Apparatus and method for estimating battery SOC

Abstract

An apparatus for estimating a battery state of charge (SOC) includes a pre-processing unit (200) configured to detect operating parameters of the battery and filter at least a portion of the parameters to generate filtered parameters; a SOC estimation unit (300) configured to calculate a nominal SOC using a first neural network based on the operating parameters and the filtered parameters; a validity estimation unit (400) configured to estimate a validity value indicative of the validity of the nominal SOC output by the SOC estimation unit (300) based on the operating parameters and the filtered parameters; and a SOC Kalman filter (500) configured to perform a Kalman filtering algorithm based on the operating parameters, the transmitted nominal SOC and the validity value, thereby outputting an estimated SOC of the battery.

Description

Cross-reference of related applications

[0001] This disclosure claims priority under patent application number 17 / 750,654, filed with the United States Patent and Trademark Office on 23 May 2022, and is titled "Apparatus and Method for Estimating Battery State of Coffee (SOC)". [Technical Field]

[0002] Embodiments of this disclosure relate generally to the field of battery technology, and more particularly to apparatus and methods for estimating the state of charge (SOC) of a battery. [Background technology]

[0003] This section presents information that may be helpful in understanding this disclosure. Therefore, the statements in this section should be read from that perspective and should not be interpreted as recognition of what is included in or not included in the prior art.

[0004] Rechargeable batteries are used in personal portable electronic devices, electric vehicles, and energy storage applications in power systems. Battery State of Charge (SOC) is defined as the ratio of remaining capacity to maximum capacity and is a critical value calculated by a Battery Management System (BMS). BMSs use SOC to indicate when the battery needs recharging, to extend battery life by preventing overcharging and over-discharging, and, in electric vehicle applications, to indicate the remaining driving range. Since battery SOC cannot be measured directly, it must be estimated using available measurements.

[0005] One common method for estimating battery SOC is the Coulomb counting method using open-circuit voltage (OCV) and charge termination correction. In this method, the BMS calculates the SOC based on the net charge extracted from the battery during operation by integrating the current drawn from the battery (i.e., counting the extracted Coulombs). If the battery voltage reaches equilibrium after a sufficiently long rest period, the SOC can be corrected using the equilibrium voltage (OCV) based on the OCV-SOC relationship. Alternatively, the SOC can be corrected by setting it to 100% when the battery voltage reaches its rated value at the end of a given charging sequence. The advantage of this method is its relative simplicity.

[0006] Much research has been conducted on developing battery SOC estimation methods based on Kalman filter algorithms used in combination with equivalent circuit models (ECMs) of batteries. Researchers have reported methods based on extended Kalman filters (EKF), unscented Kalman filters (UKF), and sigma-point Kalman filters (SPKF), which are collectively referred to as "xKF". Generally, these methods can provide accurate estimations as long as the covariance parameters of the ECM and the corresponding xKF are accurate.

[0007] The researchers also report on a battery SOC estimation method using an xKF approach combined with a neural network SOC model. The neural network SOC model can be developed efficiently and reliably using commercially available neural network optimization packages. [Overview of the project] [Problems that the invention aims to solve]

[0008] The inventors of this disclosure have found that each of the existing battery SOC estimation methods has several limitations.

[0009] The Coulomb counting method is known to be inaccurate under certain conditions. If the initial SOC estimate is incorrect, the SOC estimation error persists until the conditions for appropriate OCV or end-of-charge SOC correction are met. Furthermore, integrating the current over a long period without SOC correction can lead to an inaccurate SOC due to the accumulation of current measurement errors.

[0010] Methods based on xKF used in conjunction with ECM may require significant engineering resources to achieve accuracy targets. First, the required ECM prediction accuracy can be obtained by scheduling ECM parameters based on the operating zone. The operating zone may be defined using detected physical quantities such as temperature and current, and calculated values ​​such as SOC. Second, if it is known that ECM prediction accuracy varies depending on the operating zone, the covariance parameters of xKF may be scheduled according to the operating zone. The process of constructing such parameter schedules often requires considerable engineering development resources.

[0011] Regarding the use of the xKF approach in combination with neural network SOC models, the covariance parameters of xKF may be scheduled according to the operating zone to achieve accuracy targets, as the model's predictive accuracy may vary depending on the operating zone (for example, it may be inaccurate in "flat" voltage zones where the SOC does not correlate well with the available measurements). Developing such parameter schedules may require considerable engineering resources. [Means for solving the problem]

[0012] Generally, embodiments of this disclosure provide apparatus for battery SOC estimation, methods for battery SOC estimation, and methods for calculating parameters used in battery SOC estimation. This disclosure aims to accurately estimate battery SOC using a systematic and efficient development method.

