EQUIPMENT AND METHODS FOR PREDICTING BATTERY CAPACITY
Patent Information
- Authority / Receiving Office
- VN · VN
- Patent Type
- Applications
- Current Assignee / Owner
- LG ENERGY SOLUTION LTD
- Filing Date
- 2025-07-09
- Publication Date
- 2026-06-15
AI Technical Summary
Existing methods for predicting battery capacity at specific temperatures face challenges in accuracy and speed due to multicollinearity issues when incorporating multiple process factors, leading to delayed data processing times.
A battery capacity prediction system utilizing a machine learning-based model, comprising multiple base models (RF, SVM, PCR) and a meta model (SVM), which calculates predefined statistical values from activation process data to quickly and accurately predict battery capacity at specific temperatures.
The system achieves more accurate and rapid prediction of battery capacity at specific temperatures using a stacked ensemble model, reducing processing delays and improving estimation precision.
Smart Images
Figure VN1202603889_0
Abstract
Description
Battery capacity prediction device and method
[0001] This application claims the benefit of Korean Patent Application No. 10-2024-0111012 filed with the Korean Intellectual Property Office on August 20, 2024, the entire contents of which are incorporated herein by reference.
[0002] The present invention relates to a battery capacity prediction device and method, and more particularly, to a battery capacity prediction device and method for predicting the capacity of a battery using a machine learning-based prediction model.
[0003] Secondary batteries are batteries that can be reused by charging even after discharge, and can be used as an energy source for small devices such as mobile phones, tablet PCs, and vacuum cleaners, and are also used as an energy source for medium and large devices such as automobiles and ESS (Energy Storage Systems) for smart grids.
[0004] Secondary batteries are batteries that can be reused by charging even after discharge, and can be used as an energy source for small devices such as mobile phones, tablet PCs, and vacuum cleaners, and are also used as medium- to large-scale energy sources such as personal mobility, automobiles, and ESS (Energy Storage Systems) for smart grids.
[0005] Battery cells are manufactured through assembly and activation processes. Because battery cells are assembled in a discharged state, an activation process is required after the assembly process. This activates the positive electrode active material and forms a surface film (SEI, Solid Electrolyte Interface) on the negative electrode, enabling the battery to function. This activation process is called the formation process.
[0006] The activation process may include a charge / discharge process in which SEI is formed by charging / discharging the battery cell through a charge / discharge device, and an aging process in which SEI is stabilized by maintaining the battery cell at specific conditions of temperature, pressure, and humidity through an aging device.
[0007] To assess battery cell quality, the performance values of each battery cell can be measured before or after the activation process. Since a comprehensive inspection of all battery cells is impossible, a method of estimating battery capacity is typically used based on status values measured before and after the activation process.
[0008] Meanwhile, capacity at a specific temperature is sometimes required as a performance indicator for batteries. Typically, battery capacity at a specific temperature differs from the capacity determined during the activation process, and accurate and rapid calculation of this performance indicator presents limitations.
[0009] A related prior literature is KR 10-2683336 B.
[0010] The purpose of the present invention to solve the above problems is to provide a battery capacity prediction device for predicting the capacity of a battery using a machine learning-based prediction model.
[0011] Another object of the present invention to solve the above problems is to provide a battery capacity prediction method using such a battery capacity prediction device.
[0012] Another object of the present invention to solve the above problems is to provide a battery capacity prediction system including such a battery capacity prediction device.
[0013] A battery capacity prediction device according to one embodiment of the present invention for achieving the above purpose may include at least one processor; and a memory storing at least one command executed through the at least one processor.
[0014] The at least one command may include a command for collecting activation process data, which includes a state value for each of a plurality of batteries measured during a battery activation process; a command for calculating a plurality of predefined statistical values using the activation process data; and a command for inputting the statistical values into a pre-learned prediction model and predicting a battery capacity at a specific temperature using a prediction value output by the prediction model.
[0015] The above activation process data may include temperature data, voltage data, charge capacity data, and discharge capacity data for each of the batteries.
