SOH prediction system for batteries and SOH prediction method

Abstract

The present invention interprets information regarding the battery state provided by a Battery Management System (BMS) to extract features with clear physical meaning, and performs State of Health (SOH) prediction through an artificial intelligence model based on the extracted features, wherein the correlation between the standard deviation of voltage between battery cells and SOH is utilized to predict the State of Health (SOH) of the battery in real time while achieving high reliability and stability for the prediction of the artificial intelligence model; A system for predicting the State of Health (SOH) of a battery comprises: a computing device including a memory, at least one processor, and a prediction model that outputs the SOH of the battery through artificial intelligence computation; and a BMS module for managing the battery state of a battery pack composed of at least one battery cell; wherein the processor comprises: a data collection module that collects raw data defined as values ​​for the monitoring status of the battery cell over time from the BMS module; a feature extraction module that receives the raw data collected from the data collection module, performs preprocessing, extracts features used as input to the prediction model based on the preprocessed raw data, and converts them into vector data; and a prediction module that predicts the SOH through artificial intelligence computation using the vector data as input to the prediction model.

Description

Technology Field

[0001] The present invention relates to a system and method for predicting the State of Health (SOH) of a battery, and more specifically, to a system and method for predicting the SOH of a battery by extracting battery characteristics based on data collected from a Battery Management System (BMS) and performing calculations using an artificial intelligence model. Background Technology

[0002] Recently, industries based on batteries used in electric vehicles (EVs), energy storage systems (ESS), and drones have been growing rapidly, and the importance of management technology to maintain battery performance and safety over the long term is increasing significantly.

[0003] In particular, technology for accurately predicting the State of Health (SOH) of a battery is recognized as an essential element for ensuring the stable operation and economic viability of the Battery Management System (BMS).

[0004] The 'State of Health (SOH)' of the above battery is generally defined as the percentage of the current capacity relative to the initial capacity, and is also referred to as the battery's life state, aging state, or degradation state.

[0005] For example, in Energy Storage Systems (ESS), failure to predict the State of Health (SOH) can immediately lead to safety accidents such as thermal runaway, overheating, and fire, so the importance of this is being increasingly emphasized.

[0006] The above Energy Storage System (ESS) is a comprehensive system composed of a battery pack (secondary battery), a Power Conversion System (PCS), an Energy Management System (EMS / PMS), cooling and firefighting equipment, and various sensors that detect temperature, smoke, gas, voltage, and current. Both the 'stationary ESS' and the 'mobile ESS' are implemented to have a mechanical structure using the same components, but the mobile ESS is implemented to the extent that mechanical design considering vibration and shock environments is added.

[0007] In particular, ESS based on lithium-ion batteries is at risk of thermal runaway occurring under conditions of overvoltage, overcurrent, and overheating. There is a high likelihood of fire accidents due to insufficient or weak thermal management systems that precisely detect and control temperature changes inside the battery pack, and there is a risk of large-scale damage occurring in the event of an accident due to inadequate design of detection and initial response systems that detect fire early and prevent its spread. Furthermore, not only defects in the battery itself but also design and integration defects at the overall system level, such as power conversion systems (PCS), battery management systems (BMS), wiring, and protection devices, act as major causes of fire.

[0008] In other words, an ESS consists of multiple battery cells that are assembled to form a pack and connected to form the entire system. However, as the battery's usage period increases, imbalances in the State of Charge (SoC) and aging speed between cells or packs occur, and conventional manual balancing circuits cause high power consumption and heat generation, which reduces system efficiency and increases the burden of thermal management.

[0009] These conventional problems have acted as factors hindering the safety and economic viability of ESS, and accordingly, in order to detect the performance and safety of batteries, technology capable of accurately measuring the State of Health (SOH) of batteries in real time is required.

[0010] However, conventional methods for predicting (measuring) SOH include electrochemical model-based methods, data-based statistical models, and deep learning-based methods, but each has its own limitations.

[0011] Specifically, while the above electrochemical model-based method can precisely simulate physical and chemical reactions within a battery, there was a problem in that it was difficult to apply to actual operating environments due to the high complexity of the artificial intelligence model, as well as issues regarding the model's capacity and computational speed.

[0012] Furthermore, despite the advantage of low computational cost (typical computational speed), the aforementioned data-based statistical model had a problem of reduced accuracy under various operating conditions.

[0013] In addition, while the above-mentioned deep learning-based method can achieve a high level of prediction performance by utilizing large-scale datasets, it requires massive computational resources (the degree of computational occupancy of processors) during the training process, and the interpretability of the artificial intelligence model is low, so there were problems with limitations in terms of cost and efficiency for its application in actual industrial settings.

