. A learning method characterized in that it creates a deep neural network learning model that estimates the state of the battery through the learning data labeling method of the deep neural network according to at least one of claims 1 to 6.

8. A learning system characterized by comprising: a memory storing a program configured to generate a deep neural network learning model for estimating the state of the battery through the learning data labeling method of the deep neural network according to at least one of claims 1 to 6; and a processor configured to execute the program stored in the memory.

9. A battery state estimation method characterized in that the state of the battery is estimated by measuring the characteristic value of the battery and inputting it into the deep neural network learning model that estimates the state of the battery through the learning data labeling method of the deep neural network according to at least one of claims 1 to 6.

10. A battery state estimation system, comprising: a memory storing a program configured to estimate the state of the battery by measuring a feature value of the battery and inputting it into the deep neural network learning model, wherein the memory stores a program configured to estimate the state of the battery by inputting the measured feature value of the battery into the deep neural network learning model, which is generated through the learning data labeling method of the deep neural network according to at least one of claims 1 to 6; and a processor configured to execute the program stored in the memory.

Images