[0016] The command for calculating the above plurality of statistical values may include a command for calculating a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity using the activation process data.
[0017] The above prediction model may include a plurality of base models that output a prediction value when the statistical values are input; and a meta model that outputs a final prediction value when the prediction value of each of the base models is input.
[0018] The command for predicting the battery capacity may include a command for stacking the predicted values of each of the base models to create a data stack; a command for inputting the data stack into the meta model; and a command for determining the final predicted value output by the meta model as the battery capacity at the specific temperature.
[0019] The above prediction model may include a first base model based on a RF (RamdomForest) algorithm, a second base model based on a SVM (Support Vector Machine) algorithm, and a third base model based on a PCR (Principal Components Regression) algorithm.
[0020] The above prediction model may include any one of a first meta model based on an RF algorithm, a second meta model based on an SVM algorithm, and a third meta model based on a PCR algorithm.
[0021] The above meta model may be a second meta model based on the SVM algorithm.
[0022]
[0023] According to one embodiment of the present invention for achieving the above-described other object, a battery capacity prediction method may include a step of collecting activation process data including a state value for each of a plurality of batteries measured during a battery activation process; a step of calculating a plurality of predefined statistical values using the activation process data; and a step of inputting the statistical values into a pre-learned prediction model and predicting a battery capacity at a specific temperature using a prediction value output by the prediction model.
[0024] The above activation process data may include temperature data, voltage data, charge capacity data, and discharge capacity data for each of the batteries.
[0025] The step of calculating the above plurality of statistical values may include a step of calculating a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity using the activation process data.
[0026] The above prediction model may include a plurality of base models that output a prediction value when the statistical values are input; and a meta model that outputs a final prediction value when the prediction value of each of the base models is input.
[0027] The step of predicting the battery capacity may include a step of stacking the predicted values of each of the base models to create a data stack; a step of inputting the data stack into the meta model; and a step of determining the final predicted value output by the meta model as the battery capacity at the specific temperature.
[0028]
[0029] According to one embodiment of the present invention for achieving the above-described further object, a battery capacity prediction system may include a battery capacity prediction device that predicts the capacity of a battery at a specific temperature using a machine learning-based prediction model; and a prediction model generation device that generates the prediction model and provides it to the battery capacity prediction device.
[0030] Here, the prediction model generation device can use battery activation process data to produce a plurality of predefined statistical values, and use the statistical values as learning data to train the prediction model.
[0031] According to the above-described embodiment of the present invention, the battery capacity at a specific temperature can be predicted more accurately and quickly.
[0032] Figure 1 shows a typical battery manufacturing process.
[0033] FIG. 2 is a block diagram of a battery capacity prediction system according to an embodiment of the present invention.
[0034] FIG. 3 is a block diagram of a battery capacity prediction device and a prediction model generation device according to an embodiment of the present invention.
[0035] Figure 4 is an operational flowchart of a method for generating a prediction model according to an embodiment of the present invention.
[0036] Figure 5 is a block diagram of a prediction model according to an embodiment of the present invention.
[0037] Figure 6 is a graph for explaining variables applied to a prediction model according to an embodiment of the present invention.
[0038] Figure 7 is an operation flowchart of a battery capacity prediction method according to an embodiment of the present invention.
[0039] FIG. 8 is a graph for explaining the prediction accuracy of a battery capacity prediction method according to an embodiment of the present invention.
[0040] Figure 9 is a block diagram of a prediction model generation device according to an embodiment of the present invention.
[0041] Fig. 10 is a block diagram of a battery capacity prediction device according to an embodiment of the present invention.
[0042] 10: Process equipment
[0043] 20: Data collection device
[0044] 100: Battery Capacity Prediction System
[0045] 110: Battery capacity prediction device
[0046] 120: Prediction model generation device
[0047] 900: Predictive Model Generation Device
[0048] 1000: Battery capacity prediction device
[0049] The present invention is susceptible to various modifications and embodiments. Specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and technical scope of the present invention. Throughout the description of each drawing, similar reference numerals have been used to designate similar components.