[0014] Due to the limitations of these conventional technologies, if the battery's State of Health (SOH) cannot be accurately determined during operation, imbalances between battery cells or packs intensify, increasing the burden on the thermal management system. Consequently, coupled with the lack of fire detection and initial response systems, a risk of the situation escalating into a large-scale accident has arisen.

[0015] Therefore, with the expansion of battery-based industries, there is a need for new technological approaches that can simultaneously ensure both the accuracy and efficiency of SOH forecasting. Prior art literature

[0016] Korean Patent Publication No. 10-2016-0085529 (July 18, 2016) Korean Patent Publication No. 10-2019-0090993 (August 5, 2019) Korean Patent Publication No. 10-2020-0109092 (September 22, 2020) Korean Patent Publication No. 10-2025-0121270 (August 12, 2025) The problem to be solved

[0017] The objective of the present invention is to provide a system and a prediction method capable of predicting the State of Health (SOH) of a battery in real time while achieving high reliability and stability in the prediction of the AI ​​model by interpreting information on the battery status provided by a Battery Management System (BMS) to extract features with clear physical meaning and performing SOH prediction through an AI model based on the extracted features, while utilizing the correlation between the standard deviation of voltage between battery cells and SOH. means of solving the problem

[0018] A battery SOH prediction system according to the present invention for achieving the above-described objective of the present invention is characterized by comprising: a system for predicting the SOH of a battery; a computing device including a memory, at least one processor, and a prediction model that outputs the SOH of a battery through artificial intelligence computation; and a BMS module for managing the battery state of a battery pack composed of at least one battery cell.

[0019] The processor is characterized by comprising: a data collection module that collects raw data defined as values ​​regarding the monitoring status of a battery cell over time in the BMS module; a feature extraction module that receives the raw data collected by the data collection module, performs preprocessing, extracts features used as input to the prediction model based on the preprocessed raw data, and converts them into vector data; and a prediction module that predicts SOH through artificial intelligence computation using the vector data as input to the prediction model.

[0020] The above processor is characterized by further comprising a performance diagnostic module that classifies and outputs the performance status of the battery based on the SOH predicted by the above prediction model.

[0021] The vector data of the feature extraction module is characterized by being generated through a sequential operation of receiving the raw data transmitted from the data collection module, extracting features having physical meaning based on the raw data to generate feature data, converting the values ​​of each feature into a vector structure based on the feature data to generate the vector data, and normalizing the values ​​of the vector data.

[0022] The above feature data is characterized by including data regarding the characteristics of the degree of voltage imbalance.

[0023] The above-mentioned voltage imbalance degree is characterized by being composed of the standard deviation value of the cell voltage.

[0024] The above feature data is characterized by further including data for at least one of electrical features, battery capacity features, electrical resistance features, and time features.

[0025] The above electrical characteristics are characterized by including at least one of 'voltage', 'current', 'SOC', and 'deviation of maximum / minimum voltage'.

[0026] The capacity characteristics of the above battery are characterized by including at least one of 'charge capacity', 'discharge capacity', and 'number of cycles'.

[0027] The thermal state characteristics of the above-mentioned battery are characterized by including at least one of 'temperature', 'temperature standard deviation', 'maximum temperature', and 'minimum temperature'.

[0028] The above electrical resistance is characterized by including at least one of 'internal resistance' and 'impedance'.

[0029] The above time features include at least one of 'charging time' and 'discharging time'.

[0030] The above prediction model is characterized by comprising: a first model consisting of a machine learning model based on Random Forest that outputs a value for State of Health (SOH) through artificial intelligence computation using the vector data as input; and a second model consisting of a machine learning model based on Gradient Boosting that outputs a value for State of Health (SOH) through artificial intelligence computation using the vector data as input.

[0031] The above prediction model is characterized by being implemented in a double ensemble model structure in which the output of the first model is designated as the first SOH, the output of the second model is designated as the second SOH, and the arithmetic mean value of the first SOH and the second SOH is the output of the prediction model.

[0033] A method for predicting the SOH of a battery according to the present invention for achieving the above-described objective of the present invention is performed in a computing device (1000) which includes a memory, at least one processor, and a prediction model that outputs the SOH of a battery through artificial intelligence computation, wherein the data collection module of the processor collects raw data defined as values ​​for the monitoring status of a battery cell over time from the BMS module (S100); a feature extraction module of the processor receives the raw data collected from the data collection module and performs preprocessing (S200); a feature extraction module of the processor extracts features used as input to a prediction model based on the preprocessed raw data and converts them into vector data (S300); and a prediction module of the processor predicts the SOH through artificial intelligence computation using the vector data as input to the prediction model (S400). The method is characterized by comprising: when the prediction module of the processor determines that the reliability of SOH is less than 97% (S500), the processor branches to the step of collecting raw data (S100); and when the prediction module of the processor determines that the reliability of SOH is greater than or equal to 97% (S500), the performance diagnosis module of the processor classifies and outputs the performance status of the battery based on the SOH predicted by the prediction model (S600).