[0013] As the first aspect, A preprocessing unit (200) is configured to detect battery operating parameters and filter at least some of the parameters to generate filtered parameters, A SOC estimation unit (300) is configured to calculate the nominal SOC using a first neural network based on the operating parameters and filtered parameters transmitted from the preprocessing unit, A performance estimation unit (400) is configured to estimate a performance value indicating the performance of the nominal SOC output by the SOC estimation unit (300) based on the operating parameters and filtered parameters transmitted from the preprocessing unit, An apparatus for estimating the state of charge (SOC) of a battery is provided, which includes an SOC Kalman filter (500) configured to execute a Kalman filtering algorithm based on operating parameters transmitted from a preprocessing unit, nominal SOC transmitted from an SOC estimation unit, and effectiveness values ​​transmitted from an effectiveness estimation unit, thereby outputting an estimated SOC of the battery.

[0014] As a second aspect, The steps include detecting the battery's operating parameters, filtering at least some of the parameters to generate filtered parameters, and The steps include: calculating the nominal SOC using a first neural network based on the operating parameters and filtered parameters; A step of calculating an effectiveness value that indicates the effectiveness of the nominal SOC based on the operating parameters and filtered parameters, A method for estimating the state of charge (SOC) of a battery is provided, comprising the steps of: running a Kalman filtering algorithm using an SOC Kalman filter based on operating parameters, nominal SOC, and effectiveness values, thereby outputting an estimated SOC of the battery.

[0015] As a third aspect, A step of calculating a nominal SOC based on experimental data, A step of calculating a squared nominal SOC error based on experimental data, Identifying a subset of data from a dataset where the squared nominal SOC error is less than a first predetermined value, A step of calculating the variance of the difference between the true SOC and the nominal SOC for the identified subset of data, wherein the calculation result is the parameter R used in the Kalman filtering algorithm, the step, and a method for calculating parameters used in estimating the state of charge (SOC) of a battery is provided.

[0016] According to various embodiments of the present disclosure, a first neural network is used to estimate a nominal SOC, an effective value of the nominal SOC is calculated, and an SOC Kalman filter calculates an estimated SOC based on the nominal SOC and the effective value. Therefore, by using a systematic and efficient development method, the battery SOC can be accurately estimated.

Brief Description of the Drawings

[0017] The above and other aspects, features, and advantages of various embodiments of the present disclosure will become more fully apparent by referring to the following detailed description in conjunction with the accompanying drawings as examples. Here, like reference numerals or letters are used to designate like or equivalent elements. The drawings are shown to better understand the embodiments of the present disclosure and are not necessarily drawn to scale. <​​​​​​​​​​​​This figure schematically shows an SOC Kalman filter according to a preferred embodiment of the present disclosure. [Figure 6] This is a schematic block diagram illustrating an apparatus for estimating the State of Charge (SOC) of a battery according to another embodiment of the present disclosure. [Figure 7] This flowchart sequentially illustrates the steps for calculating the parameters used in SOC estimation according to a preferred embodiment of the present disclosure. [Figure 8] This is a flowchart of a method for estimating the battery charge state according to a certain embodiment. [Figure 9] This is a schematic diagram of the structure of an electronic device in an embodiment of the present disclosure. [Modes for carrying out the invention]

[0018] This disclosure is described with reference to several embodiments. These embodiments are discussed for the purpose of assisting the articulate to better understand and thus implement this disclosure, and should be understood not to imply any limitation on the scope of this disclosure.

[0019] The terms “first” and “second” as used herein refer to different elements. The singular forms “a” and “an” are interpreted as including the plural unless the context clearly indicates otherwise. The terms “include,” “contains,” “have,” “possess,” “include,” and “contains” as used herein specify the existence of the described features, elements, and / or components, but do not exclude the existence or addition of one or more other features, elements, components, and / or combinations thereof. The term “based on” shall be interpreted as “based on at least part.” The terms “one embodiment” and “a certain embodiment” shall be interpreted as “at least one embodiment.” The term “another embodiment” shall be interpreted as “at least one other embodiment.” Other definitions may be explicitly or implicitly included below.

[0020] (First aspect of the embodiment) In a first embodiment of the embodiment, an apparatus for estimating the state of charge (SOC) of a battery is provided.