[0050] Terms such as "first," "second," "A," and "B" may be used to describe various components, but these components should not be limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present invention, the first component could be referred to as the "second component," and similarly, the second component could also be referred to as the "first component." The term "and / or" includes any combination of multiple related items listed or any one of multiple related items listed.
[0051] When a component is referred to as being "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components intervening. Conversely, when a component is referred to as being "directly connected" or "connected" to another component, it should be understood that there are no other components intervening.
[0052] The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, it should be understood that the terms "comprise" or "have" indicate the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but do not exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.
[0053] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and shall not be interpreted in an idealized or overly formal sense unless explicitly defined herein.
[0054]
[0055] Figure 1 shows a typical battery manufacturing process.
[0056] A battery can be manufactured by sequentially performing multiple unit processes. More specifically, the battery manufacturing process can be categorized into N unit processes, and the battery can be manufactured by sequentially performing the first through Nth processes.
[0057] For example, a battery cell can be manufactured by sequentially performing unit processes classified as an electrode coating process (first process), an electrode rolling process (second process), an assembly process (third process), an activation process (fourth process), and an EOL (End Of Line) process (fifth process).
[0058] Performance testing can be conducted during individual unit processes or after battery manufacturing is complete to ensure the battery is performing as intended. For example, various performance parameters, such as the open-circuit voltage, charge / discharge voltage, charge / discharge capacity, and internal resistance of the battery cell, can be measured before or after the activation process.
[0059] In the case of the capacity (energy) of a battery cell, since it is impossible to conduct a full survey of all battery cells, a method of estimating the capacity of the battery based on the state values measured before and after the activation process is generally used.
[0060] Meanwhile, sometimes, at the customer's request, capacity at a specific temperature (e.g., 40 degrees Celsius) is required as a battery performance indicator. In this case, a capacity prediction model defined with various process factors, including temperature, as variables, along with process data for each of these process factors, can be utilized. However, if many process factors are defined as variables in the capacity prediction model to improve estimation accuracy, multicollinearity can lower model reliability and increase system load, potentially delaying data processing times.
[0061] The present invention was conceived to address these issues, and proposes a technology capable of more accurately and quickly predicting battery capacity (energy) at a specific temperature. Below, various embodiments of the present invention will be described in detail with reference to the attached drawings.
[0062]
[0063] FIG. 2 is a block diagram of a battery capacity prediction system according to an embodiment of the present invention.
[0064] Referring to FIG. 2, a battery capacity prediction system (100) according to an embodiment of the present invention may include a battery capacity prediction device (110) and a prediction model generation device (120).
[0065] The battery capacity prediction system (100) can receive data generated during the battery process from a data collection device (20) that is linked to the process device (10).
[0066] The process device (10) may correspond to an activation process device and may include a charging / discharging device and an aging device.
[0067] The data collection device (20) can collect process data including state values measured and / or calculated during the battery manufacturing process and / or after the completion of the process. For example, the data collection device (20) can collect activation process data including temperature data, voltage data, charge capacity data, and discharge capacity data for each of a plurality of batteries.
[0068] The data collection device (20) can provide the collected process data to the battery capacity prediction system (100).
[0069] The battery capacity prediction system (100) can generate a prediction model for predicting battery capacity at a specific temperature using process data provided from the data collection device (20), or can predict battery capacity at a specific temperature. For example, the prediction model generation device (120) can train a predefined machine learning-based prediction model using activation process data. In addition, the battery capacity prediction device (110) can input activation process data into a trained prediction model to estimate battery capacity at a specific temperature.
[0070]
[0071] FIG. 3 is a block diagram of a battery capacity prediction device and a prediction model generation device according to an embodiment of the present invention.
[0072] Referring to FIG. 3, the battery capacity prediction system may include a battery capacity prediction device (110) and a prediction model generation device (120).
[0073] The battery capacity prediction device (110) can predict the battery capacity at a specific temperature using a machine learning-based prediction model (111). Here, the battery capacity prediction device (110) can receive a prediction model (121) that has completed learning from the prediction model generation device (120).