[0034] The step of converting to vector data (S300) comprises: a step of receiving raw data transmitted from the data collection module of the processor (S310); a step of generating feature data by extracting features having physical meaning based on the raw data (S320); a step of generating vector data by converting the values ​​of each feature into a vector structure based on the feature data (S330); and a step of normalizing the values ​​of the vector data (S340).

[0035] The above feature data is characterized by including data regarding the characteristics of the degree of voltage imbalance.

[0036] The above-mentioned voltage imbalance degree is characterized by being composed of the standard deviation value of the cell voltage.

[0037] The above feature data is characterized by further including data for at least one of electrical features, battery capacity features, electrical resistance features, and time features.

[0038] The above electrical characteristics are characterized by including at least one of 'voltage', 'current', 'SOC', and 'deviation of maximum / minimum voltage'.

[0039] The capacity characteristics of the above battery are characterized by including at least one of 'charge capacity', 'discharge capacity', and 'number of cycles'.

[0040] The thermal state characteristics of the above-mentioned battery are characterized by including at least one of 'temperature', 'temperature standard deviation', 'maximum temperature', and 'minimum temperature'.

[0041] The above electrical resistance is characterized by including at least one of 'internal resistance' and 'impedance'.

[0042] The above time features include at least one of 'charging time' and 'discharging time'.

[0043] The above prediction model is characterized by comprising: a first model consisting of a machine learning model based on Random Forest that outputs a value for State of Health (SOH) through artificial intelligence computation using the vector data as input; and a second model consisting of a machine learning model based on Gradient Boosting that outputs a value for State of Health (SOH) through artificial intelligence computation using the vector data as input.

[0044] The step of predicting the SOH (S400) comprises: a step (S410) of designating the SOH output by artificial intelligence computation with vector data as input to the first model of the prediction model as the first SOH; a step (S420) of designating the SOH output by artificial intelligence computation with vector data as input to the second model of the prediction model as the second SOH; and a step (S430) of designating the value derived by calculating the arithmetic mean of the first SOH and the second SOH as the output for the SOH prediction of the prediction model. Effects of the invention

[0045] The battery SOH prediction system and SOH prediction method according to the present invention, as described above, have the effect of predicting the battery's SOH (State of Health) in real time while achieving high reliability and stability in the prediction of the artificial intelligence model by interpreting information on the battery status provided by the BMS (Battery Management System) and performing artificial intelligence calculations using the correlation between the standard deviation of voltage between battery cells and SOH. Brief explanation of the drawing

[0046] FIG. 1 is a block diagram showing the components of a battery SOH prediction system according to one embodiment of the present invention. FIGS. 2 to 5 are schematic examples showing the results of verifying the correlation between the voltage standard deviation and SOH between battery cells. FIG. 6 is a block diagram showing the operation and data configuration of a battery SOH prediction system according to the present embodiment. FIG. 7 is a flowchart showing a method for predicting the SOH of a battery according to the present embodiment. FIG. 8 is a flowchart showing the detailed process of step (S300) in the battery SOH prediction method according to the present embodiment. FIG. 9 is a block diagram showing the operation structure of a prediction model constituting a battery SOH prediction system according to the present embodiment. FIG. 10 is a flowchart showing the detailed process of step (S400) in the battery SOH prediction method according to the present embodiment. Specific details for implementing the invention

[0047] Hereinafter, with reference to the attached drawings, a battery SOH prediction system and a battery SOH prediction method according to a preferred embodiment of the present invention will be described in detail as follows.

[0049] Embodiments of the present invention may be modified in various forms, and the scope of the present invention should not be interpreted as being limited to the embodiments described in detail below. These embodiments are provided to more fully explain the present invention to those with average knowledge in the art. Accordingly, the shapes of elements in the drawings may be exaggerated to emphasize clearer explanations. It should be noted that in each drawing, identical components may be depicted with the same reference numeral. Detailed descriptions of known functions and configurations that are deemed to unnecessarily obscure the essence of the present invention are omitted.

[0051] In the description of the invention as described in the specification, terms such as "~module," "module," and "~part" expressed in the components refer to a unit that processes at least one function or operation, and may be implemented in hardware (computer hardware) or software (computer software), or a combination of hardware and software.

[0053] In the description of the present invention, the terms "~model" or "-model" expressed as components may refer to an "artificial intelligence model" that recognizes specific patterns based on input data and outputs predicted data as a conclusion.

[0054] Furthermore, the terms used in the specification of this invention have been selected to be as widely used as possible; however, in specific cases, terms have been arbitrarily selected by the applicant. Since the meanings of such terms are described in detail in the relevant description of the invention, this invention should be understood based on the meaning of the terms rather than their mere names.