[0021] Figure 1 is a schematic block diagram showing an apparatus 10 for estimating battery SOC according to one embodiment of the present disclosure. As shown in Figure 1, the apparatus 10 includes a preprocessing unit 200, an SOC estimation unit 300, an effectiveness estimation unit 400, and an SOC Kalman filter 500.

[0022] In at least one embodiment, the preprocessing unit 200 is configured to detect operating parameters of the battery 100 (shown in Figure 1) and filter at least a portion of these operating parameters to generate filtered parameters. For example, the operating parameters detected by the preprocessing unit 200 may include current, voltage, and temperature, while the filtered parameters may include filtered current and filtered voltage. However, the embodiments are not limited thereto, and the operating parameters and filtered parameters may be of other types.

[0023] The SOC estimation unit 300 is configured to calculate the nominal SOC using a first neural network based on the operating parameters and filtered parameters transmitted from the preprocessing unit 200.

[0024] The effectiveness estimation unit 400 is configured to estimate the effectiveness value based on the operating parameters and filtered parameters transmitted from the preprocessing unit 200. The effectiveness value represents the effectiveness of the nominal SOC output by the SOC estimation unit 300.

[0025] The SOC Kalman filter 500 is configured to execute a Kalman filtering algorithm based on operating parameters (e.g., current) transmitted from the preprocessing unit 200, the nominal SOC transmitted from the SOC estimation unit 300, and the effectiveness value transmitted from the effectiveness estimation unit 400, and to output an estimated SOC of the battery 100.

[0026] According to a first aspect of this disclosure, a first neural network is used to estimate the nominal SOC, an effectiveness value of the nominal SOC is calculated, and an SOC Kalman filter calculates the estimated SOC based on the nominal SOC and the effectiveness value. Thus, by using a systematic and efficient development method, the battery SOC can be accurately estimated.

[0027] Figure 2 is a schematic diagram showing a preprocessing unit 200 according to one embodiment of the present disclosure. As shown in Figure 2, the preprocessing unit 200 includes a current detection unit 210, a voltage detection unit 220, a temperature detection unit 230, a current low-pass filter 240, and a voltage low-pass filter 250.

[0028] In at least one embodiment, a method for detecting the current, voltage, and temperature of the battery 100 can be found by referring to related technology.

[0029] The current low-pass filter 240 can filter current using a time constant of T seconds and output the filtered current. The voltage low-pass filter 250 can filter voltage using a time constant of T seconds and output the filtered voltage.

[0030] In a preferred embodiment, the current low-pass filter 240 and the voltage low-pass filter 250 may be implemented according to the following discretized filter calculation.

number

[0031] Here, the calculation cycle time for the current low-pass filter 240 and / or the voltage low-pass filter 250 is T. s The calculation cycle is k, and the input signal is u k The filtered signal is x k . Also, the filtered signal for the next calculation cycle is x k+1 Regarding the current low-pass filter 240, the input signal u k x is the current, filtered signal x k This is the filtered current. For the voltage low-pass filter 250, the input signal u k x is the voltage and the filtered signal x k This is the filtered voltage.

[0032] Figure 3 is a schematic diagram of a SOC estimation unit 300 according to one embodiment of the present disclosure. A first neural network (i.e., the SOC neural network) is used in the SOC estimation unit 300. As shown in Figure 3, 310 is the input layer of the SOC neural network, to which, for example, current, voltage, temperature, filtered current, and filtered voltage are input. 320 is the hidden layer of the SOC neural network. 330 is the output layer of the SOC neural network. The operating principle of the SOC neural network can be found by referring to related technologies.

[0033] Figure 4 is a schematic diagram showing an effectiveness estimation unit 400 according to one embodiment of the present disclosure. As shown in Figure 4, the effectiveness estimation unit 400 includes a squared SOC error estimation unit 410 and an effectiveness classification circuit 450.

[0034] The squared SOC error estimation unit 410 can estimate the squared nominal SOC error using a second neural network (i.e., the squared SOC error neural network) based on the operating parameters and filtered parameters. The squared nominal SOC error represents the square of the difference between the true SOC and the nominal SOC.

[0035] As shown in FIG. 4, 420 is the input layer of the second neural network (i.e., the squared SOC error neural network). For example, current, voltage, temperature, filtered current, and filtered voltage are input into the input layer 420. 430 is the hidden layer of the squared SOC error neural network. 440 is the output layer of the squared SOC error neural network. For example, the squared nominal SOC error is output from the output layer 440. The operating principle of the squared SOC error neural network can be referred to the related art.

