Method for generating learning model for estimating battery state through conditional learning

Abstract

The present invention relates to a method for generating a learning model for estimating a battery state through conditional learning. The method for generating a learning model solves the problem wherein complexity increases exponentially when learning massive training data via one learning network. The problem is solved by: classifying feature values of batteries according to a plurality of conditions to configure conditional training data; and generating a learning model for estimating a battery state according to each of the conditions by learning the configured conditional training data via a separate independent learning network.

Description

A method for creating a learning model for battery state estimation using conditional learning.

[0001] The present invention relates to a method for generating a learning model for estimating a battery state through conditional learning, and more specifically, to a learning model generation method that solves the problem of exponential increase in complexity when learning a large amount of learning data through a single learning network by classifying characteristic values ​​of a battery according to a plurality of conditions to configure learning data for each condition, and generating a learning model for estimating a battery state according to each condition by learning the configured learning data for each condition through a separate, independent learning network.

[0002] As batteries (secondary batteries) that can be used for a long time until their lifespan ends by repeatedly charging and discharging are developed and commercialized, batteries are being used as an essential power source for various application devices such as electric vehicles and smartphones.

[0003] A battery is a device that stores external electrical energy in the form of chemical energy and generates electricity when needed, and has the effect of supplying electrical energy to application devices without a continuous connection to a commercial power source.

[0004] Estimating the battery state is crucial for efficiently managing the battery state and improving safety in various application devices.

[0005] Here, the battery status may include SOC (state of charge), SOH (state of health), SOP (state of power), etc.

[0006] Meanwhile, with the recent rapid development of artificial intelligence (AI) technology, research is being conducted on devices that use AI to estimate the condition of batteries.

[0007] In order to estimate the state of a battery using artificial intelligence, a learning network must be trained using the battery's characteristic values ​​to create a learning model for estimating the state of the battery.

[0008] Creating a conventional artificial intelligence-based learning model is structured to train a learning network by extensively applying large-scale feature values.

[0009] Typically, creating AI-based learning models relies on large amounts of data. Furthermore, conventional methods for creating learning models to estimate battery condition also rely on large amounts of data, as measurements are taken throughout the battery's lifespan.

[0010]

[0011] *If you train a large amount of training data like this through a single training network, the training time will be long, and the complexity of the calculations and the number of training parameters will increase exponentially with the size of the training data, and the training model created through this will inevitably become very heavy.

[0012] Accordingly, the present invention proposes a method to reduce complexity and improve performance by optimizing the size and depth of each battery state estimation learning model by generating a learning model for battery state estimation for each condition through conditional learning in which the learning data is configured according to conditions and the learning network is trained according to the conditions, rather than learning a large amount of learning data for battery state estimation through a single learning network.

[0013] That is, the present invention proposes a method of classifying feature values ​​according to conditions to configure learning data, and training each learning network corresponding to the conditions with the learning data configured according to the conditions to create a learning model for estimating a battery state according to the conditions.

[0014] 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.

[0015] First, Korean Patent No. 2171807 (October 23, 2020) relates to a system for predicting product defects during a process and a method for creating a learning model for predicting defects. The system collects measurement information, which is metadata for each process, including voltage, current, and accumulated gas volume, and accumulates the collected measurement information for each process to create a learning model for predicting defects using pattern learning through a deep neural network (DNN), an artificial neural network (ANN), etc., and relates to a system for predicting product defects during a process and a method for creating a learning model for predicting defects.

[0016] That is, Korean Patent No. 2171807 trains a learning network using learning data that has accumulated measurement information for each process to create a learning model for predicting defects. However, there is a problem in that the complexity of the learning network increases exponentially because a single learning network is trained using a large amount of learning data that has accumulated all measurement information for each process.

[0017] On the other hand, the present invention improves the complexity and performance of each battery state estimation learning model by creating a battery state estimation learning model conditionally through conditional learning in which learning data is configured for each condition and each learning network corresponding to each condition is trained using each learning data configured for each condition. Korean Patent Registration No. 2171807 does not describe, suggest, or imply any technical features of the present invention.

[0018] In addition, Korean Patent Publication No. 2021-0041483 (April 15, 2021) relates to a method and device for generating learning data for an artificial intelligence model, which collects virtual environment sensing data using a virtual sensor in a simulation environment, converts the virtual environment sensing data into real environment sensing data, and then generates learning data for learning an artificial intelligence model.