[0056] FIG. 1 conceptually illustrates the components of a battery SOH prediction system according to one embodiment of the present invention.

[0057] Referring to FIG. 1, the battery SOH prediction system according to the present embodiment includes a memory (1020), a computing device (1000) including at least one processor (1010) and a prediction model (1030); and a BMS module (2000).

[0058] At this time, a data transmission interface is configured to be connected between the computing device (1000) and the BMS module (2000).

[0059] The above BMS module (2000) is configured to manage the battery status of a battery pack comprising at least one battery cell, and may be a device to which a Battery Management System (BMS) is applied.

[0060] As an example, the BMS module (2000) may be a BMS software component or a hardware component embedded in the Electronic Control Unit (ECU) of the vehicle.

[0061] The structure of the battery managed by the BMS module (2000) may be implemented as a nickel-cadmium (Ni-Cd) battery, nickel-hydrogen (Ni-MH) battery, lithium-ion (Li-Ion) battery, lithium-polymer (Li-Poly) battery, etc., but is not limited thereto, and may be implemented as a battery in various ways with a secondary battery structure.

[0062] It is preferable that the configuration and structure of the BMS applied to the above BMS module (2000) be a configuration and structure selected by the user from among the conventionally known technologies.

[0063] The above computing device (1) may refer to a physical computing device capable of performing computer operations, configured to include an interface for input, output, and data transmission / communication.

[0064] Examples of the above computing device (1) include a desktop, tablet PC, notebook, netbook, laptop, tablet computer, smartphone, etc., which can be applied to various computer environments, but are not limited thereto.

[0065] The processor (1010) of the computing device (1000) is structured to interpret and execute instructions, controls functions operating in the computing device (1000), and is configured to process data transmitted and received with other devices or servers through a data transmission interface.

[0066] That is, the processor (1010) performs the function of operating the computing device (1000).

[0067] The above processor (1010) may be configured to perform the execution and operation of instructions by incorporating a control unit composed of a digital circuit that performs arithmetic and bit operations on binary numbers on a logic circuit and an arithmetic logic unit composed of a digital circuit that performs arithmetic and bit operations on binary numbers on a logic circuit.

[0068] That is, the processor (1010) can be implemented to interpret each instruction to compute data and to directly issue commands to electrically connected components to perform control operations.

[0069] The above instructions may be composed of instruction functions that the processor (1010) can interpret and execute.

[0070] The above instructions may consist of instructions of a computer program implemented in at least one of a set of 'instruction codes' consisting of a set of machine language consisting of sequences of micro-operations generated through a compiler, and a set of 'instruction codes' consisting of code snippets executed through an interpreter arranged in a series of sequences.

[0071] The memory (1020) may be configured to store instructions executed in the processor (1010) of the computing device (1), and may be composed of a volatile memory device (computer data storage) that operates electromagnetically and is read / write capable.

[0072] In addition, the memory (1020) can temporarily (volatilely) store data that is operated and managed in the computing device (1).

[0073] An example of the above volatile memory device may be implemented by including a temporary recording medium readable by a computer, which has the characteristic of storing data for a short period of time, such as a register, cache, or Random Access Memory (RAM).

[0074] At this time, the sub-components included as functions operated in the processor (1010) in the description of the present specification (including components included as sub-components) can be understood as being implemented as a set of interpretable instructions, stored in memory (1020), and executed by the processor (1010).

[0075] The above prediction model (1030) is an artificial intelligence model configured to output the SOH of a battery through artificial intelligence computation using data on the state of a battery cell provided by the above BMS module (2000) as input, and can be applied as a machine learning model with a supervised learning structure.

[0076] That is, the BMS module (2000) is configured to provide data regarding the battery status to the computing device (1000), and to use the data analyzing the provided battery status as input to the prediction model (1030) to perform the operation of predicting the battery's SOH through artificial intelligence computation.

[0078] Figures 2 to 5 are graphs showing the correlation between the voltage standard deviation between cells and SOH verified and plotted through 1,550 simulations.

[0079] Figure 2 illustrates the correlation between 'Cell Voltage Std Dev' and 'SOH' by plotting the 'Pearson Correlation Coefficient,' a statistical indicator that represents the strength and direction of the linear relationship between two continuous variables as a value between -1 and 1, using scatter plots and linear regression. It shows a Pearson Correlation Coefficient r = -0.880 and a coefficient of determination R² = 0.774.

[0080] Figure 3 shows the distribution by SOH range, showing that the 'Cell Voltage Std Dev' increases as the SOH decreases. In a healthy battery (where SOH ≥ 95%), the standard deviation is 3 to 5 mV, and in a deteriorated battery (where SOH ≤ 75%), the standard deviation is 12 to 16 mV.

[0081] Figure 4 shows a comparison between a healthy battery (where SOH≥95%, σ=4.43±1.34mV) and a degraded battery (where SOH≤75%, σ=14.28±2.25mV), showing that there is a difference of about 3.2 times between the healthy and degraded states.