[0074] The battery capacity prediction device (110) can receive activation process data from the data collection device (20) of FIG. 2, process the activation process data, and generate input data to be applied to the prediction model (111). Thereafter, the battery capacity prediction device (110) can input the input data into the prediction model (111), and determine the prediction value output from the prediction model (111) as the battery capacity at a predetermined specific temperature.
[0075] The prediction model generation device (120) can generate a prediction model (121) and provide it to the battery capacity prediction device (110).
[0076] The prediction model generation device (120) can train a machine learning-based prediction model (121) using training data stored in a storage device (122). Here, the prediction model (121) may correspond to an ensemble model including a plurality of base models and one meta model. When training of the prediction model (121) is completed, the prediction model generation device (120) can provide the trained prediction model (121) to the battery capacity prediction device (110).
[0077]
[0078] Figure 4 is an operational flowchart of a method for generating a prediction model according to an embodiment of the present invention.
[0079] The prediction model generation device can collect activation process data (S410).
[0080] The activation process data may include status values for each of a plurality of batteries measured during the battery activation process. Here, the activation process data may include temperature data, voltage data, charge capacity data, and discharge capacity data for each of the batteries.
[0081] The predictive model generation device can generate a plurality of predefined statistical values using the activation process data (S420). Here, the statistical values can include one or more of the first quartile temperature value (temperature value located in the bottom 25%), the average voltage value, the first quartile voltage value (voltage value located in the bottom 25%), the third quartile voltage value (voltage value located in the bottom 75%), the sum of charge capacities, and the sum of discharge capacities.
[0082] Thereafter, the prediction model generation device can train the prediction model by using the statistical values produced in S420 as learning data (S430).
[0083] The prediction model can be predefined to output a predicted value for battery capacity at a specific temperature when statistical values are input.
[0084] The training data may include multiple statistical values produced by S420 and actual battery capacity values. For example, the training data may include a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity, all produced using activation process data, and may also include an actual battery capacity value at 40 degrees Celsius.
[0085] Once training for the prediction model is completed, the prediction model generation device can transfer the trained prediction model to the battery capacity prediction device.
[0086]
[0087] Figure 5 is a block diagram of a prediction model according to an embodiment of the present invention.
[0088] A prediction model according to an embodiment of the present invention may include multiple base models and a meta model. For example, as illustrated in FIG. 5 , the prediction model may include three base models (510, 520, 530) and one meta model (540).
[0089] Each of the base models (510, 520, 530) can be trained to output a predicted value of the battery capacity at a specific temperature when statistical values produced by S420 are input as training data.
[0090] The meta model (540) can be trained to output the battery capacity at a specific temperature as the final predicted value when the predicted values of each of the base models (510, 520, 530) are input as learning data.
[0091] More specifically, the prediction model generation device can train each of the first to third base models (510, 520, 530) using the statistical values produced in S420 as learning data. For example, the prediction model generation device can train each of the first to third base models (510, 520, 530) using the first quartile temperature value, the average voltage value, the first quartile voltage value, the third quartile voltage value, the sum of the charge capacity, and the sum of the discharge capacity as learning data.
[0092] In addition, the prediction model generation device can generate a data stack by stacking prediction values (first to third prediction values) output from each of the first to third base models (510, 520, 530), and use the data stack as training data to train the meta model (540). Here, the final prediction value output by the meta model (540) can be utilized as the battery capacity at a specific, predefined temperature.
[0093] The base model can be defined based on any one of the RandomForest (RF) algorithm, the Support Vector Machine (SVM) algorithm, and the Principal Components Regression (PCR) algorithm. For example, in FIG. 5, the prediction model can include a first base model (510) based on the RF algorithm, a second base model (520) based on the SVM algorithm, and a third base model (530) based on the PCR algorithm.
[0094] Additionally, the meta-model may be defined based on any one of the RF algorithm, the SVM algorithm, and the PCR algorithm. For example, in FIG. 5, the prediction model may include a meta-model (540) based on the SVM algorithm.
[0095]
[0096] Figure 6 is a graph for explaining variables applied to a prediction model according to an embodiment of the present invention.