[0019] In other words, Korean Patent Publication No. 2021-0041483 simply generates learning data for training an artificial intelligence model using all real-world sensing data.

[0020] However, the present invention is for estimating the state of a battery, and for conditional learning, feature values ​​are classified by condition and learning data is constructed for training a learning network corresponding to each condition.

[0021] Therefore, there are significant differences between Korean Patent Publication No. 2021-0041483 and the present invention in terms of their technical composition, purpose, and effect.

[0022] The present invention was created to solve the above problems, and its purpose is to provide a method for generating a learning model for estimating a battery state through conditional learning, which generates a plurality of learning models for estimating a battery state through conditional learning.

[0023] In addition, the present invention provides a method for generating a learning model for estimating a battery state by reducing complexity and improving performance for each of a plurality of conditions by classifying feature values ​​according to conditions, constructing learning data for each condition using the classified feature values, and training a learning network for each condition using the learning data constructed for each condition.

[0024] In addition, the present invention aims to provide a method for estimating the state of an actual battery at high speed / high performance for each of the above conditions using the learning model for estimating the state of the battery generated above.

[0025] A method for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention includes a feature value classification step for classifying feature values ​​measured from a battery according to preset conditions, a learning data configuration step for configuring learning data according to the conditions using the classified feature values, and a learning step for generating a learning model for estimating a battery state for each condition by training a learning network corresponding to each condition using the learning data configured according to the conditions, and is characterized in that the size and depth of the learning model for estimating a battery state are optimized according to the conditions through the learning, thereby improving complexity and performance compared to configuring the learning model for estimating a battery state as a single model.

[0026] In addition, the above characteristic values ​​include voltage (V), current (I), temperature (T), charge / discharge rate (C-rate) or a combination thereof, and the battery state is characterized by including state of charge (SOC), state of energy (SOE), state of health (SOH), state of power (SOP), remaining useful life (RUL) or a combination thereof.

[0027] In addition, the above conditions are characterized in that they set the voltage, current, temperature, charge / discharge rate, or a combination and range thereof of the battery constituting the above characteristic values.

[0028] In addition, the above learning data configuration step is characterized by configuring learning data according to each of the above conditions by arranging the voltage, current, temperature, charge / discharge rate, or a combination thereof of the battery, which constitute feature values ​​separated by the above conditions so as to be suitable for the input of the learning network corresponding to each of the above conditions, and labeling the actual battery state.

[0029] In addition, a method for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention is characterized by including a feature value classification step of classifying, in an agent, feature values ​​measured from a battery according to preset conditions, a learning data configuration step of configuring learning data according to the conditions using the classified feature values ​​in the agent, and a reinforcement learning step of generating a learning model for estimating a battery state according to the conditions by performing reinforcement learning on a learning network corresponding to each condition using the learning data configured according to the conditions in the agent.

[0030] In addition, the learning model creation method is characterized in that it further includes a condition setting step of setting conditions including voltage, current, temperature, charge / discharge rate, or a combination and range thereof of the battery constituting the feature value in the agent.

[0031] In addition, the above reinforcement learning step includes a reinforcement learning result output step for inputting learning data for each of the above conditions into each learning network corresponding to each of the above conditions in the above reinforcement learning process to output a reinforcement learning result which is the output of each of the above learning networks, and a reward calculation step for calculating a reward according to the reinforcement learning result output for each of the above learning networks, and the reinforcement learning is characterized in that the reward calculated for each of the above learning networks is applied to each of the above learning networks to output a reinforcement learning result in which the reward is maximized.

[0032] In addition, the reinforcement learning step is characterized in that the reinforcement learning is performed when the reward exceeds a predetermined threshold value, and the reinforcement learning is not performed when the reward is below the predetermined threshold value.

[0033] Meanwhile, a device for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention is characterized in that it generates a learning model for estimating a battery state according to the method for generating a learning model for estimating a battery state through conditional learning.

[0034] Meanwhile, a method for estimating a battery state through conditional learning according to another embodiment of the present invention is characterized in that it is configured to estimate a battery state through a learning model for estimating a battery state generated according to a method for generating a learning model for estimating a battery state through conditional learning.

[0035] Meanwhile, a battery state estimation device using conditional learning according to another embodiment of the present invention is characterized in that it is configured to estimate a battery state using a learning model for battery state estimation generated according to a method for generating a learning model for battery state estimation using conditional learning.