[0082] Figure 5 illustrates the trends of the mean and standard deviation (Cell Voltage Std Dev) by SOH.

[0083] Referring to Figures 2 through 5, it can be seen that a machine learning model with a supervised learning structure that performs learning by identifying a linear association between input data (Feature) and output data can be appropriately applied based on the results of verifying the correlation between the battery's 'SOH' and the standard deviation between the battery cells.

[0085] FIG. 6 conceptually illustrates the data transmission structure used in the processor when predicting the SOH of a battery and the operation of the sub-components of the processor (1010), and FIG. 7 illustrates the sequential processing steps in which the SOH prediction method is performed by the operation of the processor (1010).

[0086] Additionally, FIG. 8 shows a process of converting raw data of a BMS module (2000) into vector data suitable for application as input to a prediction model (1030), FIG. 9 conceptually shows the operational structure of a prediction model (1030), and FIG. 10 shows a process of predicting SOH using a prediction model (1030).

[0087] Referring to FIGS. 1, 6, and 7, the processor (1010) may be configured to include a data collection module (1011), a feature extraction module (1012), a prediction module (1013), and a performance diagnosis module (1014).

[0088] The above data collection module (1011) performs the function of collecting raw data defined as values ​​for the monitoring status of the battery cell over time in the above BMS module (2000) (S100).

[0089] The above raw data is data regarding the battery status provided by the BMS module (2000), and may consist of the battery voltage V(t), current I(t), temperature T(t), and remaining charge SOC(t) according to time (t).

[0090] The above SOC (State of Charge) may refer to data in percentage (%) units regarding the remaining capacity indicator representing the current charge state of the battery.

[0091] The feature extraction module (1012) receives the raw data collected from the data collection module (1011), performs preprocessing (S200), and performs the function of extracting features used as input to the prediction model (1030) based on the preprocessed raw data and converting them into vector data (S300).

[0092] At this time, referring to FIG. 8, the vector data of the feature extraction module (1012) in step (S300) can be generated through a sequential operation of receiving the raw data transmitted from the data collection module (1011) (S310), extracting features having physical meaning based on the raw data to generate feature data (S320), converting the values ​​of each feature into a vector structure based on the feature data to generate the vector data (S330), and normalizing the values ​​of the vector data (S340).

[0093] The above feature data may be configured to include data on at least one of the following: a feature of the degree of voltage imbalance, an electrical feature, a battery capacity feature, an electrical resistance feature, and a time feature.

[0094] The above characteristics of the degree of voltage imbalance can be determined by the standard deviation value of the cell voltage, and the standard deviation of the voltage can be calculated through [Equation 1] below.

[0095]

[0096] The above [Mathematical Formula 1] is a formula for calculating the standard deviation (σ) that statistically expresses the voltage imbalance between battery cells.

[0097] 'N' in [Equation 1] is the total number of cells that make up the battery pack.

[0098] 'σ_cell' in [Equation 1] is the 'standard deviation of cell voltage' which indicates how spread out each cell voltage is from the average voltage of all cells for all cells.

[0099] 'V_i' in [Equation 1] is the voltage value of the i-th cell, which is an individual battery cell. (Here, i is 0 ≤ i < N-1)

[0100] 'V_mean' in [Equation 1] is the average voltage value obtained by summing the voltages of all cells and dividing by the number of cells.

[0101] Using the value of 'σ_cell' calculated through the above [Equation 1], the degree of voltage imbalance between cells of the battery can be calculated as a quantitative value.

[0102] At this time, the lower the value of the standard deviation of the cell voltage (σ_cell), the lower the degree of imbalance, so the battery condition is close to good; conversely, the higher the value, the higher the degree of imbalance, so the battery condition can be determined to be in a degraded state.

[0103] The above electrical characteristics may include at least one of 'voltage', 'current', 'SOC (state of charge)', and 'deviation between maximum and minimum voltage'.

[0104] That is, by receiving data regarding the voltage, current, SOC (remaining charge), and temperature of the battery according to time (t) from the BMS module (2000), extracting features, and forming feature data, the SOH prediction of the prediction model (1030) is performed while reflecting the electrical characteristics of the battery state.

[0105] The capacity characteristics of the above battery may include at least one of 'charge capacity', 'discharge capacity', and 'number of cycles'.

[0106] That is, by receiving data regarding the battery's charging capacity, discharging capacity, and number of cycles according to time (t) from the BMS module (2000), extracting features, and forming feature data, the SOH prediction of the prediction model (1030) is performed while reflecting the features regarding the capacity of the battery state.

[0107] The above charging capacity may refer to the total amount of electricity that the battery can store.

[0108] The above discharge capacity may refer to the total amount of electricity released when a fully charged battery is discharged to a discharge termination voltage under specific conditions.

