EP4427062A1

EP4427062A1 - Method and apparatus for estimating state of health of battery

Info

Publication number: EP4427062A1
Application number: EP22801836.2A
Authority: EP
Inventors: Debu ZHANG; Yaxi ZHANG; Xiang Long GU
Original assignee: Stellantis Auto SAS
Current assignee: Stellantis Auto SAS
Priority date: 2021-11-03
Filing date: 2022-10-18
Publication date: 2024-09-11
Also published as: CN116068444A; WO2023078673A1

Abstract

A method for estimating state of health of a battery includes: receiving operating data of a vehicle; determining whether the operating data can be used to estimate the state of health of the battery; and if yes, calculating a value of each of a plurality of health features based on the operating data, and providing the calculated values to a trained neural network model of the vehicle to estimate the state of health of the battery, wherein the trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery. The SOH of the battery obtained by this method is more accurate because factors of usage process and environment of the battery are taken into account. Moreover, when the above method is executed on-line, i.e., the SOH of the battery is estimated at a cloud server of the original equipment manufacturer by using the operating data transmitted by an in-vehicle network terminal, the original equipment manufacturer can dynamically grasp the battery health of the sold vehicles and dynamically monitor the battery usage of the users, and consequently provide after-sales and maintenance services, thus improving the user experience.

Description

METHOD AND APPARATUS FOR ESTIMATING STATE OF HEALTH OF BATTERY

[0001] This invention claims the priority of the Chinese patent application 202111294820.2 filed on November 03, 2021, of which the content (Text, Drawings and Claims) is incorporated by reference.

RELATED FIELD

[0002] The present disclosure generally relates to the technical field of batteries, in particular to a method and apparatus for estimating state of health of a battery, computing device, computer-readable storage medium and computer program product.

BACKGROUND

[0003] With the development of technology, pure electric vehicles and hybrid vehicles are more and more widely used. To provide drive energy, a power battery (or battery pack) consisting of multiple individual cells connected in series is generally installed in the vehicle body. The power battery will inevitably age during long-term use, so the aging condition of the power battery needs to be predicted. The aging condition of the power battery is generally predicted by calculating the State of Health (SOH) of the battery.

[0004] Currently, the SOH of the power battery is calculated mainly by off-line measurement including off-line measurement of battery capacity and off-line measurement of impedance. Although some on-line ways of calculating SOH may exist, building of their calculation models relies heavily on experimental data and results. In addition, the SOH can also be calculated in a Battery Management System (BMS) of the vehicle and subsequently collected by an in-vehicle network terminal of the vehicle (e.g., T-BOX).

[0005] However, the SOH of the power battery is affected by many factors, such as battery materials, manufacturing, usage and environment. The existing calculation methods of SOH haven't taken into account some important factors, especially the actual usage process and environment of batteries which have a large impact on SOH, making inaccurate the SOH calculated by the off-line measurement or by the model based on experimental data and results. The way of calculating SOH in the BMS depends mainly on the data in the BMS, which also has the problem of lack of accuracy due to insufficient data. [0006] In addition, the existing SOH calculation models have a narrow application range because they can only be used for certain types of batteries and cannot be applied to other types of batteries. Besides, optimization and improvement of calculation models require extensive experiments or recalculations, and the calculation models are therefore costly to maintain.

SUMMARY

[0007] As described above, the existing calculation methods of SOH haven't taken into account some important factors, especially the actual usage process and environment of batteries which have a large impact on SOH, making inaccurate the SOH calculated by the off-line measurement or by the model based on experimental data and results. The way of calculating SOH in the BMS also has the problem of lack of accuracy due to insufficient data. Besides, the existing SOH calculation models have a narrow application range and are costly to maintain.

[0008] In view of the above technical problems, a method for estimating state of health of a battery is proposed according to a first aspect of the present disclosure, which includes: receiving operating data of a vehicle; determining whether the operating data can be used to estimate the state of health of the battery; if yes, calculating a value of each of a plurality of health features based on the operating data, and providing the calculated values to a trained neural network model of the vehicle to estimate the state of health of the battery, wherein the trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery.

[0009] According to the above method, a plurality of predefined health features can be extracted from the operational data of the vehicle, and the SOH of the battery can be estimated by using the values of the health features and a pre-trained neural network model. The SOH of the battery obtained by this method is more accurate because factors of usage process and environment of the battery are taken into account. Moreover, when the above method is executed on-line, i.e., the SOH of the battery is estimated at a cloud server of the original equipment manufacturer by using the operating data transmitted by the in-vehicle network terminal, the original equipment manufacturer can dynamically grasp the battery health of the sold vehicles and dynamically monitor the battery usage of the users, and consequently provide after-sales and maintenance services, thus improving the user experience. In addition, this estimation method is independent of vehicle service time, vehicle type and battery type and is applicable to any electric or hybrid vehicle.

[0010] According to some optional embodiments, the determining whether the operating data can be used to estimate the state of health of the battery further comprises: determining whether the operating data satisfies a charge and discharge cycle; if it is satisfied, determining whether state of charge at the start of charging in the operating data is less than a first threshold and whether state of charge at the end of charging is greater than a second threshold, wherein when state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold, the operating data can be used to estimate the state of health of the battery.

[0011] According to some optional embodiments, the neural network model is a long shortterm memory (LSTM) neural network model.

[0012] According to some optional embodiments, before calculating the value of each of the plurality of health features based on the operating data, the method further comprises: performing data cleaning on the operating data and dividing the operating data according to charge and discharge cycles.