[0036] As shown in FIG. 4, the validity classification circuit 450 can output a validity value based on that the magnitude of the squared nominal SOC error is less than a first predetermined value, and that the elapsed execution time from the initialization of the current low-pass filter or the initialization of the voltage low-pass filter exceeds a second predetermined value. For example, the first predetermined value is E 2 is. In a preferred embodiment, the parameter E is 1%. The second predetermined value may be 3T, where T is the time constant of the current low-pass filter 240 and the voltage low-pass filter 250.

[0037] For example, the validity of the nominal SOC is set to 1 (valid) when the squared nominal SOC error is smaller than E 2 and the elapsed execution time from the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter is greater than 3T. Otherwise, the validity of the nominal SOC is set to 0 (invalid).

[0038] As shown in Figure 4, in a preferred embodiment, the effectiveness classification circuit 450 includes a first determination unit 451, a second determination unit 452, and a logic unit 453. The first determination unit 451 can determine whether the squared nominal SOC error is below a first predetermined value. For example, if the squared nominal SOC error is below the first predetermined value, the first determination unit 451 can output 1. The second determination unit 452 can determine whether the elapsed execution time from the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter exceeds a second predetermined value. For example, if the elapsed execution time from the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter exceeds a second predetermined value, the second determination unit 452 can output 1. The logic unit 453 can output an effectiveness value according to the determination result of the first determination unit 451 and the determination result of the second determination unit 452. For example, logic unit 453 is an AND gate and can output 1 when both the first decision unit 451 and the second decision unit 452 output 1. However, the embodiment is not limited thereto, and the effectiveness classification circuit 450 can have other structures.

[0039] Figure 5 is a schematic diagram showing an SOC Kalman filter 500 according to one embodiment of the present disclosure. The SOC Kalman filter 500 includes a gain calculator 510 and an estimated SOC calculator 520.

[0040] The gain calculator 510 can apply the gain selection unit 560 to calculate the gain value based on the effectiveness value, parameter Q(530), and parameter R(540).

[0041] In a preferred embodiment of the gain selection unit 560, if the effectiveness value indicates that the nominal SOC is effective (for example, if the effectiveness value is 1), the gain value is calculated according to the formulation of the Kalman filter.

number

[0042] In a preferred embodiment of the gain selection unit 560, the gain is set to 0 if the effectiveness value indicates that the nominal SOC is invalid (for example, if the effectiveness value is 0). Mathematically, if the nominal SOC is invalid, this is equivalent to setting the covariance related to the error between the nominal SOC and the true SOC to infinity.

[0043] The parameter Q(530) is the covariance associated with the rate of increase of the SOC covariance when the SOC is not corrected by other means. The parameter Q can be estimated based on a combination of factors such as the accuracy of the current sensor, uncertainty in the Coulomb efficiency, uncertainty in the SOC Kalman filter calculation cycle time, and uncertainty in the cell capacitance.

[0044] The parameter R(540) is the covariance associated with the error between the nominal SOC and the true SOC, where the effectiveness value indicates that the nominal SOC is effective (e.g., the effectiveness value is 1).

[0045] The estimated SOC calculator 520 can calculate the estimated SOC based on the nominal SOC, current, parameter B(550), and gain value.

[0046] Parameter B is the predicted change in SOC for a current of 1 ampere over the time period from the previous calculation cycle of the SOC Kalman filter to the current calculation cycle. Parameter B may be determined based on the calculation cycle time (Ts) of the SOC Kalman filter and the rated capacity (Q) of the battery 100. In a preferred embodiment, parameter B(550) is calculated according to the following formula:

number

[0047] In Figure 1, the parameters used by the device 10 may be pre-set, and the device 10 can directly use these parameters. These parameters may include the weights of the first neural network, the weights of the second neural network, and parameter R. These parameters may be calculated by a computer independent of the device 10.

[0048] As shown in Figure 6, in at least other embodiments, the apparatus 10a may include a preprocessing unit 200, a SOC estimation unit 300, an effectiveness estimation unit 400, and a SOC Kalman filter 500. The apparatus 10a further includes a parameter calculator 600 that calculates parameters used in the apparatus 10a. Parameters may include the weights of the first neural network, the weights of the second neural network, and the parameter R, etc. The same parts of apparatus 10a and apparatus 10 will not be described, but the parameter calculator 600 will be described below.