[0097] To improve both the accuracy and processing speed of a predictive model, it is crucial to identify the optimal combination of variables to be applied to the model. To identify variables with a high correlation with the desired outcome (battery capacity), we conducted a comparative test as described below.
[0098] No.RMSECombination of statistical values10.13861Quartile temperature value, average voltage value, 1st quartile voltage value, 3rd quartile voltage value, sum of charge capacity, sum of discharge capacity20.1401Minimum temperature value, 3rd quartile temperature value, lowest capacity value, 1st quartile capacity value, average voltage value, 1st quartile voltage value30.14161Quartile temperature value, average capacity value, sum of charge capacity, lowest voltage value, 1st quartile voltage value, 3rd quartile voltage value40.1419Mean temperature value, median temperature value, capacity deviation value, sum of charge capacity, median voltage value, 1st quartile voltage value50.14231Quartile temperature value, median capacity value, sum of charge capacity, sum of discharge capacity, average voltage value, 3rd quartile voltage value........1009470.3912Mean temperature value, median temperature value, temperature deviation value, 1st quartile temperature value, lowest capacity value, lowest voltage value
[0099] Specifically, a prediction model including a first base model based on the RF algorithm, a second base model based on the SVM algorithm, a third base model based on the PCR algorithm, and a meta model based on the SVM algorithm was defined, and as shown in Table 1, each of 100,947 variable combinations was applied to the prediction model, and the RMSE (Root Mean Square Error) for each variable combination was calculated. As a result of the comparative test, it was confirmed that the prediction accuracy was the highest when [1st quartile temperature value, average voltage value, 1st quartile voltage value, 3rd quartile voltage value, sum of charge capacity, sum of discharge capacity] was defined as variables of the prediction model.
[0100] Figure 6 shows the scatter plot matrix, Spearman correlation coefficient (r), and p-value for each of the battery capacity (40°C Energy) and six optimal variables. Referring to Figure 6, it can be confirmed that the six variables (1st quartile temperature value, average voltage value, 1st quartile voltage value, 3rd quartile voltage value, sum of charge capacity, and sum of discharge capacity) derived through the comparison test have a statistically significant correlation with the battery capacity.
[0101]
[0102] Figure 7 is an operation flowchart of a battery capacity prediction method according to an embodiment of the present invention.
[0103] The battery capacity prediction device can collect activation process data (S710).
[0104] The activation process data may include status values for each of a plurality of batteries measured during the battery activation process. Here, the activation process data may include temperature data, voltage data, charge capacity data, and discharge capacity data for each of the batteries.
[0105] The battery capacity prediction device can use the activation process data to produce a plurality of predefined statistical values (S720). Here, the statistical values can include one or more of a first quartile temperature value (a temperature value located in the bottom 25%), an average voltage value, a first quartile voltage value (a voltage value located in the bottom 25%), a third quartile voltage value (a voltage value located in the bottom 75%), a sum of charge capacities, and a sum of discharge capacities.
[0106] Thereafter, the battery capacity prediction device inputs the statistical values produced in S720 into a pre-learned prediction model, and can predict the battery capacity at a specific temperature using the prediction values output by the prediction model (S730).
[0107] The predictive model can be pre-trained to output a predicted value for battery capacity at a specific temperature when statistical values are input.
[0108] A prediction model may include multiple base models and a meta model. For example, as illustrated in FIG. 5 , the prediction model may include three base models (510, 520, 530) and one meta model (540).
[0109] Each of the base models may be trained to output a predicted value for battery capacity at a specific temperature when statistical values produced by S720 are input as input data.
[0110] The meta model may be trained to output the battery capacity at a specific temperature as the final predicted value when the predicted values of each of the base models are input as input data.
[0111] More specifically, the battery capacity prediction device can input the statistical values calculated in S720 into each of the first to third base models. For example, the battery capacity prediction device can input the first quartile temperature value, the average voltage value, the first quartile voltage value, the third quartile voltage value, the sum of the charge capacity, and the sum of the discharge capacity into each of the first to third base models.