[0013] According to some optional embodiments, the method further comprises: obtaining historical operating data of the vehicle in a preset historical period, the preset historical period comprising a plurality of charge and discharge cycles; performing data cleaning on the historical operating data and dividing the historical operating data according to charge and discharge cycles; and training the neural network model based on the divided historical operating data.

[0014] According to some optional embodiments, the training the neural network model based on the divided historical operating data further comprises: selecting the charge and discharge cycles in which state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold from the plurality of charge and discharge cycles; for each selected charge and discharge cycle, calculating its historical state of health of the battery and value of each health feature in a preset health feature set based on the historical operating data; and training the neural network model using the calculated values of the health feature set and the historical states of health of the battery.

[0015] According to some optional embodiments, the historical state of health of the battery is calculated by the following formulas:

[0016] wherein 77 is the battery coulomb efficiency, is the start time of parking charging of a parking charging phase, t₂ is the end time of parking charging of the parking charging phase, I is the current during parking charging, SOC_start is state of charge at the start of charging, and SOC_end is state of charge at the end of charging, Q_t is the full charge capacity of the battery, Q_o is the rated capacity of the battery, and SOH is state of health of the battery.

[0017] According to some optional embodiments, before training the neural network model, the method further comprises: respectively calculating degree of correlation between each health feature in the health feature set and the historical state of health of the battery; and selecting a preset number of health features from the health feature set as the plurality of health features in a descending order of degree of correlation.

[0018] According to some optional embodiments, the degree of correlation between each health feature and the historical state of health of the battery is calculated by the following formula:

[0019] wherein n is the number of the selected charge and discharge cycles, X_t is the value of each health feature in the i-th charge and discharge cycle, X is the average value of each health feature of n charge and discharge cycles, Y_t is the historical state of health of the battery of the i-th charge and discharge cycle, Y is the average historical state of health of the battery of n charge and discharge cycles, and p_xy is the degree of correlation between each health feature and the historical state of health of the battery.

[0020] According to some optional embodiments, training the neural network model using the calculated values of the health feature set and the historical states of health of the battery further comprises: using the values of the plurality of health features of each selected charge and discharge cycle as input samples, and using the historical states of health of the battery of each selected charge and discharge cycle as output samples to train the neural network model.

[0021] An apparatus for estimating state of health of a battery is proposed according to a second aspect of the present disclosure, which includes: a data receiving unit configured to receive operating data of a vehicle; a data determination unit configured to determine whether the operating data can be used to estimate the state of health of the battery; and an SOH estimation unit configured to calculate value of each of a plurality of health features based on the operating data and provide the calculated values to a trained neural network model of the vehicle to estimate the state of health of the battery when the data determination unit determines that the operating data can be used to estimate the state of health of the battery, wherein the trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery.

[0022] According to the above apparatus, a plurality of predefined health features can be extracted from the operational data of the vehicle, and the SOH of the battery can be estimated by using the values of the health features and a pre-trained neural network model. The SOH of the battery obtained by this method is more accurate because factors of usage process and environment of the battery are taken into account. Moreover, when the above method is executed on-line, i.e., the SOH of the battery is estimated at a cloud server of the original equipment manufacturer by using the operating data transmitted by the in-vehicle network terminal, the original equipment manufacturer can dynamically grasp the battery health of the sold vehicles and dynamically monitor the battery usage of the users, and consequently provide after-sales and maintenance services, thus improving the user experience. In addition, this estimation apparatus is independent of vehicle service time, vehicle type and battery type and is applicable to any electric or hybrid vehicle.

[0023] According to some optional embodiments, the data determination unit is further configured to: determine whether the operating data satisfies a charge and discharge cycle; if it is satisfied, determine whether state of charge at the start of charging in the operating data is less than a first threshold and whether state of charge at the end of charging is greater than a second threshold, wherein when state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold, the operating data can be used to estimate the state of health of the battery.

[0024] According to some optional embodiments, the neural network model is a long short- term memory (LSTM) neural network model.

[0025] According to some optional embodiments, the apparatus further comprises: a first preprocessing unit configured to perform data cleaning on the operating data and divide the operating data according to charge and discharge cycles before calculating, by the SOH estimation unit, the value of each of the plurality of health features based on the operating data.

[0026] According to some optional embodiments, the apparatus further comprises: a data obtaining unit configured to obtain historical operating data of the vehicle in a preset historical period, the preset historical period comprising a plurality of charge and discharge cycles; a second preprocessing unit configured to perform data cleaning on the historical operating data and divide the historical operating data according to charge and discharge cycles; and a model training unit configured to train the neural network model based on the divided historical operating data.

[0027] According to some optional embodiments, the model training unit is further configured to: select the charge and discharge cycles in which state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold from the plurality of charge and discharge cycles; for each selected charge and discharge cycle, calculate its historical state of health of the battery and value of each health feature in a preset health feature set based on the historical operating data; and train the neural network model using the calculated values of the health feature set and the historical states of health of the battery.

[0028] According to some optional embodiments, the historical state of health of the battery is calculated by the following formulas: r_t

[0029] wherein 7] is the battery coulomb efficiency, is the start time of parking charging of a parking charging phase, t₂ is the end time of parking charging of the parking charging phase, I is the current during parking charging, SOC_start is state of charge at the start of charging, and SOC_end is state of charge at the end of charging, Q_t is the full charge capacity of the battery, Q_o is the rated capacity of the battery, and S()H s state of health of the battery.

[0030] According to some optional embodiments, the apparatus further comprises: a health feature selection unit configured to: respectively calculate degree of correlation between each health feature in the health feature set and the historical state of health of the battery; and select the plurality of health features from the health feature set in a descending order of degree of correlation.

