Machine learning based battery charge / discharge current prediction method and apparatus for charge / discharge strategy of battery thermal management

Abstract

The present invention provides a method of predicting battery charge / discharge current by a computing device, the method comprising the steps of: obtaining training data and validation data comprising temperature profiles from charge / discharge current profiles of a battery; training a machine learning model using the training data; validating and re-training the trained machine learning model using the validation data; and predicting a current profile from a temperature profile of the battery using the machine learning model.

Description

Technical Field

[0001] This disclosure relates to a method and apparatus for predicting battery charge / discharge current based on machine learning for a charge / discharge strategy for battery thermal management. Background Technology

[0002] Unlike non-rechargeable primary batteries, secondary batteries are rechargeable and dischargeable. Smaller capacity secondary batteries are used in portable small electronic devices such as smartphones, feature phones, laptops, digital cameras, and camcorders. Larger capacity secondary batteries are widely used as motor drive power sources and energy-saving batteries in hybrid vehicles and electric vehicles. A secondary battery includes an electrode assembly containing negative and positive electrodes, a casing housing them, and electrode terminals connected to the electrode assembly.

[0003] Extensive research has been conducted on forward prediction techniques, such as predicting the state of charge (SoC) of a battery (including secondary batteries) using parameters such as the battery's current, voltage, or temperature. However, research on backward prediction techniques regarding battery charge and discharge strategies, and on the next operating strategy based on the current battery temperature, is insufficient.

[0004] The information disclosed above in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may also include information that does not constitute prior art. Summary of the Invention

[0005] Technical issues The technical objective of this disclosure is to provide a method and apparatus for predicting battery charge / discharge current based on machine learning for battery thermal management charge / discharge strategies.

[0006] However, the technical objectives to be achieved by this disclosure are not limited thereto, and those skilled in the art can clearly understand other unmentioned technical objectives from the description below.

[0007] Technical solution To achieve the technical objective, according to embodiments of the present disclosure, a method for predicting battery charge and discharge current executed by a computing device includes the following steps: acquiring training data and validation data, wherein the training data includes a temperature curve based on the battery charge and discharge current curve; training a machine learning model using the training data; validating and retraining the trained machine learning model using the validation data; and predicting a current curve based on the battery temperature curve using the machine learning model.

[0008] According to the example, the step of predicting the current curve based on the battery temperature curve may include: predicting the change of the battery charge and discharge current over time, which satisfies the change of the battery temperature over time.

[0009] According to another example, the steps of obtaining training data and validation data may include: obtaining data segments on the changes in battery charging and discharging current over time and the changes in battery temperature based on the changes in battery charging and discharging current.

[0010] According to another example, the steps of obtaining training and validation data may include: obtaining data segments by random sampling to extract combinations of variables that have a uniform distribution over a given combination and range of variables.

[0011] According to another example, the steps of obtaining training data and validation data may include: obtaining training data and validation data that have the same distribution but respectively include different types of current data and temperature data.

[0012] According to another example, the steps of training a machine learning model may include: training a machine learning model configured to receive, over time, the battery temperature T(t), the first rate of change of the battery temperature ΔT(t), and the second rate of change of the battery temperature ΔT(t). 2 The battery's time-dependent state of charge (SoC) is used as input values, and the battery's time-dependent current is used as output values.

[0013] According to another embodiment, the first-order rate of change of battery temperature ΔT(t) and the second-order rate of change of battery temperature Δ 2 T(t) can have the following correlation with the battery temperature T(t) over time.

[0014] First-order rate of change: ; Second-order rate of change: .

[0015] According to another embodiment, based on the battery current I'(t) predicted according to time, the battery state of charge SoC(t) according to time can be expressed as the following equation.

[0016] .

[0017] According to another embodiment, the steps of validating and retraining the machine learning model may include: using validation data not included in the training data, using the machine learning model to predict the predicted value of the battery current based on the battery temperature, comparing the predicted value of the battery current with the current data of the validation data, and comparing the battery temperature data obtained from the predicted value of the battery current with the battery temperature data of the validation data.

[0018] As a technical solution to achieve the aforementioned technical objectives, a computer program stored in a recording medium is provided to execute the aforementioned method using a computing device.

[0019] To achieve the above technical objectives, an apparatus for predicting battery charge and discharge current according to an embodiment of the present disclosure is provided. The apparatus includes: a memory configured to store training data and validation data, wherein the training data includes a temperature curve based on a battery charge and discharge current curve; and a processor configured to acquire the training data and validation data, train a machine learning model using the training data, validate and retrain the trained machine learning model using the validation data, and predict a current curve based on a battery temperature curve using the machine learning model.