[0049] Figure 7 is a flowchart showing, in sequence, blocks for calculating parameters used in SOC estimation by the parameter calculator 600, and follows one embodiment of the present disclosure.

[0050] In block S010, current, voltage, and temperature data are extracted from experimental results (i.e., experimental data) from battery tests designed to represent the operating conditions for battery applications.

[0051] In block S020, the current is filtered using a low-pass filter in a manner consistent with the operation of the current low-pass filter 240, the voltage is filtered in a manner consistent with the operation of the voltage low-pass filter 250, and the true SOC is calculated using the Coulomb counting method. For example, the parameter calculator 600 can control the preprocessing unit 200 to filter the current and voltage, or the parameter calculator 600 can filter the current and voltage itself.

[0052] In block S030, the first neural network is trained. For example, the weights of the SOC neural network (first neural network) are optimized using a training algorithm such as the Levenberg-Marquardt algorithm, with the goal of minimizing the RMS error between the true SOC value calculated in block S020 and the nominal SOC value calculated by the SOC neural network (first neural network) during training. These are evaluated using the corresponding current, voltage, temperature, filtered current, and filtered voltage values ​​collected in blocks S010 and S020.

[0053] In block S040, the nominal SOC value is calculated by evaluating the trained SOC neural network using the current, voltage, temperature, filtered current, and filtered voltage values ​​collected in blocks S010 and S020. For example, the parameter calculator 600 controls the SOC estimation unit 300 to calculate the nominal SOC using the trained SOC neural network (first neural network).

[0054] In block S050, the squared SOC prediction error is calculated as the square of the difference between the true SOC value and the corresponding nominal SOC value.

[0055] In block S060, a second neural network is trained. For example, the weights of the squared SOC error neural network (the second neural network) are optimized using a training algorithm such as the Levenberg-Marquardt algorithm, with the goal of minimizing the RMS error between the squared SOC prediction error calculated in block S050 and the squared nominal SOC error value calculated by the squared SOC error neural network (the second neural network) during training. These are evaluated using the corresponding current, voltage, temperature, filtered current, and filtered voltage values ​​collected in blocks S010 and S020.

[0056] In block S070, the nominal squared SOC error value is calculated by evaluating the trained squared SOC error neural network using the current, voltage, temperature, filtered current, and filtered voltage values ​​collected in blocks S010 and S020. For example, the parameter calculator 600 controls the effectiveness estimation unit 400 to calculate the nominal squared SOC error using the trained squared SOC error neural network (the second neural network).

[0057] In block S080, the squared nominal SOC error value is E 2 A subset of the original dataset that is less than 1% is identified. In one embodiment, the parameter E is set to 1%.

[0058] In block S090, the parameter R is calculated as the variance of the difference between the true SOC value and the corresponding nominal SOC value for the data subset identified in block S080. The parameter R is used in the Kalman filtering algorithm.

[0059] In Figure 7, blocks S010, S020, and S030 represent the processes for calculating the weights of the first neural network, blocks S040, S050, and S060 represent the processes for calculating the weights of the second neural network, and blocks S070, S080, and S090 represent the processes for calculating the parameter R.

[0060] In at least one embodiment, the method shown in Figure 7 is performed to calculate parameters, which are used in a method for estimating the battery SOC performed by the device 10 or 10a.

[0061] In at least one other embodiment, the method shown in Figure 7 and the method for estimating the battery SOC performed by the device 10 or 10a may be performed independently. For example, the method shown in Figure 7 is provided for calculating parameters, which are used to estimate the battery SOC in a different way than those performed by the device 10 or 10a. In another embodiment, the parameters used by the device 10 or 10a may be calculated in a different way than the method shown in Figure 7.

[0062] As can be seen from the above embodiment, the first neural network is used to estimate the nominal SOC, the effectiveness value of the nominal SOC is calculated, and the SOC Kalman filter calculates the estimated SOC based on the nominal SOC and the effectiveness value. Therefore, by using a systematic and efficient development method, the battery SOC can be accurately estimated.

[0063] In this disclosure, a second neural network is used to estimate the effectiveness of the nominal SOC based on the operating zone. The KF gain is set to zero in the operating zone where the nominal SOC is predicted to be ineffective, and otherwise calculated according to the KF gain formula.

[0064] This disclosure provides a method for calculating parameters used in battery SOC estimation. This method utilizes a commercially available neural network optimization package and enables high accuracy in neural networks with relatively few engineering resources.