[0031] According to some optional embodiments, the degree of correlation between each health feature and the historical state of health of the battery is calculated by the following formula:

[0032] wherein n is the number of the selected charge and discharge cycles, X_t is the value of each health feature in the i-th charge and discharge cycle, X is the average value of each health feature of n charge and discharge cycles, Y_t is the historical state of health of the battery of the i-th charge and discharge cycle, Y is the average historical state of health of the battery of n charge and discharge cycles, and p_xy is the degree of correlation between each health feature and the historical state of health of the battery.

[0033] According to some optional embodiments, the model training unit is further configured to: use the values of the plurality of health features of each selected charge and discharge cycle as input samples, and use the historical states of health of the battery of each selected charge and discharge cycle as output samples to train the neural network model.

[0034] According to a third aspect of the present disclosure, a computing device is proposed, where the computing device includes: a processor; and a memory for storing computerexecutable instructions, wherein the computer-executable instructions, when executed, cause the processor to perform the method according to any one of the above embodiments.

[0035] According to a fourth aspect of the present disclosure, a computer-readable storage medium is proposed, where the computer-readable storage medium has computer-executable instructions stored thereon, and the computer-executable instructions are used to perform the method according to any one of the above embodiments.

[0036] According to a fifth aspect of the present disclosure, a computer program product is proposed, where the computer program product is stored on a tangible computer-readable storage medium and includes computer-executable instructions that, when executed, cause at least one processor to perform the method according to any one of the above embodiments.

BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The features, advantages and other aspects of various embodiments of the present disclosure will become more apparent in connection with the accompanying drawings and with reference to the following detailed description. Several embodiments of the present disclosure are illustrated herein in an exemplary and non-restrictive manner. In the drawings, the same reference numerals indicate the same or similar components.

[0038] FIG. 1 is a general flow chart of a method for estimating state of health of a battery proposed by the present disclosure;

[0039] FIG. 2 is a flow chart of a training method of the neural network model in FIG. 1;

[0040] FIG. 3 illustrates several sub-steps of step 23 of FIG. 2;

[0041] FIG. 4 is a flow chart of the method for estimating state of health of a battery according to an embodiment of the present disclosure;

[0042] FIG. 5 illustrates an apparatus for estimating state of health of a battery according to an embodiment of the present disclosure; and

[0043] FIG. 6 illustrates a computing device for estimating state of health of a battery according to an embodiment of the present disclosure.

DETAILED DESCRIPTION

[0044] Exemplary embodiments of the present disclosure are described in detail hereafter with reference to the accompanying drawings. Although the exemplary methods, devices described below include software, executed on hardware of other components, and/or firmware, it should be noted that these examples are merely illustrative and should not be construed as restrictive. For example, it is possible that any or all hardware, software, and firmware components can be implemented exclusively in hardware, exclusively in software, or in any combination of hardware and software. Thus, although exemplary methods and apparatus have been described below, it should be readily appreciated by those skilled in the art that the examples provided herein are not intended to limit the manner used to implement these methods and apparatus.

[0045] In addition, the flowchart and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that the functions indicated in the blocks may also occur in a different sequence than that indicated in the accompanying drawings. For example, two consecutive blocks as illustrated may be substantially concurrently executed, or they may sometimes be executed in a reverse sequence, depending on the functions involved. It should also be noted that each block in the flowchart and/or the block diagrams, and the combination of blocks in the flowchart and/or the block diagrams, may be implemented using a dedicated hardware-based system used to perform the specified function or operation, or may be implemented using a combination of dedicated hardware and computer instructions.

[0046] For ease of description, some of the terms in the present disclosure are described below. It should be understood that the terms used herein should be interpreted as having a meaning consistent with their meaning in the context of the specification of this application and in the field to which they pertain. The terms "includes", "comprises" and the like used in the present disclosure should be interpreted as open-ended terms, i.e., "includes/comprises but is not limited to", indicating that other elements may also be included.

[0047] In the embodiments of the present disclosure, the phrase "based on" means "at least partially based on".

[0048] In the embodiments of the present disclosure, the phrase "an embodiment" means "at least one embodiment".

[0049] In the embodiments of the present disclosure, the phrase "another embodiment" means "at least one another embodiment", etc.

[0050] In the embodiments of the present disclosure, the term "vehicle" may be a car, truck, SUV, van, bus, or any other rolling platform, and the vehicle may be a pure electric vehicle or a hybrid vehicle.

[0051] In the embodiments of the present disclosure, the term "battery" or "power battery" may include a battery group or pack consisting of a single cell or multiple cells. [0052] At present, the existing calculation methods of SOH haven't taken into account some important factors, especially the actual usage process and environment of batteries which have a large impact on SOH, making inaccurate the SOH calculated by the off-line measurement or by the model based on experimental data and results. The way of calculating SOH in the BMS also has the problem of lack of accuracy due to insufficient data. Besides, the existing SOH calculation models have a narrow application range and are costly to maintain.

[0053] Therefore, a method for estimating state of health of a battery is proposed by the embodiments of the present disclosure, this method can extract a plurality of predefined health features from the operational data, and estimate the SOH of the battery by using the values of the health features and a pre-trained neural network model. The SOH of the battery obtained by this method is more accurate because factors of usage process and environment of the battery are taken into account. Moreover, when the above method is executed on-line, i.e., the SOH of the battery is estimated at a cloud server of the original equipment manufacturer by using the operating data transmitted by the in-vehicle network terminal, the original equipment manufacturer can dynamically grasp the battery health of the sold vehicles and dynamically monitor the battery usage of the users, and consequently provide after-sales and maintenance services, thus improving the user experience. In addition, this estimation method is independent of vehicle service time, vehicle type and battery type and is applicable to any electric or hybrid vehicle.