[0109] The above cycle may refer to the number of times power equivalent to 100% of the total battery capacity is discharged and recharged.

[0110] The thermal state characteristics of the above battery may include at least one of 'current temperature', 'temperature standard deviation', 'maximum temperature', and 'minimum temperature'.

[0111] That is, by receiving data regarding the current temperature, temperature standard deviation, maximum temperature, and minimum temperature of the battery according to time (t) from the BMS module (2000) and extracting features to form feature data, the SOH prediction of the prediction model (1030) is performed while reflecting the characteristics of the thermal state of the battery.

[0112] The above current temperature may refer to the average temperature of the cells inside the battery pack.

[0113] The above temperature standard deviation may refer to a standard deviation value indicating how far the temperature distribution among the cells inside the battery pack deviates from the average temperature.

[0114] The above maximum temperature may refer to the maximum temperature value detected during the temperature detection of the cells inside the battery pack.

[0115] The above minimum temperature may refer to the value at which the minimum temperature is detected during the temperature detection of the cells inside the battery pack.

[0116] The above characteristics of electrical resistance may include at least one of 'internal resistance' and 'impedance'.

[0117] That is, by receiving data on the internal resistance and impedance of the battery according to time (t) from the BMS module (2000), extracting features, and forming feature data, the SOH prediction of the prediction model (1030) is performed while reflecting the features of the electrical resistance of the battery state.

[0118] The above time characteristics may include at least one of 'charging time' and 'discharging time'.

[0119] That is, by receiving data regarding the charging time and discharging time of the battery according to time (t) from the BMS module (2000) and extracting features to form feature data, the SOH prediction of the prediction model (1030) is performed while additionally reflecting the feature that the state of SOH can change according to the charging time or discharging time of the battery.

[0121] The above vector data is generated by using each feature constituting the above feature data as a variable and converting the value of each feature into a vector structure.

[0122] In one embodiment, if the number of features constituting the feature data consists of 1 feature of voltage imbalance degree, 4 electrical features, 3 capacitance features, 4 thermal state features, 2 electrical resistance features, and 2 time features, and is designated as 16 features, it is converted into a vector X=[x_1,x_2,...,x_{16}].

[0123] That is, it is converted into data with a 16-dimensional vector structure.

[0124] StandardScaler Normalization of the above vector data is a process of making the length (magnitude) 1 while maintaining the direction of the given vector, and feature scaling can be performed using [Equation 2] below to scale each feature to mean 0 and variance 1.

[0125]

[0126] X_scaled in [Equation 2] is the value of the normalized feature.

[0127] X in [Equation 2] is the original feature value.

[0128] μ in [Equation 2] is the average value of the corresponding feature.

[0129] σ in [Equation 2] is the standard deviation value of the corresponding feature.

[0130] Using the value of 'X_scaled' calculated through the above [Equation 2], normalization (transformation) of vector data can be performed.

[0131] The above prediction module (1013) performs the function of predicting SOH through artificial intelligence computation using the above vector data as input to the above prediction model (1030) (S400).

[0132] At this time, referring to FIGS. 9 and FIGS. 10, the prediction model (1030) may be configured to include: a first model (1031) which is a machine learning model based on Random Forest that outputs a value for SOH (State of Health) through artificial intelligence computation using the vector data as input; and a second model (1032) which is a machine learning model based on Gradient Boosting that outputs a value for SOH (State of Health) through artificial intelligence computation using the vector data as input.

[0133] In addition, in step (S400), the prediction model (1030) may be operated such that the SOH output by artificial intelligence computation with vector data input to the first model (1031) of the prediction model (1030) is designated as the first SOH (S410), the SOH output by artificial intelligence computation with vector data input to the second model (1032) of the prediction model (1030) is designated as the second SOH (S420), and the arithmetic mean value of the first SOH and the second SOH derived by calculating the arithmetic mean of the first SOH and the second SOH is used as the output for the SOH prediction of the prediction model (1030) (S430).

[0134] That is, the above prediction model (1030) can be implemented as a double ensemble model structure.

[0135] The above performance diagnosis module (1014) performs the function of classifying and outputting the performance status of the battery based on the SOH predicted by the above prediction model (1030) (S600).

[0136] At this time, the prediction module (1013) can determine the reliability of SOH (S500).

[0137] That is, referring to FIG. 7, if the prediction module (1013) of the processor (1010) determines that the reliability of SOH is less than 97% (S500), the processor (1010) branches to the step of collecting raw data (S100), and if the prediction module (1013) of the processor (1010) determines that the reliability of SOH is greater than or equal to 97% (S500), the performance diagnosis module (1014) of the processor (1010) can classify and output the performance status of the battery based on the SOH predicted by the prediction model (1030) (S600).