[0065] This disclosure has proven to be extremely useful in providing accurate SOC estimations, particularly in applications such as LFP cell chemistry.

[0066] The correlation between SOC and available measurements varies significantly from one operating zone to another.

[0067] (Second aspect of the embodiment) In a second embodiment of the embodiment, a method for estimating the state of charge (SOC) of a battery is provided. This method corresponds to the apparatus for estimating the state of charge (SOC) of a battery provided in the first embodiment. Content identical to that of the first embodiment is omitted.

[0068] Figure 8 is a flowchart of a method for estimating the battery charge state according to one embodiment.

[0069] As shown in Figure 8, the method includes the following:

[0070] S801 detects the battery's operating parameters, and at least some of these parameters are filtered to generate filtered parameters.

[0071] The S802, nominal SOC, is calculated using a first neural network based on operating parameters and filtered parameters.

[0072] S803, The effectiveness value is calculated based on the operating parameters and filtered parameters, and the effectiveness value represents the effectiveness of the nominal SOC, and

[0073] Based on the S804, operating parameters, nominal SOC, and effectiveness value, a Kalman filtering algorithm is executed, which outputs the estimated SOC of the battery.

[0074] Details of each block can be found in the corresponding description of the first embodiment.

[0075] As shown in Figure 8, the method may further include the following:

[0076] S805 calculates the parameters used for battery SOC estimation.

[0077] Details of block S805 can be found in the flowchart in Figure 7 of the first embodiment.

[0078] As can be seen from the above embodiment, the first neural network is used to estimate the nominal SOC, the second neural network is used to estimate the effectiveness value of the nominal SOC, and the SOC Kalman filter calculates the estimated SOC based on the nominal SOC and the effectiveness value. Furthermore, the parameters are calculated using a clearly defined optimization-based workflow. Thus, by using a systematic and efficient development method, the battery SOC can be accurately estimated.

[0079] (Third aspect of the embodiment) A third aspect of the embodiments of this disclosure provides electronic equipment including the apparatus 10, 10a or parameter calculator 600 described in the first aspect of the embodiments.

[0080] Figure 9 is a schematic diagram of the structure of an electronic device in an embodiment of the present disclosure. As shown in Figure 9, the electronic device 900 may include a processor 910 and a memory 920, the memory 920 being connected to the processor 910. The memory 920 can store various data and can also store a program 930 for data processing and execute the program 930 under the control of the processor 910.

[0081] In one implementation, the functions of devices 10, 10a, or the parameter calculator 600 may be integrated into the processor 910. The processor 910 may be configured to perform the method of the second aspect of the embodiment.

[0082] In other implementations, the devices 10, 10a, or parameter calculator 600 and the processor 910 may be configured separately. For example, the devices 10, 10a, or parameter calculator 600 may be configured as chips connected to the processor 910, and the functions of the devices 10 or parameter calculator 600 are performed under the control of the processor 910.

[0083] Embodiments of this disclosure may further provide a computer-readable program that, when executed on a device or electronic device, causes the device or electronic device to perform the method described in a second embodiment of the embodiments of this disclosure.

[0084] Embodiments of the present disclosure further provide a computer storage medium containing a computer-readable program, the program operating a device or electronic device that performs a method described in a second embodiment of the present disclosure.

[0085] The above-described apparatus and methods of this disclosure may be implemented by hardware or hardware in combination with software. This disclosure relates to computer-readable programs that enable logical devices to execute the aforementioned apparatus or components or to perform the aforementioned methods or steps when the program is executed by the logical device. This disclosure also relates to storing the above-described programs on storage media such as hard disks, floppy disks, CDs, DVDs, and flash memory.

[0086] The methods / apparatus described in connection with embodiments of this disclosure may be implemented directly as hardware, as software modules executed by a processor, or as a combination thereof. For example, one or more functional block diagrams and / or one or more combinations of functional block diagrams shown in the drawings may correspond to software modules or hardware modules of procedures in a computer program. Each such software module may correspond to a step shown in the drawings. The hardware module may be implemented, for example, by fixing the software module using a field-programmable gate array (FPGA).

[0087] Soft modules may reside in RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disks, floppy disks, CD-ROMs, or other forms of memory media known in the art. The memory media may be connected to a processor, allowing the processor to read or write information to it. Alternatively, the memory media may be a component of the processor. The processor and memory media may reside in an ASIC. Soft modules may be stored in the memory of a mobile device, or on a pluggable memory card in a mobile device. For example, if a device (such as a mobile device) uses a relatively large-capacity MEGA-SIM card or a large-capacity flash memory device, the soft modules may be stored on the MEGA-SIM card or the large-capacity flash memory device.