[0036] As described above, the present invention generates a battery state estimation learning model for each condition through conditional learning, and optimizes the size and depth of each battery state estimation learning model according to the conditions, thereby reducing complexity and improving performance compared to configuring a single battery state estimation learning model.

[0037] In addition, the present invention has the effect of preventing damage caused by battery deterioration, etc. in advance by inputting actual feature values ​​into the learning model for battery estimation according to conditions to estimate the state of the battery accurately and quickly and providing it to the user by creating a learning model for battery estimation according to conditions.

[0038] FIG. 1 is a diagram illustrating a method for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention.

[0039] FIG. 2 is a diagram illustrating in detail a method for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention.

[0040] FIG. 3 is a diagram illustrating a method for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention.

[0041] FIG. 4 is a diagram illustrating in detail a method for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention.

[0042] FIG. 5 is a block diagram illustrating a device for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention.

[0043] FIG. 6 is a block diagram illustrating a device for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention.

[0044] Fig. 7 is a block diagram showing the configuration of a battery state estimation device according to one embodiment of the present invention.

[0045] Figure 8 is a flowchart illustrating a procedure for generating a learning model for estimating a battery state according to conditions according to one embodiment of the present invention.

[0046] Figure 9 is a flowchart illustrating a procedure for generating a learning model for estimating a battery state according to conditions according to another embodiment of the present invention.

[0047] Figure 10 is a flowchart illustrating a procedure for estimating a battery state according to one embodiment of the present invention.

[0048] [Description of symbols] 100, 200: Device for generating a learning model for estimating a battery state through conditional learning; 110, 210, 310: Feature value receiving unit; 220: Agent; 120, 222: Feature value classification unit; 221: Condition setting unit; 130, 223: Learning data composition unit; 140: Learning unit; 224: Reward calculation unit; 225: Reinforcement learning unit; 300: Battery state estimation device; 310: Feature value receiving unit; 320: Condition matching unit; 330: Learning model loading unit; 340: Input data composition unit; 350: Battery state estimation unit; 360: Estimation result providing unit.

[0049] Hereinafter, a preferred embodiment of a method for generating a learning model for estimating a battery state through conditional learning according to the present invention will be described in detail with reference to the attached drawings. The same reference numerals in each drawing represent the same elements. In addition, specific structural and functional descriptions of embodiments of the present invention are merely illustrative for the purpose of explaining 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 defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning they have in the context of the related technology, and are preferably not interpreted in an ideal or excessively formal sense unless explicitly defined herein.

[0050] FIG. 1 is a diagram illustrating a method for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention.

[0051] As illustrated in FIG. 1, a device (100) for generating a learning model for estimating a battery state through conditional learning according to an embodiment of the present invention (hereinafter referred to as a learning model generating device) classifies the characteristic values ​​of a battery according to conditions set in advance, configures learning data for each condition for learning through a separate independent learning network, and trains each learning network using each configured learning data to generate a learning model for estimating a battery state for each condition.

[0052] The above characteristic values ​​include voltage (V), current (I), temperature (T), charge / discharge rate (C-rate) measured from the battery, or a combination thereof.

[0053] At this time, the feature values ​​may further include voltage statistics, current statistics, temperature statistics, or a combination thereof. Each of the above statistics may include the mean, variance, median, standard deviation, or a combination thereof for voltage, current, temperature, and charge / discharge rate.

[0054] The above conditions refer to voltage, current, temperature, charge / discharge rate, each statistical value, or a combination and range thereof. At least one of the above conditions may be set.

[0055] For example, if condition 1 defines specific ranges for voltage, current, temperature, and charge / discharge rate, and feature value 1 is configured to include voltage, current, temperature, and charge / discharge rate and falls within each specific range, the learning model creation device (100) classifies feature value 1 as condition 1.

[0056] The above learning network is configured and equipped for each of the above conditions. For example, if Condition 1 sets voltage, current, temperature, and each range, the learning network corresponding to Condition 1 is configured to learn learning data consisting of voltage, current, and temperature corresponding to each range.

[0057] That is, each learning network is configured to learn by receiving input feature values ​​(i.e., voltage, current, temperature, charge / discharge rate, statistical values, or a combination thereof) corresponding to each condition.

[0058] The above battery state includes state of charge (SOC), state of health (SOH), state of energy (SOE), state of power (SOP), remaining useful life (RUL), or a combination thereof.

[0059] The above SOC or SOE is information that tells you how long the battery can be used, SOH is information that shows how good the battery's performance is, SOP is information that shows how powerful the battery is, and RUL is information that shows how much life the battery has left.