[0020] According to the example, the processor can be configured to predict changes in battery charge / discharge current over time, which in turn depends on changes in battery temperature over time.

[0021] According to another example, the processor can be configured to acquire data segments about the change of battery charging and discharging current over time and the change of battery temperature based on the change of battery charging and discharging current.

[0022] According to another example, the processor can be configured to train a machine learning model that is configured to receive, over time, the battery temperature T(t), the first rate of change of the battery temperature ΔT(t), and the second rate of change of the battery temperature ΔT(t). 2 The system takes T(t) and the battery state of charge SoC(t) that varies with time as input values ​​and outputs the battery current that varies with time as output value.

[0023] According to another example, the processor can be configured to: use validation data not included in the training data to predict the battery current based on the battery temperature using a machine learning model, compare the predicted battery current with the current data in the validation data, and compare the battery temperature data obtained from the predicted battery current with the battery temperature data in the validation data.

[0024] Beneficial effects According to this disclosure, a machine learning-based backward prediction model configured to predict charge / discharge curves that satisfy a given battery temperature curve can be obtained, and in the process, training efficiency can be improved by sequentially increasing the amount of training data.

[0025] However, the effects that can be obtained through this disclosure are not limited to those described above, and those skilled in the art will clearly understand from the specific embodiments other technical effects not mentioned. Attached Figure Description

[0026] The following figures accompanying this specification illustrate desired embodiments of the present disclosure, and together with the detailed description which will be described later, will help to understand the technical spirit of the present disclosure. Therefore, the present disclosure should not be construed as being limited to what is shown in the figures.

[0027] Figure 1 This is a schematic diagram illustrating a battery pack according to an embodiment of the present disclosure.

[0028] Figure 2 This is a flowchart describing a method for predicting battery charge / discharge current according to embodiments of the present disclosure.

[0029] Figure 3 It is a graph used to describe the state-of-charge curves and the current curves corresponding to the state-of-charge curves according to embodiments of the present disclosure.

[0030] Figure 4 and Figure 5 This is a diagram illustrating a method for predicting battery charge / discharge current according to embodiments of the present disclosure. Detailed Implementation

[0031] In the following, desired embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Therefore, the terms or words used in this specification and claims are not to be construed as limited to their conventional or lexical meanings, but are to be interpreted as meanings and concepts consistent with the technical spirit of the present disclosure, based on the principle that the inventor may appropriately define terminology and concepts to describe his / her invention in a maximally effective manner. Therefore, the embodiments disclosed in this specification and the (constructive) configurations shown in the drawings are merely some of the most desired embodiments in this disclosure and do not fully represent the technical spirit of the present disclosure, and it will be understood that various equivalents and modifications may be present in lieu of them at the time of application. Furthermore, when used in this specification, "comprising" and / or "including" are used to indicate the presence of the mentioned shapes, quantities, stages, operations, components, elements, and / or groups thereof, without excluding the presence or addition of one or more other shapes, quantities, operations, components, elements, and / or groups thereof. Additionally, in the description of embodiments of the present disclosure, "may be..." and "may indicate..." may include "one or more embodiments of the present disclosure".

[0032] Additionally, for ease of understanding, the accompanying drawings may not be shown to scale, and the dimensions of some components may be exaggerated. Furthermore, the same reference numerals may be assigned to the same components in different embodiments.

[0033] The phrase "equal" to describe two objects of comparison indicates "substantially equal." Therefore, substantially equal can include situations where there is a deviation that is considered low in the art, for example, a deviation within 5%. Furthermore, the statement that a parameter is uniform in a certain region can indicate that it is uniform in an average viewpoint.

[0034] Although terms such as first, second, etc., are used to describe various components, the components are not limited to these. These terms are only used to distinguish one component from other components, and unless otherwise specified, the first component can be the second component.

[0035] Throughout this specification, unless otherwise specified, each component may be singular or plural.

[0036] The expression “any component is arranged on top (or bottom) of a component” or “above (or below) a component” can indicate that another component can be arranged between the component and the arbitrary component arranged above (or below) the component, and that the arbitrary component is arranged adjacent to the top (or bottom) surface of the component.

[0037] Furthermore, when one component is "combined," "integrated," or "connected" to another component, it will be understood that these components can be directly combined or connected to each other, but the other component can be arranged "between" these components, or these components can be "combined," "integrated," or "connected" to each other through another component. Additionally, when one part is electrically combined to another part, this also includes cases where the part is combined to the other part using another element therein, as well as cases of direct combination.

[0038] Throughout this specification, unless otherwise specifically disclosed, “A and / or B” means A, B, or A and B. That is, “and / or” includes all or any combination of the listed items. Unless otherwise specifically disclosed, “C to D” means C or greater and D or less.