[0088] One or more functional blocks and / or one or more combinations of functional blocks in the drawings may be realized as a universal processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or a suitable combination for performing the functions described herein. Alternatively, one or more functional block diagrams and / or one or more combinations of functional block diagrams in the drawings may also be realized as a combination of a DSP and a microprocessor, multiple processors, one or more microprocessors communicating with a DSP, or other combinations of such configurations.

[0089] This disclosure has been described above with reference to specific embodiments. However, those skilled in the art should understand that such descriptions are illustrative and not intended to limit the scope of protection of the invention. Various modifications and alterations may be made by those skilled in the art in accordance with the spirit and principles of the invention, and such modifications and alterations fall within the scope of the invention.

Claims

1. A device for estimating the state of charge (SOC) of a battery, A preprocessing unit (200) is configured to detect battery operating parameters and filter at least a portion of the operating parameters to generate filtered parameters, An SOC estimation unit (300) is configured to calculate the nominal SOC using a first neural network based on the operating parameters and filtered parameters transmitted from the preprocessing unit (200), A validity estimation unit (400) is configured to estimate a validity value indicating the validity of the nominal SOC output by the SOC estimation unit (300) based on the operating parameters and filtered parameters transmitted from the preprocessing unit (200), The system includes an SOC Kalman filter (500) configured to execute a Kalman filtering algorithm based on the operating parameters transmitted from the preprocessing unit (200), the nominal SOC transmitted from the SOC estimation unit (300), and the effectiveness value transmitted from the effectiveness estimation unit (400), thereby outputting the estimated SOC of the battery. The operating parameters detected by the pre-processing unit (200) include current, voltage, and temperature. The filtered parameters include filtered current and filtered voltage. The operating parameters transmitted to the SOC Kalman filter (500) include the current, The aforementioned pre-processing unit (200) A current low-pass filter (240) is configured to filter the current using a time constant T seconds and output the filtered current, The system includes a voltage low-pass filter (250) configured to filter the voltage using a time constant T seconds and output the filtered voltage, The effectiveness estimation unit (400) is A squared SOC error estimation unit (410) is configured to estimate a squared nominal SOC error, which represents the square of the difference between the true SOC and the nominal SOC, using a second neural network based on the operating parameters and the filtered parameters, Apparatus comprising: an effectiveness classification circuit (450) configured to output the effectiveness value based on the magnitude of the squared nominal SOC error and the elapsed execution time since the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter.

2. The magnitude of the squared nominal SOC error is a first predetermined value (E ２ The apparatus for estimating the battery charge state according to claim 1, wherein the effective value indicates that the nominal SOC is effective if the effective value is below a second predetermined value (3T) and the elapsed execution time from the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter exceeds a second predetermined value (3T).

3. The effectiveness classification circuit (450) is A first determination unit (451) configured to determine whether the squared nominal SOC error is below the first predetermined value, A second determination unit (452) is configured to determine whether the elapsed execution time from the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter exceeds the second predetermined value, The apparatus for estimating the battery charge state according to claim 2, further comprising: a logic unit (453) configured to output the effectiveness value based on the determination result of the first determination unit (451) and the determination result of the second determination unit (452).

4. The aforementioned SOC Kalman filter (500) is A gain calculator (510) configured to calculate a gain value based on the effectiveness value, parameter Q and parameter R, wherein parameter Q is a covariance related to the rate of increase of the SOC covariance when the SOC is not modified by any other means, and parameter R is a covariance related to the error between the nominal SOC and the true SOC when the effectiveness value indicates that the nominal SOC is effective, An apparatus for estimating the battery charge state according to claim 1, comprising: an estimation SOC calculator (520) configured to calculate the estimated SOC based on the nominal SOC, the current, parameter B, and the gain value, wherein parameter B is determined according to the SOC Kalman filter calculation cycle time (Ts) and the rated capacity (C) of the battery.

5. The apparatus further includes a parameter calculator, and the parameter calculator is The SOC estimation unit (300) is controlled to calculate the nominal SOC based on the experimental data. The effectiveness estimation unit (400) is controlled to calculate the squared nominal SOC error based on the experimental data. From the dataset, identify a subset of data in which the squared nominal SOC error is smaller than a first predetermined value. The apparatus for estimating the battery charge state according to claim 1, wherein the apparatus is configured to calculate the variance of the difference between the true SOC and the nominal SOC for a subset of identified data, the calculation result being a parameter R used in the Kalman filtering algorithm.