[0054] The present disclosure is described below according to the following embodiments. FIG. 1 is a general flow chart of the method for estimating state of health of a battery proposed by the present disclosure. The method of FIG. 1 can be executed by the cloud server of the original equipment manufacturer or by a computing device on the vehicle. Referring to FIG. 1, method 100 begins at step 11. In step 11, operating data of a vehicle is received. As is well known in the art, an in-vehicle network device such as a T-BOX can communicate with the original equipment manufacturer's cloud server via a mobile network (e.g., 4G, 5G network) and send the vehicle's operating data to the cloud server at specific times (e.g., the time when the vehicle is powered up and when it is parked and charged). The operating data includes but is not limited to the operating status and battery status of the vehicle, and includes, for example, vehicle speed, mileage, gear, voltage and current of the battery, state of charge (SOC), motor data, etc. Thus, the cloud server can receive vehicle operating data from the in-vehicle network device. In addition, the operating data can also be received directly by the computing device on the vehicle. [0055] Next, in step 12, it is determined whether the operating data can be used to estimate the state of health of the battery. In some embodiments, step 12 further includes: determining whether the operating data satisfies a charge and discharge cycle; if it is satisfied, determining whether SOC at the start of charging in the operating data is less than a first threshold and whether SOC at the end of charging is greater than a second threshold, wherein when SOC at the start of charging is less than the first threshold and SOC at the end of charging is greater than the second threshold, the operating data can be used to estimate the state of health of the battery. The process between two parking charging of the vehicle is referred to as a charge and discharge cycle. In other words, one charge and discharge cycle includes a parking charging phase and a vehicle driving phase. It is first determined if the operating data satisfies a charge judgment loop. In the present disclosure, a neural network model is used to estimate the state of health of the battery, and in training this neural network model, in order to make the model more accurate, the historical operating data in the charge and discharge cycles, where SOC at the start of charging is less than the first threshold and SOC at the end of charging is greater than the second threshold, are selected for model training (described in detail later). Therefore, when using the trained model, if the operating data satisfies the charge judgment loop, it is then determined whether SOC at the start of charging in the operating data is less than the first threshold and whether SOC at the end of charging is greater than the second threshold. The values of the first threshold and the second threshold can be set as needed. For example, the first threshold may be set to 40% and the second threshold to 60%.

[0056] In step 13, if it is determined that the operating data can be used to estimate the state of health of the battery, value of each of a plurality of health features based on the operating data is calculated, and the calculated values are provided to a trained neural network model of the vehicle to estimate the state of health of the battery. The trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery. The health features may be features that have an impact on the state of health of the battery, such as the accumulated use time of the battery, the accumulated mileage of the vehicle, parameters such as current, voltage and power during battery charging, battery temperature and discharge power during driving, etc. Their values can be obtained from the operating data. In some embodiments, the neural network model is a long short-term memory (LSTM) neural network model. In some other embodiments, other forms of neural network models may also be used.

[0057] According to the above method, a plurality of predefined health features can be extracted from the operational data of the vehicle, and the SOH of the battery can be estimated by using the values of the health features and a pre-trained neural network model. The SOH of the battery obtained by this method is more accurate because factors of usage process and environment of the battery are taken into account. Moreover, when the above method is executed on-line, i.e., the SOH of the battery is estimated at a cloud server of the original equipment manufacturer by using the operating data transmitted by the in-vehicle network terminal, the original equipment manufacturer can dynamically grasp the battery health of the sold vehicles and dynamically monitor the battery usage of the users, and consequently provide after-sales and maintenance services, thus improving the user experience. In addition, this estimation method is independent of vehicle service time, vehicle type and battery type and is applicable to any electric or hybrid vehicle.

[0058] In some embodiments, the method 100 further includes: performing data cleaning on the operating data and dividing the operating data according to charge and discharge cycles before calculating the value of each of the plurality of health features based on the operating data. Data cleaning may include sorting the operating data according to chronological order and accumulated mileage order, removing erroneous and abnormal values from the data, and filling in missing values in the data by means of linear interpolation. Then, the operating data is divided according to the charge and discharge cycles based on flag bits.

[0059] In some embodiments, the method 100 further includes building a neural network model and training it. FIG. 2 is a flow chart of a training method of the neural network model in FIG. 1. It should be understood by those of skill in the art that the training of the model should be performed before using the model to estimate the state of health of the battery, such as before step 13, but may be performed sequentially or concurrently with steps 11 and 12. The frequency and time of model training can be set as needed. For example, when the service time of the vehicle is short, the model can be trained repeatedly at a higher frequency due to lack of historical operating data. As the service time of the vehicle increases, the frequency of model training can be reduced. Referring to FIG. 2, in step 21, historical operating data of the vehicle in a preset historical period are obtained, where the preset historical period includes a plurality of charge and discharge cycles. The historical period can be set, such as 6 months, 1 year, etc., as needed.

[0060] In step 22, data cleaning is performed on the historical operating data and the historical operating data is divided according to charge and discharge cycles. Similarly, data cleaning includes sorting the historical operating data according to chronological order and accumulated mileage order, removing erroneous and abnormal values from the historical operating data, and filling in missing values in the historical operating data by means of linear interpolation. Then, the historical operating data is divided according to the charge and discharge cycles based on flag bits. As described above, one charge and discharge cycle includes a parking charging phase and a vehicle driving phase.

[0061] Then, in step 23, the neural network model is trained based on the divided historical operating data. As described above, in some embodiments, the neural network model may be a long short-term memory (LSTM) neural network model. In some other embodiments, other forms of neural network models may also be used. The trained neural network model has high accuracy because the data used to train the neural network model is sufficient and relevant to the actual use of the battery.