[0039] Figure 1 A battery pack according to an embodiment of the present disclosure is illustrated schematically.

[0040] Reference Figure 1 The battery pack 100 includes a battery module 110, a processor 150, a memory 160, a voltage measurement unit 120, a current measurement unit 130, and a temperature measurement unit 140.

[0041] The battery pack 100 includes a housing, in which at least one battery module 110 is formed and a receiving space for accommodating the at least one battery module 110. The battery module 110 may include a plurality of battery cells 111 and a module housing. The battery cells 111 may be stacked and housed in the module housing. Each battery cell 111 may include a negative electrode cap and a positive electrode cap. Depending on the battery form, circular, polygonal, and pouch-shaped battery cells may be used as the battery cells 111.

[0042] In the battery pack 100, instead of battery modules, stacked cells can form modules. The stacked cells can be housed in the housing space of the pack casing or in housing spaces divided by frames, separators, etc.

[0043] The battery cell 111 generates a large amount of heat during charging / discharging. This heat accumulates in the battery cell 111 and accelerates its degradation. Therefore, the battery pack 100 also includes a cooling member to suppress the degradation of the battery cells. The cooling member is disposed at the bottom of the receiving space in which the battery cells 111 are disposed, but is not limited thereto, and may also be disposed on the top surface or side surface depending on the battery pack 100.

[0044] In the battery cell 111, exhaust gases generated in the battery cell under abnormal operating conditions, referred to as thermal runaway or thermal events, can be discharged to the outside of the battery cell. The battery pack 100 or battery module 110 may be provided with exhaust ports or the like to suppress damage to the battery or module caused by exhaust gases.

[0045] The battery module 110 may include at least one battery cell 111, and the battery cell 111 may be a rechargeable secondary battery. For example, the battery cell 111 may include at least one selected from the group consisting of nickel-cadmium batteries, lead-acid batteries, nickel-metal hydride (NiMH) batteries, lithium-ion batteries, lithium polymer batteries, etc.

[0046] The number and combination type of battery cells 111 included in the battery module 110 can be determined based on the amount of electrical power and voltage required by the battery pack 100. Although Figure 1 The diagram shows battery cells 111 included in battery module 110 connected in series, but this is for conceptual purposes only; battery cells 111 can be connected in parallel with each other, and can also be connected in series and parallel. Although Figure 1 The diagram shows a battery pack 100 comprising a single battery module 110, but this is for conceptual purposes only. The battery pack 100 may also include multiple battery modules 110 connected in series, parallel, or series-parallel. A battery module 110 may also include only a single battery cell 111.

[0047] Battery module 110 may include multiple battery modules, each battery module consisting of multiple battery cells 111. Battery pack 100 includes a pair of terminals 101 and 102, to which an electrical load or charging device may be connected.

[0048] Battery pack 100 according to embodiments of the present disclosure may include a switch. The switch may be coupled between battery module 110 and one of group terminals 101 and 102. The switch may be controlled by processor 150. Although in Figure 1 Although not shown in the diagram, the battery pack 100 may also include a battery protection circuit, a fuse, a current sensor, etc.

[0049] An apparatus for predicting battery charge / discharge current according to embodiments of the present disclosure includes a processor 150 and a memory 160.

[0050] Processor 150 is configured to control the general operation of a device for predicting battery charge / discharge current. For example, to perform the above operations, processor 150 may be implemented in the form of processors, application-specific integrated circuits (ASICs), other chipsets, logic circuits, registers, communication modems, and / or data processing devices known in the art.

[0051] Processor 150 can be configured to perform basic computations, logic, and input / output operations; for example, it can execute program code stored in memory 160. Processor 150 can be configured to store data in memory 160 or load data stored in memory 160.

[0052] The memory 160, serving as a recording medium readable by the processor 150, may include a permanent mass storage device such as random access memory (RAM), read-only memory (ROM), and a disk drive. An operating system and at least one program or application code may be stored in the memory 160. According to embodiments of this disclosure, program code for predicting a current curve based on a battery temperature curve using a machine learning model may be stored in the memory 160. For example, training data and validation data, including thermal curves based on the battery charge / discharge current curve, may be stored in the memory 160. For example, data generated by measuring at least one parameter of the battery module 110 may be stored in the memory 160. Additionally, the nominal capacity of the battery module 110 may be stored in the memory 160. Furthermore, the state of charge (SOC)-open circuit voltage (OCV) data of the battery module 110 may be stored in the memory 160. For example, the data may include the battery's charge / discharge current, terminal voltage, and temperature. Additionally, the data may include the battery's discharge rate. Program code for estimating the SOC of a battery by using data generated by measuring at least one parameter of the battery module 110 and SOC-OCV data can be stored in memory 160. At least one parameter of the battery module 110 indicates components or variables of the battery module 110 (such as terminal voltage, charge / discharge current, and / or ambient temperature).