[0112] Thereafter, the battery capacity prediction device can generate a data stack by stacking the prediction values (first to third prediction values) output from each of the first to third base models, and input the data stack into the meta model. Here, the battery capacity prediction device can determine the final prediction value output by the meta model as the battery capacity at a specific, predefined temperature.
[0113] The base model can be defined based on any one of the RandomForest (RF) algorithm, the Support Vector Machine (SVM) algorithm, and the Principal Components Regression (PCR) algorithm. For example, the prediction model may include a first base model based on the RF algorithm, a second base model based on the SVM algorithm, and a third base model based on the PCR algorithm.
[0114] Additionally, the meta-model can be defined based on any one of the RF algorithm, the SVM algorithm, and the PCR algorithm.
[0115] To derive the optimal prediction model that exhibits the highest prediction accuracy, a comparative test was conducted as follows.
[0116] Model RMS E E Comparative Example 1 - RF 0.1968 0.1524 Comparative Example 2 - SVM 0.2125 0.1707 Comparative Example 3 - PCR 0.2098 0.1686 Example 1 - Stacking RF 0.1942 0.1534 Example 2 - Stacking SVM 0.1848 0.1489 Example 3 - Stacking PCR 0.1877 0.1501
[0117] Comparative Examples 1 to 3 are single prediction models, wherein Comparative Example 1 is a model based on the RF algorithm, Comparative Example 2 is a model based on the SVM algorithm, and Comparative Example 3 is a model based on the PCR algorithm. Examples 1 to 3 are ensemble models including three base models (a first base model based on the RF algorithm, a second base model based on the SVM algorithm, and a third base model based on the PCR algorithm) and one meta-model.
[0118] Here, Example 1 is an ensemble model to which a meta-model based on the RF algorithm is applied, Example 2 is an ensemble model to which a meta-model based on the SVM algorithm is applied, and Example 3 is an ensemble model to which a meta-model based on the PCR algorithm is applied.
[0119] As a result of the comparative test, it was confirmed that all of Examples 1 to 3 had higher prediction accuracy than Comparative Examples 1 to 3.
[0120] In particular, it was confirmed that the prediction model showed the highest prediction accuracy when it was composed of a first base model based on the RF algorithm, a second base model based on the SVM algorithm, a third base model based on the PCR algorithm, and a meta model based on the SVM algorithm (Example 2).
[0121]
[0122] FIG. 8 is a graph for explaining the prediction accuracy of a battery capacity prediction method according to an embodiment of the present invention.
[0123] Specifically, FIG. 8 is a graph comparing the results of battery capacity prediction using the prediction model according to Example 2 of [Table 2] by the battery capacity prediction device with the actual battery capacity (energy).
[0124] A battery capacity prediction device according to an embodiment of the present invention can predict a battery capacity at a specific temperature with higher accuracy using activation process data, and can quickly derive a prediction result without delaying data processing speed because it uses only six statistical values.
[0125]
[0126] Figure 9 is a block diagram of a prediction model generation device according to an embodiment of the present invention.
[0127] A predictive model generation device (900) may include at least one processor (910), a memory (920) that stores at least one command executed through the processor, and a transmission / reception device (930) that is connected to a network and performs communication.
[0128] The at least one command may include a command for collecting activation process data, which includes a state value for each of a plurality of batteries measured during a battery activation process; a command for calculating a plurality of predefined statistical values using the activation process data; and a command for training a machine learning-based prediction model, which uses the statistical values as learning data to output a battery capacity at a specific temperature as a predicted value.
[0129] The above activation process data may include temperature data, voltage data, charge capacity data, and discharge capacity data for each of the batteries.
[0130] The command for calculating the above plurality of statistical values may include a command for calculating a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity using the activation process data.
[0131] The above prediction model may include a plurality of base models that learn the statistical values and output a prediction value; and a meta model that learns the prediction value of each of the base models and outputs a final prediction value.
[0132] The command for training the above prediction model may include a command for generating a data stack by stacking the prediction values when the prediction values of each of the base models are output; and a command for training the meta model by using the data stack as training data.