[0060] Below, the process of creating a learning model for estimating battery status through learning is described in detail with reference to Fig. 2.

[0061] FIG. 2 is a diagram illustrating in detail a method for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention.

[0062] As illustrated in FIG. 2, the learning model generation device (100) according to one embodiment of the present invention classifies the received feature values ​​according to pre-set conditions when the feature values ​​measured from the battery are received.

[0063] At this time, in order to classify the feature values ​​according to each condition, each data (voltage, current, temperature, etc.) that constitutes the feature values ​​must all meet the conditions.

[0064] For example, if condition 1 sets the ranges of voltage, current, temperature, and charge / discharge rate as conditions, and the data constituting feature value 1 is composed of voltage, current, temperature, and charge / discharge rate, and the data constituting feature value 2 is composed of voltage and temperature, and each data satisfies all of the above ranges, then feature value 1 is classified as condition 1, but feature value 2 is not classified as condition 1 because voltage and temperature satisfy condition 1, but current and charge / discharge rate do not exist.

[0065] Conversely, in the above example, if condition 1 sets each range of voltage and temperature as a condition, feature values ​​1 and 2 can be classified as condition 1 because they satisfy condition 1, and at this time, the current and charge / discharge rate of feature value 1 can be deleted.

[0066] Additionally, the learning model generation device (100) configures learning data for each condition using feature values ​​classified according to the conditions.

[0067] The above learning data is composed by arranging feature values ​​according to the input of the learning network corresponding to each condition and labeling the actual battery status.

[0068] The actual battery status described above is added based on feature values ​​and can be calculated using a predetermined method based on the feature values. Since the present invention only requires configuring the learning data conditionally, there are no restrictions on the method for labeling the actual battery status with feature values.

[0069] In addition, the learning network illustrated in Fig. 2 illustrates a DNN (deep neural network). However, the present invention is not limited thereto, and may be configured with various artificial intelligence learning networks other than DNN, such as a DCNN (deep convolutional neural network), a transformer, a TCNN (temporal convolution neural network), a CNN (convolutional neural network), an RNN (recurrent neural network), and a GRNN (general regression neural network).

[0070] For example, if a DNN configured including multiple input nodes is assumed as a learning network, the data input to each input node is determined. That is, if Condition 1 is about voltage, current, and temperature, and the first input node, second input node, and third input node of the DNN corresponding to Condition 1 are set to input current, voltage, and temperature, respectively, then the learning model creation device (100) arranges the feature values ​​classified according to Condition 1 into current, voltage, and temperature to configure learning data corresponding to Condition 1.

[0071] Organizing these learning data by condition means converting the feature values ​​classified by condition into a data format suitable for the learning network.

[0072] In addition, the learning model generation device (100) trains a learning network according to each condition using learning data for each condition, thereby generating a learning model for estimating the battery status for each condition.

[0073] The above learning is performed in a way that minimizes the error between the output of each learning network (learning result) and the label of the learning data (i.e., the correct answer) when learning data is input during the learning process.

[0074] At this time, the learning model generation device (100) performs the learning by updating the weights of each learning network to minimize the error through the back propagation method.

[0075] The input of the learning model for estimating battery status by condition generated in this way is the feature value for each condition, and the output is the battery status according to the input.

[0076] FIG. 3 is a diagram illustrating a method for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention.

[0077] As illustrated in FIG. 3, a learning model generation device (200) according to another embodiment of the present invention is configured to include an agent (220).

[0078] The above agent (220) sets conditions, and when a characteristic value for a battery is received, the characteristic value is classified according to the set conditions.

[0079] Additionally, the learning model generation device (200) configures learning data for each condition using feature values ​​classified according to the conditions.

[0080] The above classification and training data configuration are performed in the same manner as described with reference to Fig. 2, and a detailed description thereof will be omitted.

[0081] In addition, the learning model generation device (100) uses learning data for each condition to perform reinforcement learning on a learning network for each condition, thereby generating a learning model (i.e., a reinforcement learning model) for battery state estimation for each condition.

[0082] The input of the learning model for estimating the battery status according to each condition generated through the above reinforcement learning is the feature value according to each condition, and the output is the battery status according to the input.

[0083] FIG. 4 is a diagram illustrating in detail a method for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention.