[0053] The apparatus for predicting battery charge / discharge current according to embodiments of the present disclosure may further include a voltage measuring unit 120, a current measuring unit 130, and a temperature measuring unit 140 configured to measure at least one parameter of the battery module 110. The apparatus for predicting battery charge / discharge current according to embodiments of the present disclosure may further include a communication module configured to communicate with other devices, such as electronic control devices for vehicles, controllers for charging devices, etc.

[0054] The voltage measurement unit 120 can be configured to measure the voltage of the battery module 110. For example, as Figure 1 As shown in the configuration, the voltage measurement unit 120 can be electrically connected to both ends of the battery module 110 and / or the battery cell 111. Additionally, the voltage measurement unit 120 can be electrically connected to the processor 150 to exchange electrical signals. Furthermore, the voltage measurement unit 120 can be configured, under the control of the processor 150, to measure the voltage across the battery module 110 and / or the battery cell 111 at regular time intervals, and output a signal to the processor 150 indicating the magnitude of the measured voltage. Here, the processor 150 can determine the voltage of the battery module 110 and / or the battery cell 111 based on the signal output from the voltage measurement unit 120. For example, the voltage measurement unit 120 can be implemented using voltage measurement circuits commonly used in the art.

[0055] Additionally, the current measurement unit 130 can be configured to measure the battery current. For example, as Figure 1 As shown in the configuration, the current measurement unit 130 can be electrically coupled to a current sensor disposed on the charging / discharging path of the battery module 110 and / or the battery cell 111. Additionally, the current measurement unit 130 can be electrically coupled to the processor 150 to exchange electrical signals. Furthermore, the current measurement unit 130 can be configured to repeatedly measure the magnitude of the charging current or discharging current of the battery module 110 and / or the battery cell 111 under the control of the processor 150, and output a signal indicating the magnitude of the measured current to the processor 150. Here, the processor 150 can determine the magnitude of the current based on the signal output from the current measurement unit 130. For example, the current sensor can be implemented using a Hall sensor or a sensing resistor commonly used in the art.

[0056] The temperature measurement unit 140 can be configured to measure the battery temperature. For example, as... Figure 1 As shown in the configuration, the temperature measurement unit 140 can be coupled to the battery module 110 and / or the battery cell 111, and measures the temperature of the secondary battery disposed in the battery module 110 and / or the battery cell 111. Additionally, the temperature measurement unit 140 can be electrically coupled to the processor 150 to exchange electrical signals. Furthermore, the temperature measurement unit 140 can be configured to repeatedly measure the temperature of the secondary battery at regular time intervals and output a signal indicating the measured temperature value to the processor 150. Here, the processor 150 can determine the temperature of the secondary battery based on the signal output from the temperature measurement unit 140. For example, the temperature measurement unit 140 can be implemented using a thermocouple commonly used in the art.

[0057] Additionally, the processor 150 can be configured to monitor and calculate the state of the battery module (voltage, current, temperature, SOC, state of health (SOH), etc.) by using at least one of the voltage measurement value, current measurement value, and temperature measurement value of the battery module 110 received from the voltage measurement unit 120, current measurement unit 130, and temperature measurement unit 140. Here, SOC can be obtained as a value corresponding to the remaining charge of the battery module 110. Furthermore, the processor 150 can be configured to perform control functions (e.g., temperature control, balance control, charge / discharge control, etc.) and protection functions (e.g., functions to prevent over-discharge, overcharge, overcurrent, short circuit, power failure, etc.) based on the state monitoring results. Moreover, the processor 150 can also be configured to perform wired or wireless communication functions with external devices (e.g., monitoring controller, vehicle, charger, or personal communication service (PCS)).

[0058] The processor 150 can also be configured to control the charging / discharging operation and protection operation of the battery. For this purpose, the processor 150 may include a charge / discharge controller, a balance controller, and a protector.

[0059] The battery management system, including processor 150 (which is a system configured to monitor the state of the battery and perform diagnostic, control, communication and protection functions), can be configured to calculate the state of charge / discharge, calculate the battery's state of harmonics (SOH), interrupt (relay control) battery power when necessary, control thermal management (cooling, heating, etc.), perform high-voltage interlock functions, and detect or calculate the state of insulation and short circuits.