[0133] The above prediction model may include a first base model based on a RF (RamdomForest) algorithm, a second base model based on a SVM (Support Vector Machine) algorithm, and a third base model based on a PCR (Principal Components Regression) algorithm.
[0134] The above prediction model may include any one of a first meta model based on an RF algorithm, a second meta model based on an SVM algorithm, and a third meta model based on a PCR algorithm.
[0135] The at least one command may further include a command for transmitting the prediction model to a battery capacity prediction device when learning of the prediction model is completed.
[0136] The predictive model generation device (900) may further include an input interface device (940), an output interface device (950), a storage device (960), etc. Each component included in the predictive model generation device (900) may be connected to each other by a bus (970) and may communicate with each other. The storage device (960) may store a diagnostic model, a defect detection model, and an anomaly detection model.
[0137] Here, the processor (910) may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. The memory (or storage device) may be composed of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory may be composed of at least one of a read-only memory (ROM) and a random access memory (RAM).
[0138]
[0139] Fig. 10 is a block diagram of a battery capacity prediction device according to an embodiment of the present invention.
[0140] A battery capacity prediction device (1000) may include at least one processor (1010), a memory (1020) that stores at least one command executed through the processor, and a transmission / reception device (1030) that is connected to a network and performs communication.
[0141] The at least one command may include a command for collecting activation process data, which includes a state value for each of a plurality of batteries measured during a battery activation process; a command for calculating a plurality of predefined statistical values using the activation process data; and a command for inputting the statistical values into a pre-learned prediction model and predicting a battery capacity at a specific temperature using a prediction value output by the prediction model.
[0142] The above activation process data may include temperature data, voltage data, charge capacity data, and discharge capacity data for each of the batteries.
[0143] The command for calculating the above plurality of statistical values may include a command for calculating a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity using the activation process data.
[0144] The above prediction model may include a plurality of base models that output a prediction value when the statistical values are input; and a meta model that outputs a final prediction value when the prediction value of each of the base models is input.
[0145] The command for predicting the battery capacity may include a command for stacking the predicted values of each of the base models to create a data stack; a command for inputting the data stack into the meta model; and a command for determining the final predicted value output by the meta model as the battery capacity at the specific temperature.
[0146] The above prediction model may include a first base model based on a RF (RamdomForest) algorithm, a second base model based on a SVM (Support Vector Machine) algorithm, and a third base model based on a PCR (Principal Components Regression) algorithm.
[0147] The above prediction model may include any one of a first meta model based on an RF algorithm, a second meta model based on an SVM algorithm, and a third meta model based on a PCR algorithm.
[0148] The above meta model may be a second meta model based on the SVM algorithm.
[0149] The battery capacity prediction device (1000) may further include an input interface device (1040), an output interface device (1050), a storage device (1060), etc. Each component included in the battery capacity prediction device (1000) may be connected to each other by a bus (1070) and may communicate with each other. The storage device (1060) may store a diagnostic model, a failure detection model, and an anomaly detection model.
[0150] Here, the processor (1010) may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. In addition, the memory may be configured with at least one of a volatile / transitory storage medium and a non-volatile / non-transitory storage medium. For example, the memory may be configured with at least one of a read-only memory (ROM) and a random access memory (RAM), and may include an Electrically Erasable Programmable Read-only Memory (EEPROM).
[0151]
[0152] The operations of the method according to an embodiment of the present invention can be implemented as a computer-readable program or code on a computer-readable recording medium. The computer-readable recording medium may include any type of recording device that stores data readable by a computer system. The computer-readable recording medium may also be distributed across network-connected computer systems, allowing the computer-readable program or code to be stored and executed in a distributed manner.
[0153] The operation of the method according to an embodiment of the present invention may be implemented in various forms related to a program, such as a computer program or code itself or a computer program product.
[0154] Additionally, the computer-readable recording medium may include one or more of a volatile / transitory recording medium and a non-volatile / non-transitory recording medium.