[0138] The performance diagnosis module (1014) can classify the SOH value predicted by the prediction module (1013) as Good if SOH ≥ 90%, Caution if 80% ≤ SOH < 90%, Inspect if 70% ≤ SOH < 80%, and Replace if SOH < 70%.

[0140] An experiment was conducted to predict the SOH of an electric vehicle battery through a battery cycle test by applying the battery SOH prediction system of the present invention, and the process and results of the experiment are as shown in [Experiment 1] below.

[0141] In this case, the battery cycle test mentioned above refers to a test that evaluates the battery condition by repeatedly charging and discharging the battery and measuring factors such as capacity reduction, internal resistance, and voltage changes.

[0142] [Experiment 1] Electric vehicle battery

[0143] < Battery Specifications >

[0144] - Configuration: 96 cells connected in series (NMC Lithium-ion)

[0145] - Nominal voltage: 355.2V

[0146] - Nominal capacity: 50Ah

[0147] - Cycle: 1520 times

[0148] < Post-cycle Measurement and Prediction Results >

[0149] - Standard deviation of inter-cell voltage: σ_cell = 3.6mV

[0150] - SOH prediction of the first model based on Random Forest: SOH_RF = 96.42%

[0151] - SOH prediction of Model 2 based on Gradient Boosting: SOH_GB = 96.18%

[0152] - Final SOH Calculation: SOH_fin = (SOH_RF / SOH_GB) / 2 = 96.30%

[0153] - Confidence : 100 - |SOH_RF - SOH_GB| = 99.76%

[0154] < Result Verification >

[0155] - Actual capacity measurement: SOH_val = 48.15Ah / 50Ah = 96.3%

[0156] - Prediction error : |SOH_fin - SOH_val| = |96.30 - 96.3| = 0.00%

[0158] Through the above [Experiment 1], experimental results were obtained showing that the error predicted by the SOH prediction system of the battery of the present invention was calculated to be 0%, making it possible to make a prediction that is practically close to perfect.

[0160] The embodiment of the present invention described above is merely illustrative, and those skilled in the art will readily understand that various modifications and equivalent alternative embodiments are possible therefrom. Therefore, it will be understood that the present invention is not limited only to the forms mentioned in the above detailed description. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims. Furthermore, the present invention should be understood to include all modifications, equivalents, and substitutions within the spirit and scope of the present invention as defined by the appended claims. Explanation of the symbols

[0161] 1000: Computing device 1010: Processor 1011: Data collection module 1012: Feature Extraction Module 1013: Prediction Module 1014 : Performance Diagnosis Module 1020 : Memory 1030 : Prediction model 1031: Model 1 1032: Model 2 2000 : BMS module