6. The parameter calculator further: Based on the aforementioned experimental data, the true SOC was calculated using the Coulomb counting method. The first neural network is trained with the aim of minimizing the root mean square error between the true SOC and the nominal SOC calculated by the first neural network during training. The squared SOC prediction error is calculated as the square of the difference between the true SOC and the nominal SOC calculated by the trained first neural network. The device for estimating the battery charge state according to claim 5, wherein the second neural network is trained with the aim of minimizing the root mean square error between the squared SOC prediction error and the squared nominal SOC error calculated by the second neural network under training.

7. A method for estimating the state of charge (SOC) of a battery, The steps include detecting the battery's operating parameters, filtering at least a portion of the operating parameters to generate filtered parameters, A step of calculating the nominal SOC using a first neural network based on the operating parameters and the filtered parameters, A step of calculating an effectiveness value indicating the effectiveness of the nominal SOC based on the operating parameters and the filtered parameters, The step includes executing a Kalman filtering algorithm using an SOC Kalman filter (500) based on the operating parameters, the nominal SOC, and the effectiveness value, thereby outputting the estimated SOC of the battery, The detected operating parameters include current, voltage, and temperature. The filtered parameters include filtered current and filtered voltage. The operating parameters used to execute the Kalman filtering algorithm include the current, The filtered current is generated by a current low-pass filter filtering the current using a time constant T seconds. The filtered voltage is generated by a voltage low-pass filter filtering the voltage using a time constant T seconds. The step of calculating the effectiveness value is: The steps include: using a second neural network based on the operating parameters and the filtered parameters to estimate the squared nominal SOC error, which represents the square of the difference between the true SOC and the nominal SOC; A method comprising the step of outputting the effectiveness value based on the magnitude of the squared nominal SOC error and the elapsed execution time since the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter.

8. The magnitude of the squared nominal SOC error is a first predetermined value (E ２ A method for estimating the battery charge state according to claim 7, wherein the effective value indicates that the nominal SOC is effective if the effective value is below a second predetermined value (3T) and the elapsed execution time from the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter exceeds a second predetermined value (3T).

9. The aforementioned effectiveness value is, A step of determining whether the squared nominal SOC error is below the first predetermined value, A step of determining whether the elapsed execution time since the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter exceeds the second predetermined value, A method for estimating the battery charge state according to claim 8, comprising the steps of: outputting the effectiveness value according to the determination result of the squared nominal SOC error and the determination result of the elapsed execution time since the initialization of the current low-pass filter and / or the initialization of the voltage low-pass filter.

10. The aforementioned estimated SOC is, A step of calculating a gain value based on the effectiveness value, parameter Q and parameter R, wherein parameter Q is a covariance related to the rate of increase of the SOC covariance when the SOC is not modified by any other means, and parameter R is a covariance related to the error between the nominal SOC and the true SOC when the effectiveness value indicates that the nominal SOC is effective, A method for estimating the battery charge state according to claim 7, comprising the steps of: calculating the estimated SOC based on the nominal SOC, the current, parameter B and the gain value, wherein parameter B is determined according to the SOC Kalman filter calculation cycle time (Ts) and the rated capacity (C) of the battery.

11. The above method further, The steps include: calculating the nominal SOC based on experimental data, The steps include: calculating the squared nominal SOC error based on the aforementioned experimental data; From the dataset, identify a subset of data in which the squared nominal SOC error is smaller than a first predetermined value. A method for estimating the battery charge state according to claim 7, comprising the step of calculating the variance of the difference between the true SOC and the nominal SOC for a subset of identified data, wherein the calculation result is a parameter R used in the Kalman filtering algorithm.

12. The above method further, The steps include: calculating the true SOC using the Coulomb counting method based on experimental data, The steps include training the first neural network with the aim of minimizing the root mean square error between the true SOC and the nominal SOC calculated by the first neural network during training, The steps include: calculating the squared SOC prediction error as the square of the difference between the true SOC and the nominal SOC calculated by the trained first neural network; A method for estimating the battery charge state according to claim 7, wherein the second neural network is trained with the aim of minimizing the root mean square error between the squared SOC prediction error and the squared nominal SOC error calculated by the second neural network under training.

Images