[0155] A computer-readable recording medium may include hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory, and may include, for example, various types of servers located on a network. Program instructions may include not only machine language codes, such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.
[0156] While some aspects of the present invention have been described in the context of a device, they may also represent a description of a corresponding method, wherein a block or device corresponds to a method step or a feature of a method step. Similarly, aspects described in the context of a method may also be described as a corresponding block or item or a feature of a corresponding device. Some or all of the method steps may be performed by (or using) a hardware device, such as, for example, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, one or more of the most significant method steps may be performed by such a device.
[0157] Although the present invention has been described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various modifications and changes may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below.
Claims
1. At least one processor; and A memory that stores at least one instruction to be executed through at least one processor, At least one of the above commands, A command to collect activation process data, which includes a status value for each of a plurality of batteries measured during a battery activation process; A command for producing a plurality of predefined statistical values using the above activation process data; and A battery capacity prediction device, comprising a command for inputting the above statistical values into a pre-learned prediction model and predicting the battery capacity at a specific temperature using the predicted value output by the prediction model.
2. In claim 1, The above activation process data is, A battery capacity prediction device comprising temperature data, voltage data, charge capacity data, and discharge capacity data for each battery.
3. In claim 1, The command to produce the above multiple statistical values is: A battery capacity prediction device comprising a command for calculating a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity using the above activation process data.
4. In claim 3, The above prediction model is, When the above statistical values are input, a plurality of base models outputting predicted values; and A battery capacity prediction device including a meta model that outputs a final prediction value when the prediction value of each of the above base models is input.
5. In claim 4, The command to predict the above battery capacity is: When the predicted values of each of the above base models are output, a command to stack the predicted values to create a data stack; The data stack is a command that the meta model inputs; and A battery capacity prediction device, comprising a command for determining the final predicted value output by the meta model as the battery capacity at the specific temperature.
6. In claim 4, The above prediction model is, A battery capacity prediction device comprising a first base model based on an RF (RamdomForest) algorithm, a second base model based on an SVM (Support Vector Machine) algorithm, and a third base model based on a PCR (Principal Components Regression) algorithm.
7. In claim 6, The above prediction model is, A battery capacity prediction device comprising any one of a first meta model based on an RF algorithm, a second meta model based on an SVM algorithm, and a third meta model based on a PCR algorithm.
8. In claim 7, The above meta model is, A battery capacity prediction device, a second meta-model based on the SVM algorithm.
9. A step of collecting activation process data, which includes a status value for each of a plurality of batteries measured during the battery activation process; A step of calculating a plurality of predefined statistical values using the above activation process data; and A battery capacity prediction method, comprising the step of inputting the above statistical values into a pre-learned prediction model and predicting the battery capacity at a specific temperature using the predicted value output by the prediction model.
10. In claim 9, The above activation process data is, A battery capacity prediction method, comprising temperature data, voltage data, charge capacity data, and discharge capacity data for each battery.
11. In claim 9, The step of calculating the above multiple statistical values is: A battery capacity prediction method, comprising a step of calculating a first quartile temperature value, an average voltage value, a first quartile voltage value, a third quartile voltage value, a sum of charge capacity, and a sum of discharge capacity using the above-mentioned activation process data.
12. In claim 11, The above prediction model is, When the above statistical values are input, a plurality of base models outputting predicted values; and A battery capacity prediction method, comprising a meta model that outputs a final prediction value when the prediction value of each of the above base models is input.
13. In claim 12, The step of predicting the above battery capacity is: When the predicted values of each of the above base models are output, a step of stacking the predicted values to create a data stack; A step of inputting the above data stack into the above meta model; and A battery capacity prediction method, comprising a step of determining the final predicted value output by the meta model as the battery capacity at the specific temperature.
14. A battery capacity prediction device that predicts the capacity of a battery at a specific temperature using a machine learning-based prediction model; and A prediction model generation device is included, which generates the above prediction model and provides it to the battery capacity prediction device, The above prediction model generation device, A battery capacity prediction system that uses battery activation process data to produce a plurality of predefined statistical values and uses the statistical values as learning data to train the prediction model.