Claims

Claim 1 A system for predicting the state of health of a battery comprises: a computing device (1000) including a memory (1020), at least one processor (1010), and a prediction model (1030) that outputs the state of the battery through artificial intelligence computation; and a BMS module (2000) for managing the battery state of a battery pack configured to include at least one battery cell; wherein the processor (1010) comprises: a data collection module (1011) that collects raw data defined as values ​​for the monitoring state of the battery cell over time in the BMS module (2000); and a feature extraction module (1012) that receives the raw data collected by the data collection module (1011), performs preprocessing, extracts features used as input to the prediction model (1030) based on the preprocessed raw data, and converts them into vector data. The system is configured to include a prediction module (1013) that predicts SOH through artificial intelligence computation using the above vector data as input to the prediction model (1030); the vector data of the feature extraction module (1012) is generated through a sequential operation of receiving the raw data transmitted from the data collection module (1011), extracting features having physical meaning based on the raw data to generate feature data, converting the values ​​of each feature into a vector structure based on the feature data to generate the vector data, and normalizing the values ​​of the vector data; the feature data includes data regarding features of the degree of voltage imbalance; the features regarding the degree of voltage imbalance consist of standard deviation values ​​of cell voltage; and the prediction model (1030) is a first model (1031) that is based on Random Forest and is a machine learning model that outputs a value for SOH (State of Health) through artificial intelligence computation using the above vector data as input.A battery SOH prediction system comprising: a second model (1032) based on Gradient Boosting, wherein the second model is a machine learning model that outputs a value for SOH (State of Health) through artificial intelligence computation using the vector data as input; wherein the prediction model (1030) is implemented in a double ensemble model structure in which the output of the first model (1031) is designated as the first SOH, the output of the second model (1032) is designated as the second SOH, and the arithmetic mean value of the first SOH and the second SOH is the output of the prediction model (1030). Claim 2 A battery SOH prediction system according to claim 1, characterized in that the processor (1010) further comprises a performance diagnosis module (1014) that classifies and outputs the performance status of the battery based on the SOH predicted by the prediction model (1030). Claim 3 delete Claim 4 delete Claim 5 A battery SOH prediction system according to claim 1, wherein the feature data further includes data for at least one of electrical features, battery capacity features, electrical resistance features, and time features. Claim 6 A battery SOH prediction system according to claim 5, wherein the electrical characteristics include at least one of 'voltage', 'current', 'SOC', and 'deviation of maximum / minimum voltage'. Claim 7 A battery SOH prediction system according to claim 5, wherein the capacity characteristics of the battery include at least one of 'charge capacity', 'discharge capacity', and 'number of cycles'. Claim 8 A battery SOH prediction system according to claim 5, wherein the thermal state characteristics of the battery include at least one of 'temperature', 'temperature standard deviation', 'maximum temperature', and 'minimum temperature'. Claim 9 A battery SOH prediction system according to claim 5, wherein the characteristics of the electrical resistance include at least one of 'internal resistance' and 'impedance'. Claim 10 A battery SOH prediction system according to claim 5, wherein the characteristics of the above time include at least one of 'charging time' and 'discharging time'. Claim 11 delete Claim 12 A method performed in a computing device (1000) comprising a BMS module (2000) for managing battery status and a data transmission interface connected thereto, and including at least one processor (1010) and a prediction model (1030) that outputs the SOH of a battery through artificial intelligence computation; wherein the data collection module (1011) of the processor (1010) collects raw data defined as values ​​for the monitoring status of a battery cell over time from the BMS module (2000) (S100); the feature extraction module (1012) of the processor (1010) receives the raw data collected from the data collection module (1011) and performs preprocessing (S200); and the feature extraction module (1012) of the processor (1010) extracts features used as input to the prediction model (1030) based on the preprocessed raw data and converts them into vector data (S300). The prediction module (1013) of the above processor (1010) predicts SOH through artificial intelligence computation using vector data as input to the prediction model (1030) (S400); And if the prediction module (1013) of the processor (1010) determines that the reliability of SOH is a % value less than 97% (S500), the processor (1010) branches to the step of collecting raw data (S100); and if the prediction module (1013) of the processor (1010) determines that the reliability of SOH is a % value greater than or equal to 97% (S500), the performance diagnosis module (1014) of the processor (1010) classifies and outputs the performance status of the battery based on the SOH predicted by the prediction model (1030) (S600); the step of converting to vector data (S300) comprises: the step of receiving raw data transmitted from the data collection module (1011) of the processor (1010) (S310); and the step of generating feature data by extracting features having physical meaning based on the raw data (S320). A step of generating vector data by converting the value of each feature into a vector structure based on feature data (S330);A method for predicting the State of Health of a battery, comprising: a step (S340) of normalizing the values ​​of vector data; wherein the feature data includes data regarding the characteristics of the degree of voltage imbalance; wherein the characteristics regarding the degree of voltage imbalance are composed of the standard deviation values ​​of the cell voltage; wherein the feature data further includes data regarding at least one of electrical characteristics, battery capacity characteristics, electrical resistance characteristics, and time characteristics; and wherein the prediction model (1030) comprises: a first model (1031) which is based on Random Forest and is a machine learning model that outputs a value for SOH (State of Health) through artificial intelligence computation using the vector data as input; and a second model (1032) which is based on Gradient Boosting and is a machine learning model that outputs a value for SOH (State of Health) through artificial intelligence computation using the vector data as input. Claim 13 delete Claim 14 delete Claim 15 delete Claim 16 A method for predicting the SOH of a battery according to claim 12, wherein the electrical characteristics include at least one of 'voltage', 'current', 'SOC', and 'deviation of maximum / minimum voltage'. Claim 17 A method for predicting the SOH of a battery according to claim 12, wherein the capacity characteristics of the battery include at least one of 'charge capacity', 'discharge capacity', and 'number of cycles'. Claim 18 A method for predicting the SOH of a battery according to claim 12, wherein the thermal state characteristics of the battery include at least one of 'temperature', 'temperature standard deviation', 'maximum temperature', and 'minimum temperature'. Claim 19 A method for predicting the SOH of a battery according to claim 12, wherein the characteristics of the electrical resistance include at least one of 'internal resistance' and 'impedance'. Claim 20 A method for predicting the SOH of a battery according to claim 12, wherein the characteristics of the above time include at least one of 'charging time' and 'discharging time'. Claim 21 delete Claim 22 A method for predicting the SOH of a battery according to claim 12, wherein the step of predicting the SOH (S400) comprises: a step (S410) of designating the SOH output by artificial intelligence computation with vector data input to the first model (1031) of the prediction model (1030) as the first SOH; a step (S420) of designating the SOH output by artificial intelligence computation with vector data input to the second model (1032) of the prediction model (1030) as the second SOH; and a step (S430) of designating the value derived by calculating the arithmetic mean of the first SOH and the second SOH as the output for the SOH prediction of the prediction model (1030).

Images