[0062] FIG. 3 illustrates several sub-steps of step 23 of FIG. 2. As shown in FIG. 3, in step

231, the charge and discharge cycles in which SOC at the start of charging is less than the first threshold and SOC at the end of charging is greater than the second threshold are selected from the plurality of charge and discharge cycles. The values of the first threshold and the second threshold can be set as needed. For example, the first threshold may be set to 40% and the second threshold to 60%. The accuracy of the trained neural network model can be further improved by selecting the charge and discharge cycles that meet the above conditions. In step

232, for each selected charge and discharge cycle, its historical state of health of the battery and value of each health feature in a preset health feature set are calculated based on the historical operating data. That is, the historical state of health of the battery and value of each health feature in a preset health feature set corresponding to each selected charge and discharge cycle are calculated by using the historical operating data of this charge and discharge cycle.

[0063] In this embodiment, the historical state of health of the battery is calculated by the following formulas ( 1 )-(3) : [0064] wherein 7] is the battery coulomb efficiency, is the start time of parking charging of the parking charging phase, t₂ is the end time of parking charging of the parking charging phase, I is the current during parking charging, SOC_start is state of charge at the start of charging, and SOC_end is state of charge at the end of charging, Q_t is the full charge capacity of the battery, Q_o is the rated capacity of the battery, and S()H s state of health of the battery.

[0065] In this embodiment, 26 health features are predefined as the health feature set. These 26 health features are illustrated in Table 1 below.

Table 1

[0066] It should be noted that the health feature set illustrated in Table 1 is only used for example. In other embodiments, other numbers and/or defined health features may be set. The values of the above health features can be extracted or calculated from the historical operating data.

[0067] Next, in step 233, degree of correlation between each health feature in the health feature set and the historical state of health of the battery is calculated. In this embodiment, the degree of correlation between each health feature and the historical state of health of the battery is calculated by the following formula (4):

[0068] wherein n is the number of the selected charge and discharge cycles, X_t is the value of each health feature in the i-th charge and discharge cycle, X is the average value of each health feature of n charge and discharge cycles, Y_t is the historical state of health of the battery of the i-th charge and discharge cycle, Y is the average historical state of health of the battery of n charge and discharge cycles, and p_xy is the degree of correlation between each health feature and the historical state of health of the battery. The closer the value calculated by formula (4) is to 1, the higher the degree of correlation between the respective health feature and the historical state of health of the battery, and vice versa.

[0069] In step 234, a preset number of health features are selected from the health feature set as the plurality of health features in a descending order of degree of correlation. The number of the selected health features can be set as needed, for example, 10. In step 235, the values of the plurality of health features of each selected charge and discharge cycle are used as input samples, and the historical states of health of the battery of each selected charge and discharge cycle are used as output samples to train the neural network model. That is, when training the neural network model, only the values of the plurality of health features of each selected charge and discharge cycle and the corresponding historical states of health of the battery are used as the sample set. In this embodiment, the sample set is divided in a ratio of 8: 2 into a training set and a test set to train the model.

[0070] It should be understood by those skilled in the art that in this embodiment, a portion of the health features having a high degree of correlation is selected from the health feature set according to the degree of correlation with the historical state of health of the battery, thereby improving the accuracy of the model. However, in other embodiments, the model may be trained without considering the correlation with historical state of health of the battery, and instead the values of all health features in the health feature set are used as sample inputs to train the model. [0071] FIG. 4 is a flow chart of the method for estimating state of health of a battery according to an embodiment of the present disclosure. The method 400 for estimating state of health of a battery illustrated in FIG. 4 is performed by a cloud server in communication with an in-vehicle network device. In method 400, steps 41-42 are used to train the model and are therefore executed off-line. Steps 43-47 are used to on-line estimate the state of health of the battery and are therefore executed on-line. In step 41, operating data of a vehicle is obtained. In step 42, the LSTM model is trained and the model parameters are determined. The process of training the model has been described in detail in the above section and will not be repeated here. In this embodiment, steps 41-42 are performed before steps 43-46. However, in other embodiments, steps 41-42 may also be performed concurrently with steps 42-46. After the model has been trained, the cloud server stores the trained model in the memory for subsequent use. Those skilled in the art can understand that a model for estimating state of health of the battery can be trained for each vehicle.

[0072] When it is required to on-line estimate state of health of the battery, first in step 43, the operating data of the vehicle is received. Then, step 44 includes determining whether the received operating data can be used to estimate the state of health of the battery. If yes, method 400 directs to step 45, otherwise it returns to step 43 to continue receiving the operating data. In step 45, data cleaning is performed on the accepted operating data and the accepted operating data is divided according to charge and discharge cycles. Then, step 46 includes calculating the values of the plurality of health features and, in step 47, the state of health of the battery is estimated based on the trained model for estimating state of health of the battery. The estimated state of health of the battery can be used for after-sales and maintenance services. For example, by analyzing the correlation between each health feature and the state of health of the battery, better vehicle usage habits can be recommended or suggested to the user to extend battery life. For another example, a warning can be issued when the state of health of the battery falls below a certain threshold to remind the user to replace the battery in time. Yet for another example, the state of health of the battery can be used as a reference factor when assessing the value of a used car.