[0084] As illustrated in FIG. 4, the agent (220) of the learning model generation device (100) according to one embodiment of the present invention sets at least one condition, and when a feature value is received, classifies the received feature value according to the pre-set conditions.

[0085] Additionally, the agent (220) configures learning data for each condition using feature values ​​classified by condition.

[0086] The above agent (220) uses learning data configured for each condition to reinforce learning of a learning network for each condition, thereby generating a learning model for estimating the battery status for each condition.

[0087] The above agent (220) calculates a reward based on the reinforcement learning result (output result) of each learning network for the learning data input during the reinforcement learning process.

[0088] The above reward can be calculated based on the error between the reinforcement learning results of the learning network and the actual battery status. For example, the agent (220) can be configured to produce a higher reward as the error approaches zero, and a lower reward as the error increases.

[0089] Additionally, the agent (220) performs reinforcement learning for each learning network to output a reinforcement learning result that maximizes the reward by utilizing the reward.

[0090] At this time, the agent (220) is configured to not perform reinforcement learning if the reward is below a predetermined value, and to perform reinforcement learning if the reward exceeds the predetermined value. Performing reinforcement learning in this manner has the advantage of being able to estimate the optimal battery condition by adapting to changing feature values ​​in real time.

[0091] FIG. 5 is a block diagram illustrating a device for generating a learning model for estimating a battery state through conditional learning according to one embodiment of the present invention.

[0092] As illustrated in FIG. 5, a learning model generation device (100) according to one embodiment of the present invention is configured to include a feature value receiving unit (110), a feature value classification unit (120), a learning data configuration unit (130), and a learning unit (140).

[0093] The above characteristic value receiving unit (110) receives the characteristic value of the battery measured from the battery.

[0094] The above characteristic values ​​are measured according to the life cycle of the battery in various usage environments, and may be received from a database (not shown) storing the characteristic values, or from a user's application device (e.g., electric vehicle, smartphone, etc.).

[0095] The above-described feature value classification unit (120) classifies feature values ​​according to pre-set conditions. Classifying the above-described feature values ​​has been described with reference to FIG. 2, and thus is omitted here.

[0096] The above learning data configuration unit (130) configures learning data for each condition to train a learning network corresponding to each condition using feature values ​​classified according to the conditions.

[0097] The above learning data is converted into data for training a learning network according to the conditions by converting feature values ​​classified by conditions. The configuration of the above learning data by conditions has been described with reference to Fig. 2, so it is omitted here.

[0098] The above learning unit (140) trains a learning network corresponding to each condition using learning data configured for each condition, thereby generating a learning model for estimating the battery status according to the conditions.

[0099] FIG. 6 is a block diagram illustrating a device for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention.

[0100] As illustrated in FIG. 6, a learning model generation device (200) according to another embodiment of the present invention is configured to include a feature value receiving unit (210) and an agent (220).

[0101] The above-described feature value receiving unit (210) receives feature values. The above-described feature value receiving unit (210) is provided to perform the same function as the feature value receiving unit (110) described with reference to FIG. 5.

[0102] The above agent (220) is configured to generate a learning model for estimating a battery state according to conditions by performing reinforcement learning on a condition-specific learning network using the above feature values, and includes a condition setting unit (221), a feature value classification unit (222), a learning data composition unit (223), a reward calculation unit (224), and a reinforcement learning unit (225).

[0103] The above condition setting unit (221) sets at least one condition. The above conditions are set to classify feature values.

[0104] The above condition setting unit (221) sets the above conditions by setting the voltage, current, temperature, charge / discharge rate, voltage statistical values, temperature statistical values, charge / discharge rate statistical values, or a combination and range thereof that constitute the characteristic values.

[0105] The above-mentioned feature value classification unit (222) classifies the feature values ​​according to the above-mentioned set conditions, and the learning data configuration unit (223) configures learning data according to the above-mentioned conditions using the above-mentioned classified feature values.

[0106] The above reinforcement learning unit (225) is for reinforcement learning learning data classified by condition, inputting each learning data classified by condition into a learning network corresponding to each condition, and providing the reinforcement learning result, which is the output of each learning network, to the reward calculation unit (224).

[0107] At this time, the reward calculation unit (224) calculates a reward according to the reinforcement learning result for each learning network and provides it to the reinforcement learning unit (225).

[0108] Afterwards, the reinforcement learning unit (225) performs reinforcement learning for each learning network to output a reinforcement learning result that maximizes the reward by applying a reward according to the input learning data to each learning network.