[0060] In one aspect of this disclosure, the processor 150 can be configured to estimate the state of charge (SOC) of the battery module 110 by integrating the charging / discharging current of the battery module 110. Here, when charging or discharging of the battery module 110 begins, an initial value of the state of charge (OCC) can be determined using the initial value of the OCV of the battery module 110 measured before charging or discharging begins. To this end, the processor 150 can map the SOC-OCV data to the corresponding OCV of the battery module 110 based on SOC-OCV data that defines the state of charge according to each OCV.

[0061] Equation 1 SoC = SoC i-1 +dSoC Among them, SOC indicates the state of charge of the battery, SoC i-1 It indicates the initial value of the battery's state of charge, and dSOC indicates the value obtained by integrating the battery's charge / discharge current.

[0062] Equation 2

[0063] Here, dSOC represents the value obtained by integrating the battery's charge / discharge current from the point when charging or discharging begins, and the nominal capacity represents the battery's design capacity.

[0064] In another aspect of this disclosure, the processor 150 may be configured to estimate the state of charge (SOC) of the battery module 110 using an extended Kalman filter. The extended Kalman filter refers to a mathematical algorithm used to adaptively estimate the state of charge of the secondary battery using the voltage, current, and temperature of the secondary battery.

[0065] In addition to the current integration method or extended Kalman filter mentioned above, the SOC of the battery module 110 can also be determined by other known methods that can estimate the state of charge by selectively using the voltage, current and temperature of the secondary voltage.

[0066] Figure 2 This is a flowchart describing a method for predicting battery charge / discharge current according to embodiments of the present disclosure. The method for predicting battery charge / discharge current according to embodiments of the present disclosure can be... Figure 1 The processor 150 shown executes this.

[0067] Reference Figure 2 The method for predicting battery charging / discharging current according to this disclosure includes: acquiring training data and validation data, including temperature curves, based on the battery's charging / discharging current curve (S110); training a machine learning model using the training data (S120); validating the trained machine learning model using the validation data and retraining the machine learning model (S130); and predicting the current curve based on the battery temperature curve using the machine learning model (S140).

[0068] According to an embodiment, acquiring training data and validation data (S110) may include acquiring data segments regarding the change of battery charging / discharging current over time and the change of battery temperature based on the change of battery charging / discharging current.

[0069] Additionally, according to an embodiment, acquiring training data and validation data (S110) may include acquiring data segments through random sampling to extract combinations of variables that have a uniform distribution within a given combination and range of variables.

[0070] Additionally, according to the embodiment, obtaining training data and validation data (S110) may include: obtaining training data (S111) and validation data (S112) that have the same distribution but respectively include different types of current data and thermal data.

[0071] The processor 150 according to embodiments of the present disclosure can be configured to acquire a temperature curve T(t) based on various types of battery charge / discharge curves I(t). Here, t is a time exponent and can indicate information about time points at specific intervals. For example, in embodiments of the present disclosure, curves with 50-second intervals and a total of 3600 seconds (t = 1, 2, ..., 72) can be used.

[0072] For example, the method for acquiring training data (S111) and validation data (S112) according to embodiments of this disclosure can be used to obtain data segments that will be used for both training and validation of a machine learning model. Here, although battery temperature for any current curve can be obtained through various methods, the characteristics of a machine learning model require a large number of samples; therefore, in this disclosure, data can be collected using COMSOL simulation. However, this disclosure is not limited thereto, and it does not limit the specific method for obtaining the temperature curve T(t) based on the current curve I(t).

[0073] Regardless of the method used to acquire training data, the method for building machine learning-based models proposed in this disclosure is expected to work when there is a consistent correlation between the inputs (temperature, etc.) and outputs (current, etc.) used for training. However, in order to meet the number of training data segments required in commonly performed machine learning methods, it is beneficial to apply a method for obtaining numerical data through thermofluid analysis.

[0074] In embodiments of this disclosure, battery operation is simulated using COMSOL (i.e., a commercial thermal fluid analysis program), and the current-temperature data obtained through simulation is validated. The COMSOL battery model is a computational model validated by experimental measurements. For example, the experimentally validated battery temperature is the temperature of the outer wall of the battery canister, and the temperature in embodiments of this disclosure can indicate the temperature at the center of the battery's internal structure.

[0075] Furthermore, since it is important to train a model for charge-discharge curves over a wide range of conditions, training data (S111) and validation data (S112) can be obtained without inter-sample bias by using the Latin-hypercube sampling method. For example, typically, the number of validation data segments can be set to 1% to 5% of the number of training data segments.

[0076] For example, the current curve I(t) can be extracted using Latin-hypercube sampling. For instance, Latin-hypercube (LH) sampling can represent a sampling method that minimizes deviation within a given range for any combination of values ​​and variables. For example, processor 150 can set the shape of the current curve to a step function shape and extract the variables of the simulated step function using the LH sampling method to avoid deviations within the current range.