[0073] FIG. 5 illustrates an apparatus for estimating state of health of a battery according to an embodiment of the present disclosure. The units in FIG. 5 can be implemented using software, hardware (for example, integrated circuits, FPGAs, etc.), or a combination of software and hardware. Referring to FIG. 5, the apparatus 500 includes a data receiving unit 501, a data determination unit 502, and a SOH estimation unit 503. The data receiving unit 501 is configured to receive operating data of a vehicle. The data determination unit 502 is configured to determine whether the operating data can be used to estimate the state of health of the battery. The SOH estimation unit 503 is configured to calculate a value of each of a plurality of health features based on the operating data and provide the calculated values to a trained neural network model of the vehicle to estimate the state of health of the battery when the data determination unit 502 determines that the operating data can be used to estimate the state of health of the battery. The trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery.

[0074] In some embodiments, the data determination unit 502 is further configured to: determine whether the operating data satisfies a charge and discharge cycle; if it is satisfied, determine whether state of charge at the start of charging in the operating data is less than a first threshold and whether state of charge at the end of charging is greater than a second threshold, wherein when state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold, the operating data can be used to estimate the state of health of the battery.

[0075] In some embodiments, the neural network model is a long short-term memory (LSTM) neural network model.

[0076] In some embodiments, the apparatus 500 further includes a first preprocessing unit (not shown in FIG. 5). The first preprocessing unit is configured to perform data cleaning on the operating data and divide the operating data according to charge and discharge cycles before calculating, by the SOH estimation unit 503, the value of each of the plurality of health features based on the operating data.

[0077] In some embodiments, the apparatus 500 further includes a data obtaining unit, a second preprocessing unit, and a model training unit (not shown in FIG. 5). The data obtaining unit is configured to obtain historical operating data of the vehicle in a preset historical period, where the preset historical period includes a plurality of charge and discharge cycles. The second preprocessing unit is configured to perform data cleaning on the historical operating data and divide the historical operating data according to charge and discharge cycles. The model training unit is configured to train the neural network model based on the divided historical operating data.

[0078] In some embodiments, the model training unit is further configured to: select the charge and discharge cycles in which state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold from the plurality of charge and discharge cycles; for each selected charge and discharge cycle, calculate its historical state of health of the battery and value of each health feature in a preset health feature set based on the historical operating data; and train the neural network model using the calculated values of the health feature set and the historical states of health of the battery.

[0079] In some embodiments, the historical state of health of the battery is calculated by the following formulas:

[0080] wherein 7] is the battery coulomb efficiency, is the start time of parking charging of the parking charging phase, t₂ is the end time of parking charging of the parking charging phase, I is the current during parking charging, SOC_start is state of charge at the start of charging, and SOC_end is state of charge at the end of charging, Q_t is the full charge capacity of the battery, Q_o is the rated capacity of the battery, and SOH is state of health of the battery.

[0081] In some embodiments, the apparatus 500 further includes a health feature selection unit (not shown in FIG. 5). The health feature selection unit is configured to: respectively calculate degree of correlation between each health feature in the health feature set and the historical state of health of the battery; and select a preset number of health features from the health feature set as the plurality of health features in a descending order of degree of correlation.

[0082] In some embodiments, the degree of correlation between each health feature and the historical state of health of the battery is calculated by the following formula: [0083] wherein n is the number of the selected charge and discharge cycles, X_t is the value of each health feature in the i-th charge and discharge cycle, X is the average value of each health feature of n charge and discharge cycles, Y_t is the historical state of health of the battery of the i-th charge and discharge cycle, Y is the average historical state of health of the battery of n charge and discharge cycles, and p_xy is the degree of correlation between each health feature and the historical state of health of the battery.

[0084] In some embodiments, the model training unit is further configured to: train the neural network model with the values of the plurality of health features of each selected charge and discharge cycle as input samples, and the historical states of health of the battery of each selected charge and discharge cycle as output samples.

[0085] FIG. 6 illustrates a computing device for estimating state of health of a battery according to an embodiment of the present disclosure. It should be understood that the computing device 600 may be implemented to perform the functions of the method 100 in FIG. 1 for estimating state of health of a battery. As can be seen in FIG. 6, the computing device 600 includes a central processing unit (CPU) 601 (e.g., a processor) that can perform various appropriate actions and processes based on computer program instructions stored in a read-only memory (ROM) 602 or loaded into a random access memory (RAM) 603 from a memory unit 608. Various programs and data required for the operation of the computing device 600 may also be stored in the RAM 603. The CPU 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input/output (VO) interface 605 is also connected to the bus 604.

[0086] Multiple components in the computing device 600 are connected to the I/O interface 605, including: an input unit 606, an output unit 607, a storage unit 608, and a communication unit 609. The communication unit 609 allows this computing device 600 to exchange information/data with other devices via a computer network such as the Internet and/or via various telecommunication networks.

[0087] The various methods described above such as the method 100 for estimating state of health of a battery may be executed by the CPU 601. For example, in some embodiments, the method 100 for estimating state of health of a battery may be implemented as a computer software program that is tangibly contained in a machine-readable medium such as the storage unit 608. In some embodiments, part or all of the computer program may be loaded into and/or installed on the computing device 600 via the ROM 602 and/or the communication unit 609. One or more of the actions or steps in the method 100 for estimating state of health of a battery described above may be performed when the computer program is loaded into the RAM 603 and executed by the processor CPU 601.

[0088] Therefore, in another embodiment, a computer-readable storage medium is proposed according to the present disclosure, where the computer-readable storage medium has computer-executable instructions stored thereon, and the computer-executable instructions are used to perform the method for estimating state of health of a battery in the embodiments of the present disclosure.

[0089] In another embodiment, a computer program product is proposed according to the present disclosure, where the computer program product is tangibly stored on a computer- readable storage medium and includes computer-executable instructions that, when executed, cause at least one processor to perform the method for estimating state of health of a battery in the embodiments of the present disclosure.