[0109] The process of calculating the above reward and performing reinforcement learning is described in detail with reference to Fig. 4, so it is omitted here.

[0110] Additionally, estimating the battery state using a learning model for battery estimation can be performed through a separate battery state estimation device (300).

[0111] Additionally, the learning model generation device (100, 200) described with reference to FIGS. 5 and 6 is preferably constructed as a server on the cloud to generate a learning model for battery state estimation. However, the present invention is not limited thereto and may be constructed on a user application device equipped with a battery.

[0112] Additionally, estimating the battery state using a learning model for battery estimation can be performed through a separate battery state estimation device (300). The battery state estimation device (300) can also be built in the form of a server on the cloud or built locally (i.e., on a user application device).

[0113] Meanwhile, a device for generating a learning model for estimating a battery state through conditional learning according to another embodiment of the present invention is characterized by including a memory for storing a program for generating a learning model for estimating a battery state according to the method for generating a learning model for estimating a battery state through conditional learning, and a processor for loading the program stored in the memory to generate the learning model for estimating a battery state.

[0114] Hereinafter, a battery state estimation device (300) for estimating a battery state using a learning model for estimating a battery state generated through a learning model generation device (100, 200) of the present invention will be described.

[0115] Fig. 7 is a block diagram showing the configuration of a battery state estimation device according to one embodiment of the present invention.

[0116] As illustrated in FIG. 7, a battery state estimation device (300) according to one embodiment of the present invention is configured to include a feature value receiving unit (310), a condition matching unit (320), a learning model loading unit (330), an input data configuration unit (340), a battery state estimation unit (350), and an estimation result providing unit (360).

[0117] The above-mentioned feature value receiving unit (310) receives feature values ​​from a user application device equipped with an actual battery.

[0118] Among the above characteristic values, statistical values ​​(voltage, current, temperature, charge / discharge rate) can be received from a user application device or calculated from a battery state estimation device (200).

[0119] That is, the battery state estimation device (200) can receive voltage, current, temperature, charge / discharge rate, or a combination thereof from a user application device, and calculate statistical values ​​for the same for a predetermined period of time.

[0120] The above condition matching unit (320) matches a specific condition that matches the feature value among at least one condition set in advance.

[0121] The above learning model loading unit (330) loads a learning model for battery state estimation corresponding to a specific condition from a database (or memory).

[0122] The above input data configuration unit (340) configures the input data of the battery state estimation learning model by converting the feature values ​​into a data format suitable for input of the loaded battery state estimation learning model.

[0123] That is, the input data configuration unit (340) configures the input data by arranging (arranging) each data that constitutes the feature value to fit the input of the learning model for battery status estimation.

[0124] The above battery state estimation unit (350) inputs the configured input data into a learning model for battery state estimation and estimates the battery state according to the output of the learning model for battery state estimation.

[0125] The above estimation result provision unit (360) provides the result of estimating the battery status to the user application device.

[0126] Meanwhile, if the battery status estimation device (300) determines that there is no matching condition as a result of matching through the condition matching unit (320), it determines that there is a problem with the battery and provides a notification to the user application device.

[0127] The above-mentioned characteristic values ​​of voltage, current, temperature, and charge / discharge rate are interrelated and change as the battery is used. However, if at least one of the voltage, current, temperature, and charge / discharge rate values ​​is extremely high or extremely low, it is clear that a problem has occurred in the battery. Therefore, the battery condition estimation device (300) determines this through condition matching and determines whether a problem has occurred in the battery.

[0128] Meanwhile, a battery state estimation device through conditional learning according to another embodiment of the present invention is characterized by including a memory storing a program configured to estimate a battery state using a learning model for battery state estimation generated according to a method for generating a learning model for battery state estimation through conditional learning, and a processor estimating the battery state by loading the program stored in the memory.

[0129] Figure 8 is a flowchart illustrating a procedure for generating a learning model for estimating a battery state according to conditions according to one embodiment of the present invention.

[0130] FIG. 8 illustrates a procedure for generating a learning model for battery state estimation according to conditions based on the learning model generation device (100) described with reference to FIG. 5.

[0131] As illustrated in Fig. 8, the procedure for generating a learning model for estimating a battery state conditionally is as follows: First, the learning model generation device (100) performs a feature value receiving step for receiving the feature values ​​of the battery (S110).

[0132] Here, the feature values ​​are received for learning and can be received from a database (not shown) or a user application device.