[0077] Furthermore, based on the determination of the time-varying state-of-charge curve SoC(t), and then the corresponding current curve, the processor 150 can extract four variables of the simulated current-state curve through LH sampling: the initial state of charge SoC0, the state of charge at the point of change SoC0, and the current curve at the time of change. c Final state of charge SoC1 and change point t c These four variables are then transformed into current curves for use. For example, each variable can have a range as shown in Equation 3.

[0078] Equation 3 0≤SoC0,SoC t ,SoC1≤1 300 seconds ≤ t c ≤3300 seconds Here, the point of change t of the current step function cIt is set between 300 and 3300 seconds to prevent rapid changes in the state of charge (i.e., excessive increase in current magnitude).

[0079] Reference Figure 3 , Figure 3 The diagram schematically illustrates the state-of-charge curve and its corresponding current curve as they vary with time, and current curves based on combinations of variables can be prepared according to the corresponding interaction formulas. For example, the current can range from 0 seconds to t seconds. c Seconds flow in increments of I0' and can be derived from t c It flows at a size of I1' from 3600 seconds to 3600 seconds.

[0080] According to an embodiment, training the machine learning model (S120) may include training a model configured to receive the battery temperature T(t) changing over time, the first rate of change of the battery temperature ΔT(t), and the second rate of change of the battery temperature ΔT(t). 2 A machine learning model that takes T(t) and the time-varying state of charge of the battery SoC(t) as inputs and outputs the time-varying battery current as output.

[0081] Furthermore, according to the embodiment, the first-order rate of change of battery temperature ΔT(t) and the second-order rate of change of battery temperature Δ 2 T(t) can have a correlation with the battery temperature T(t) as it changes over time, as shown in Equation 4.

[0082] Equation 4 First-order rate of change:

[0083] Second-order rate of change:

[0084] Additionally, according to the embodiment, based on the battery current I'(t) predicted over time, the battery state of charge SoC(t) that varies over time can be expressed as in Equation 5.

[0085] Equation 5

[0086] The processor 150 according to an embodiment of the present disclosure can be configured to train an LTSM-based machine learning model based on training data S111. For example, embodiments of the present disclosure can be executed using the Deep Learning Toolbox of MATLAB. Additionally, the trained machine learning model (S125) can be stored and used in a subsequent validation process.

[0087] For example, a machine learning model can receive the charge SoC(t), temperature T(t), the first rate of temperature change T(t) - T(t-1), and the second rate of temperature change T(t) + T(t-2) - 2T(t-1) as input curves. For instance, the second rate of temperature change can include physical information for detecting rapid current changes that significantly affect temperature. Additionally, the charge SoC(t) can be obtained as a cumulative value of the current over time.

[0088] According to the machine learning model (S125) of an embodiment of this disclosure, a current I'(t) curve that satisfies the input charge and temperature can be predicted as an output value. For example, the predicted current curve can be used as a charge / discharge strategy that satisfies a given charge and temperature curve.

[0089] For example, given that the temperature curve T(t) and the initial state of charge SoC(0) among the variables used as input values ​​are known, the processor 150 can obtain a current curve for the temperature curve.

[0090] Additionally, the processor 150 can be configured to obtain the first and second rates of change of temperature in Equation 4 using the acquired temperature profile. For example, as in Equation 6, the state of charge at the next time point can be updated based on the predicted current.

[0091] Equation 6

[0092] The reason for multiplying by 50 / 3600 here is that the unit of current is C-rate, so the charge / discharge is performed in 50-second increments per hour.

[0093] Processor 150 can predict the current I'(t) at each time point based on the time exponent t. For example, processor 150 can obtain the final current curve I'(t) for each time exponent (t = 1, 2, ..., 72).

[0094] According to an embodiment, the validation and retraining of the machine learning model (S130 and S140) may include: using validation data not included in the training data, using the machine learning model to predict the predicted value of the battery current based on the battery temperature, comparing the predicted value of the battery current with the current data of the validation data, and comparing the battery temperature data obtained from the predicted value of the battery current with the battery temperature data of the validation data.

[0095] For example, processor 150 can predict the charge / discharge curve I'(t) and verify its authenticity using verification data (S112). For example, the verification method may include: 1) comparing the difference between the current curve I(t) that causes the input curve used for prediction and the predicted current curve I'(t) by operating the method of S110; or 2) re-acquiring the temperature curve T'(t) based on the predicted current curve and comparing the re-acquiring temperature curve with the initial temperature curve.

[0096] For example, processor 150 can acquire verification data (i.e., real values) by operating method S110. For example, processor 150 can acquire verification data including current curves and temperature curves.