[0090] According to the above embodiments, a plurality of predefined health features can be extracted from the operational data of the vehicle, and the SOH of the battery can be estimated by using the values of the health features and a pre-trained neural network model. The SOH of the battery obtained by this method is more accurate because factors of usage process and environment of the battery are taken into account. Moreover, when the above method is executed on-line, i.e., the SOH of the battery is estimated at a cloud server of the original equipment manufacturer by using the operating data transmitted by the in-vehicle network terminal, the original equipment manufacturer can dynamically grasp the battery health of the sold vehicles and dynamically monitor the battery usage of the users, and consequently provide after-sales and maintenance services, thus improving the user experience. In addition, this estimation method is independent of vehicle service time, vehicle type and battery type and is applicable to any electric or hybrid vehicle.

[0091] The computer readable program instructions or the computer program product for performing various aspects of the present disclosure may also be stored in a cloud server, and the user can access the computer readable program instructions stored in the cloud server for performing one aspect of the present disclosure via a mobile internet, a fixed telephone network, or other network when called upon, so as to implement the technical solutions disclosed in accordance with various aspects of the present disclosure.

[0092] In general, various exemplary embodiments of the present disclosure can be implemented in hardware, dedicated circuit, software, firmware, logic, or any combination thereof. Some aspects can be implemented in the hardware, while others can be implemented in the firmware or software that can be executed by a controller, microprocessor, or other computing device. When various aspects of the embodiments of the present disclosure are illustrated or described as block diagrams, flow charts, or are represented by some other graphs, it will be appreciated that the blocks, devices, systems, techniques, or methods described herein may be implemented as non-restrictive examples in hardware, software, firmware, dedicated circuit or logic, general purpose hardware or controller, or other computing device, or some combination thereof.

[0093] Although it has been described above that the various exemplary embodiments of the present disclosure can be implemented in hardware or dedicated circuit, the computing device described above can be implemented in the form of hardware, and can also be implemented in the form of software. That is because: in the 1990s, it was easy to distinguish whether a technical improvement was a hardware improvement (e.g., an improvement in the structure of a circuit such as a diode, transistor, or switch) or a software improvement (e.g., an improvement in the method). However, as the technology continuously evolves, many of today's method improvements can almost always be achieved by programming the improved method into a hardware circuit. In other words, a corresponding hardware circuit structure can be obtained by programming a different program for the hardware circuit, achieving a change in the hardware circuit structure. Therefore, such method improvements can also be considered as direct improvements to the hardware circuit structure. Therefore, it is inappropriate to say that a method improvement cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) such as a field programmable gate array (FPGA) is one such integrated circuit whose logic functions are determined by user programming for the device. A digital system is "integrated" to a programmable logic device by programing from the designer without the need for a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of making integrated circuit chips manually, this programming is mostly done with a "logic compiler" software which is similar to the software compiler used for program development, and the original code has to be written in a specific programming language before compiling. This specific programming language is called Hardware Description Language (HDL). And there is not just one HDL, but many such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently the most commonly used is VHDL (Very -High- Speed Integrated Circuit Hardware Description Language) and Verilog. It is also apparent to those skilled in the art that a hardware circuit implementing the logical method can be easily obtained by programming the method into the integrated circuit in simple logic programming with the hardware description languages mentioned above.

[0094] Although the embodiments of the present disclosure have been described with reference to several specific embodiments, it should be understood that embodiments of the present disclosure are not limited to the disclosed specific embodiments. Embodiments of the present disclosure are intended to cover a variety of modifications and equivalent arrangements included within the spirit and scope of the appended claims. The scope of the claims conforms to the broadest interpretation and thus encompasses all such modifications and equivalent structures and functions.

Claims

1. A method for estimating state of health of a battery, comprising: receiving operating data of a vehicle; determining whether the operating data can be used to estimate the state of health of the battery; and if yes, calculating a value of each of a plurality of health features based on the operating data, and providing the calculated values to a trained neural network model of the vehicle to estimate the state of health of the battery, wherein the trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery.

2. The method of claim 1, wherein determining whether the operating data can be used to estimate the state of health of the battery further comprises: determining whether the operating data satisfies a charge and discharge cycle; if it is satisfied, determining whether state of charge at the start of charging in the operating data is less than a first threshold and whether state of charge at the end of charging is greater than a second threshold, wherein when state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold, the operating data can be used to estimate the state of health of the battery.

3. The method of claim 1, wherein the neural network model is a long short-term memory (LSTM) neural network model.

4. The method of claim 1, wherein before calculating the value of each of the plurality of health features based on the operating data, the method further comprises: performing data cleaning on the operating data and dividing the operating data according to charge and discharge cycles.

5. The method of claim 1, wherein before providing the calculated values to the trained neural network model of the vehicle, the method further comprises: obtaining historical operating data of the vehicle in a preset historical period, the preset historical period comprising a plurality of charge and discharge cycles; performing data cleaning on the historical operating data and dividing the historical operating data according to charge and discharge cycles; and training the neural network model based on the divided historical operating data.

6. The method of claim 5, wherein training the neural network model based on the divided historical operating data further comprises: selecting charge and discharge cycles in which state of charge at the start of charging is less than a first threshold and state of charge at the end of charging is greater than a second threshold from the plurality of charge and discharge cycles; for each selected charge and discharge cycle, calculating its historical state of health of the battery and a value of each health feature in a preset health feature set based on the historical operating data; and training the neural network model using the calculated values of the health feature set and the historical states of health of the battery.