[0133] Next, the learning model generation device (100) performs a feature value classification step of classifying feature values ​​according to at least one condition set in advance (S120), and performs a learning data configuration step of configuring learning data for each condition using the classified feature values ​​(S130).

[0134] Classifying the above feature values ​​and configuring learning data have been described with reference to Fig. 2, so a detailed description is omitted.

[0135] Next, a learning step is performed by inputting the configured learning data into a learning network equipped with conditions to learn (S140).

[0136] Afterwards, when training for each conditional learning network is completed using all training data (S150), the completed learning network for each condition becomes a learning model for estimating the battery status for each condition. As described above, the learning is performed using the backpropagation method.

[0137] Figure 9 is a flowchart illustrating a procedure for generating a learning model for estimating a battery state according to conditions according to another embodiment of the present invention.

[0138] FIG. 9 illustrates a procedure for generating a learning model for battery state estimation according to conditions based on the learning model generation device (200) described with reference to FIG. 6.

[0139] As illustrated in Fig. 9, the procedure for generating a learning model for estimating a battery state conditionally is as follows: First, the learning model generation device (200) performs a feature value receiving step of receiving feature values ​​in the feature value receiving unit (210) (S110).

[0140] As described above, the feature values ​​here are received for learning. Next, the learning model creation device (200) performs a feature value classification step (S220) in which the feature values ​​are classified according to at least one condition set in advance in the agent (220), and performs a learning data configuration step in which learning data is configured for each condition using the classified feature values ​​(S230).

[0141] Classifying the above feature values ​​and configuring learning data have been described with reference to Fig. 4, so a detailed description is omitted.

[0142] The above learning model creation device (200) can set conditions through a condition setting step of setting at least one condition through an agent (220).

[0143] Next, the learning model creation device (200) performs a learning result output step (S240) in which the configured learning data is input into a learning network equipped with conditions in the agent (220) to output learning results, and a reward calculation step (S250) in which a reward is calculated according to the output learning results.

[0144] Here, the learning result refers to the output result of each learning network according to the learning data input during the reinforcement learning process.

[0145] Next, a reinforcement learning step is performed in which the agent (220) applies the reward calculated according to the learning result of each learning network to each learning network, thereby performing reinforcement learning so that the learning result that maximizes the reward according to the input is output in the conditional learning network (S260).

[0146] Thereafter, the learning model generation device (200) performs reinforcement learning on each conditional learning network until all learning data is used (S270). At this time, each conditional learning network for which reinforcement learning has been completed becomes a learning model for estimating the battery status for each condition.

[0147] Figure 10 is a flowchart illustrating a procedure for estimating a battery state according to one embodiment of the present invention.

[0148] As illustrated in FIG. 10, in a procedure for estimating a battery state according to one embodiment of the present invention, first, a battery state estimation device (300) performs a feature value receiving step of receiving a feature value (S310).

[0149] Here, the feature value refers to an actual feature value for estimating the battery status and is received from the user application device.

[0150] Additionally, statistical values ​​that constitute the feature values ​​may be produced by the battery state estimation device (300).

[0151] Next, the battery state estimation device (300) performs a condition matching step of matching conditions corresponding to the characteristic values ​​(S320).

[0152] That is, the condition matching step performs the matching by checking whether a condition to which the feature value belongs exists among multiple conditions set in advance.

[0153] Next, if there is a condition matching the feature value as a result of the matching (S330), the battery state estimation device (300) performs a learning model loading step of loading a learning model for battery state estimation corresponding to the matched condition (S340).

[0154] Next, the battery state estimation device (300) performs an input data configuration step of configuring input data to be input into the loaded battery state estimation learning model (S350).

[0155] Configuring the above input data means arranging each data that constitutes the feature value to fit the input of the learning model for battery state estimation.

[0156] Next, the battery state estimation device (300) performs a battery state estimation step of estimating the battery state by inputting the configured input data into a learning model for battery state estimation (S360), and performs an estimation result provision step of providing the estimation result to the corresponding user application device (S370).

[0157] Meanwhile, if the condition corresponding to the characteristic value is not matched (S340), the battery status estimation device (300) performs a notification provision step of notifying the user application device that an abnormality has occurred in the battery (S341).

[0158] As described above, the present invention organizes learning data by classifying feature values ​​according to conditions, and trains each learning network corresponding to the conditions through the organized learning data to create a learning model for estimating a battery state according to the conditions, thereby solving the problem of exponential increase in complexity due to learning the entire learning data through a single learning network, thereby completing learning in a short period of time, reducing the complexity of each learning model, making it lightweight, and improving performance.