[0097] For example, processor 150 can use validation data not included in the training data to predict the battery current based on battery temperature using a machine learning model. Next, processor 150 can compare the predicted battery current with the current data from the validation data.

[0098] Additionally, the processor 150 can compare battery temperature data obtained from predicted values ​​of battery current with battery temperature data from verification data.

[0099] When the convergence of the machine learning model is achieved using method 1) or 2), the processor 150 according to an embodiment of the present disclosure can derive a charging / discharging strategy using the machine learning model. Optionally, when the convergence of the machine learning model is insufficient, the processor 150 according to an embodiment of the present disclosure can retrain the machine learning model. For example, the convergence of the machine learning model can be determined as convergence when the error between the compared data segments corresponds to a preset range.

[0100] For example, when convergence of the machine learning model cannot be achieved, processor 150 can acquire additional training and validation data. For instance, processor 150 can acquire another segment of training data using the same method as in operation S110, and acquire validation data when needed. For example, when the number of validation data samples is identified as having decreased from the previously maintained ratio (1%-5%), additional validation data can be acquired. Next, processor 150 can re-train and validate the machine learning model using the enhanced training and validation data.

[0101] According to an embodiment, predicting the current curve based on battery temperature may include predicting the change of battery charge / discharge current over time that satisfies the change of battery temperature over time.

[0102] For example, when the convergence of the machine learning model is achieved during its validation, the processor 150 can ultimately store the trained machine learning model and subsequently use it to derive a battery charging / discharging strategy. For instance, the trained machine learning model can reverse-engineer and derive a current profile (i.e., a charging / discharging strategy) that satisfies a given temperature profile.

[0103] Figure 4 and Figure 5 This is a diagram illustrating a method for predicting battery charge / discharge current according to embodiments of the present disclosure.

[0104] Figure 4 and Figure 5 The following diagram illustrates a battery current curve, a battery temperature curve, a current curve predicted based on the battery current curve, and a temperature curve recalculated based on the predicted current curve, according to embodiments of this disclosure. Here, the battery temperature curve can be derived from the current curve using the method described in S110. For example, embodiments of this disclosure use a total of 4397 training data points and 77 validation data points.

[0105] For example, Figure 4 Figures (a) through (d) show examples of validation data. The machine learning model can be validated by comparing the battery current curve with a current curve predicted based on the battery current curve. Alternatively, the machine learning model can be validated by comparing the battery temperature curve with a temperature curve recalculated from the predicted current curve. Here, the horizontal axis can indicate time (in seconds), and the vertical axis can indicate the current magnitude (in C-rate) for current and the internal temperature of the battery (in K) for temperature.

[0106] For example, Figure 5 (a) and (b) show examples of validation data where a different charge / discharge strategy was obtained using predictions from a machine learning model than the true value, but the recalculated temperature was different from the true value.

[0107] There may be cases where the current curve I'(t) predicted by the method proposed in this disclosure differs from the true value I(t). However, referring to... Figure 5 The temperature curve T'(t) re-obtained by operating S110 can be equal to the true value T(t). Based on this, the method proposed in this disclosure presents a charge / discharge strategy and a simple backward prediction of the charge / discharge strategy from various aspects, thereby providing users with an effective battery usage method.

[0108] This disclosure relates to a method for backward prediction of charge / discharge information, which can output any temperature of the battery from the charge / discharge information by training with temperature data that can be acquired during the driving process of the battery.

[0109] Unlike forward prediction, which predicts the current state (charge, etc.) based on battery output data, this disclosure can derive a battery charge / discharge strategy that satisfies any limitations (temperature, etc.) on the battery's use through backward prediction.

[0110] Unlike predictive forecasting of future usage patterns based on actual user usage patterns, this disclosure can establish a strategy for driving the battery within normal ranges to propose real-time battery usage strategies to the user based on current usage.

[0111] Unlike existing technologies that require additional optimization procedures to establish battery usage strategies, this disclosure can rapidly predict battery usage strategies through backward prediction of the learned model.

[0112] By using this disclosure, a machine learning-based backward prediction model configured to predict charge / discharge curves that satisfy a given battery temperature profile can be obtained, and in the process, training efficiency is also improved by sequentially increasing the amount of training data.

[0113] According to embodiments of this disclosure, charge / discharge curves can be directly predicted based on a given battery temperature profile, and by doing so, charge / discharge strategies required for real-time dynamic battery thermal management can be proposed quickly and accurately.

[0114] According to embodiments of this disclosure, the obtained predictive model can also propose alternative charge / discharge curves and initial charge / discharge curves that cause battery temperature curves, thereby helping to establish various charge / discharge strategies.