7. The method of claim 6, wherein the historical state of health of the battery is calculated by the following formulas:

Q r_t wherein rj is the battery coulomb efficiency, is the start time of parking charging of a parking charging phase, t₂ is the end time of parking charging of the parking charging phase, I is the current during parking charging, SOC_start is state of charge at the start of charging, and SOC_end is state of charge at the end of charging, Q_t is the full charge capacity of the battery, Q_o is the rated capacity of the battery, and S()H s state of health of the battery.

8. The method of claim 6, wherein before training the neural network model, the method further comprises: respectively calculating degree of correlation between each health feature in the health feature set and the historical state of health of the battery; and selecting a preset number of health features from the health feature set as the plurality of health features in a descending order of degree of correlation.

9. The method of claim 8, wherein the degree of correlation between each health feature and the historical state of health of the battery is calculated by the following formula: wherein n is the number of the selected charge and discharge cycles, X_t is the value of each health feature in the i-th charge and discharge cycle, X is the average value of each health feature of n charge and discharge cycles, Y_t is the historical state of health of the battery of the i-th charge and discharge cycle, Y is the average historical state of health of the battery of n charge and discharge cycles, and p_xy is the degree of correlation between each health feature and the historical state of health of the battery.

10. The method of claim 8, wherein training the neural network model using the calculated values of the health feature set and the historical states of health of the battery further comprises: using the values of the plurality of health features of each selected charge and discharge cycle as input samples, and using the historical states of health of the battery of each selected charge and discharge cycle as output samples to train the neural network model.

11. An apparatus for estimating state of health of a battery, comprising: a data receiving unit configured to receive operating data of a vehicle; a data determination unit configured to determine whether the operating data can be used to estimate the state of health of the battery; and an SOH estimation unit configured to calculate a value of each of a plurality of health features based on the operating data and provide the calculated values to a trained neural network model of the vehicle to estimate the state of health of the battery when the data determination unit determines that the operating data can be used to estimate the state of health of the battery, wherein the trained neural network model is used to represent the relationship between the plurality of health features and the state of health of the battery.

12. The apparatus of claim 11, wherein the data determination unit is further configured to: determine whether the operating data satisfies a charge and discharge cycle; if it is satisfied, determine whether state of charge at the start of charging in the operating data is less than a first threshold and whether state of charge at the end of charging is greater than a second threshold, wherein when state of charge at the start of charging is less than the first threshold and state of charge at the end of charging is greater than the second threshold, the operating data can be used to estimate the state of health of the battery.

13. The apparatus of claim 11, wherein the neural network model is a long short-term memory (LSTM) neural network model.

14. The apparatus of claim 11, further comprising a first preprocessing unit configured to perform data cleaning on the operating data and dividing the operating data according to charge and discharge cycles before calculating, by the SOH estimation unit, the value of each of the plurality of health features based on the operating data.

15. The apparatus of claim 11, further comprising: a data obtaining unit configured to obtain historical operating data of the vehicle in a preset historical period, the preset historical period comprising a plurality of charge and discharge cycles; a second preprocessing unit configured to perform data cleaning on the historical operating data and dividing the historical operating data according to charge and discharge cycles; and a model training unit configured to train the neural network model based on the divided historical operating data.

16. The apparatus of claim 15, wherein the model training unit is further configured to: select charging and discharging cycles in which state of charge at the start of charging is less than a first threshold and state of charge at the end of charging is greater than a second threshold from the plurality of charge and discharge cycles; for each selected charge and discharge cycle, calculate its historical state of health of the battery and a value of each health feature in a preset health feature set based on the historical operating data; and train the neural network model using the calculated values of the health feature set and the historical states of health of the battery.

17. The apparatus of claim 16, wherein the historical state of health of the battery is calculated by the following formulas: SOH = — x l00%

2o wherein T] is the battery coulomb efficiency, is the start time of parking charging of a parking charging phase, t₂ is the end time of parking charging of the parking charging phase, I is the current during parking charging, SOC_start is state of charge at the start of charging, and SOC_end is state of charge at the end of charging, Q_t is the full charge capacity of the battery, Q_o is the rated capacity of the battery, and SOH is state of health of the battery.

18. The apparatus of claim 16, further comprising a health feature selection unit configured to: respectively calculate degree of correlation between each health feature in the health feature set and the historical state of health of the battery; and select a preset number of health features from the health feature set as the plurality of health features in a descending order of degree of correlation.

19. The apparatus of claim 18, wherein the degree of correlation between each health feature and the historical state of health of the battery is calculated by the following formula: wherein n is the number of the selected charge and discharge cycles, X_t is the value of each health feature in the i-th charge and discharge cycle, X is the average value of each health feature of n charge and discharge cycles, Y_t is the historical state of health of the battery of the i-th charge and discharge cycle, Y is the average historical state of health of the battery of n charge and discharge cycles, and p_xy is the degree of correlation between each health feature and the historical state of health of the battery.

20. The apparatus of claim 18, wherein the model training unit is further configured to use the values of the plurality of health features of each selected charge and discharge cycle as input samples and use the historical states of health of the battery of each selected charge and discharge cycle as output samples to train the neural network model.

21. A computing device, comprising: a processor; and a memory used to store computer-executable instructions, wherein the computerexecutable instructions, when executed, cause the processor to execute the method according to any one of claims 1 to 10.

22. A computer-readable storage medium having computer-executable instructions stored thereon, the computer-executable instructions being used to execute the method according to any one of claims 1 to 10.

23. A computer program product, wherein the computer program product is stored on a computer-readable storage medium, and comprises computer-executable instructions, the computer-executable instructions, when executed, cause at least one processor to execute the method according to any of claims 1 to 10.