[0159] 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.

[0160] As described above, the present invention can reduce complexity and improve performance compared to configuring a single battery state estimation learning model by generating a battery state estimation learning model for each condition through conditional learning and optimizing the size and depth of each battery state estimation learning model according to the conditions, and further, by generating a learning model for battery estimation for each condition, the actual feature value is input into the learning model for battery estimation for each condition, thereby accurately estimating the state of the battery at high speed and providing it to the user, thereby preventing damage caused by battery deterioration in advance, and thus has industrial applicability.

Claims

1. A feature value classification step that classifies feature values measured from a battery according to preset conditions; A learning data composition step for composing learning data according to the above conditions using the above-classified feature values; and A learning step for generating a learning model for estimating a battery state for each condition by training a learning network corresponding to each condition using learning data configured according to the above conditions; A method for generating a learning model for battery state estimation through conditional learning, characterized in that the size and depth of the learning model for battery state estimation are optimized according to the above conditions through the above learning, thereby improving complexity and performance compared to configuring the learning model for battery state estimation as a single model.

2. In claim 1, The above characteristic values include voltage (V), current (I), temperature (T), charge / discharge rate (C-rate), or a combination thereof. A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the battery state includes SOC (state of charge), SOE (state of energy), SOH (state of health), SOP (state of power), RUL (remaining useful life), or a combination thereof.

3. In claim 1, The above conditions are, A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the voltage, current, temperature, charge / discharge rate, or a combination and range thereof of the battery constituting the above characteristic values are set.

4. In claim 1, The above learning data configuration step is: A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the voltage, current, temperature, charge / discharge rate, or a combination thereof of a battery, which constitute feature values separated by the above conditions to be suitable for the input of a learning network corresponding to each of the above conditions, is arranged, and the actual battery state is labeled, thereby configuring learning data according to each of the above conditions.

5. A feature value classification step in which the feature values measured from the battery are classified according to preset conditions in the agent; In the above agent, a learning data configuration step for configuring learning data according to the above conditions with the classified feature values; and In the above agent, a reinforcement learning step is included to generate a learning model for estimating the battery status for each condition by performing reinforcement learning on a learning network corresponding to each condition using learning data configured according to the above conditions; A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the size and depth of the learning model for estimating a battery state are optimized according to the above conditions through the above reinforcement learning, thereby improving complexity and performance compared to configuring the learning model for estimating a battery state as a single unit.

6. In claim 5, The above characteristic values include voltage (V), current (I), temperature (T), charge / discharge rate (C-rate), or a combination thereof. A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the battery state includes SOC (state of charge), SOE (state of energy), SOH (state of health), SOP (state of power), RUL (remaining useful life), or a combination thereof.

7. In claim 6, The above learning model creation method is: A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the method further comprises a condition setting step for setting conditions including voltage, current, temperature, charge / discharge rate, or a combination and range thereof of the battery constituting the feature value in the agent.

8. In claim 6, The above learning data configuration step is: A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the voltage, current, temperature, charge / discharge rate, or a combination thereof of a battery, which constitute feature values separated by the above conditions to be suitable for the input of a learning network corresponding to each of the above conditions, is arranged, and the actual battery state is labeled, thereby configuring learning data according to each of the above conditions.

9. In claim 6, The above reinforcement learning step is, In the process of performing the above reinforcement learning, a reinforcement learning result output step of inputting learning data for each of the above conditions into a learning network corresponding to each of the above conditions and outputting a reinforcement learning result which is the output of each of the above learning networks; and It includes a reward calculation step for calculating the reward according to the reinforcement learning result output for each of the above learning networks; A method for generating a learning model for estimating a battery state through conditional learning, characterized in that the reinforcement learning is performed to output a reinforcement learning result in which the reward calculated for each of the above learning networks is applied to each of the above learning networks to maximize the reward.

10. In claim 9, The above reinforcement learning step is, A method for generating a learning model for estimating a battery state through conditional learning, characterized in that when the above reward exceeds a predetermined threshold value, the above reinforcement learning is performed, and when the above reward is below the predetermined threshold value, the above reinforcement learning is not performed.

11. A device for generating a learning model for estimating a battery state through conditional learning, which generates a learning model for estimating a battery state through conditional learning according to the method for generating a learning model for estimating a battery state through conditional learning described in any one of claims 1 to 10.

Images