[0115] Although this disclosure has been described with reference to limited embodiments and drawings, it is not limited thereto, and various modifications and variations can be made by those skilled in the art within the equivalent scope of the technical spirit of this disclosure and the appended claims.

[0116] The various embodiments described above can be implemented in the form of a computer program, which can be executed by various components of a computer, and such a computer program can be recorded on a computer-readable medium. Here, the medium can persistently or temporarily store a computer-executable program for execution or download. Furthermore, the medium can include various types of recording units or storage units in the form of a single piece of hardware or a combination of multiple pieces of hardware, and can be distributed across a network, not limited to media that directly access a particular computer system. Examples of media can include media configured to store program instructions, including magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floppy disks; ROM; RAM; flash memory, etc. Another example of media can include recording or storage media managed by application stores that distribute applications, sites that provide and distribute various other software, servers, etc.

[0117] In this specification, terms such as "unit" and "module" can refer to hardware components such as processors or circuits and / or software components executed by hardware components such as processors. For example, "unit" and "module" can be implemented by components such as software components, object-oriented software components, class components and task components, processes, functions, features, procedures, subroutines, program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables.

Claims

1. A method for predicting battery charge / discharge current performed by a computing device, the method comprising the following steps: Acquire training and validation data, including temperature curves, based on the battery charge / discharge current curve; Machine learning models are trained using training data. The trained machine learning model is validated and retrained using validation data. as well as The current curve is predicted based on the battery temperature curve using a machine learning model.

2. The method according to claim 1, wherein, The steps for predicting the current curve based on the battery temperature curve include: predicting the change of battery charging / discharging current over time that satisfies the change of battery temperature over time.

3. The method according to claim 1, wherein, The steps for obtaining training and validation data include: acquiring data segments on the changes in battery charging / discharging current over time and the changes in battery temperature based on the changes in battery charging / discharging current.

4. The method according to claim 3, wherein, The steps for obtaining training and validation data include: obtaining data segments through random sampling to extract combinations of variables that have a uniform distribution over a given range of variables.

5. The method according to claim 4, wherein, The steps for obtaining training and validation data include: obtaining training and validation data with the same distribution but including different types of current and temperature data respectively.

6. The method according to claim 1, wherein, The steps of training a machine learning model include: training a machine learning model configured to receive, over time, battery temperature T(t), the first rate of change of battery temperature ΔT(t), and the second rate of change of battery temperature ΔT(t). 2 The system takes T(t) and the battery state of charge SoC(t) over time as inputs and outputs the battery current that varies with time as output.

7. The method according to claim 6, wherein, The first-order rate of change of battery temperature ΔT(t) and the second-order rate of change of battery temperature Δ 2 T(t) has the following correlation with the battery temperature T(t) over time, where First-order rate of change: ； Second-order rate of change: 。 8. The method according to claim 6, wherein, Based on the time-predicted battery current I'(t), the time-varying battery state of charge SoC(t) is expressed by the following equation: 。 9. The method according to claim 1, wherein, The steps for validating and retraining a machine learning model include: By using validation data not included in the training data, a machine learning model is used to predict the battery current based on the battery current. Compare the predicted battery current with the current data from the validation data; and The battery temperature data obtained from the predicted battery current is compared with the battery temperature data from the validation data.

10. A computer program stored in a recording medium, the computer program being used to perform the method according to any one of claims 1 to 9 by using a computing device.

11. An apparatus for predicting battery charge / discharge current, the apparatus comprising: The memory is configured to store training data and validation data, wherein the training data includes a temperature profile based on the battery charge / discharge current profile; and At least one processor is configured to acquire training data and validation data, train a machine learning model using the training data, validate the trained machine learning model using the validation data and retrain the machine learning model, and predict a current curve based on a battery temperature curve using the machine learning model.

12. The apparatus according to claim 11, wherein, The processor is configured to predict the change in battery charge / discharge current over time, which is consistent with the change in battery temperature over time.

13. The apparatus according to claim 11, wherein, The processor is configured to acquire data segments about the change of battery charging / discharging current over time and the change of battery temperature based on the change of battery charging / discharging current.

14. The apparatus according to claim 11, wherein, The processor is configured to train a machine learning model, which is configured to receive, over time, the battery temperature T(t), the first rate of change of the battery temperature ΔT(t), and the second rate of change of the battery temperature ΔT(t). 2 The system takes T(t) and the battery state of charge SoC(t) as input values ​​and outputs the battery current as the battery current that varies with time.

15. The apparatus according to claim 11, wherein, The processor is configured to compare the predicted battery current based on battery temperature with the current data in the validation data using validation data not included in the training data, and to compare the battery temperature data obtained from the predicted battery current with the battery temperature data in the validation data.

Images